Environment Variables

In this episode of Backstage, we go backstage with the Awesome Green Software (AGS) project, a pioneering initiative from the Green Software Foundation’s Open Source Working Group. Project leads Ahmad Antar and Abid Ali, along with contributor Josh Rauvola, share how AGS is building a centralized platform for developers to access tools, libraries, and frameworks to create greener, more sustainable software. They discuss the innovative tech behind AGS, the project’s progress toward launch, and its potential to make green software development accessible to organizations worldwide.

Learn more about the Awesome Green Software Project:

• Awesome Green Software
• Digital Emissions 
  
  

Learn more about our people:

• Ahmad Antar: LinkedIn 
• Abid Ali: LinkedIn 
• Josh Rauvola: LinkedIn

Find out more about the GSF:

• The Green Software Foundation Website 
• Sign up to the Green Software Foundation Newsletter

Resources:

• Algolia [4:03]
• Notion [4:04]
• Scaphandre [5:07]
• CodeCarbon.io [5:15]

Events:

• Green Gaming with STRATEGIES (November 9 at 9:00 am CET · Katowice) [07:39]
• Implications of AI-Related Data Center Electricity Use and Emissions (November 12-13 · Washington DC [07:58]
• Digital Impact Pitch (November 20 at 6:30 pm CET · Virtual) [08:09]
• Green Waves Hackathon Closing Day (November 22 at 8:30 am CET · TU Delft) [08:30]

If you enjoyed this episode then please either:

• 
  Follow, rate, and review on Apple Podcasts
• Follow and rate on Spotify
• Watch our videos on The Green Software Foundation YouTube Channel!
• Connect with us on Twitter, Github and LinkedIn!

TRANSCRIPT BELOW:
Chris Skipper: Welcome to Environment Variables, where we bring you the latest news and updates from the world of sustainable software development. I'm your host, Chris Skipper. On this episode, we have another edition of Backstage. Here, we'll take you behind the scenes at the Green Software Foundation to highlight ongoing projects and share insights from the members directly involved.

Today, we're diving into the Awesome Green Software Project, an exciting initiative curated by the Green Software Foundation's open source working group. Today, project leads Ahmad Antar and Abid Ali have provided us with some informative content and along with some soundbites from contributor Josh Rauvola to give us the lowdown on the collection of tools, libraries, and frameworks designed to help developers create greener, more sustainable software.

So, what makes this amazing, nay awesome, collection of green software resources tick? Let's explore that in this backstage episode. The GSF is all about building software that emits less carbon. Awesome Green Software fulfills that by creating a one stop shop which compiles research, tools, code, libraries, and training from volunteers that will be used for building applications that emit less carbon into our atmosphere.

There aren't too many tools available today that organizations can use to ensure their software is following sustainability principles. But, Awesome Green Software aims to fill that void. The GSF has the Impact Framework and the Software Carbon Intensity Specification, the SCI. And now, Awesome Green Software to provide that support.

As previously mentioned, Awesome Green Software brings research and resources that will be available at one place for organizations to access, hence reducing their efforts and time to find these tools and rather focus their efforts on building impactful software. Awesome Green Software was an idea that the GSF wanted to start for a while.

GSF member Digital Emissions took the initiative and is bringing it to reality. They have made a lot of progress over the last few months. The front end and back end are almost ready, and they are getting ready to start testing soon. Awesome Green Software is still in the developmental phases, but once it's ready to launch, We are expecting many contributors from the GSF and beyond to start contributing impactful code, tools, and utilities.

It will have a great impact on the overall community because currently it is not easy to find such tools in one place. Next, I'd like to play you a soundbite from Josh. We asked him the question, what are some of the key challenges you've encountered while developing the Awesome Green Software project?

And how did you and your team work to overcome them?

Josh Rauvola: Well, one of the main challenges we've faced is consolidating a vast amount of information and resources into a single user friendly platform. There's so much valuable content out there, but it's scattered across various sources. Organizing it in a way that's easily accessible to developers and other people was a significant task.

We also had to ensure that integrating technologies like Notion for content management and Algolia for search, which we can get into later, would provide a seamless user experience. To overcome these challenges, our team focused on building a solid backend and frontend infrastructure. We've been meticulous in our approach, and we're excited to get this project out into the hands of people that truly care about the environment.

Chris Skipper: With such a comprehensive list of tools and resources, the Awesome Green Software Project provides a bird's eye view of everything related to sustainable software out there on the internet. Can you explain some of the methods that you used to compile such a comprehensive list?

Josh Rauvola: Certainly. One innovative approach we're taking is leveraging powerful AI search capabilities through Algolia. This allows users to quickly find the tools and resources they need to build sustainable software. We're also using Notion as a content management system, which simplifies the process, for contributors to add and update resources in the directory.

By combining these technologies with modern front end frameworks, we're creating a responsive and efficient platform that not only compiles existing resources, but also encourages community engagement and contributions. We're all about being as efficient as possible, so we utilized some of the tools provided on our platform to align better with our awesome green software goals.

Chris Skipper: Could you highlight some of the specific tools, technologies, or frameworks included in the Awesome Green Software Project? What are some of your favorites? And how do they support the project's sustainability goals?

Josh Rauvola: That's a great question. The project includes a variety of tools like the Carbon Aware SDK, which helps developers make their applications carbon aware by optimizing energy consumption based on grid carbon intensity. We also feature tools like Scaphandre, a power monitoring agent that helps measure the energy consumption of servers.

Personally, I'm a big fan of the CodeCarbon.io library. It allows developers to track and reduce CO2 emissions from their computing by integrating seamlessly within their code. These tools support our sustainability goals by providing practical solutions that developers can implement to reduce the carbon footprint of their software.

Chris Skipper: The final question we put to Josh was, throughout the development of the Awesome Green Software project, What valuable lessons or insights have you gained that could benefit other organizations aiming to promote sustainable software?

Josh Rauvola: Well, one of the most valuable lessons we've learned is the importance of accessibility and community engagement. Only by the amazing support of the Green Software Foundation and the team at Digital Emissions have we been able to create such a wonderful platform. Creating a centralized platform where resources are shared.

Green software, easily accessible, significantly lowers the barrier for organizations to adopt sustainable practices. By making it easier for developers to find and use these tools, we encourage wider adoption of green software practices. Now all we need to do is get the word out. I'm sure this podcast will go a long way in doing so.

Chris Skipper: Yes, hopefully it does. I'd like to extend a big thank you to Ahmad, Abid, and Josh for sharing their valuable insights on the Awesome Green Software project. It's evident that this initiative goes beyond just providing resources. Thank you. It's about creating a comprehensive platform where developers can easily access tools, libraries, and frameworks to build greener, more sustainable software.

From their use of innovative technologies like Algolia and Notion to the thoughtful selection of carbon tracking tools, Ahmad, Abid, and Josh have shown us how AGS will make a real difference for organizations like yours. Aiming to reduce their software's carbon footprint. Their dedication to making the Awesome Green Software project a one stop shop is inspiring.

And it's exciting to hear about the milestones reached so far. With the front and back end nearly ready, and testing just around the corner, the team's focus on accessibility and community engagement will undoubtedly drive broader adoption of sustainable software practices as more contributors join after the launch.

We can't wait to see how AGS will shape the green software landscape and empower developers worldwide. slide. Now, before we finish this episode of Backstage, we have a few events to read out. Firstly, a free in person event, Green Gaming with Strategies, on November the 9th at 9am CET in Katowice, Poland.

Learn how the gaming industry contributes to and can mitigate its environmental footprint. Next up, we have Implications of AI related data center electricity use and emissions. A hybrid event happening on November the 12th to the 13th in Washington DC. Explore how to map, measure and mitigate the impacts of AI data center electricity usage.

Next up, a virtual event titled Digital Impact Pitch on November the 20th at 6. 30pm CET. This virtual event will help you discover what is the impact of our daily use of digital technologies. And finally, the Green Waves Hackathon closing day on November the 22nd at 8. 30am CET at TU Delft in the Netherlands.

Network, get talks and celebrate the ideas created during the hackathon. This event is your opportunity to experience the most innovative solutions developed by the Green Waves Hackathon participants. Whether you're passionate about sustainable IT or water research, This closing day will showcase the cutting edge ideas that can help reduce environmental impact.

Join us for a day packed with insightful presentations, live demos, and the excitement of the awards ceremony. Links to all the events and the resources mentioned in the show will be in the show notes below this podcast episode. So, we've reached the end of this special backstage episode. I hope you enjoyed the podcast.

To listen to more podcasts about Green Software, please visit podcast. greensoftware.foundation and we'll see you on the next episode. Bye for now!

In this episode, we honor the memory of Abhishek Gupta, who was an instrumental figure in the Green Software Foundation and a Co-Chair of the Standards Working Group. Abhishek's work was pivotal in the development of the Software Carbon Intensity (SCI) Specification, now adopted globally. His tireless efforts shaped the future of green software, leaving an indelible mark on the industry. As we remember Abhishek, we reflect on his legacy of sustainability, leadership, and friendship, celebrating the remarkable impact he had on both his colleagues and the world. We are airing an old episode that featured Abhishek Gupta, Episode 5 of Environment Variables. Where host Chris Adams is joined by Will Buchanan of Azure ML (Microsoft), Abhishek Gupta; the chair of the Standards Working Group for the Green Software Foundation and Lynn Kaack, assistant professor at the Hertie School in Berlin to discuss how artificial intelligence and machine learning impact climate change. They discuss boundaries, Jevons paradox, the EU AI Act, inferencing and supplying us with a plethora of materials regarding ML and AI and the climate!

In Memoriam: Abhishek Gupta:

• Honoring Abhishek Gupta | GSF 
  
  

Learn more about our guests:

• Chris Adams: LinkedIn / GitHub / Website
• Will Buchanan: LinkedIn
• Abhishek Gupta: LinkedIn
• Lynn Kaack: LinkedIn / Latest Paper

Find out more about the GSF:

• The Green Software Foundation Website 
• Sign up to the Green Software Foundation Newsletter

Episode resources:

• ClimateAction.tech [3:44]
• Green Web Foundation [3:49]
• Green Software Foundation’s Standards and Innovation Working Group [4:14]
• Montreal AI Ethics Institute [4:43]
• Hertie School Berlin [5:50]
• Aligning Artificial Intelligence with Climate Change Mitigation [6:32]
• The IPCC [7:11]
• Paper: Green AI | Roy Schwartz, Emma Strubell, Jesse Dodge [8:37]
• Project: Pachama [9:33]
• Montreal Institute for Learning Algorithms [10:34]
• Project: This Climate Does Not Exist [10:48]
• Austrian Institute of Technology | Infrared [11:32]
• Jevons Paradox [20:19]
• The GHG Protocol [23:27]
• Legislation: The EU AI Act [25:08]
• Paper; Measuring the Carbon Intensity of AI in Cloud Instances | Will Buchanan et al. [30:08]
• ONNX Runtime [37:02]
• TinyML [37:09]
• GitHub: Dynamic Batch Inferencing - Taylor Prewitt & Ji Hoon Kang of UW
• GitHub: NVIDIA Triton server on AzureML & Model Analyzer
• Green Software Foundation Summit

If you enjoyed this episode then please either:

• Follow, rate, and review on Apple Podcasts
• Follow and rate on Spotify
• Watch our videos on The Green Software Foundation YouTube Channel!

Connect with us on Twitter, Github and LinkedIn!

TRANSCRIPT BELOW:

Chris Skipper: This week on Environment Variables, we have some sad news.

We have to report the untimely passing of Abhishek Gupta. Abhishek was a key part of the Green Software Foundation. He was the co chair of the Standards Working Group from the GSF's formation to his passing on September 30th of this year. I would like to read out an in memoriam that was posted by Asim Hussain, honoring the great man, his legacy of leadership, his collaboration, and the heart that he put into the Green Software Foundation.

When the Green Software Foundation was formed over three years ago, Abhishek offered to help. The standards working group was looking for a co chair, and he jumped in to help without hesitation. He led the standards working group for over three years, with the full support of everyone involved. Leading a work group isn't about imposing your will on others, It's about finding common ground, nurturing discussions, and searching for the truth.

Abhishek excelled in all of those areas. He leaves a powerful legacy with the Software Carbon Intensity SCI specification. His tireless efforts over the years led to a consensus on the specification, which was later published to ISO in late 2023, and is now being adopted globally.

The impact of the SCI is truly global, with academics and enterprises worldwide adopting it. This widespread adoption is a testament to Abhishek's vision and dedication and his influence will be felt in every software product that bears an SCI score. His legacy is not just a memory, but a living testament to his work.

He has left an unforgettable mark on the community. And will be remembered for his contributions for years to come. Aho brother, I'll see you on the other side, Asim Hussain. To honor Abhishek, we're going to revisit an older episode of Environment Variables from the very start of the podcast, episode five, How Does AI and ML Impact Climate Change?

In this episode, host Chris Adams is joined by Abhishek Gupta, Lynne Kaack, and Will Buchanan to discuss these topics. So, without further ado, here's the episode.

Abhishek Gupta: We're not just doing all of this accounting to produce reports and to, you know, spill ink, but it's to concretely drive change in behavior. And this was coming from folks who are a part of the standards working group, including Will and myself, who are practitioners who are itching to get something that helps us change our behavior, change our team's behaviors when it comes to building greener software.

Asim Hussain: Hello and welcome to Environment Variables brought to you by the Green Software Foundation. In each episode we discuss the latest news and events surrounding green software. On our show you can expect candid conversations with top experts in their field who have a passion for how to reduce the greenhouse gas emissions of software.

I'm your host Asim Hussain.

Chris Adams: Hello there and welcome to the Environment Variables podcast, the podcast about green software. I'm Chris Adams filling in for Asim Hussain, the regular host while he's on paternity leave with a brand new baby. I met Asim on climateaction.tech, an online community for climate aware techies. And I work for the Green Web Foundation where we work towards a fossil free internet by 2030, as well as working as the co chair for the Green Software Foundation's policy group. Today, we're talking about climate change, AI, and green software. And I'm joined by Lynn, Will, and Abhishek.

Will Buchanan: Thanks for having me. My name is Will. I'm a product manager on the Azure Machine Learning team. also a member of the Green Software Foundation's Standards and Innovation Working Group. Within Microsoft, I foster the Green AI community, which now has a few hundred members, and I'm also a climate activist that's focused on pragmatic solutions to complex environmental issues.

Recently, I shipped energy consumption metrics within Azure Machine Learning, and we are about to publish a paper titled Measuring Carbon Intensity of AI in Cloud Instances, which I think we'll touch on today.

Abhishek Gupta: Well, thanks for having me. I'm Abhishek Gupta. I'm the founder and principal researcher at the Montreal AI Ethics Institute. I also work as a senior responsible AI leader and expert at the Boston Consulting Group, BCG, and I serve as the chair for the Standards Working Group at the Green Software Foundation.

So I've got a few hats on there. Most of my work, as it relates to what we're going to talk about today, It runs at the intersection of responsible AI and green software. In particular, what's of interest to me is looking at how the intersections of social responsibility, the environmental impacts of software systems, in particular AI systems, can be thought about when we're looking to make a positive impact on the world while using technology in a responsible fashion. I also, as a part of the Green Software Foundation, help through the standards working group, come up with the software carbon intensity specification, where we're trying to create an actionable way for developers and consumers of software systems to better assess and mitigate the environmental impacts of their work.

Chris Adams: Okay. And Lynn, last but not least, joining us from Berlin. Thank you very much for joining us. 

Lynn Kaack: Yeah thank you so much. I am an assistant professor at a policy school, public policy school, called Hertie School in Berlin. And I am also a co founder and a chair of an organization called Climate Change AI. with Climate Change AI, we facilitate work at the intersection of machine learning and And different kinds of climate domains, focusing on climate change mitigation and adaptation. And in my work, in my research, I am looking at how we can use machine learning as a tool to address different problems related to energy and climate policy. I'm also interested in the policy of AI and climate change. And today, actually, since we're talking about papers, I have a paper coming up. That is called Aligning Artificial Intelligence with Climate Change Mitigation, where we look at the different impacts from machine learning and how they affect greenhouse gas emissions.

Chris Adams: So we actually have some decent deep domain expertise and I'll try to keep this quite accessible, but we might drop into a little bits of like data science nerder on here, but the podcast has done that previously and it turns out to be something that we've got some decent feedback from because there aren't that many podcasts covering this. Okay so let's, uh, get into this topic of green AI and climate change. As we we know, IT is a significant driver of emissions in its own right. When we think about the climate crisis this year, the IPCC, which is the Intergovernmental Panel on climate Change in their big reports, which synthesized literally thousands of papers explicitly called out digital as a thing we, we should be talking about and thinking about.

And if you're a responsible technologist, it seems like a thing that we should be taking into account here. Now, I found it helpful to think about IT IT a little bit like how we think about the shipping industry, partly because they're similar in terms of emissions, which is around between 1 and 3%, depending on what you look at it, but also in that both of these acts like kind of connective tissue for society.

We also think of IT as a kind of force multiplier for The existing forms of activity. So if you use it, which is in line with the recommendations of the science, that's a good thing. But if you use it to do something, which is kind of rejecting some of the science, it might not be such a good thing. And within technology, AI and machine learning in particular is one of the fastest growing sectors and often seen as one of the biggest levers of all. So we're going to highlight some interesting projects we'll start off with. And out of that, we'll probably dive into some specifics about that or some other things you might want to take into account. If you're a technologist wanting to incorporate an awareness of climate into how you work and build greener software, then finally, we'll hopefully leave you with some actionable tips and techniques or projects that you may contribute to or use in your daily practice.

There's another term that we might be touching on here when you're making AI greener, and that's specifically Green AI. Is that the case, Will?

Will Buchanan: Correct. And that actually was coined by researchers a few years ago, Roy Schwartz, Emma Struble, Jesse Dodge. And it's really focused on making the development of the AI system itself more sustainable. And it's to be disambiguated from the term using AI for sustainability.

Chris Adams: Okay, so that's something we'll touch on both today. We'll talk about some of the external impacts and some of the internal impacts. We're going to start with something quite easy first because, well, why not? I'm just going to ask each of the people here to kind of point to maybe one project that they've seen that's using ML in quite an interesting fashion to ideally come up with some kind of measurable win. Well, if there was one project you'd actually look to that you think is kind of embodying these ideas of like green AI something which is really helping us essentially face some of the challenges, maybe you could tell us about what's catching your eye at the moment or what you'd look at.

Will Buchanan: I've been thinking a lot about natural capital recently, and I learned about a startup called Pachama, which combines remote sensing data with machine learning to help measure and monitor the carbon stored in a forest. I think it's really, really valuable because they're providing verification and insurance of carbon credits at scale. They've protected about a million hectares of forest. That's really when you have IoT and remote sensing and machine learning combining to help nature restore itself.

Chris Adams: Okay, cool. So if I understand that, they're using satellites to basically track forests and track deforestation. Is that the idea that they're doing there? 

Will Buchanan: Yes, Yes, and and also to verify the amount of carbon that a forest can sequester.

Chris Adams: Okay, cool. All right. I know there's a few other projects related to this. If I just hand over to Abhishek, can you let us know what's caught your eyes recently, and then we'll see what other projects come out of this.

Abhishek Gupta: Yeah, absolutely. I think one of the projects, and I don't know, I mean, what the measurable impact has been so far. In fact, it's, it's something that's come out of MILA, which is, or, you know, called the Montreal Institute for Learning Algorithms, which is Dr. Bengio's lab in, in Montreal. In fact, one of the people who led that project is a part of Climate Change AI as well, who I'm sure Lynn can talk more about too, which is SASA. And she's done this project called This Climate Does Not Exist, which I think was a fascinating use of machine learning to visualize the impact climate change will have on, you know, places around you in a very arresting and visually capturing fashion, which I think when we think about.

What impact climate change is going to have around us. Sometimes it, it feels quite distant because it's a, it's a slow rolling thing that's coming our way. And this puts it in, in, in a way that's quite immediate, quite visually arresting. And I think stirs people to action. I, as I said, I'm not sure what the measurable impact of that has been yet, but I, I certainly feel that those are the kinds of creative users of AI when we want to galvanize people into action around climate change.

Lynn Kaack: I'm happy to also talk about a application, which is also kind of difficult in terms of measuring impact, but I think it's an another interesting component of what AI can do. And this is something that researchers at the Austrian Institute of Technology do on a project called Infrared. And they use machine learning to help design new districts and cities.

And especially at the moment in many countries, a lot of new urban districts are being built and how we build these has a huge impact on energy consumption in cities, both in terms of transportation, but also how buildings are heated or cooled. And by the use of machine learning, they can drastically improve design choices because now they can approximate their very computationally heavy models and run them much faster. which means that they can also have more runs and can try out more design configurations. So this is a rather indirect application, but it has huge implications also on emissions for, for many decades to come.

Chris Adams: So essentially it's using of housing policy as climate policy there, because there's just a huge amount of emissions built into how people live and whether they need to drive everywhere in a car and stuff like that. That's some of the stuff that it's doing and making that part easier?

Lynn Kaack: So it's not really looking at housing policy, but it's looking at how districts are designed. So they take a group of, of houses, like if the new district is to be built and then they simulate the wind flow going through the cities, which are very expensive simulation models. they take the outputs of their model and approximate it with a machine learning model, which makes it much, much faster.

So from hours or days, you go to milliseconds or below seconds for one run. And then you can try out different design configurations and understand better how The built infrastructure affects natural cooling, for example, in cities or walkability, energy impacts generally of the microclimate on, on the built environment.

Chris Adams: Wow, I had no idea that was, it was actually possible. That's really, really cool.

Will Buchanan: That's very cool. That's similar to generative design.

Chris Adams: design. This is a phrase I haven't heard actually, Will. Maybe you could elucidate or share something there, actually. 

Will Buchanan: It's similar similar to to some software that Autodesk has built where you can try out many different iterations of a design and come up with optimum solutions. I think it's really cool that you're just consolidating it and running these models more efficiently.

Chris Adams: cool. And that's a bit like following, say, a fitness function saying, I have a chair, or, you know, I want to have something that works like a chair. It needs four legs and a seating pattern. Then it essentially comes up with some of the designs or iterates through some of the possibilities, something like that?

Will Buchanan: Exactly. 

Chris Adams: Oh, wow. Okay. That's cool. All right. And so we've spoken about AI, and there's a few exciting, interesting ones that we can add to the show notes and list from, and for people to look into and see how that might relate to what they do, I suppose. I want to ask a little bit about measuring impact from these projects, because quite a few different ways that you can actually measure impact here.

And in many times, it can be quite a difficult thing to kind of pin down. And this is continually a thing that's come up. When I know that. So, people have tried to come up with specs like the Software Carbon Intensity, and I'm sure Abhishek, you've had some experiences here. Will, you've mentioned a little bit about, um, Actually measuring impact internally, and it sounds like you've just had to do a bunch of this work on the ML team right now and expose some of these numbers to people consuming these services in the first place. Could you talk about some of that part a bit, perhaps?

Will Buchanan: Certainly. And so, as I mentioned, we have shipped energy consumption metrics for both training and inference within Azure machine learning. And that's really complex when you think of the infrastructure required to just report that,but that doesn't necessarily account for the additional power that's consumed in the data center, such as the idle power for devices or for the utilization of your servers.

There's so many different factors there. So you really, you could encounter scope creep when you come to your measurement methodology. So it's really necessary to put boundaries around that.

Chris Adams: and when you use the term boundaries here, you're saying, I'm going to measure the environmental impact of the servers, but not the environmental impact of building the building to put the servers in. Is that the idea of when you're referring to a boundary here? Yeah.

Will Buchanan: Yes, that's a great example.

Chris Adams: Okay. All right. I think this is actually something we've come across quite a few times in other places as well, actually, because maybe it's worth asking about this kind of boundary issue that we have had here, because automatically that sounds complicated here.

And I know that Abhishek, you've had some issues at your end as well with defining this style for deciding what's in or out, because I think this is one thing that we've had to explicitly do for the software carbon intensity spec, right?

Abhishek Gupta: Exactly. And, and I think when we talk about boundaries, it's, it's, it's trying to get a sense for what are the actual pieces that are consumed, right? From an operational standpoint, from an embodied emission standpoint, and how you make those, you know, allocations across you know, what, what your system is consuming.

And I use the word system because I think again, when we talk about software, we're not just talking about a specific piece, but we're talking about really everything that it touches, be that, you know, network, be that bandwidth consumption, be that, you know, as, as Will was saying, idle power. Even when we're looking at cloud computing, it becomes even more complicated when you have your pieces of software that are sharing tenancy across the pieces of hardware and how different consumers are perhaps sharing that piece of hardware with you and thinking about whether you've booked the resource ahead of time or not, whether it's hot or cold in terms of its availability and what implications that has.

I mean, there are so many different facets to it. And each of those decisions, what I want to highlight here is that it comes with a trade off, right? So we also don't have any standards in terms of how we should go about measuring that and what should be included, what shouldn't be included. And so the way people report out these numbers today also doesn't really make it actionable for folks who are consuming or who want to consume these reports, these metrics in taking decisions as to, you know, whether something is green or not.

And I think that's one of the places that The Software Carbon Intensity Specification is trying to help folks is to help standardize it first and foremost, but also to make it actionable so that if you're someone who's environmentally conscious, you can make the right choice by being informed about what the actual impacts are.

Chris Adams: Okay, this is question that I'm curious here, because so far we've only been speaking internally about, okay, what is the environmental impact of IT itself, like it's direct emissions. But the assumption that I have here is that there are ways we might talk about the impact that it has on the outside world in terms of activity we're speeding up or accelerating or supporting there.

Is that the only issue that we need to think about, or are there any other things to take into account about like this system boundary part that we've just been talking 

Lynn Kaack: Yeah. So the system effects are really important to look at and to consider. Maybe just to give an example, like if you use machine learning in, let's say the oil and gas sector to make a small part of the operation is more energy efficient, that first sight looks like something that could be considered sustainable and green, but you also have to realize that often then you are reducing costs as well, and that might change the way that oil and gas in this particular example is competitive, or the particular company is competitive, and that actually might shift also how much oil and gas we are able to use in the short run, how the price is changed.

So, these indirect effects can actually then have much larger impacts than the immediate effects of such an application. So drawing boundaries is really important and also opening this up to to have a broader system level view and really try to understand how does the technology also change then, then to larger consumption and production patterns. It's important.

Chris Adams: So if I understand that correctly, that's talking almost like the consequences of an intervention that we might make here. So even though we've might have reduced the emissions of say the drilling part by putting a wind turbine on an oil rig, for example, that might change the economics and make people more likely to use the oil, in which many cases they might burn, for example, or stuff like that.

Is that basically what you're saying there?

Lynn Kaack: Essentially what I'm saying is that efficiency improvements in particular, and often they can be done with data science or with machine learning or AI systems. They often come with cost reductions and then those cost reductions do something and change something. And often this is also considered under rebound effects, but it's not only rebound effects.

So there's systemic, the system level impacts that come from more small scale applications that need to be considered.

Will Buchanan: That's such a good good point, and I think I've also heard it called Jevon's paradox, too.

Chris Adams: Yes. Jevons paradox. This is stuff from like the 1800s with steam engines, right? Like my understanding of the Jevons paradox was when people had steam engines and they made steam engines more efficient, this led to people basically burning more coal because it suddenly became more accessible to more people and you end up using them in a greater number of factories.

So there's a kind of rebound, I think, that we need to take into account. This is something I think has been quite difficult to actually capture with existing ways of tracking the environment impact of particular projects. We have like an idea of, say, an attribution based approach and a consequence based approach. And maybe it's worth actually talking about here, about how Some of the complexities we might need to wrestle with when you're designing a system here. I mean, Abhishek, I think this was one of the things that was an early decision with the software carbon intensity part to not try to have an attribution approach versus a marginal approach. And if we're not diving too deeply into jargon here, maybe you might be able to kind of share a bit more information on that part there, because It sounds like it's worth expanding or explaining to people, to the audience, a bit better here.

Abhishek Gupta: Indeed, you know, the reason for making that choice was again, our emphasis on being action oriented, right? So as we had started to develop the software carbon intensity specification, one of the early debates that we had to wrestle with and, and, you know, Will and Will was of course a crucial part of that as well as whether folks who were a part of the standards working group was figuring out how.

For example, the GHG way of going about doing that, you know, accounting doesn't really translate all that well for software systems and how perhaps adopting a slightly different approach would lead to more better. More actionable outcomes for the folks who want to use this ultimately to change behavior because, you know, without getting into specifics of, you know, what marginal is and what consequential approaches are, and, and if we want, I mean, I'm, I'm sure you know, will, would, would be happy to dive into all of those details as would I.

But the, the, the thing that we were seeing was that we're doing all of this great work around, you know, talking about scope one, two, three emissions, et cetera, but it's not really helping to drive behavior change. And that's really the crux of all of this, right, is that we're not just doing all of this accounting to produce reports and to, you know, spill ink, but it's to concretely drive change in behavior.

And that's where we found that adopting a consequential, adopting a marginal approach actually helped make it more actionable. And this was coming from folks who are a part of the standards working group, including Will and myself, who are practitioners, who, who are itching to get something that helps, helps us change our behavior, change our team's behaviors when it comes to building greener software, broadly speaking.

Chris Adams: Okay. So that helps with explaining the difference between a consequential approach and a marginal approach, as in the consequences of me building this thing will mean that this is more likely to happen. And if I understand it, the GHG protocol that you mentioned, which is the greenhouse gas protocol, and this scoped emissions approach, this is the kind of standard way that an organization might report It's kind of climate responsibility, as it were, when, and when you say scoped emissions, that's like scope one, which is burning burning say that's, that's emissions from fossil fuels burned on site or in your car.

For example, scope two is electricity and scope three is your supply chain. If I understand what you're saying, there's like a kind of gap there that doesn't account for the impacts of this, perhaps. I mean, some people who've referred to this as scope zero or scope four, which might be, what are the impacts an organization is happening to. Essentially, we mentioned before, do something around this systemic change, or as Lynn mentioned, like this is changing the price of a particular commodity to make it more likely to be used or less likely to be used. And this is what I understand the SCI is actually trying to do. It's trying to address some of this consequential approach, because the current approach doesn't capture all of the impacts an organization might actually have at the moment. Right?

Will Buchanan: Yeah, that's a good summary. One challenge that I have noticed is that until it's required in reporting structures like the greenhouse gas protocol, then organizations don't have an incentive to really take the action that they need to avoid climate disaster. Something I encounter on a daily basis, and I think broadly we need to bring this into the public discourse.

Chris Adams: I think you're right. I think it's worth actually, Lynn, I think that when I've seen some of the work that you've done, you've done previously, this is something that's come into some of the briefings that I think that you've shared previously with the climate change AI work and some of the policy briefings for governments as well. Is there something that you might be able to add on here?

Lynn Kaack: Yeah, so something that comes to mind is, for example, like a concrete legislation that's currently being developed is the EU AI Act. And that's a place where, for the first time, AI systems are being regulated also that scale. And climate change almost didn't play a role for that regulation in the first draft.

So here it's also really evident that if we don't write in climate change now as a criterion for evaluating AI systems. It will probably be ignored for the next years to come. So the way that legislation works is by classifying certain AI systems as high risk and also just outright banning some other systems, but as high risk systems could, the original legislation stood, weren't really explicitly classified as high risk, even if they had like a huge environmental or climate change impact. And that's that I talked about a lot with policymakers and trying to encourage them to more explicitly make environmental factors and climate change a factor for evaluating systems. So that'd be a very concrete case where making climate change more explicit in the AI context is important also in terms of legislation.

Abhishek Gupta: So there's, there's a lot to be said about the EU AI Act, right? And a ton of ink has been spilled everywhere, I think, as, as, you know, it's, it's, it's, it's called the Brussels effect for a reason, right? Where the, whatever happens in the EU is taken as gospel and sort of spread across the world, which I think, as already Lynn has pointed out there, It's not, it's not perfect, right?

I think one of the things that I've seen being particularly problematic is, is the rigid categorization of what, you know, high risk use cases are and, and whether the EU AI act, as we'll see, hopefully with some, with some you know, revisions that are coming down the pipe is whether we'll have the ability to add new categories and, and not just update subcategories within the existing identified high risk categories.

And I think that's where things like considerations for environmental impacts and really tying that to this you know, societal impacts of AI, where we're talking about bias, privacy, and all the other areas, is going to be particularly important because we need multiple levers to, to try to account for, or to push on getting people to consider the environmental impacts.

And given that there is such a great momentum already in terms of privacy considerations, bias considerations. I think now is the time where we really push hard to make environmental considerations an equally first class citizen when it comes to, you know, thinking about the societal impacts of AI.

Will Buchanan: This is something I'm incredibly passionate about. I think needs to encompass the full scope of harms that are caused by an AI system. So that could be the hidden environmental impacts of either the development or the application. The application could vastly outweigh the good that you're doing, even just expanding oil and gas production by a certain percentage amount. I think it just must account for all of the harms for both ecosystems and people.

Chris Adams: And there's this idea of like a risk thing. Does this categorization actually include this stuff right now? What counts as a high risk use case, for example, when mentioned here?

Lynn Kaack: So I haven't seen the latest iteration. I think that there has been some update on, there's been a lot of feedback on the version that was published in April last year. I haven't seen the latest iteration. I think a lot of things changed in, yeah, in the first version, there was It's high risk systems where when, that affect personal safety, like human rights in the sense of, of personal wellbeing, but the completely overlooked environmental protection aspects aspects of human rights.

Chris Adams: Wow, that's quite a large one, especially when you take into account the human rights. Okay. We've spoken about the external impact, but I am led to believe there is also an internal impact from this as well. Like the AI has, has some direct impact that we might want to talk about as well. As I understand it, we spoke about 2 to 3 percent of emissions here, but if we know there's an external impact, why would we care about any of the internal impacts of AI we might be doing or why you might want to care about the internal impacts of AI as well, for example, like the direct emissions.

Will Buchanan: So by direct emissions, you're talking about, let's say, the scope, too, of the operational cost of the model.

Chris Adams: Yeah, there'll be things that we have, there's an external impact or there is a, we use this phrase scope 4, for example, to talk about all the other things that induce in the world. But there is a, a kind of stuff which happens inside the system boundary that we've spoken about. And presumably that's something we should be caring about as well, right? So there'll be steps that we can take to make the, the use of AI, particularly like say the, the models more efficient and more effective and more, all these parts here, this is something that we should be looking at as well, presumably right?

Will Buchanan: Totally. And so in our paper, which is going to be published, I think on Monday, we've calculated the emissions of several different models. And one of them was a 6 billion parameter transformer model. And the operational carbon footprint was equivalent to about a rail car of coal. And that's just for training. So it's really imperative that we address this and provide transparency this

Lynn Kaack: Is that Is that for developing a model or for training at once? I mean, is that with grid search, architecture search?

Will Buchanan: For for a single training run. So it does not account for sweeps or deployments.

Chris Adams: All right, so there's a, there's some language that we haven't heard for here, so, but maybe it might be worth asking, maybe Will, could you maybe talk about, just briefly, you said a rail car full of coal, and I don't actually know what that is, I mean, in metric terms, what does that look like?

Will Buchanan: A hundred million grams. I don't have the conversion handy, but we took the U.S EPA greenhouse gas equivalencies. And I should add the methodology that we applied was the Green Software Foundation's SCI. So we calculated the energy consumed by the model and multiplied it by the carbon intensity of the grid that powers that data center.

Chris Adams: Okay, cool, and that was per training run? So that wasn't the, in the, the equation of the entire model, is that correct?

Will Buchanan: Correct. 

Abhishek Gupta: That's the other interesting part as well, right? When you're thinking about the life cycle is, or life cycle of the model, I should say, because life cycle has multiple meanings here, which is that once that model is out there, what are the inference costs, right? And are we, if this is something that's going to be used you know, hundreds, thousands, tens of thousands of times, if it's something, you know, if it's, if it's a large model that's, you know, now being used as a pre trained model and is going to be fine tuned on by, by other folks downstream, are we able to then, you know, talk about amortization of that cost across all of those use cases?

And again, I think what becomes interesting and, and is how do we account for that stuff as well, right? Because we, we don't have complete visibility on that as well. And, and I know Lynn's nodding here because her paper that's, I think coming out, being released in an hour and a half, actually the embargo gets lifted on our paper, actually talks about some of those system level impacts.

So maybe, maybe Lynn, you want to chime in and talk a little bit about that as well?

Lynn Kaack: Yeah, thank you so much. Exactly. So I think what's a crucial number that we're currently still missing is not what is emitted from a single model in a well known setting, but what is emitted overall from applying machine learning? So what are the usage patterns and practice? Like how often do people develop models from scratch?

How often do they train or retrain them? People I mean, of course, organizations and typically large organizations and companies. And how do they perform inference on how much data, how frequently? And there are some numbers out there from Facebook and Google and in their large scale applications actually inference outweighs their training and development costs in terms of greenhouse gas emissions.

So inference might become a bigger share depending on the application. So we really need to understand better how machine learning is being used in practice also to understand the direct emissions that come from it.

Chris Adams: An inference is a use of a model once it's in the wild. Is that what an inference is in this case? So there's an environment, so you could think of the making part, and then there is the usage part from the inference, right? So is that how that part works?

Lynn Kaack: Exactly. So if you use a model on a data point, we call that inference. So you feed in the data and it gives you a result. Then training means you sort of train a single configuration of the model once on your training data set, and then development is what I refer to as if you search over different configurations of the model.

So there are lots of hyperparameters that you can adjust to achieve better performance. And if new models are being developed, then there's an extensive search over those hyperparameters and architecture configurations that then of course gets really energy intensive because we are training the model thousands of times essentially.

Will Buchanan: Yeah, one of the figures that really resonated with me, I think Nvidia posted on their blog that inferencing for about 80 to 90 percent of the carbon cost of a model. think Lynn on one of your papers, it was, Amazon had also claimed around 90 percent. So these are really non trivial costs, and I'm not aware of any framework to measure this.

Lynn Kaack: Yeah. So that Amazon number just to be clear is costs or monetary costs that came from a talk, but there are numbers now published by Google and Facebook, but they look at some applications of theirs where inference outweighs training in terms of energy consumption. They're not exact numbers. It's not entirely clear which applications those are, but there is some data at least that shows that.

And I think it just highly depends on the application that you're looking at. And sometimes, you know, you build a model and then you do inference once and you have the data set that you, and then in other types, you build a model and then you apply it billion times a day. So, of course, that can then add up to a lot more energy consumption.

Chris Adams: Wow, I didn't realize that was actually an issue, because most of the numbers I've seen have been focusing on the training part. So, Will, I think this is something we spoke about before, that training, there's a kind of trend in the use, in the energy use from training already. Is this something, because I've seen figures from OpenAI, but my assumption was that basically computers are generally getting more efficient. About twice as efficient every two years, or so with like Moore's Law or Kumi's Law or things like that. But if you're seeing an uptick in usage here, is, does that mean that they're staying about the same? Or is there, is there a trend that we should be taking into account there?

Will Buchanan: So I think the computational costs of training have been doubling every 3.4 months or so, and so I think the trend is only accelerating. The models are just getting larger and larger, and you've got, I think, GPT 3 is one of the largest ones around at this point. We might challenge Moore's Law.

Chris Adams: Okay. So if Moore's Law is doubling every once every two years, I mean, what is the impact of doubling every 3.4 months? I mean, over a few years, what does that work out to be? Because I don't think I could do the exponential numbers, the exponential math, but it sounds like it's, it sounds like a pretty big number, basically, if something is doubling on every three or four months, right?

Will Buchanan: I also don't have the math handy, but I think it's important to note here, and Abhishek was talking about this earlier, models are very flexible, so you can train them once and then provide some fine tuning or transfer learning approach on top of them, and then repurpose these models for a number of different applications. And then you can even compress them, let's say using ONNX Runtime. You can be very efficient. You can really amortize the cost of that model.

Abhishek Gupta: So yeah, just building on Will's point, there's a lot of work on quantizing the weights of a trained network, applying distillation approaches, using teacher student model approaches that actually helps to shrink down the model quite a bit, especially with the whole push for TinyML, trying to shrink down models so that they can be deployed on edge devices has been something that's helped to manage to a great extent, the computational impacts.

One of the other things that I wanted to highlight as, as, you know, Will was talking about more models getting larger is, there's this almost fetish in the world today to continuously scale and keep pushing for ever larger models and in chasing SOTA, as they would say, so chasing state of the art, you know, which is great for academic publications where you get to show, "Hey, I improved state of the art performance on this benchmark data set by 0.5 percent or whatever," right? And in performance, I think what's being ignored is that that has a tremendous, tremendous computational cost. In fact, one of the hidden costs that I think doesn't get talked about enough is there's this statistic out there that, you know, 90 percent of the models don't make it into production.

And that kind of relates to things like, you know, neural architecture search and, you know, hyper parameter tuning, where you're constantly trying to refine a model to achieve better performance. A lot of that actually goes to waste because that stuff doesn't make it into production. So it's actually not even used.

And so there's a whole bunch of computational expenditure that is done that actually never sees the light of day, never becomes useful. That obviously has environmental impacts, right? Because of the operational and embodied carbon impacts, but none of that actually gets talked about, reported, documented anywhere because, well, who wants to know that, hey, I trained, you know, 73 different, you know, combinations to get to where I'm at.

You just talk about the final results.

Chris Adams: Okay, let's say that if you don't want to go down one of those rabbit holes, what should you be using or where would you start if you wanted to start applying some of these ideas about greener AI in your work on a daily basis? Does anyone have anything that they would lead with, for example?

Will Buchanan: Bigger is not always better. Sometimes you really should choose the right tool for the job. We've had some really great graduate student projects from the University of Washington's Information School, and they built some case studies and samples around green AI. As an example, a project led by Daniel Chin was comparing a sparse or a dense model of a green AI model to a dense model for an anomaly detection setting.

And they found that using sparse, meaning less trees and a shallow, smaller depth per tree, random forest would save a massive amount of carbon and provide the equivalent accuracy. So I think it saved about 98 percent in terms of the monetary cost and energy.

Chris Adams: Okay, wow, that's bigger than I was expecting. What would you say to people if they're in production, they're trying to do something now?

Lynn Kaack: I think a big goal should be to not only develop more energy efficient machine learning models, but then also ensure that those are actually being used. And surprisingly, even sometimes within the same company, certain model developments are not being passed on to other parts of the company. So really trying to develop standard models that then are also being used in practice is important.

So interoperability of energy efficient machine learning models.

Chris Adams: If people, someone does want to look at this stuff, and they do want to apply some of these ideas, you spoke a little bit about using some other models. would you suggest people look if they wanted to operationalize some of the kind of wins or some of the better ways to make green software greener, for example?

I realize you've got a paper coming out and you work on this day to day. So yeah, what would you point us to?

Lynn Kaack: So, I mean, as I understand, there's a lot of ongoing research in the machine learning community for more energy efficient machine learning. So I don't have any names on top of my head in terms of workshops or community resources where one can see what are the most energy efficient model types. For a specific application.

I know that there are some very comprehensive papers also that summarize all the different research approaches that are being taken, but I would encourage if you are looking for using like a deep learning model of some kind, just inform yourself quickly if there's also a leaner version of it. So much of the like widely used models like BERT, for example, smaller versions that can almost do the same thing.

And maybe your performance doesn't suffer much. If you're using a much lighter model.

Chris Adams: Okay, so light up models and looking around what we have there. And Will, is there a paper or a source you might point to?

Will Buchanan: I was actually going to talk about the Carbon Aware paper that we're about to publish. I think that's a slightly different track.

Chris Adams: That's up next week, right? So that'll be the 13th or 14th of June. That's when that'll be visible, correct?

Will Buchanan: Exactly. 

Chris Adams: Okay, cool. All right, then. There's a load more that we could dive into. We've got copious, copious, copious show notes here. So what I'm gonna do is I'm gonna say thank you everyone for coming in and sharing your wisdom and your experiences with us, and hopefully we'll have more conversations about green software in future. Thank you folks.

Asim Hussain: Hey everyone, thanks for listening. Just a reminder to follow Environment Variables on Apple Podcasts, Spotify, Google Podcasts, or wherever you get your podcasts. And please do leave a rating and review if you like what we're doing. It helps other people discover the show and of course we want more listeners.

find out more about the Green Software Foundation, please visit greensoftware. foundation. Thanks again and see you in the next episode.

In this episode of Environment Variables, host Chris Skipper introduces "Backstage," a new segment offering behind-the-scenes insights into the Green Software Foundation's key projects. The episode highlights the "Transforming Organizations for Sustainable Software" (TOSS) project, led by Pindy Bhullar, which aims to embed sustainable practices across all aspects of business operations. The TOSS framework, divided into four pillars—strategy, implementation, operations, and compliance—provides organizations with a decision tree approach to drive sustainability in software development and IT. With collaborative efforts, workshops, and real-time feedback, the project is set to release version 1.0, with the ultimate goal of reducing carbon emissions across the industry. 
Learn more about our people:

• Chris Skipper: LinkedIn | Website
• Pindy Bhullar: LinkedIn | Website

Find out more about the GSF:

• The Green Software Foundation Website 
• Sign up to the Green Software Foundation Newsletter

Resources:

• GitHub - Green-Software-Foundation/toss: Transforming Organisational Software Sustainability (TOSS) [01:28]
• Transforming Organizational Software Sustainability | GSF
•  Software Carbon Intensity (SCI) Specification Project | GSF [03:47] 
• GitHub - Green-Software-Foundation/sci: A specification that describes how to calculate a carbon intensity for software applications.
• Impact Framework [03:54] 
• GitHub - Green-Software-Foundation/if: Impact Framework 

Events:

• Sustainable AI: Myth vs Reality (October 15 at 6 pm CEST · Amsterdam) [12:49]
• CNCF Sustainability Week - Cloud Native for Green Software (October 15 at 6:30 pm CEST · Hybrid · Karlsruhe) [13:02]
• Code Green London October Meetup (October 15 at 7:00 pm · London) [13:25]

If you enjoyed this episode then please either:

• Follow, rate, and review on Apple Podcasts
• Follow and rate on Spotify
• Watch our videos on The Green Software Foundation YouTube Channel!

Connect with us on Twitter, Github and LinkedIn!

TRANSCRIPT BELOW: 

Chris Skipper: Welcome to Environment Variables, where we bring you the latest news and updates from the world of sustainable software development. I'm your host, Chris Skipper. Today, we're launching an exciting new segment: 

Backstage. Here, we'll take you behind the scenes at the Green Software Foundation to highlight ongoing projects and share insights from the members directly involved.

In the first episode of Backstage, you'll hear about innovative projects shaping the future of green software and the hard work behind it all. It's easy to overlook the complexity and dedication required to drive sustainability in the tech industry. But today, you'll hear more about the passion, thoughtfulness, and expertise that fuels these initiatives.

From collaborative projects to cutting edge research on sustainable computing, the Green Software Foundation is the conversation you need to be part of, and we're bringing you Deep Inside. Join us as we look into the stories, challenges, and triumphs of the people making a real impact on our environment through software.

We're no longer gatekeeping what it takes to set new standards and norms to green software. So, what does it take to inject sustainability into every aspect of software development? How are organizations and passionate climate activists in software turning ambitious goals into practical, actionable results?

Let's find out on this special Backstage episode. For this initial episode, we'll be looking at the Transforming Organizations for Sustainable Software project, better known as the TOSS project. Leading this project is the inimitable Pindy Bhullar, who provided us with some soundbites. This initiative is really important in the Green Software Foundation's broader mission to reduce carbon emissions through software.

It's all about providing changemakers in organizations with the tools, methodologies, and frameworks they need to embed green software practices and drive real, measurable change. Firstly, we asked Pindy to give us an overview of the TOSS project. It's goals and why it's so crucial for the foundation.

Pindy Bhullar: Transforming Organizations for Sustainable Software is the acronym for TOSS. Businesses will be able to utilize the TOSS framework as a guide to lay the groundwork for managing change and also improving software operations in the future. Software practices within organizations can be integrated with sustainability in a cohesive and agile manner, rather than addressing green software practices in an isolated approach.

For a company to fully benefit from sustainable transformation of their software development processes, we need to review all aspects of technology. The TOSS framework is designed to be embedded across multiple aspects of its business operations. Dividing the task framework along four pillars has allowed for simultaneous top down and bottom up reinforcement of sustainable practices, as well as the integration of new tools, processes, and regulations that emerge over time.

The four pillars include aim to foster a dynamic foundation for companies to understand where to act now, to adjust later, and expand with an organizational sustainable software transformation. The four pillars are strategy, implementation, operational, compliance, and regulations. And within each of the pillars, we have designed a decision tree that will be constructed to guide organizations in transforming their software journey. 

 The Green Software Foundation has created the specification to calculate carbon emissions from software, the SCI, and also developed the Impact Framework to enable firms to calculate their emissions. But in order for firms to transform their technology operations, we need to embed best practices across the wider organization.

Chris Skipper: It's clear that by integrating sustainability across multiple levels of business operations, the TOSS project is setting up organizations for long term success in their green transformation journey. Now, let's talk about the impact this project can have. Reducing the carbon footprint in software development is a huge focus for many companies today.

So, how exactly does the TOSS project help achieve that, and what kind of sustainability improvements can organizations expect by adopting this framework? 

Pindy Bhullar: The TOSS project views sustainability in software development and IT operations as a holistic approach, and the primary goal is not to view software and IT as an isolated function, but integrate the carbon footprint into the functions across the organization.

The framework will help organizations to ask the questions and adapt their responses based on their needs, but also connect strategy, implementation, operations, and compliance and regulations. 

Chris Skipper: It's insightful to see how the TOSS framework fosters a more integrated approach, making sustainability a core part of a company's strategy, operations, and compliance.

Now, as with any large scale initiative, developing a framework like TOSS must come with its challenges. We asked Pindy to share some of the key obstacles the team has encountered while bringing this project to life, and how have they navigated these challenges to keep the project moving forward. 

Pindy Bhullar: We're all busy in our day jobs, so just being able to focus on the TOSS framework has been a challenge.

We've been having regular calls since the project inception January this year, which has enabled us to keep the discussions moving. We have overcome obstacles by building a version 0.5 of the framework and then asking volunteers from the GSF member organizations to join a series of workshops to review, question, challenge, and brainstorm. It enabled us to develop the framework further by incorporating all the great feedback. 

Chris Skipper: It's great to hear how the team has managed to build the toss framework through collaborative efforts and feedback from GSF member organizations. It sounds like the workshops have played a crucial role in refining the framework.

Now, Pindy mentioned that the TOSS project incorporates a decision tree strategy. We asked Pindy to walk us through how this approach is helping organizations embed green software practices more effectively. How does this decision tree guide organizations in making sustainable choices throughout their software development and IT operations?

Pindy Bhullar: We use the decision tree approach for the TOSS project for a number of reasons, particularly in contexts where decisions involve multiple options and varying outcomes. It's also a great way to visually represent complex decisions into manageable parts, and it helps decision makers visualize the entire process in a structured, analogical format.

It also enables the evaluation of multiple potential paths and their respective outcomes. So each decision point can have different branches, each representing a different course of action. One of the benefits of decision trees is the ability to easily update or expand as new information becomes available.

This flexibility is critical in the space of green IT and green software, as this topic is constantly evolving. 

Chris Skipper: It's clear that the decision tree approach adds flexibility and adaptability, making it a perfect fit for an evolving field like green IT and software. Now, let's dive into the specific tools and technologies that are supporting the TOSS project.

We asked Pindy to highlight some of the key tools or frameworks being used and explain how they align with the project's sustainability goals. How do they contribute to helping organizations achieve lower emissions and more sustainable software practices? 

Pindy Bhullar: To build the framework of decision trees, we are using mirrorboards.

Initially, it helped us to brainstorm and use the ideation tools to create mind maps to generate and organize our ideas in an interactive way. We also benefited from the real time collaboration and, more importantly, the ability to create the visual charts. The tool is enabling us to progress with the TOSS framework and build the sustainable software decision trees.

Chris Skipper: Now, let's explore the collaborative side of this initiative. We asked Pindy if there have been any significant partnerships or collaborations with other organizations or teams that have enhanced the success of the TOSS project. How have these partnerships contributed to its development and progression?

Pindy Bhullar: There have been excellent collaborations during this project. The workshop sessions were critical in partnering with organizations because it enabled us to further build the TOSS decision trees. It also helped us to make faster progress, while allowing the group to discuss new ideas, make revisions, and also to bring new insights.

Chris Skipper: It's clear that collaboration has played a crucial role in pushing the TOSS project forward, with the workshops really driving the development of those decision trees. Now, let's shift focus to the progress the TOSS project has made over time. We asked Pindy to walk us through some of the significant achievements or milestones the project has achieved so far, and how has it evolved since its inception earlier this year?

Pindy Bhullar: At the start of the project, we created a business case which outlined the TOSS project and what we wanted to achieve. Then we moved into a phase of brainstorming and mind mapping, really trying to capture key points of how we wanted to develop the framework and decision trees. We started then to break down the functions that would be involved in organizations and we created the four pillars that I mentioned earlier.

And out of that, then we started to get into the detail and start to create the decision trees. We have progressed since the workshops, and now we're working towards version 1.0 of the decision trees. We have redeveloped the strategy and implementation pillar, and we're still working on the operational and compliance and regulation pillars.

From the initial ideas, we've added further content and details to the framework, and it's constantly evolving. 

Chris Skipper: It's truly fascinating to hear how the TOSS project has progressed. From the initial business case to the detailed decision trees, and the continual refinement of the strategy and implementation pillars.

Now, as we look to the future, we'd love to know more about what's on the horizon for TOSS. What are the next steps in the project, and what long term impact will it have on the widespread adoption of green software practices across the industry? 

Pindy Bhullar: Over the course of the next few months, we want to complete version 1. 0 of the four decision trees. Also, create the guidance documentation that will sit alongside it. And in addition, create a white paper documenting the framework. We want the long term impact of the framework to help organizations implement tools, methods for calculating their emissions, processes, ensuring that we link the strategic level of the organization all the way through to software development and operations.

Chris Skipper: I'd like to thank Pindy for sharing such valuable insights into the TOSS project. It's clear that this framework is not only about integrating sustainable practices, but transforming the entire approach organizations take to software operations. From her detailed explanation of the four pillars, strategy, implementation, operations, and compliance, it's evident that TOSS provides a holistic solution that embeds sustainability at every level.

The collaborative workshops and iterative feedback process she described have played a crucial role in refining the decision trees, and it's exciting to hear that version 1.0 is on the horizon. As the project evolves, it's clear TOSS will have a lasting impact, empowering organizations to seamlessly incorporate green software practices and meet their carbon reduction goals.

We truly look forward to seeing how the framework will continue to grow and shape the future of sustainable software. Now, we've come to the end of the episode, but before we finish off, we have a few events to read out. Firstly, we have Sustainable AI, Myth vs Reality, Gain Insights and Actionable Strategies on the Intersection of IT and Environmental Responsibility at this in-person event hosted by Sustainable IT Netherlands at Accenture in Amsterdam on October 15th, starting at 6pm Central European Standard Time.

Next up, we have CNCF Sustainability Week, Cloud Native for Green Software. Share and explore ideas, patterns, best practices, and tools in the context of the green software principles. This hybrid event is hosted by Green Software development, Karlsruhe at Blue Hands in Karlsruhe on October the 15th at 6:30 PM CEST.

And finally, we have Code Green London October Meetup. Learn about the latest in green software and network with London's green tech community. Also on October the 15th at 7:00 PM BST. This in-person event at the Red Badger in London will feature past environment variables guest Tereze Gail. Who will be talking about Salesforce's sustainable AI approach.

Hear how the world's largest enterprise software company is balancing AI innovation with environmental responsibility. So, we've reached the end of this special backstage episode. I hope you enjoyed the podcast. To listen to more episodes about Green Software, please visit podcast.greensoftware.foundation and we'll see you on the next episode. Bye for now.

This Week in Green Software, the affable Anne Currie is joined by Sara Bergman, Senior Software Engineer at Microsoft and co-author of Building Green Software. Together, they dive into the complexities of sustainable data in relation to AI and cloud computing. They explore the environmental impact of managing and storing vast quantities of data, and question the feasibility of making these processes more eco-friendly. The discussion touches on cloud providers' carbon reporting, the importance of using AI responsibly, and how businesses can optimize their cloud use to minimize their environmental footprint. Tune in for an insightful conversation on balancing technological advancements with sustainability in the age of AI.

Learn more about our people:

• Anne Currie: LinkedIn | Website
• Sara Bergman: LinkedIn

Find out more about the GSF:

• The Green Software Foundation Website 
• Sign up to the Green Software Foundation Newsletter

News:

• Report: Thinking about using AI? - Green Web Foundation [03:54]
• Why Cloud Provider GHG Reporting isn’t enough: The Case for Product-Level Accountability [14:16]
• Is sustainable data storage a paradox? | TechRadar [33:03]

Events:

• GSF Global Summit in London Summit (October 1 at 6 pm BST · London) [37:56]
• GSF Global Summit inGlobal Summit Munich Summit 2024 (October 1 at 1:30 pm CEST · Munich):
• GSF Global Summit in Green Software Summit Hamburg (October 1 at 1:30 CEST · Hamburg)
• GSF Global Summit in Dublin (October 8 at 9:00 am BST · Dublin): 
• Green Software Summit Berlin (October 9 at 6:30 pm · Berlin): 
• GSF Global Summit in Singapore Edition 2024 (October 9 and 9:00 am SGT · Singapore):
• GSF Global Summit in2024 Tokyo (October 9 at 6:00 pm JST · Tokyo):     

Resources:

• Building Green Software [Book] [02:04]
• Greening Digital and the Rebound Effect | Environment Variables [11:36]
• GitHub - Green-Software-Foundation/real-time-cloud [21:35]
• Green Software Maturity Matrix [21:55]

If you enjoyed this episode then please either:

• Follow, rate, and review on Apple Podcasts
• Follow and rate on Spotify
• Watch our videos on The Green Software Foundation YouTube Channel!

Connect with us on Twitter, Github and LinkedIn!

TRANSCRIPT BELOW:

Sara Bergman: What data do we need to take in order to take meaningful action? Like, what is the level that, of course, yeah, if I could get minute by minute, like there's tons of stuff we could do and correlations we could draw, but what is the level of data that we would need to start taking meaningful action? And I think that could unlock a lot of good things.

Chris Adams: Hello, and welcome to Environment Variables, brought to you by the Green Software Foundation. In each episode, we discuss the latest news and events surrounding green software. On our show, you can expect candid conversations with top experts in their field who have a passion for how to reduce the greenhouse gas emissions of software.

I'm your host, Chris Adams.

Anne Currie: Hello, and my name is Anne and welcome to 

The Week in Green Software. So this week, you won't be hearing the usual dulcet tones of Chris Adams. I'll be joining you, Anne Currie, and we'll be delving into the tricky, but interesting world of sustainable data, whether it's possible to store and manage huge quantities of data, which we will need now, particularly for AI, in a way that's eco friendly.

Or is that impossible? Well, look, I'm going to leap through to the end and say, we have to do it. And therefore it is not impossible. It just has to be done. We'll find a way of doing it. And there are ways of doing it, which we'll be talking about today. we'll also be exploring why big cloud providers' carbon reporting isn't really telling us the full story.

Or, well, we are not interpreting it in the way that it was designed to be interpreted. And we need to be more careful about that. We're reading more into it than is true. And we need to be much, much more explicit about what the carbon reports from people like AWS and Azure actually mean.

We'll also be talking about data centers and AI. And that's something that my guest today is an expert in. And talking about my guest, joining me today is my co-author on Building Green Software, the book on what we need to do to make the tech industry green, and also Environment Variables regular, the lovely Sara Bergman.

So Sara, do you want to introduce yourself?

Sara Bergman: Yeah, hi. Thanks for having me on again. Always lovely to be here. I'm so excited to have a chat with you, Anne. Yeah, my name is Sara Bergman. I am a senior software engineer at Watttime. Microsoft, author of Building Green Software. And like here, I'm always asked, what have you been up to recently? Nowadays, I'm like, what have I been up to recently?

I did a fun thing, though. I had the talk for the Norwegian, because I live in Norway, and the Norwegian tax authorities about green software. And that was really fun. I love it when Because they are actually very far ahead in their journey, they're one of the most innovative companies when it comes to IT, so it was really fun to come out and have a chat with them.

Anne Currie: That's great. And I should introduce myself as well. My name is Anne Currie and I am co-author of O'Reilly's new Building Green Software along with Sara and our other co-author, Sara Hsu. And I also do a lot of training. So I've been busy at the moment doing loads of training courses. So, workshops on building green software and also an experts training course, which is all quite good.

So, if you want to get involved in any of that stuff, you can follow me on LinkedIn. So, as usual, today, we will be talking about a couple of interesting articles, publications that have come out over the past week around green software, all things green, and as usual, all the links to the articles will be in the show notes, so you'll be able to read them yourselves afterwards.

But I'll give you a little bit of a summary about what they say. So the first article we'll be talking about today is from the Green Web Foundation. And it was written by our normal Environment Variables host, Chris Adams. So that's where he is, or that's what he's been up to today in his work at the Green Web Foundation.

And he wrote it with his colleague, Hannah Smith. And the report is all around AI's environmental impacts. And it's got some interesting figures in there. Basically, AI uses a load of electricity and at the moment, as we don't yet have a completely green grid, that means that a lot more CO2 is being emitted into the atmosphere as a result of the fact that we're training a lot of models, doing a lot of inference.

So it is an interesting report and it's, I'm going to kind of summarize, they have some actionable things, some questions for you to ask yourself at the end of the report. So I will go over those now and then Sara and I can discuss them. The first is that you should always question your use of AI.

That's kind of part of using the right tool for the job. Is AI the right tool for what you're doing? Is it overkill? Could you use a spreadsheet? That's If you are using AI and you decide it is the right tool for the job, are you using it properly? Are you using it well? Are you using the right AI tools for the job?

And the third is to try and get ideas of your footprint, of the footprints of the work that you're doing with the AI, so you actually have an idea about what impact it is having and what you're going to need to do about it in the future. So Sara, this is kind of your area. Did you enjoy the report? What did you think?

Sara Bergman: I did. It's a very long report. It's very well written. Obviously, I mean, with Chris and Hannah, you're going to get something that's well written, of course. So no surprises there. No, it's good. And I think also for people who are maybe newer to the field of green software and green AI in general, there was a lot of good background to like really help understand the intricacies of this area. And something that I particularly find interesting in the shift we are now is that they talk about different phases where your emissions kind of stem from, it's like manufacturing, training and inference. And now, like you said, we talk a lot about inferencing, like using AI, that the use phase is what we talk about.

But back when I started, sounds like I'm really old, only four years ago, not that long ago, but when I first started talking about green AI. Yeah, a lot of, a few people, not so many were talking about green AI. A lot of people were researching, but not so many people were like discussing it. And then it was a lot of focus on the training.

There was a lot of great research being done on how to minimize the impact on training. I think in the research community, that's maybe the easiest, not the easiest thing, but a good first thing to research, right? And now we're seeing more focus on. on the production side, not like inferencing. And I think that shift has been very interesting to follow.

Anne Currie: But yeah, it is fascinating, isn't it? Cause there's loads, and we've had this conversation offline, because obviously we a book together and therefore we talk together quite a lot. But we've had this conversation a lot in that it feels like there's a load of stuff to be learned about inference.

So how you actually get the answers back as a user for models from the world of things like CDNs, how do you get fast answers and easy answers to things all over the world from data that is not necessarily by default, wasn't created close to the user who's querying it?

So yeah, there's, loads of prior art there to learn from it. It's a really interesting field.

Sara Bergman: It is a very interesting field and I think an additional like spiciness of this is that now a lot of people equate AI to large language models. That's not all of the AI that exists in the world, it's just the right now the most popular and what we think maybe will be the most interesting use case, or for like productivity and stuff.

And for that we require a lot of specialized hardware, like GPUs or TPUs on CPUs, et cetera, which is another thing to kind of reason over. And, maybe if you're very good at handling your production workloads on a CPU, like this might be a new area for you where you, we need to learn as an industry. I thought that was an interesting part of the article.

Anne Currie: Yes. Yeah, it is. It is all very interesting, isn't it? it's like something that all three of us say in the book a lot, is that there's a load of clever stuff going on in the tech industry, and it's usually, there's a desire to make things efficient because it's cheaper. Otherwise, everything gets very expensive.

So if you use things as they were intended to be used and use the right tools for the job, usually that's significantly more energy efficient and therefore greener. And the same is very much true of AI, isn't it? If you use the right tools, the right hardware, the right models,

Sara Bergman: Yeah, exactly. Exactly. So that and then I think that ties into right, that was it number two on their like, use AI responsibly, which was interesting. I also like the first question that they had. Now I'm scrolling and losing it was it think about when you should use AI? Was that the phrasing?

Yeah, question your use, know when to use it. Like, do you have a favorite use case?

Anne Currie: Well, I have to say, I do love generative AI. 

Sara Bergman: Yeah. 

Anne Currie: it's, an interesting one, cause quite often... well, it's amazing how often it comes up where people think that it's kind of like you're either doing AI or you're green. That's it. Those are the two, two options. And they cannot be the two options.

It can't be either don't do AI and therefore it's green or do AI and therefore it's not green. They have be brought together because if there's one thing that we know, two things we know are coming in the future for humanity. One is climate change. The other is AI and use of AI and AI systems.

They cannot be mutually exclusive. They have to be brought together. I mean, and AI is just computing. It's the same kind of things that we do when we talk about in Building Green Software for all computing just needs to be applied to AI. It's not a, it's not a new thing on its own, but

Sara Bergman: and I agree. And I think that's also so interesting with this, like, larger discourse, as you say, I think not only that it exists in other areas of life, where there are some people who would be like, "no, we should just stop doing this, like, stop advancing technology, and then everything will solve itself," but that doesn't work. We cannot and does not want to go back to like a farmer society where everyone grows your own food. Like if you're into growing your own food, that's fantastic. I'm happy for you. I grow radishes on my balcony and I enjoy that thoroughly. But there's so much advantage and good for humanity and our planet that comes from technology as well.

So we need to do both. We must do both. Like, it's non negotiable. So it's more like how and when? And 'when' is hopefully starting now.

Anne Currie: Cause the trouble is, if you have that conversation, if you say, "Oh, well, I want to do AI, therefore I can't do green." People will choose AI because the businesses, if they don't, if they don't try these new tools and services as they come available, they will go out of business. So, if you say, "Oh, well, if you're going to be green, you can't use AI," then what you're telling people is don't be green.

And it's crazy. You're just making the wrong arguments. But anyway.

Sara Bergman: I fully agree for sure. And I think it was something we were in a different, or like this podcast together, but another episode and you said something that I've been carrying with me for so long. Sometimes people say something and it just get like something clicks. And what you said, and I'm going to maybe paraphrase, but things that are limited are sort of less concerning. Like, for example, if you build a super efficient washing machine, I still have only so much washing up to do. There is an end to how much I will wash it. Like this is in response to Jevons paradox, of course. And the same goes for this. Like, yeah, AI will make us more efficient, more productive.

Okay. But we'll then just do more. It's like, yeah, but the working day is sort of eight hours and we are only so many people on the planet. There is a limit to when we'll be done, whereas for crypto, for example, which always comes up when we talk about this, of course, there's no limit. I will never say I have enough money, I'm done mining. But there is a limit where I say, thank you, co-pilot for GitHub, I don't need more code suggestions.

I'm happy with my feature now, sort of.

Anne Currie: Oh, yeah. That's an issue, because, I always tend to think that one of the, one of the things about AI is that, that we're not,, we haven't yet reached the limits of what we want to do with it. But, you're right. There's, it's nowhere near. I mean, we've already come up with the thing that is the most limitless.

You've heard the use of CPU and it's CPU use that's really the problem that's, that uses a lot of electricity, is crypto. It is literally boundless how much we might want to do with it. So we've kind of already created the worst 30.

Sara Bergman: yes.

Anne Currie: AI is so

Sara Bergman: not it.

Anne Currie: it's a comparison. At least there's some people, 

I don't know, we shouldn't be controversial because there are a whole load of people who say, well, there's loads of benefits of cryptos, but I personally do not, I'm not a crypto bro.

Sara Bergman: Same. No, I did write my master's thesis on blockchain. I think blockchain is a cool invention. Like, it's a cool technology. And there are use cases that are, of course, bounded. And then there are use cases that are unbounded. And I think people can use their own head to figure out what's what.

Anne Currie: We've now got ourselves into enough trouble talking about good and bad uses of technology. And to a certain extent, it's very hard to say what's a good or bad use of technology. I tend in my head to think it was a bad use of technology, I, not everyone agrees with me, but we will move on to the next article, which is one, now...

So this one, cloud provider Greenhouse Gas reporting isn't enough. The case for product level accountability. Which is, again, the link is included in the show notes. Now I know, because this talks about Azure, Sara, you, might be in a position where I'm saying things to you and you have to say, you may say that "Anne, but I cannot possibly comment."

So in which case you'll just have to leave me to talk, but that's fine. Cause I can talk endlessly about this stuff because I find it really, I think this is absolutely fascinating and there's loads of stuff to learn from this article. So the article is all about, it's highlighting the limits of cloud providers and data, the carbon footprint reports.

And it emphasizes the need for companies to track emissions at product, at data product level. And it argues that these broader reports mask the true environmental costs of specific processes like data and that things would be better if you broke the emissions down to product level. And I think that's great, but I don't think that's actually the problem.

I think it's well worth reading the article, but I think it skips over, I think, what is the more significant problem that we're having, that we're seeing, because I'm out there talking to people a lot and I know this is a significant problem at the moment, that the cloud providers' reports are doing something, a lot of the new cloud providers' reports, particularly the recent AWS releases are doing something which I can't really shout at them for because I demanded it myself some years ago and I am somewhat hoist by my own petard for this.

So many years ago, me and a fellow, a collaborator called Paul Johnston ran a, back in 2018, we ran a campaign called Sustainable Servers by 2024. And what we were campaigning for was that all of the cloud providers would commit, and this was some time ago, would commit to being carbon neutral by 2024.

And it was really aimed entirely at AWS because Google and Azure were already carbon neutral at that point and AWS was not. So, we ran this big campaign and we had petitions and all kinds of things for saying that they should be carbon neutral, collateral and tradeful. And of course, carbon neutrality is quite a limited demand.

It's basically saying, "I want you to be carbon neutral. I want you to know how much your carbon emissions are. And then buy offsets that match the carbon emissions." Now, we all know that offsets are good in the past, but they're not time matched. They don't mean that the carbon savings that you've created are time matched with the carbon emissions of your systems.

So your systems can still be throwing off carbon dioxide into the atmosphere whilst you are carbon neutral. So it was a useful, a useful measure in 2018, it was still useful. Now in 2024, it's actually not as good as it could, it's not enough. it's the bare bones. It's the least we can ask for.

But AWS have done it on time, 2024, and they are producing these reports that they're giving to AWS customers that say, "you're carbon neutral, your systems are net, not producing any carbon." But the problem is that a report that says, and that's great, but it is nowhere near enough.

It's not enough. It's great information for your finance team because your finance team don't want that data. They need that data, particularly if you're going to be, if you're in the EU or you're selling into the EU. Because the EU is now demanding carbon taxes, and carbon taxes are another form of offsetting, really.

You pay for the carbon you're emitting into the atmosphere. So if AWS produced a report saying, look, there's no net carbon you're producing into the atmosphere at the moment, which they are generally at the moment for most AWS systems, it seems. That is really a report for your finance team, so they don't have to, so they can go, "oh, that's great.

The offsetting has already been done for me." That is not a report for your tech team because they're not saying that no carbon is actually being produced by your systems. Your systems are still producing loads of carbon and you still need to adjust your systems. to reduce carbon, which you can do, but the report is kind of giving you the impression that you don't need to.

So is that something you can comment on or?

Sara Bergman: Yeah, this gets me going, right? Cause I'm all about enabling like people with feet on the ground, hands on the code is what I say. So I do agree a lot of the reporting that comes out, it's much better than it used to be. This difference, and maybe now we're slipping into the next paper a little bit, but there, of course, difference between market-based and location-based reporting in the Greenhouse Gas Protocol.

That is a feature, a fundamental feature of the Greenhouse Gas Protocol, which every single industry makes use of. Now, not only, 

Anne Currie: I'm going to interrupt you at this point, just because this is something that I realized when talking to people. One of the issues is that people don't understand the difference between market-based and location-based. What does that mean? So market-based, I think that the word that people really understand is offset.

If you're market-based, it's saying you're still putting carbon into the atmosphere, but we are offsetting it, which was great 10 years ago, nowhere near good enough nowadays. location-based means actually, it's about reducing, it's about not putting carbon into the atmosphere. So that's what we want. Market offsets are a step to getting there, but they are not there.

So sorry, go on.

Sara Bergman: yeah, no, but that's, good. I also like it's like the electrons you pay for versus the electrons you use, if you want to get really down to the socket level. So yeah, I think that's something to talk about. Also, something that I kept thinking about after reading this article, and I just want to read the comment from it because I think the comment was amazing.

And the comment is, "conceptually, I agree. More data is better. However, I've never met anyone advocating for product level data who has actual operational experiences of running shared service platforms and therefore would understand the complexity of delivering these metrics. Because yes, if we're talking about, we could have one report for the finance team, that's fine, but should there be an additional one for, the people who write the code, who maintain the service? And then the question becomes, okay, but what data do we need to take in order to take meaningful action? Like, what is the level that, of course, yeah, if I could get minute by minute, like, there's tons of stuff we could do, correlations we could draw, but what is the level of data that We would need to start taking meaningful action? And I think that could unlock a lot of good things.

Maybe we don't need the world. Do you know what I mean?

Anne Currie: Yeah. I know what you mean. I totally know what you mean. I think we should step down our emissions. Well, having said that, within the Green Software Foundation, there is a project, the Real Time Metrics Project. And that is looking forward to when we can have second, millisecond by millisecond data about, so we can tune applications to, to get there.

But then we also have another project, which I run, so their real time metrics project is run by Adrian Cockcroft and I run a project called the Maturity Matrix Project, which, comes out of our book. So it's, the penultimate chapter in our book. And the Maturity Matrix Project is start, is, it says that all that real time stuff is really quite advanced.

It's way further advanced than pretty much anybody in the world currently is. What we actually need at the moment is quite simple stuff, like "just turn off machines." Turn off machines when you're not using them. And you can, you don't need real time metrics for that. The other thing that we can do that doesn't require real time metrics is, so one of the things that kind of annoys me about the new AWS Cloud Footprint report, which is, it's fine, it's not a lie, it's just giving you offsets.

It's just telling you what your offsets are. It's totally fine. But people are misusing it and misreading it as thinking it's about carbon emissions. There's another thing that AWS have said, which I really like, which is that, "look, we all do a load of work."

It's called the shared responsibility model. "We will take responsibility for the sustainability of the cloud, if you take responsibility for the sustainability in the cloud. Now" I like it, except that it's also very confusing. What they are saying is that "we will build tools that can be aligned with actual location-based zero carbon operations.

We will build those tools." Those tools in the book, we call them green platforms. They're things like serverless, spot instances, really clever instance types. You see, it kind of aligned with their whole modernization strategy, how you get into the cloud and use it well, as it was intended to be used.

So they're saying that, "but it's your responsibility to use those tools. We can build those tools, but if you don't use them, you won't be green. If you just sit in dedicated instances, you will, there's nothing we can do about it. You will never be green. We'll offset you," which is why, the reports say, "well, we'll pay for your bad behavior.

But it is still bad behavior." You'll get to carbon neutral, but you'll never get to carbon zero. And I think that those messages, which are quite complicated, can both exist, but they need to be quite clearly communicated. And at the moment, I don't think we are being so clearly communicated. What's your thoughts?

Sara Bergman: Yeah, this is also, I think this is also something I thought a lot about reading this article. Like, where is the line? Because many cloud providers, they do have clearly green ambition. They are financially incentivized in many ways to be more energy efficient, use less resources of course, because all of those things cost money for any cloud provider, even if that cloud provider is your local on-prem.

So that is one thing, they wanna make cheaper stuff that are often greener, almost always greener, but it is your responsibility to get on them. And how do you do that? Now, all three of the cloud providers do have architected frameworks that have a dedicated sustainability section.

And there for some scenarios, it's actually incredibly good. Like it's very detailed. So you can go and just like, "Oh, my scenario fits into this." I would like those sections to be longer for all of the big three and for the smaller ones, maybe to include more. But I also now maybe skipping randomly ahead, back and forth, something, a mental image that I think helps because sometimes I think the key message is getting lost.

"Okay. But the cloud is green. Why do I need to take action?" and that is a tricky conversation to have because Yeah, if you're selling the cloud, of course, you don't want to say "no, but it's not green." It's like, it just becomes a bit messy. So a mental model I like to use, which I heard from another wonderful woman here in Norway, which is if you have a car and you have a ski box, because right, we have lots of skis here.

So they can think of the car as the cloud and it's the responsibility of the car manufacturer to make that as efficient and green as possible. And they do, right? And but you, as a user, you choose how do I transport my skis in the car? Do I transport them inside the car? Yeah, maybe they that would be greener right because then there's no additional wind catch of the ski box on top, but maybe they don't fit, so you put them on the ski box on the roof. Fine, you accept that additional wind cost and thus increase the energy. But, once ski season is over, and you don't need the skis anymore, what does eeveryone do? You've remove the ski box. I don't see anyone riding around with the ski box in May, just because, right?

And the same goes for the cloud. So yeah, the cloud has lots of work, a lot of stuff in it to make it greener. But if you choose to use a ski box, that's fine. But once the ski season is over, remove the ski box. So once you're not using your test environment, remove it, shut it down.

Anne Currie: Yeah. Yeah. I like the analogy. It's very Scandinavian.

Sara Bergman: It's very Scandinavian, I'm sorry, but I am very Scandinavian, so that's what you get. I guess the same applies if you have a surfing board, I suppose you put them on the roof as well, 

or a canoe,

Anne Currie: A canoe!

Sara Bergman: yeah, or a kayak, yeah. They have longer seasons though, but, so yeah, to try to translate, I feel like I was very, poor Chris who has to edit this, all my ramblings.

But yeah, as a cloud user, you have a responsibility and I think there could be two reports of showing this is what we as a cloud provider, I think all cloud providers are pretty good at this. This is what we took responsibility for. That's awesome. So much better than it was like 10, 5 years ago.

But then also how do we enable people using the cloud to take green actions? Because they want to take green actions, we want them to take green actions, like how do we enable that at the same time? Yeah.

Anne Currie: Yeah. I mean, it's, kind of like "is the cloud green?" The trouble is the, answer is as always in tech, it depends. Oh yeah. It's like, it depends. Are you using it as it was intended to be used? And if you, and the, one of the reasons why you sometimes see Gartner reports and things saying that the cloud is really poorly used,

it's because there's two stages in moving into the cloud. One, is you just kind of lift and shift. For most people, they lift and shift. Although you might argue that's actually never a really good way to go into the cloud. I've written books about this. It's not, but it's, it is a way, a common way for people to go into the cloud.

But once you've lifted and shifted, you have to then actually use the cloud, as it were, you see, you have to go beyond that. If you stop at lift and shift, it's really ungreen. It's worse than being on prem', because in the cloud, it's so easy to over provision. So you have to actually, you have to adopt the cloud, it's called cloud native and it's not really cloud native, but then you have to adopt the cloud way of doing things because otherwise you will over provision and you will then make more carbon than you did on prem', because it's harder to over provision on prem'.

Sara Bergman: Yeah, And that's a journey. And that's something I think all cloud providers are pretty good at supporting, right? That's their bread and butter in many ways.

Anne Currie: It is, they just want people to do it, but still it's hard because it's really so hard to lift and shift into the cloud that once people get there, they go, "thank God for that." And they don't want to go and look at what happens next, which is kind of one of the issues of lift and shift as opposed to kind of just slicing up and moving bits and bits one by one into the cloud, is that

these big projects, they're so painful that once you've done them, you never want to, I've done so many big projects in the past, and you always think you're going to go on and do the next stage, but you're so destroyed by that project. You just think, "Oh my God, we're all burned out. We don't even want to think about it again."

It's, yeah. Yes.

Sara Bergman: We need to glamorize DevOps or operations. It needs to be something there, more and more afterworks, I don't know. Something to make them more glamorous.

Anne Currie: These big projects are awful. I never want to do one again. But anyway. Yeah. And they're, ungreen because you then go on to the next bit where you start to actually refine and improve. So yeah, it's, we need to step back and think about how we're doing that. So, which is, I think a lot, I was signing books at a conference about kind of, DevOps and, CICD and team topologies and moving faster releases, last week.

And I think that's really green because if you can't move, you can't adopt these better tools and services that remain.

Sara Bergman: Yeah, exactly. Yeah, I also said someone, or heard someone who said like, what we need now is like performance engineering, we need to go back to basics in many senses, right? And because a lot of these learnings are not new. It's the same when people are like, "ooh, if I," now I'm going on a tangent here, but if people want to be greener, they're like, "oh, I should just rewrite, like, make my code more efficient."

And like, by all means do. But if you haven't had like a reason to do that yet, I don't think you're going to convince your management chain that sustainability is going to be the reason why you go implement it because high performance computing is not new at all. Like we've known about all these things for a very long time.

So if you haven't done it yet, I mean, kudos to you if sustainability is the thing that makes you implement it. Kudos to you. I just think it's going to be a tough sell.

Anne Currie: I've mentioned a few times on environment variables, but it always interests me, is on that note of efficient code is expensive and we don't do it anymore, sara now works on Microsoft Exchange and 25 years ago, I also worked on Microsoft Exchange, but in Microsoft Exchange 25 years ago, we had to have everything written in C because the hardware was, it was just not possible to do it using anything other than the most optimized C code, which meant that it was really expensive to do things.

It took a long time. Yeah. The world has changed. we've got better hardware and we use it. We use it to go, to not have to write everything in C.

Sara Bergman: Exactly. And we see the same on the mobile industry, right? They have more apps that do more things now because their phones can handle it. They couldn't when the first smart modes came out and just, again, going back to what we talked about earlier, I don't think we can stop technology progressing in order to be greener.

We need it to progress and be green at the same time.

Anne Currie: Yeah. Yeah. We need it all. 

Sara Bergman: People are going to call us greedy.

Anne Currie: We want it all. So we're going to zip onto our final article link today, which is, 

Is Sustainable Data Storage a Paradox? So it's a piece in TechRadar by Jon Howes about environmental challenges posed by cloud storage and AI, and the rising energy demands of data centers that are associated with AI and storage. So thoughts? Is green data storage possible? Sarah?

Sara Bergman: To quote you, it has to be. No, but I, okay. So there are many things here, but I think also he said that cloud is the least wasteful storage solution because it is highly optimized, blah, blah, blah, blah. So I don't know if the paradox is. Could there be other ways? Or if the paradox is, I don't know, it was a catchy title, and I'm not sure I understood what the paradox was.

But also, what I really kept thinking about, is storage really what we should be concerned about here? I (naively, maybe) would guess that compute would be much more resource intensive than the actual, like, store at rest. Then, of course, training the model, you would need to access the storage.

But again, I would be, in these days, more concerned about inference. And then, again, naively maybe, I would expect the CPU and GPU usage to be what we're worried about, not our disks.

Anne Currie: Yeah, I tend to agree. I would say with data, the opportunity there is there's quite a lot of low hanging fruit. And it's a really, good example of the shared responsibility model as well, that it's up to you, to us as users to not be wasteful in our use of data. And by waste, I don't mean storing more than we need, because I think that telling people to throw the stuff away in the world of AI is just crazy talk.

No one's going to do it. You're going to keep everything, right? We just need to see if the general AI can make use of it. The wastage is having it in a medium that's where it can be accessed more quickly than it ever needs to be accessed. So that the longer, if you say, look, I'm going to access this once a year, you can stick it all on tape and it's practically free in terms of carbon.

You can save as long as you like, there's this, that kind of like never underestimate the bandwidth of a station wagon full of tapes driving down the freeway. Tapes are pretty good. You can put a load of stuff on tapes and most data, especially data for AI, it doesn't matter if it takes you a couple of hours or even a couple of days to get it back.

You can, so you don't need to keep it on sSD, where it's just really much more carbon intensive than tape. So it's just, it's quite easy. You just need to think about it and not store things in the wrong medium. Use the tools that are best suited to the job.

Sara Bergman: yeah, be diligent about hot, warm, and cold storage. Like, what do you need, when, and how much? And yeah, also people say, "should I delete all my photos and emails?" I'm like, "no, you want them, right?" You're going to look at your photos again, you're going to maybe search or read your emails again. So don't remove them.

Maybe you know, unsubscribe to the email you never look at. That's like an option if you always get an email from your build pipeline, but you never look at it. Yeah, maybe don't need to get those emails even. Again, low hanging fruits exists. But yeah, storage mediums are not energy intensive if they're cold.

Anne Currie: Yes. Yeah. Don't keep it hot when it can be cold. Cold is the new hot. It's the new hot.

Sara Bergman: and the new green.

Anne Currie: And the new green, yeah. Right, so yeah, we've talked through all of the things. So is there anything else you want to say, coming out of what we've talked about today?

Sara Bergman: Oh, I don't think so. I think it was an interesting discussion. We are, what I love about this industry is that we're constantly on our learning curve. New things always come up, we have to adapt and adjust, and I get to put on my engineering thinking hat. And I love that. I think, we should be excited about that.

Like we have opportunities to be green, "ooh, I have opportunities to learn, opportunities to explore," not like, "ooh, this dreadful thing I have to go do." It can be fun. I mean, yeah, make it glamorous.

Anne Currie: Well, on that incredibly positive note, I will, sign us off. That's fantastic. So before I sign off today, just a reminder, all the links are in the show notes, so have a read through the articles yourself. And to tell you that this week, it is the Green Software Foundation Global Summit event.

So there are live events in quite a few cities. There's one on the first in London. There are ones in Munich and Hamburg, and there's one in Dublin, there's Berlin and Singapore. And, do feel free, I can't actually make, it's a real shame because I do really love to meet people in person.

And I think flying is one of those things where we, flying is an incredibly valuable thing. I don't tell people not to fly, but for me, the reason to not fly all the way over to London if you're in the US, for example, to go and meet people in person is, that you should set up your own local meetups, which are in person, which people can get to and from.

And because having people locally that you can talk with is incredibly useful. So set up your own summits if you can't make any of the summits that are out there.

Sara Bergman: Or your own meetup, if a summit seems a bit much, then set up your own GSF meetup.

Anne Currie: Yeah, that is an excellent idea. You'll meet a load of lovely, like-minded people, and you can chat away about saving the world, which is always a good thing to chat about. So one of the nice things about the green community is we're all aligned on trying to make positive change without turning everything off and going back to the stage.

So we've come to the end of our podcast and oh, it looks like Chris, our editor, has given me a final fun question for you, which is, if you were to design an AI that had zero environmental mental impact for a totally non-serious purpose, what would it do?

Sara Bergman: I'm embarrassed to say that I thought so much about this, because my mind went, ooh, what would I do? And then I'm like, would I optimize my life? Would I make something silly? Would I make something useful? And then I just, there were so many options that I don't know. But some of the things that like, popped up was like, organize all my photos.

I have a small child. I take a lot of photos of him. I would like for there to be some nice organization going on there. That would be very nice. Also, an AI that designed cute nails that I could do at home with like the stuff I have. Because I am not artsy, but I like fun nails. So yeah, like, like, if I am bad at doing my nails and I have four colors, what can I do that's cute?

Anne Currie: And of course, all of those things could be done zero impact because the, key thing about them is that they are not latency insensitive. You could say, well, actually, I'll wait to run my models and do my inference till the sun's shining and the wind's blowing and there's excess electricity on the grid.

The, I don't, I would, to make it so it's there, I wouldn't say, well, it's in the middle of the night and it's still night, but I'm really desperate to get my fancy nails done.

Sara Bergman: Yeah. 

Anne Currie: Just wait. Just say, well, I'll, find out tomorrow morning what the AI is storming. It's all about making things less on demand, so we align with when the sun's shining and the wind's blowing. 

Sara Bergman: Thanks for taking me back to technical.

Anne Currie: So thank you very much for being on. It's been, as always, delightful to talk to you.

Sara Bergman: Likewise. I had a blast. Thank you so much.

Anne Currie: It's nice. And we don't say, obviously, while we're writing the book, we talked together all the time, but now that the book's finished, then we talk together less often. So for all listeners today, links are in the show notes. And if you haven't read Building Green Software from O'Reilly, you are missing a trick because that really, it's a fun read.

Everybody seems to be enjoying it.

Sara Bergman: Yeah. I think the top thing people are like, "Oh, it's fun." I'm like, "Yeah, it's funny. It's good."

Anne Currie: So if you haven't read it, crazy! Go out and read it, or join one of my training courses, which are also fun. So thank you very much. And I'm sure we will both be back on the Environment Variables sometime in the not-too-distant future. So goodbye from me. 

Sara Bergman: Bye. 

Anne Currie: Cheerio.

Chris Adams: Hey everyone, thanks for listening! Just a reminder to follow Environment Variables on Apple Podcasts, Spotify, Google Podcasts, or wherever you get your podcasts. And please, do leave a rating and review if you like what we're doing. It helps other people discover the show, and of course, we'd love to have more listeners.

To find out more about the Green Software Foundation, please visit greensoftware.foundation. That's greensoftware.foundation in any browser. Thanks again and see you in the next episode!

Host Chris Adams sits down with CEO Olivier Corradi and Tech Lead Íngrid Munné Collado of Electricity Maps, a company that leverages data to enable decarbonization of electricity grids. They discuss the complexities of carbon intensity data, the role of accurate forecasting in renewable energy, and how this data helps optimize electricity usage for sustainability. Olivier explains the origins of Electricity Maps, their goal of providing real-time carbon footprint insights, and their theory of change—targeting citizens, corporations, and institutions to create a greener future. Tune in to hear about the history, frontiers, and future of this engaging field.

Learn more about our people:

• Chris Adams: LinkedIn | GitHub | Website
• Olivier Corradi: LinkedIn | Website | X
• Íngrid Munné Collado: LinkedIn | Website | X

Find out more about the GSF:

• The Green Software Foundation Website 
• Sign up to the Green Software Foundation Newsletter

News:

• Our Road to Impact: How we contribute to fixing climate change [04:56]
• [1812.06679] Real-Time Carbon Accounting Method for the European Electricity Markets [20:24]
• SCI Specification Achieves ISO Standard Status | GSF [21:24]
• Electricity Maps | Client Story: Monta [29:52]
• How to save costs and emissions with a flexible electricity load? [33:32]
• How the prices can change by location | X [34:28]
• Another example in Texas | X 
• Marginal vs average: which one to use for real-time decisions? | Electricity Maps [39:21]
• eCO₂grid | 50Hertz [41:58]
• Investigating Apple's Clean Grid Forecast 

Resources:

• Energy, Business, Climate & Geopolitics | Commons For Future [09:58]
• Commons For Future [10:02]
• The Week in Green Software: Mapping Green Software on the Grid [10:32]
• Open Source | Electricity Maps [10:53]
• How to trace back the origin of electricity [15:28]
• GitHub - electricitymaps/electricitymaps-contrib: A real-time visualisation of the CO2 emissions of electricity consumption [19:26]
• Online Browsing Platform (OBP) [21:33]
• Finland extends nuclear reactor outage, sees power prices soar | Reuters [35:15]
• Breaking borders: The future of Europe’s electricity is in interconnectors | Ember [35:49]
• Carlos Pérez Linkenheil on LinkedIn: #energymarket #electricityprices #epexspot #dynamictariff | 23 comments
• Increasing renewables without regulation or curtailment mechanisms - Lion Hirth | LinkedIn 
• Giant Batteries Are Transforming the Way the U.S. Uses Electricity 
• Solar will get too cheap to connect to the power grid. [35:57] 
• The reasons for negative prices - by Julien Jomaux [37:41]
• Making Testbeds for Carbon Aware Computing [38:04} 
• Avoid restricting the SCI by prescribing a specific metric. [38:33]
• MARGINAL EMISSIONS RATE – A PRIMER [40:00]
• On the Implications of Choosing Average versus Marginal Carbon Intensity Signals on Carbon-aware Optimizations | Proceedings of the 15th ACM International Conference on Future and Sustainable Energy Systems [45:19]
• On the Implications of Choosing Average versus Marginal Carbon Intensity Signals on Carbon-aware Optimizations
• On the Limitations of Carbon-Aware Temporal and Spatial Workload Shifting in the Cloud | Proceedings of the Nineteenth European Conference on Computer Systems [45:41]
• [2306.06502] On the Limitations of Carbon-Aware Temporal and Spatial Workload Shifting in the Cloud
• Tammy Sukprasert | Move Your Workloads To Sweden! | #53 - Disseminate | Acast
• Methodology + Validation - WattTime
• VERACI-T      

If you enjoyed this episode then please either:

• Follow, rate, and review on Apple Podcasts
• Follow and rate on Spotify
• Watch our videos on The Green Software Foundation YouTube Channel!
• Connect with us on Twitter, Github and LinkedIn!

TRANSCRIPT BELOW:

Olivier Corradi: The dream is to get to street-level granularity, and this is why flow tracing, by the way, is so important, because as you increase the resolution of what you're looking at, then obviously if I'm looking at just a city, most of the electricity is actually produced outside of the city, and that's also why, by the way, the whole forecasting challenge has just become massive. Try to forecast how clean the electricity is going to be at every location on the planet. Obviously, you need really good renewable forecasts as well. And that's what civilization needs if it wants to rely significantly on these renewables that are intermittent.

Chris Adams: Hello, and welcome to Environment Variables, brought to you by the Green Software Foundation. In each episode, we discuss the latest news and events surrounding green software. On our show, you can expect candid conversations with top experts in their field who have a passion for how to reduce the greenhouse gas emissions of software.

I'm your host, Chris Adams. Hello, and welcome to Environment Variables, where we bring you the latest news and updates from the world of sustainable software development. I'm your host, Chris Adams. We often talk about carbon intensity, or how clean electricity is on this podcast, as one of the levers for making software more sustainable.

To know this, you need to get the data from somewhere, in a form that's easy to understand and consume. So today we're sitting down with two people who live and breathe this data from Electricity Maps. Electricity Maps is a company at the forefront of enabling a data-driven decarbonization of electricity and today we're joined by both Íngrid and Olivier from the firm.

Hey folks, how's it going?

Íngrid Munné Collado: Hi.

Olivier Corradi: Hi, Chris.

Chris Adams: Cool. Folks, we'll talk about the origins of Electricity Maps in a second, but before I do, I just want to give you a bit of space to introduce the two of you, actually. We normally go by surname first, so I think Olivier Corradi, you're ahead of Íngrid, so can I give you some space to introduce yourself first, and then we'll do the same thing for Íngrid, alright?

Olivier Corradi: Of course, and thank you so much for the invitation and for inviting us to the show. So, my name is Olivier, I'm originally a machine learning engineer and data scientist. My background has been academic, I've worked as well in the research industry with IBM Research and Simulating Electricity Grids. And I would say the most relevant thing that I'm doing now is Electricity Maps that I founded eight years ago.

Chris Adams: Okay, thanks for that, Olivier. And Íngrid, over to you, I suppose.

Íngrid Munné Collado: Hi. Hi, Chris. I'm Íngrid. I'm the Tech Lead at Electricity Maps and I'm very happy to be here. More specifically, I'm part of the grid forecast team where we built, as we have said, forecast models for renewable energy sources and other signals such as carbon intensity. I've been recently involved in replacing and improving the weather forecast data that we use at Electricity Maps to achieve better forecasts.

And yeah, outside of work, when I'm not working, I'm either going to CrossFit where I'm trying to learn how to do a pull up, which I don't know, or knitting and just making sweaters. So quite, quite contrast.

Chris Adams: Cool, thank you for telling me. I'm glad you mentioned weather, actually. I mean, as an English person, we talk about weather the same way that we breathe, I suppose, and as I understand it, you folks are both calling from Denmark, right? We're a little bit chilly today than it otherwise would be, right?

Íngrid Munné Collado: Yeah, it is quite cold for September. I think it's quite usual, but it would be a bit on the colder side.

Chris Adams: Yeah, same here. We, it's, Berlin has, someone has switched off the summer, so we now had to, like, I've cycled in with my coat for the first time today, and did not enjoy that. Okay, so I should introduce myself, actually, folks, if you're new to this podcast. So my name is Chris. I work at the Green Web Foundation as the executive director there.

I also work in the Green Software Foundation as one of the co chairs of the policy working group where we Basically, work on policy to see about coming laws and things that we might actually want to respond to or help members understand what the implications of might actually be. The other thing I'll share with you is that we try our best to have quite helpful show notes for this.

So we're going to mention various projects and papers along, and if you are viewing this in Spotify or YouTube podcasts or something, you might not see them. So be sure to look up podcast. greensoftware.foundation to see the full links and the transcript. Okay, you two folks sitting comfortably?

Íngrid Munné Collado: Yeah.

Olivier Corradi: We are, sure.

Chris Adams: Brilliant, okay, then I'll begin.

Okay, first question I'll put to you, Olivier, because I've actually been a fan of Electricity Maps for quite a while, and when I first heard about Electricity Maps back in 2017, it wasn't Electricity Maps, it was electricitymap.org. And this is something that is reserved for non profit organizations. So it's 2024 now.

And I now know that the same kind of cool map that I started talking about at conferences and so on, is now available under electricitymaps.com. And as someone working in a non profit, I end up talking and thinking about theories of change all the time. So there was a really interesting post I saw on the blog, on the Electricity Maps blog post, particularly talking about this theory of inaction.

And it's rare to see start ups talk about this kind of stuff with a fluency, really. So, Olivier, can you tell me a little bit about basically, what this is, and how this informs you spending all the time on Electricity Maps, because I know there's a few other things that you've worked on before.

Olivier Corradi: Yeah, I'm happy that you actually give me the opportunity to articulate this. So in the early days when Electricity Map was started in 2016, there was this question that remained for a couple of years afterwards, which was, "should we actually be an NGO or should it be a for profit company?" And honestly, the conclusion is that it depends on how you want to impact the world and it depends on how you want to be funded and so on.

But the way that we've been looking at things is basically to say when a company will pay you for something you're delivering, that represents a change that is happening in the world that is sufficiently valuable that someone will pay you to do it. And people are taking you seriously when they're implementing it.

So this was a little bit what tipped it over towards more of the, let's say, for profit world. But in order to make sure that we never lost track of the actual impact we want to have, we started having like this framework that we stole from someone else, honestly, and adapted afterwards. But first of all the vision we have here is to imagine a world where we have low carbon electricity that's delivered everywhere across the world, every hour of the year.

And this was really why Electricity Maps was created, out of almost a frustration of folks looking at the electricity grid from a yearly perspective, instead of looking at it hour by hour. And of course it made sense before we had renewables. Now that we, the wind doesn't always blow and the sun doesn't always shine, we need that.

And the angle we're taking here is data-driven decarbonization, as you mentioned, because we're a mixture of folks who come from academia, who know how to build machine learning systems and tech. So this is our angle. So with that vision in mind, providing clean electricity every hour of the day, we realized there's three pressure points that we need to apply if we want to create that transformation.

The first one is, we identify them as citizens. Second one, corporations. And the third one, institutions. And there is this framework that, that depicts a triangle of inaction where each of these corners are pointing at the other as an excuse to not change. For example, citizens are saying, "well I actually would love to be greener, but the government is just not putting the right incentives for me to be cleaner," like the public infrastructure is not helping. Another piece of it is the citizens are pointing at corporations and saying, "well, I, too bad, like, flying is actually cheaper than taking the train. Like, I don't have a clean offering here."

And then the corporations on the other side will say, "well, but everyone wants to fly, right? I mean, the citizens are not ready for that change." They'll say, "well, I'll just keep operating like I am." Corporations will point at institutions as well and say, "well, the right tax schemes are not here for us to make flying cheaper, actually, so we'll just continue what we're doing."

And then institutions will be, "we don't have" like the, let's say, "social acceptance from the citizens to actually, exactly, to put these green tariffs. And on the other hand we're just a small organization and institution, sorry," and you have corporations like Microsoft, OpenAI, Google, that are just like, more powerful than governments.

And so you can basically map these out in a small triangle. And we said, okay, so how are we going to impact this? And then most of our initiatives can actually be mapped to this triangle, where on the citizen side, the app that we have of Electricity Maps is creating this awareness, creating the debates as well to make sure we have a factual understanding and can challenge our politicians in the right way.

There's the piece on corporations where we are working with them in order to ensure that green offerings can actually hit the ground, like that we have electric vehicles that can use electricity at the optimal time, all these things. And it's a win because we get a financial cut of this and we can grow our company and grow the impact.

And then finally, on the institutional side of things, we're trying to make sure that the right carbon accounting methodologies and so on are being pushed. We basically want to live in a world where we have an accounting that represents what's physically happening in the world. And one example is we have a data portal where our historical data can be used by any company who wants to do granular carbon accounting.

Again, moving away from this world where you're doing it on a yearly basis to an hourly basis. So this is the framework we've put in place and the way we're articulating internally as well how we allocate resources and prioritize and make sure that what we do leads to impact ultimately.

Chris Adams: Cool. Thank you for that. I have one question following up from this, because when I was doing a bit of research, I looked up this triangle of inaction, and I think I might need one of your help, one of your help in pronouncing the person's name. Is Pierre, is it Peyretou?

Olivier Corradi: Not too bad. Yeah. Well done.

Chris Adams: Okay, so for people who are listening, what we'll do is we'll share a link to some of the theory behind it.

He's actually got an online course. I think there is, there's a French school of business which does actually have some online resources to understand this kind of theory. So if you're Curious and you want to think about, okay, how does this work? Or if you'd like to see an actual triangle rather than have us describe it, follow the link and you'll see some of it there.

Okay, thanks for that, Olivier. Okay, so we've spoken a little bit about data, and you mentioned about data being in the open, Olivier, and I can actually confirm that when we spoke to Toby before, Toby is one, sorry, Tony is one of the colleagues at Literacy Maps. We did an interview with him around about November last year, actually, and he was talking all about this, Open Data Portal.

And back then we were like, "oh, sweet, there's all this data being published." And I'll be honest, you're a startup. I was like, "are they really going to publish the next year?" And then January came around and it actually did get published. I was like, "wow, this almost never happens." So I was really pleased to actually see a startup and a company follow through with making some of this data available because it's so, so, so useful and this is so hard to find otherwise.

I can speak as someone in a non profit who's been trying to find this data. It's so, so handy to have some of that available now. So yeah, thank you on that one. Okay, so the question, I'll leave some space because I think there was a response coming up there actually, Olivier.

Olivier Corradi: Oh, thank you so much for the shout out. You know, it's important for us as well to feel like what we're doing leads to direct impact and that we have folks that are demanding this data. And sometimes these feedback loops are not always present. So I appreciate you giving us the shout out.

Chris Adams: Yeah, so if anyone is listening, this is open data, so it's something that you're able to build on and we'll share a link to the actual website that makes it very clear how you can use this information. So if you're trying to build something and you're looking at historical stuff, it's totally there.

Have at it. Okay, so we spoke a little bit about carbon intensity at different parts of the world and how you need to think about it on an hourly basis rather than just an annual basis. And we might talk about why when you talk about green energy saying green energy with certificates that came from solar and then saying, using that to make usage at night be counted as green might be conceptually a little bit challenging, we might say.

So that's some of the stuff we spoke about. Now, the thing I want to ask a little bit about, and Íngrid, if I can hand this over to you. When you're working as an engineer and you start thinking about carbon intensity. It's when you use, say, APIs, you just see a single number that goes up and down.

And like, it's fun. And it's very easy to underestimate just how much work can go into this and all the complexities around this. Because as I understand it, it's more than just like, looking at what the power will look at looking at generation from a single plant. Íngrid, can I just give you a chance to actually talk about what goes into sharing some of these current and historical figures? Because I know there's some forecasting work, and we'll talk about that later, but we've just spoken about open data and some of the historical stuff, and some of the context there might be useful for people who are considering downloading some of this or messing around with it themselves.

Íngrid Munné Collado: Yeah, absolutely. And I would say that without this amazing work of collecting data, any forecast, like forecast would not be possible at Electricity Maps because we need this data to be able to produce our forecast. So let's take a look at how we collect data. This is currently done by one of our teams at Electricity Maps, the Grid Modeling and Methodology team.

And our work starts by first trying to get as much real-time data and historical data from, that is available publicly. And here we're talking about governmental institutions, transmission system operators, that for people without electrical engineering background, those are the folks that manage the electricity grid at high voltage, and that make sure that demand and production is actually matches at every single second.

So we try to collect this data and now we have data for 228 zones. And when we think about it, we might say, okay, we just collect CSVs and everything is standardized and nice, but the data is really messy. So we might get data from like in PDF files, in TXT files. We have sometimes, I can tell you a bit of an experience.

When I joined Electricity Maps, I had to build the parsers for Japan. And I'm originally from Spain, so I know a lot about electrical system in, in Spain, but when I had to dig into the Japanese electrical system, I didn't know that they have eight different organizations that collect data and each and single one of them is in a different format.

And sometimes we even had to do some image recognition to get the power of the nuclear power plants that they have in Japan. So data is very messy. It contains outliers. We have missing values, wrong values. Some production modes might be missing as well, so there's a lot of work that we have to do behind the scenes to make sure that we collect this data every hour or even in lower resolutions to make sure that we can use this data and process it.

So once we have this data in the raw format, then we process it to make sure that we don't have these outliers or missing values. And that we have a complete power breakdown. That means that if Spain, for example, has nuclear, gas, coal, and like in total eight production modes, that the data we get has eight production modes because otherwise the carbon intensity values that we might show on the app or on the API, they are not going to make any sense.

So once we have this done, then we can actually go into maybe one of the core pipelines we have at Electricity Maps. That is the flow tracing pipeline. And Tony did an amazing job in the previous episode where he explained how this works. But for those who don't know, the flow tracing pipeline makes sure that we are able to trace back the origin of electricity.

Because if I plug my laptop here in Denmark, the electricity that I'm consuming is not only the one that is being produced in Denmark, but that one that is being produced in Denmark all the exchanges that happen between Denmark and the neighboring countries. So, we have this pipeline that makes sure that we can take into account all the neighboring countries and what happens at every single hour of the day to make sure that we can know the exact mix at a given hour of the day.

That's actually when we get the origin of electricity, the power breakdown, and then we can translate this number into carbon intensity by using emission factors that we update recurrently and that this actually helps us understand one, what's the carbon footprint of one kilowatt hour that we consume at a given country.

So emission factors are different based on the source we use and based on the country we are at. And that's what you see on the map.

Chris Adams: Wow, okay, so that's, let me just run through some of that then. Just to make sure I'm understanding it. So you're essentially getting a bunch of data. You're having to do, clean some of it up. Like, that's quite common. And most of us might be used to like, working with maybe text or CSVs. But I think you said that you're essentially like doing OCR, like optical character recognition in some on like, gIFs or pictures rather than actually having to read an actual number. Okay, and then once you've got that, you've got an idea of what the generation might be, but then you then need to do a bit of, like, working out where data is, where energy is being traded across borders, essentially, because in some ways the grid does have all this stuff, you just can't look at the production, for example. Because, like, I mean, the UK has, like, new connectors to other countries all the time, and Germany uses loads and loads of France's nuclear power, for example.

So there's all the stuff like that, and then once you've got this idea that, okay, there's probably this much coming from these places, you then need to think about, okay, well, what is the carbon intensity of power from a coal fired power station, or a new coal fired power station versus an old one, and stuff like that.

So there's all that other depth as well. Okay, and that all goes into a single number.

Íngrid Munné Collado: Exactly.

Chris Adams: Alright, I am kind of, so, I see why people do this now, because when I've looked at this before, I've looked at numbers saying, "oh yeah, it's just like generation," but no, there is, generation and production are two totally different things, and you do need to take into account some of this if you want a meaningful number.

Íngrid Munné Collado: And if I can say something just to give you a rough number, a country like France, it has interconnectors like with eight different countries. So imagine if you only consider the production or generation in France without considering all the interconnectors around, you might get a completely different picture.

Chris Adams: Ah, okay, alright, thanks for providing extra context because, yeah, I can see how complicated this gets very quickly in that case. Alright, okay, one thing that you mentioned before was that there are all these complicated scrapers, and when I looked at Electricity Maps a few years ago, I was surprised by there's quite a lot of it which is open source, so like, don't believe me?

Look at the repo, for example. There's some of this out there so you can see just how messy the data might actually be or possibly contribute if there's a gap, right? 

Íngrid Munné Collado: Yeah, actually, for example, when I had to work in the Japanese parsers, I got help from people in Japan. That they actually would point me to the right CSV file. Oh, and at some point the CSV changed the URL where it was stored. And it was not until someone, a contributor, helped me out and said, "Hey, check this link instead."

So we have, so all our parsers are open source and they are hosted on GitHub, on the contrib repo. And we are extremely happy about all the contributors that help us out, like be able to get this 228 zones that we have now, on the map.

Chris Adams: Okay, go yeah, Olivier, I was going to ask if you anything you wanted to add, because I believe you did a bit work on some of the academic underpinnings for some of this.

Olivier Corradi: I just wanted to add as well that if we tie it back a little bit to the triangle of inaction, what is happening here and why this is so exciting is that we're basically enabling some of the citizens to also act on climate change by being able to contribute with their unique expertise, which helps us.

There's no way Electricity Maps with, we're a bit more than 20 people now based in Copenhagen. There's no way we would have been able to cover the world if not for the help of all these wonderful contributors that have helped us all along, and it gives them also a way to contribute something meaningful where maybe before they didn't have that opportunity.

Chris Adams: Cool, and Olivier, we spoke a little bit about flow tracing, and I understand this is the you published a paper about this years and years ago, and that was one of the things to basically, the methodology that you're talking about, yes, there's some proprietary code, but the general approach that's being taken, it's in the public for people to understand, so they can challenge it and interrogate it and say, well, yeah, this is, I disagree with this thing and this is why I think this might need to be changed in the future, right?

Olivier Corradi: Absolutely and our philosophy has always been if there's something where we think we can move faster with the help of others, then let's open it up because then we can allow for contributions and so on. If it's something where we know that if we open it up, it's going to take us a lot of efforts to handle the contributions, then we don't open it up, and that's why some of the internal pipelines that Íngrid described are proprietary, because it's just faster for us to change a couple of things and not worry about what will happen if we open it up. But in general, we always try to be open, because we are trying to create a global consensus on how we account for things, so that can only be achieved through openness.

Chris Adams: Okay, you're singing from the same hymn sheet as us, like, we, the Green Software Foundation has a big thing about Open, the software carbon intensity metric is Open. If you want to pay to download it, you can buy it from the ISO for 63 Swiss francs. I don't know why you would, but that's available for people, and me, working at the Green Web Foundation, we use open as a lever, so we publish almost everything we can, either under open source or under open licenses. So, Íngrid, if I may, can I just come back to you, because we spoke a little bit about historical stuff, and you mentioned that forecasting is now a big thing, because the thing about renewable energy that we kind of alluded to before is that it changes over time. The sun goes up, sometimes the wind blows, storms move around, stuff like that.

So, can you just tell me a little bit about why this idea of forecasting is maybe more of a focus for you folks now and maybe explain a little bit about how, I guess, the sausage gets made and how that can be difficult.

Íngrid Munné Collado: Yeah, of course. So, as Olivier mentioned in the beginning, we want our data to be as actionable as possible. And now imagine that we are software developers and we run cloud jobs. And at the same time, we want to be aware of our carbon footprint. And we know that our cloud jobs might last five, six hours.

So, real-time data and historical data does not really help us achieve, like, be aware of what's the carbon footprint of our cloud job in the future, like, when is the right time to schedule this? And EV chargers might also think the same, like, when is the right time to charge? So, we saw that real-time and historical data would not help us achieve that.

Then we realized, okay, then we need to provide them with forecasts, and we know that we are providing a global API with data from all over the world for all bidding zones. Then, if we, let's go back to this software developer who wants to schedule a cloud job, because I think that's going to help the audience understand.

I can choose a data center in Sweden, but I can also choose a data center in Texas. So we can, we have the power of choosing which data center we want to run our cloud job. But at the same time, how do I know which one is the best? I don't have any idea. And at the same time, okay, but is it better to run my cloud job at 9 in the morning or 9 in the evening?

And then that's when we realized, okay, forecasts can actually solve that, but we need to provide it globally for all the zones in the world. So that's why we decided to focus on forecasting carbon intensity for the next 24 hours. That means, like, day-ahead. So we run the pipeline every hour and we provide forecast for the next 24 hours.

But then the next challenge came up. When we realized, "wait a second, is carbon intensity a metric that everybody can understand?" If I tell you that your cloud job used, I don't know, 200 grams of CO2, you don't, it's very difficult to relate. But if I tell you that if you schedule your cloud job at 9 am,

you're gonna run on 90 percent of renewable energy share. Instead of running 9 pm and then the renewable energy share is going to be 10%, it's very easy to understand and say, "oh, you know what, I'm going to do it at 9 in the morning because the renewable energy share is 90%." That's when we realized, okay, we can't just focus on carbon intensity forecasts, we need to do that with renewable energy forecasts and more specifically, wind and solar power forecasts.

Because that increases user engagement and the actionability of our forecasts.

Chris Adams: Ah, okay, that's quite a subtle change then, so, and I understand that because intuitively it's something that I just become so, kind of, you take it for granted, right? If someone is coming to this new, yes, explaining carbon intensity is conceptually quite complicated, but how much is running on clean energy is quite a bit easier to understand very quickly.

Íngrid Munné Collado: Exactly. if I can add on that, then when we said, let's start on renewable energy forecasts, it's a problem that has been not solved maybe because research evolves very fast, but it's a problem that has been laying around for years. I've worked on that field for many years. Because energy is traded on this day-ahead market, so people need this forecast to be able to trade energy in day-ahead markets or intra-day markets.

So when I started working at Electricity Maps on that field, I came from my previous job, and I came in, completely biased and thinking that this challenge would be easy to solve, because in my previous job, we had to do renewable energy forecast of specific assets. And then I would have the perfect setup for a machine learning engineer.

That means I knew the location of the wind turbines. I knew all the data regarding the wind turbines, like the blades, the installed capacity, if it was under maintenance or not. I also had access to multiple weather forecast data and I only had to build models for Denmark or the UK or Netherlands, so it was very focused, but when I joined Electricity Maps, the problem is completely different.

Here we focus on building a renewable energy forecast at country level or, bidding zone level without knowing the location of the assets, without knowing what's installed capacity, if the turbines are under maintenance or not, and that makes the problem very difficult because we know that renewable energy forecast is very linked to weather.

And if you don't know the exact location, then that's another challenge. So, I think in the grid forecasting, we did a really good job on finding a solution that generalizes well enough, and that allows us to generate this forecast for wind and solar for all zones in the world.

Chris Adams: Ah, okay, thanks for providing that extra context. So, I'm just going to check if I understand some of the terminology you used there. You said like a day-ahead market and an intraday market. So, so basically, as I understand it, if you maybe run a wind farm, for example, the day-ahead, you're going to say, "well, I reckon we can sell this many kilowatt hours or megawatt hours of power tomorrow," and that's what you'll make a bid in, and that's, there's consequences for either underbidding or overbidding for that kind of number, so that's why you'd care.

And the intraday is a bit like the kind of shorter term thing, so, you might make one big bit but then you might say there's a little bit of flexibility or you say well okay I, okay cool. And you mentioned this term bidding zone. Now bidding zone is a little bit like a country but it's not always a country.

So like America has all these different bidding zones because it has different grids and there's, that's the kind of, when you folks have mentioned the word zone that's kind of what you're referring to. It's not quite a country but it's more related to like, is there a kind of a unit of carbon intensity for a particular grid region, right?

Íngrid Munné Collado: Exactly. I mean, if we focus in Spain, for example, Spain, Portugal, France, Germany, a zone is the same as a country. But if we look at Sweden, it's split into four zones, into four bidding zones, and Norway into three. So if you check out our data, then you can compare between zone and country.

Chris Adams: Okay, and within a given country you can have radically different carbon intensity, and we might talk a little bit about some of that a little bit later then. Okay, cool.

Olivier Corradi: The dream is to get to street level granularity. And this is why flow tracing, by the way, is so important, because as you increase the resolution of what you're looking at, then obviously, if I'm looking at just a city, most of the electricity is actually produced outside of the city. And that's also why, by the way, the whole forecasting challenge just becomes massive.

Try to forecast how clean the electricity is going to be at every location on the planet. Obviously, you need really good renewable forecasts as well, and that's what civilization needs if it wants to rely significantly on these renewables that are intermittent.

Chris Adams: Wow, okay, I didn't realize street level was the dream for this, blimey. Okay, so we spoke a little bit about software and cloud jobs and stuff like that, but it's also worth just briefly touching, like, this is used outside of the cloud world. And I think one of you mentioned EVs as one of the examples here.

Could we just briefly touch on that? Because that might be one of the things which is concrete that lots of people might experience or might know someone who might have something which is, like, impacting them. Because yeah, EVs are becoming more popular now and it's probably one of the biggest new large uses of electricity in most people, in houses for example.

So yeah, Íngrid, maybe you could talk a little bit about that and then we can move on to some of the other questions.

Íngrid Munné Collado: Yeah, of course. So one of our clients is Monta. And we have a very great success story with them. Monta, for those who don't know them, is a global operating platform for EV. And one of the solutions is the smart charging feature, where they offer users to shift their charging according to the carbon intensity of the grid, the share of renewables, and that's by using our forecasts, or the price of the electricity grid.

And users are completely empowered to choose and to prioritize which signal they want to follow, if it's mostly price or low carbon or high share of renewables, and by doing that, those users took action in decarbonizing the grid. And the results are quite impressive because 70, 000 charging sessions were optimized for low carbon or high renewable share, and there was a 48 percent growth in user engagement, according to Monta, and that means that they optimized for low carbon charging, and in this process, 200 grams of CO2 were avoided for each charge on average.

So these are the numbers and this is the feedback that we're always willing to get because when we create, when we produce this forecast and we spend so much time building these forecast models is because we really want to know that people use our data and that they really use our data for this use cases.

So we were really happy to know that.

Chris Adams: Ah, thanks. Okay, so you mentioned one thing that I think was quite interesting. You mentioned, like, the cost being a thing that might change at different times of day and, broadly speaking, this is because, as I understand it, when energy is really when there's lots of green energy, renewable energy on the grid, it will be relatively cheap and somewhat green.

So, in the UK, for example, we have something like this Octopus Energy. I'm not in the UK, I'm in Germany, but I grew up in the UK. So, Octopus Energy is one company that's been doing a bunch of stuff like this about having agile and intelligent tariffs. So there's essentially a financial reason as well as a kind of basically an environmental reason for doing some of this.

And I think what I've heard in the UK, for example, I believe on an interview recently I heard just by doing some of this, essentially, when controlling some EVs, for example, I think the figure was something like 1.2 gigawatts of demand they had control over. Now that's basically the size of a nuclear power station in many cases, so that's like a significant amount of flexibility on the grid that would otherwise have to come from burning loads of fossil fuels.

Right? Okay, so maybe I can allow us to talk a little bit more about the cost thing because I know that when we talk about this, when we talk about green software, you can make an argument that yeah, you should do it because it's good for the planet, but there's also a real cost fact which comes into this that I think is actually growing and maybe this is something that I understand that you folks have been looking into as well as one of the ways to address more this triangle of inaction and align some of the incentives for more kind of grid complementary activity perhaps.

Íngrid Munné Collado: Yeah. So one of the issues we face is that we know that we want, we are implementing climate action and we want people to use our data for, in order to decarbonize the grid. But sometimes the entry barrier can be quite high when we just go there with, "hey, you have to save CO2." But sometimes it might be easier for users to understand that, "hey, if you look at the electricity prices, that you have also seen that they are not constant, you might be able to save money and CO2."

And that's very, that's a very good entry point as we mentioned in the Triangle of Inaction to increase user awareness because the main, for people it's very easy to understand price and how much they are going to pay at the end of the month and if they are going to save money and therefore it's also nice for us to get to them by I'm Price and then explain that, "hey, by doing that, you might also have an impact on CO2," but we have also a blog post where we mentioned that just optimizing for price doesn't mean that you're also saving CO2.

It's you need to implement some what we call co-optimization, where you want to either prioritize one or the other. And at the same time, maybe you might be able to save more CO2 if you optimize for higher renewable share, and at the same time, you can also save money by doing that. So that's why we think that price is a problem worth solving and that we also need to explore that area. 

Chris Adams: Ah, I see. And if I understand it, and if we follow the kind of path from before, we spoke about, like, a lot of us might be used to just paying a single price for power, but the price might change depending on where you are geographically as well, and that has some implications too, like, this is one thing that I guess, this is probably the newer world we're moving into, in that, yes, there might be, like, night time tariffs, but it feels like there's a lot more dynamism, both temporally and spatially for some of this.

Íngrid Munné Collado: Absolutely. And we saw that when the war in Ukraine started, that as soon as we saw that there were less gas resources, prices skyrocketed. And for example, in Spain, where there was a price cap

The situation was solved also because Spain has other sources of getting natural gas. But countries like Denmark, Germany, they really struggled with that. And the prices really doubled for some months. And it's not only political conflicts. It's also availability of the power plants. I it was two weeks ago where Finland had some unexpected maintenance in one of the, of their, of the largest nuclear power plants.

And that cost their head prices to double. And also in Texas, we have seen extreme weather events that causes prices to change from

Chris Adams: More than double,

Íngrid Munné Collado: yeah, more than double from negative prices at some point because there, there's one area in Texas that has a lot of renewables and another area in Texas that doesn't rely on renewables and due to those extreme weather events, they have to turn off some of the power plants and they, then this affects the prices.

And something I want to mention, I'm an electrical engineering at heart. So I want to talk about the power of interconnectors, and we might think that just by increasing renewables, this is going to make prices to be cheaper, but the problem is not going to be solved, and it's not going to make prices cheaper if we don't have nice interconnectors that make these flows between countries possible.

That's why prices are just changing so much because there's so much happening and going on right now, both on installing new energy sources, weather events that we're seeing, unfortunately, due to climate change and also the lack of interconnections at the moment.

Chris Adams: Okay, so you mentioned two things that I think are interesting there. So one thing was this idea of, okay, the price can change massively. Like the Texas example, I think if we look at, like, the cost of power, right, it might be between 20 to maybe 60 US dollars per megawatt hour in Texas. For example, you're one, you mentioned, like, there's a bunch of wind in the panhandle in the kind of northwest, and then there's around Houston, there's loads and loads of demand where people use all the power, right?

And I think the figures I've heard were something in the region of negative two and a half thousand dollars at one point, and then 30 minutes drive away, the cost is three and a half thousand positive. So like, you got a almost four, five, thousand dollar swing in the pricing here. And one of the reasons is just because it's a bit like network connectivity.

You know, the pipe isn't big enough. And this is one thing that we have to kind of work around, essentially, thinking about this.

Íngrid Munné Collado: And the same issue is happening in Germany, and like, Germany doesn't have these price mechanisms that we have in the US, but in Germany, there are also, like, huge problems due to transmission capacity, because the, most of the solar power production is, takes place in one part of the country, while the demand is concentrated in the completely other opposite.

And they are now experiencing with curtailments and extreme negative prices because they don't have this transmission capacity. They also don't have storage capacity, and there are no market mechanisms that control how we increase this solar power production in the grid and how we just export this to other areas.

Chris Adams: Ah, okay. All right, thanks for the extra context. For people who are still with us, we'll share a link to the previous episode, because we spoke with chap Philipp Wiesner, who was building a bunch of this work to simulate these kind of grids to give an idea of what the pricing might actually be with different services.

And if you have data centers, how adding some storage might actually change the cost and the carbon of running various software services. So we'll add a link for that. Olivier, while you're here, I want to ask one question, if I may, about basically, optimising for carbon intensity, because that's what lots of engineers are kind of trained to do somewhat.

And I, inside the Green Software Foundation, there is a kind of standard called Software Carbon Intensity. It's the thing that you're supposed to, or that you might optimise for, essentially. And it basically lets you use two different ways of thinking about the carbon intensity of electricity.

And so one of these is called a average carbon intensity and another one is called marginal intensity and they are slightly different. And it can be quite counterintuitive when you're first coming to this because it can give you somewhat different answers or incentivize different kinds of actions.

For someone who's coming to this for the first time, can you provide a little bit of like background on how to navigate some of this and how to think about some of this? Because it's something that. I think a lot of people come to and they scratch their head quite a lot because it can be a bit confusing having two numbers which can suggest you do totally different things sometimes.

Olivier Corradi: Yeah, well, we might get into some of the weeds, so I'll try to keep the, sort of, the discussion a little bit high level so we don't get too technical. But I think the way that I'm trying to explain this, so, on the average side of things, the first thing that I will say, actually, is that the word average can be a little bit misleading, because it sounds like, like you're taking an average over a period of time, which you're not.

Actually, what we're doing here is computing what we call the flow trace signal. So it's like taking the production locally, looking at what's imported, and then concluding on what is the constitution of the electricity that I'm getting at. So it's a representation of all the power plants on the system, basically.

So that's what we'll refer to as average. And on the marginal side of things, the story goes that if you're plugging an electric vehicle to charge at a particular time, then it's not all the power plants in that system that will ramp up to give you that additional electricity. It's the one that's called the marginal.

And a loose way to define it is to say it's the cheapest power plant that still has capacity to ramp up, to produce more. So, in theory, that marginal concept makes a lot of sense, because we're saying that seems to reflect what is physically happening on the system. If you actually go a little bit more into the details, and that's where the differences start to pop up quite starkly.

Well, if you are plugging, for example, your iPhone on this, on, to charge right now. Then it's not this marginal power plant that will ramp up. In all likelihood, nothing will happen on the grid. It's just the frequency of the system that will change a little bit. But you're not shoveling a little bit more coal in the coal power plant to burn off a little bit more, right?

So, emission factor would be zero. Like, no impact on the grid, you could say. If you go the other way, and you say, now I plug, imagine a metaphorical plug on a data center that's using like 90 percent of the consumption in the grid, Then you can't have only that small power plant that has spare capacity, it's just not going to be enough to ramp up this data center.

You would need all the power plants on the grid to ramp up in this what if scenario where the data center didn't exist. And then the last example is, if you have an electric vehicle and you're plugging it in, then the electricity will be delivered instantaneously to you. And that's just not a change that a coal power plant or a gas turbine can react on.

And that starts to create some additional complexity when you think about, okay, but can this be predicted? Can the plugging in of my electric vehicle, could that have been predicted by the market? And if it's already predicted, then it's part of the business as usual scenario. It's not marginal anymore.

So when you actually try to go down to all the details, it becomes hyper complicated. And we've tried really hard to talk to all the power system engineers, the electrical engineers and the TSOs in Europe as well. And one of the most fruitful discussions we had was with 50 Hertz that concluded that they don't think you can actually reliably identify or verify what the marginal power plant is because depending on what market you're looking at, if it's day-ahead, intraday, real-time, depending on the magnitude of the changes, depending on all the interconnectivity, the marginal power plant is just a concept that philosophically makes sense, but from a data perspective, it's just hard to measure, and that's also, I believe, why most of the regulations recently on the hydrogen regulation in both the US and in Europe that is documenting what is it, what signal should you be using, whatwhen you want to prove that your hydrogen is clean, then they are settling on an average signal. So if you take a step back from all of this, you're having an argument that philosophically for me makes sense. Of course, you want to make sure that the short term impact of what you're doing is minimized.

When you start looking at the data, you have a signal that is difficult to audit. And we've been working on this for eight years now, six years, sorry, creating a marginal signal and trying to verify it. And I've just seen enough that it can be manipulated in many ways. And that's why it's a little bit difficult.

So to get back and conclude a little bit and to answer your first question, which is how would we navigate this thing? I think the advice we're generally giving is get both signals, plot the data next to each other, and depend on the use cases, ask yourself "Is user acceptance important?" Are you going to show this in an interface?

Because if you are going to show it in the dashboard of an electric vehicle, in a country like France, which is majority nuclear, or in Ontario, lots of decarbonized electricity already, it's going to be a tough sell to tell them, look, your EV is being charged on gas, which is on the margin. So this user acceptance is important.

The second piece is auditability. Are you thinking of being generally directionally correct? Or do you need an auditor to prove that the data is correct from a scope 2, scope 3 perspective? And these are typically things they're going to inform depending on if you're willing to have something that is philosophically more accurate, but in practice more wobbly or if you want to have something that's just simple to explain and sort of abides by the regulation.

Chris Adams: Ah, okay, all right, so if I try to kind of summarize it, it sounds like the marginal thing it's conceptually attractive and kind of fun and it might give me some and in very ways it's basically gives me this idea that I can make relatively small changes rather than some of the systemic changes that might be needed for some of this.

But from an audit, from an auditing point of view, because you're comparing some of this to essentially a counterfactual which might not exist, it's actually very difficult to say, "well, yeah, I definitely made this impact and if it weren't for me doing this these people wouldn't have switched this stuff on," for example.

So you kind of need to have a degree of kind of clairvoyant level of information for this to actually really check this. Okay, cool. Speaking of clairvoyant, I've just realized the reason I've used this term is it makes me think of, there is a paper by a research student, I think her name, Tammy, I'm so sorry if I pronounce her name incorrectly, Thanathorn Sukprasert, she wrote all about using the SCI and using these different signals and how they, when they agree and when they do not agree, and if you are curious about this as someone listening to this, we'll share a link to that paper because it's a really good paper, she's also doing another one which is all about, okay, what are the possible savings from carbon-aware computing?

And that is also a really fun paper to read, and maybe, Tammy, if you're free, we'd love to have you on the podcast to talk about some of that, because there's quite a lot of fun to read there. Okay, cool. Thank you for that, Olivier. I realize that was quite detailed, but it, I appreciate you talking about the fact that in other sectors, there's stuff we can look at, because, as I understand it, this whole shift towards clean or green hydrogen is somewhat comparable to data centers in the fact that you have a very concentrated amount of energy being drawn in one place. So, in the same way a data center might be tens or hundreds of megawatts, you might see something similar with, like, creating hydrogen.

Is that the idea behind some of that?

Olivier Corradi: Absolutely, and if if the way that I'm thinking about this if you really want to simplify it, if we take a step back, the opportunity we have ahead of us is large, abundant renewables that are going to be the cheapest and the fastest way for us to expand the system. I mean, nuclear is great as well, and hydro and so on, but it's just, it takes more time, and it's a little bit more expensive when you put it in directly.

And so, if we put all this renewable in the system, we better well make sure that every flexible appliance out there is aligning their consumption to the time at which the renewables are creating that electricity, because else we're just hindering their deployment. You know, it's less batteries that we need.

And I think that's generally also the thinking behind the hydrogen. We want to make sure we're electrolyzing the hydrogen at the times where all these abundant renewables are producing electricity. And so that's the simplest heuristic and the fastest way for me to explain the systemic change we're undergoing and also why we're focusing on these renewable power forecasts.

There's a lot of value there. And by the way, why they also typically align with the price of electricity.

Chris Adams: Ah, okay, thank you. I do hope we still have some people with us, because that was a bit of a deep dive. But as sometimes when we're coming to new fields, working in software engineering, you do need to kind of engage with the details sometimes. Folks, I think we're coming up to the time that we had allocated for this.

And this has been loads and loads of fun. I've really enjoyed this. If people do want to know more about what either of you are doing, can I just give you a bit of space to talk about, like, Follow me on either Twitter, LinkedIn, Mastodon, whatever, like yourselves personally. And then if there's anything you would direct people's attention to, yeah please do, and then I think we'll wrap up actually, so maybe Íngrid if I hand over to you and then you Olivier, we'll wrap up with you okay?

Íngrid Munné Collado: Absolutely, so if people want to know what I do, they can check my LinkedIn, and it's Íngrid Munne, M U N N E, and I'm also on X or Twitter with the same name. So looking forward to that.

Chris Adams: Thank you, Íngrid. And Olivier, for you.

Olivier Corradi: Likewise, like Íngrid, you can reach me on LinkedIn and Twitter. If you search for my name, you'll find me. And I think in terms of resources to watch, we try to publish blog posts that are going deep into the topic, that are thoughtful. We don't publish a lot, but when we do, it's like We, we at least try to have serious research there, so check out our website, check out our blog, and we have a couple of guides on carbon accounting as well if you want to go deeper on the topic, and a few videos on YouTube if you want to nerd out more on some of the things we just discussed.

Chris Adams: Okay, we'll add a few links for all of those and the thing I'll also share is that in the Green Software Foundation, there are ongoing discussions in the standards working groups to discuss all this stuff. So you can see all the kind of back and forth around this, so if you do want to engage with this and possibly join to actually take part in that conversation, there is that available.

So you can see what discussions have come before and how people arrive at deciding which carbon signal to be following. Alright, I think that takes us to our time. Folks, I really enjoyed this trip, and we'll make sure that everything, or as many things as we remember, are in the show notes for people who want to continue this quest to learn more about building more sustainable software. Thanks a lot, folks, and have a lovely time in Denmark. Ta ra!

Íngrid Munné Collado: Thank you. Bye!

Olivier Corradi: Bye-bye.

Chris Adams: Hey everyone, thanks for listening! Just a reminder to follow Environment Variables on Apple Podcasts, Spotify, Google Podcasts, or wherever you get your podcasts. And please, do leave a rating and review if you like what we're doing. It helps other people discover the show, and of course, we'd love to have more listeners.

To find out more about the Green Software Foundation, please visit greensoftware.foundation. That's greensoftware.foundation in any browser. Thanks again, and see you in the next episode! 

Host Chris Adams is joined by Asim Hussain to dive into The Week in Green Software, exploring the environmental impacts of artificial intelligence and how the growing adoption of AI technology affects carbon emissions, as well as the growing complexities in the measurement and reduction of these. They discuss a primer on AI's direct environmental footprint, regulatory trends in Europe and the US, and the complexities surrounding the renewable energy credits tech companies use to offset emissions. The conversation touches on real-time cloud data initiatives, carbon accounting in AI, and the future challenges of balancing sustainability with technological innovation.

Learn more about our people:

• Chris Adams: LinkedIn | GitHub | Website
• Asim Hussain: LinkedIn | Website

Find out more about the GSF:

• The Green Software Foundation Website 
• Sign up to the Green Software Foundation Newsletter

News:

• The Environmental Impacts of AI -- Primer | Hugging Face[03:12]
• How Tech Companies Are Obscuring AI's Real Carbon Footprint | Bloomberg [22:25]
• AI analysed 1,500 policies to cut emissions. These ones worked | Nature [32:48] 

Events:

• Green IO Conference 2024 London Dates and City Locations - Sustainable Tech Partner for Green IT Service Providers 

Resources:

• Does the EU AI Act really call for tracking inference as well as training in AI models? | Chris Adams [12:21]
• Simon Willison on openai [14:15]
• EnergyStarAI (AI Energy Star Project) | Hugging Face [16:12]
• Meta-Llama-3.1-8B-Instruct · Hugging Face [21:28]
• Jevons paradox and greening software—why increasing efficiency makes sense | ASIM.DEV [21:51]
• Olivier Corradi [27:43]
• Real-Time Cloud | GSF [28:41]
• GitHub - Green-Software-Foundation/real-time-cloud
• Reviewing the evidence we accept for Green hosting verification | Green Web Foundation [31:06] 
• The Week in Green Software: Modeling Carbon Aware Software | TWiGS with Iegor Riepin [37:18] 

If you enjoyed this episode then please either:

• Follow, rate, and review on Apple Podcasts
• Follow and rate on Spotify
• Watch our videos on The Green Software Foundation YouTube Channel!
• Connect with us on Twitter, Github and LinkedIn!

TRANSCRIPT BELOW:
Asim Hussain: Three, four years ago, everybody treated all carbon offsets the same. They didn't realize there was nuance between them. Now that's changed. Everybody needs to now pay attention to the same thing in terms of renewable energy. If you do not pay attention to the fact that there is a lot of variability in a lot of this stuff, it's all going to get tarnished with the same brush in the future and any renewable energy claim is not going to be trusted.

Chris Adams: Hello, and welcome to Environment Variables, brought to you by the Green Software Foundation. In each episode, we discuss the latest news and events surrounding green software. On our show, you can expect candid conversations with top experts in their field who have a passion for how to reduce the greenhouse gas emissions of software.

I'm your host, Chris Adams.

Hello, and welcome to the Week in Green Software, where we bring you the latest news and updates from the world of sustainable software development. I'm your host, Chris Adams. This is our news roundup show in the Green Software Foundation podcast. So we aren't doing a domain expert news roundup. Deep dive where we go into a deep narrow subject, but rather we're taking a more broader view.

So we'll try to add some context and commentary to the stories that have been shared with us that we discuss with our guests. With me today is my friend, colleague and mushroom enthusiast Asim Hussain of the Green Software Foundation. Asim, it's really good to see you again. How are your holidays?

Asim Hussain: Yeah, it's been quite a long time off and just gently dipping my toes back into the swing of things. Glad to be on the show. It's a nice gentle introduction back into the world of green software. So, glad to be here again, Chris. Asim Hussain, I'm the executive director of the Green Software Foundation.

And yeah, I spend my day probably similar to you thinking about how to advocate for green software. What do we need? What do we want? What are the questions that need to be answered and what are the levers that we need to pull to get action taken in this space?

Chris Adams: Reduce the environmental impact of digital services. Yeah. Okay, cool. Thanks. I should introduce myself as well. Hello, folks. My name is Chris Adams. I am the executive director of the Green Web Foundation. That's a smaller, a different organization. We are a Dutch non profit focused around reaching a fossil free internet by 2030.

And we do that using open as a lever. So we do loads and loads of stuff with open source, open culture, and things like that. As a quick reminder, we're going to share all the stories that we have and any projects or things that come up, we'll add to the show notes. So if you want to continue your quest to learn more about green software and how to reduce the environmental impact of digital services yourself, just look at podcast.greensoftware.foundation to see this. If you're looking in Spotify or some of the other podcast tools, you might not be able to see the links. So please do go to the website to see the show notes and you will be rewarded with diligently prepared links and helpful notes, and a transcript as well.

Alright then, so, I guess, Asim, I've introduced our News Roundup format, we kind of know what we're going to do, we've done this a few times, I assume you're sitting comfortably, shall we begin?

Asim Hussain: Yeah. Yes. Yeah. Let's go for it.

Chris Adams: Alright, okay, so the first story I see here is actually a post from, from Sasha Luccioni, Bruna Trevelin, and Margaret Mitchell, Hugging Face the Environmental Impacts of AI - A Primer.

So, this is one thing we had shared, and, Asim, I'm going to ask you, you've had a chance to look at this, what made you think this was actually worth discussing, and what would you draw people's attention to if they'd heard about this? And how would you, like, persuade people this is worth a read, for example?

Asim Hussain: Well, I suppose if you're new to the space and I think there's a lot of people out there who, for whom are kind of surprised to find out that AI has an environmental impact. So this is, I mean, A lot of this stuff, obviously we've been talking about different components of it over the last couple of years, but I think it's a really good, it's actually a great summary of the different components of what makes environmental impacts of software.

It's also got information there about what are some of the legislation coming down the pipeline? What are the, some of the actions that you can do? And some some things there. So I think it's a really good kind of primer for people. I think it's the title, is the title Primer? It

Chris Adams: Yes,

Asim Hussain: Yeah. So it does what it says on the tin. And I think it's probably could do as a really great introductory piece of information. It's got some great links there as well.

Chris Adams: Yeah, the thing that it might be worth just focusing on briefly, or one thing that leapt out at me when I looked at this is that it talks about the direct environmental impacts of AI specifically. So rather than talking about AI for good, or like "isn't it great that you could use AI to, say, make it easier to deploy renewable energy, or do this, or do that?" Right?

They're talking about, "no, there's an environmental impact that you still need to address regardless of whether you use something for good or for bad," and it seems to be focusing primarily on that kind of stuff. So as a responsible engineer, these are the things to think about. These are the kind of, what you might call an impact criteria, like there's carbon emissions, there's water, there's other things like that.

And generally this is from one of the kind of most trusted hands-on group who are like at the coalface for all this stuff. I mean, maybe, is coalface the correct term?

Asim Hussain: Yeah. I always use that term as well. And then I'm like, actually, that's a

Chris Adams: Maybe if you're in, maybe if you're in US AWS East, it's coalface and then if it's on AWS west, it's hydro face.

I dunno. 

Asim Hussain: I use coalface and then if I don't, then I sometimes use front lines and I'm like, "actually, that's not a good term as well." We need a term which doesn't have a war Metaphor

Chris Adams: or industrial revolution connotations

Asim Hussain: Metaphor.

But anyway, it's at the cutting edge!

Chris Adams: Yes.

Asim Hussain: Cutting edge. There you go.

Chris Adams: Okay. That, that, that will do for me. Alright then. Okay. That's the first thing. I think on a following from this, when I was reading this. I quite like that it actually touched on some of the regulatory drivers that you have, because it's very common for people to talk about the AI Act, because that's probably the first piece of legislation, but it also calls out stuff taking place in Spain, and taking place in the US, and it shows that there's a kind of growing, I guess, regulatory trend to basically say, "well, If you're going to have this piece of technology in society, then we need to have a data informed discussion about what impacts it might actually have."

So we can talk about, okay, where is it responsible to deploy this? But also just like, okay, how do we actually mitigate this? Because, I don't know, it seemed like cars are useful. And the fact that cars are useful doesn't mean that we don't talk about seatbelts, right? You still have to talk about them being safer, regardless of how useful they are, basically.

So it's really nice to have a trusted organization sharing some information like this in a relatively roundabout in a, in my view, quite concise fashion. But if you look at the set of footnotes, wow, there's so much stuff that you can dive into if you wanted to kind of go down that rabbit hole, basically.

Asim Hussain: Yeah it's a very well researched, almost a state of the art paper. And also say like, I think it's also good to know, because people don't, yes, there's regulation coming down the pipeline, and some of that stuff is more mature than other regulation. But I think when you're working inside large enterprise organizations, this is the kind of stuff that gets people to pay attention.

You can be talking for ages about, "hey, look, there's the carbon impact, it's having a... we should be looking at our AI usage" and sometimes that can land and sometimes that can't, but a regulation or the threat of a future regulation is something that I've seen personally open a lot of doors. It doesn't kind of complete the internal sale of, "we need to invest in mitigating the impacts of our AI use." But it's certainly, I've seen it kind of open up a lot more doors because regulation is something that a lot of organizations pay significant attention to, and it's also something that they will, Invest time before the regulation comes out to look at it and put effort into it.

So I, I look at that, right. And it's kind of, it's great to see those regulations that I wasn't aware of. So I'll be obviously using this as it's a great primer for me as well, but this is a really good way of capturing people's attention. So then you can have that more refined conversation about, well, how are you using your AI?

Let's have a conversation about it. How is it, how is future regulation going to affect you? So it's a really good way of opening that door. And if you're inside an organization and you are a little bit concerned about the consumption of AI that you're having, I think for me personally, like pointing out the regulation that's coming down the pipeline does open a lot of doors, enables some conversations with leaders.

Chris Adams: This actually might be a kind of somewhat appropriate time to mention some of the kind of policy stuff we're doing, because I, so in addition to doing the podcast here, I help co-chair the policy working group. And we've, I think we've, we, we're likely to be getting quite a I'm going to put a policy radar out to see precisely this kind of stuff coming up.

Because, yeah, if you didn't know, I mean, okay, today is the 6th of September, and, you've seen this whole energy efficiency directive thing in Europe, right? So, in nine days time, every datacenter that uses more than 500 kilowatts of power draw has to start reporting and posting in public all of their absolute energy use, absolute water use, the amount of energy coming from renewable energy, how much of it comes from the kind of credits that you might buy which are unbundled, how much is coming from a power purchase agreement, so the kind of green energy that you've purchased directly.

There's all this new stuff. There are some caveats around this. So not every single organization will have to, we'll probably publish, but we basically have a regulation that's saying, "look, this has to happen now in nine days at the time of this recording. So when this goes out, it probably will have already happened.

And like this gives you an idea that if you didn't know this is happening, then you probably will, we do need to know this because this is written into the law in lots and lots of countries now. Well, all across Europe, for example, and I suspect this is the kind of thing we might see in other parts of the world because when you look at the figures and look at the data that people are currently basing policy on, it's really hard to figure out what the environmental impact of, say, data centers might be or what the growth is going to be.

And if you want to plan for a grid or plan for hitting some climate targets, this is the kind of stuff you need to actually be knowing about.

Asim Hussain: But it's also useful because we've been having these conversations for a while about, I think we spoke on the podcast a couple of times in the past and when we were developing the SCI specification, it came up a lot like do you include the data? How much of the data center do you include? But the biggest problem was, is that you don't know.

If you decided to put into the specification, you've got to include certain, the concrete or whatever it is that goes into data center. If that data is not public, then what's the point of putting into specification? That's why these regulations, that specific regulation is so interesting. I'm interested to see what actually happens in nine days time and the quality of data that comes out.

The conversations I've had in the past, because when this first started being discussed, I was chatting to a lot of, not data center operators, but people that worked with data center operators. And obviously it kicked up a storm and everybody's like, well, I need this data. How do I get this data?

What is the minimum level of information I can provide to like meet? And that's where it gets really interesting for me. What is the minimum level of information I need to provide to meet the regulation? And I think in nine days time, we're going to find out what is the minimum level of information that people have figured out that they

Chris Adams: Yeah, they can get away with,

Asim Hussain: that they can get away with? Because when I had that conversations, I don't know where it landed now, to be honest with you, but they were like, is it at the building level? Is it at the rack level? You know, it's, and it was like, it's at the building level is where I was left at. So I think the more and more this regulation comes along and it kind of surfaces this data to us, then we can then use that data to make more informed choices, hopefully not from a consumer level.

I think it should be from a, not from an end user level, but from the people who use data centers and make make different choices.

Chris Adams: I think you're right. Okay, what we'll do, we'll share some, a couple of links to this, because this is something we've discussed in a few places, there's one or two working groups where this has come up, in particular, because there's also, on the, just, just before we move on from this, there's a whole, there's a current kind of, in my view, an interesting discussion going on about, okay, with this, in Europe at least, with this AI act, yes, it says that you need to talk, you need to disclose the training data, the energy used for training a model, right?

But it's not totally clear if you need to also track the inference, right? So if you think about the training part, and I've shared a link to a blog post where I've basically highlighted the bits of the law that make me think that you might need to track inference, or at least disclose some information about likely inference because you can think of like the training part as like the energy going into making a car and then the inference figures as a bit like the car's mile per gallon for that model, for example. And well yeah, well it's not totally clear yet and we've seen the law passed and we'll figure out yeah.

Asim Hussain: Cause I assume the only, I'll be honest with you. I assumed included inference. The only time I did it was when I just like read your article and I was like, oh, hang on. That's and that's where these things get very interesting to me. I mean when I was in working in enterprise like organizations and I've one of the things that was always interesting to me was whenever I asked questions or got meetings together, like "let's talk about, we've got some questions about the, how do we calculate this figure to meet with this specification?" And there's almost always legal got involved and lawyers got involved. And I was always kind of, I'm like, "I don't need to speak to a lawyer. I need to speak to an engineer."

Why am I speaking to a lawyer? Because it's all about, "let me read this text. What can we interpret from this text? What do we need to give?" So I assume just because everybody just, we know inference is where most of the emissions are these days, I just assumed it was that, but you've now actually read the text and gone, the text has, is interpretable.

Chris Adams: Well, yeah, because you think about like when this initiative was written ages ago, like a few years ago, it's gone through this massive kind of gestation process, right? And a couple of years ago, when we hadn't really got to this point where AI is being deployed in the same way, like, was it November 2023?

That was like 100 million users. OpenAI had gone from zero to 100 million users in five months. And maybe last week? We should share a link to Simon Willison's blog post because he wrote about this quite eloquently. He's like, well, OpenAI have just mentioned that they're now at 200 million weekly users. So that's like doubled in a single year. So we've gone from, so inference is now a significant part of the story in a way that it wasn't previously, basically. So it may be that the law, when it was written two years ago and began that process through it, it might not have been such a concern. And this is the thing that we're, this is why it seems a bit unclear, and I think we'll probably end up with a test case that will set a precedent for people to figure out what they should be sharing or what you might need to share if you're building new foundational models in future.

Asim Hussain: I mean, I can tell you that, that preChatGPT kind of announcing, it was very well known that inference was significant, as in way more than training for any, like, I can't reveal too much, but you know, it was known very much that that was the case and a lot of effort had been put into mitigating, not from a carbon emissions perspective, just from a cost and energy, but just all of that stuff.

So it's kind of known, it might not have been in the zeitgeist, it might not have been in the kind of the public discourse because it's so much easier to talk about this big training runner. Maybe there's just more public data about that because inference in a way, if you think about it, is going to be pretty private.

Because that's inference is basically telling everybody the business end of your where you're making money from, and they'll probably keep that pretty private. So yeah, maybe it just wasn't well known, but it was true and well known, I think, to anybody in the kind of the AI space that inference was a pretty big deal prior to this.

But yeah, it makes sense. Yeah, these acts take a long time. So yeah, a couple of years ago, all we were talking about was training. It was a good, it was a good headline to discuss. Yeah.

Chris Adams: Useful insight from the inside tracker team, alright. So there's one thing you mentioned actually, we're just on the subject of inference. There is this, in my view, really interesting project right now, EnergyStarAI, which is a project which is, you see a few names associated with, so, Sasha Luccioni, Sara Hooker, Régis Pierrard, Emma Strubel, Yacine Jernite, I think, Carole-Jean Wu, and one of our own at the Green Software Foundation, Boris Gamazaychikov.

He's at Salesforce and he's been one of the people who's writing publicly about a bunch of this stuff and also about like, quite, the small models as well as large models. And it's, I was really pleased to see his name actually. So. Hi Boris, if you're listening to this. This is a really good story to look at, because this essentially is talking about inference and saying, "well, let's find ways to make this visible for people" two years later now, basically, and say, "well, let's see if we can find ways to introduce some of the incentives to go for more efficient inference," the same way we've done with Energy Star in other kind of industries, for example.

Asim Hussain: There's really interesting, I'll try and get her on the podcast, actually, she's from IBM, her name's slipping me, so I'm not as good as names with you, you can just rattle them off, but I'll make sure to put it in. Because yeah, there's Energy Star for AI, but we're also, there's conversations inside the foundation as well now, kind of looking at SCI and how do you apply SCI to AI and kind of, there's a lot of overlap with a lot of this work as well.

But what's interesting is there's this real question about what to do when it comes to inference and training. Like if you were to report that. How do you report that for a model? And the point that was raised, and I thought it was so, because I never thought of it before, which is, if you've got a foundational model, you've done the training, you've done a big training run for a foundational model, and you're now then running inference on that, when you report, let's say, I don't know how any, I think Energy Star is just going to be like a good bad kind of label, where SCI is more like a score, like a carbon per prompt or something like that.

How do you apportion the training into the, do you include the training every time you call it? Because if you do, there's a really interesting thing that happens, which is these foundational models, like if you're using an open source model and you just, that costs 10 000 to run. Do you include that in your AI solution and then just say like, "Oh, I'm three, three grammars per prompt?"

Then somebody else uses that foundational model. Do you then divide that by two and say, well, now you're, you

Chris Adams: talking about double counting kind of question, right?

Asim Hussain: And then like, if that is how the measurement eventually lands, then if you're an unscrupulous organization, all that you would do is try and get as many people to use this foundational model as possible to then dilute your numbers.

And so I think one of the, one of my little bugbears, and it comes up quite often, is the assumption that something that works in the physical realm will work in the digital realm. And one of the things I try and educate people as much as possible is that stop trying to take something that has worked in the physical realm and apply it to digital because there's so many ways it just doesn't work.

If you're thinking about training like a scope 3 embodied carbon physical device thing. You can't divide a chip in two and I give you half and you, it's like, that chip's yours, but you can do that in the digital realm. So there's this whole supply chain accounting aspects of digital emissions, which, it just needs to be thought of differently in the world that we're in. And if you don't think about it differently, you can then have, I call them unintended, we used to call them gaming, like when people were developing the SCI, like one of the conversations we'd have on the calls is, how is somebody going to game?

How is somebody going to take SCI,

Chris Adams: Carbon Intensity here, right? Yeah, okay.

Asim Hussain: how is somebody going to game the side? That's kind of a lot of the conversations we had at the start. Cause obviously everybody was like, we want to make sure if we develop a standard, people aren't going to then

Chris Adams: Abuse it,

Asim Hussain: it. And therefore the standard has no respect in the world.

And so like a lot of how, like a lot of how I kind of work with the standards projects here is I'm a bit of an annoying devil's advocate. I love it actually because I kind of walk in and go "here are 10 ways I can hack this standard to present a better score without actually doing anything." And so I think that's some of the things we need to think of as we think of SCI for AI, as you think of Energy Star, as you think of these other things is yes, there's this happy path that everybody's a good actor,

it will work and it will give you the right signal, but we need to think about the non-happy part, where people might not even necessarily be bad actors. It's just death by a thousand cuts. You're working in an organization, you've got a deadline, you've got a bonus you have to meet, there's a customer that you're going to have to get or you lose your business.

And so you're just death by a thousand cuts. So yeah, we have to be very careful as we explore like SCI for AI and Energy Star and anything really in this space, which is talking about measuring emissions. Because if you don't think through those unintended consequences, that's a problem. And that's one of the ones I have is like, is if you're including training, how do you apportion that?

You might not actually want to include training. You might actually want a separate. You actually, you might actually only want to measure the inference, because that gives a truer figure.

Chris Adams: So to cap this off, I'm going to, as we move to the next story, I'm going to link, share two links which might be useful for this. So the first is the link to the Meta Llama 3.1 8B, their model card. They literally say "the methodology used to determine energy use and greenhouse gas emissions can be found here."

They've linked to it and they said "since Meta is openly releasing these models, the training energy use and greenhouse gas emissions will not be incurred by others." So they're basically saying you don't need to count that part. That gives an example. We'll share a link to that. The other thing we'll share a link to for the show notes is actually the Asim, for this software, carbon, yeah, this is

Asim Hussain: Yeah, this is my unintended.

Chris Adams: This is a good example because this is where you've basically said, this is me, like, red teaming this approach, and these are the ways I can basically, in bad faith, try and engage with this example.

And this will probably be useful for people who are looking at this, to get an idea of, like, how some of these standards or some of these conventions are being developed. Alright, shall we move to the next story? Because it kind of does relate a little bit to, basically, tweaking numbers to present a view of the world, basically.

So this story is now, how tech companies are obscuring AI's real carbon footprint. This is a story from Bloomberg, I believe. Asim, do you want to introduce this one?

Asim Hussain: So yeah, so I thought, well, the reason this kind of popped up on my radar was, I forgot what I posted on LinkedIn, but I was posting, I started posting a bunch of information on LinkedIn about the use of, the use of RECs and the effectively like kind of renewable claims that organization makes and how it's in a really frustrating way, it actually puts us in what we're doing in kind of competition with this very important energy transition, because the argument I'm making is, look, you can either do two things and then we'll talk about AI.

Let's keep it to AI. You can either make your AI model more efficient, so it consumes less energy. Or, you can do absolutely nothing, and just buy offsets, energy offsets, RECs, whatever you want to call them, to mitigate, theoretically, your energy offsets, your energy consumption. And that's kind of like being this "Are we friends?

Are we not friends?" How do we like, we want to support the energy transition, but at the same time, like we really want to advocate for more energy efficiency. So, and I think one of the things we've spoken about is that there's, when you do make these renewable energy claims, like one of the things that you do with all types of offsets to kind of avoid a greenwashing claim, you have to have that additionality component to your offset, which for the audience means that how do you, if I'm saying this thing is offsetting your emissions.

What is it a litmus test to say that is a true statement and is basically, are you actually adding? So for a renewable energy credit, it's like, if you weren't about this renewable energy, would that thing have happened?

Chris Adams: So you're talking about the counterfactual here, right? So you're trying to compare something against. This is, you see a load of this in the hydrogen circle, in the hydrogen, in the world of hydrogen, because Just like datacenters, hydrogen electrolysis, like the electrolyzers use loads and loads of energy, right, and one way that you can do that is just by plugging them into the grid, right, and there's various people doing various things to say, well, I'm just going to buy a bunch of, say, renewable energy credits, right, and that's going to make that count as green, and there's, that's, in some ways, that's kind of somewhat problematic because, essentially,

Asim Hussain: coal to make hydrogen.

Chris Adams: Yeah, that's not exactly what all, you're, in many cases, you're a, you're burning coal to make hydrogen, so the actual net, it's a net loss in climate terms. But also, the, there's been a big fight in the kind of hydrogen world of, to have like this notion of three pillars. Where you basically, if you're going to have something, if you're going to count something as green hydrogen, then you need to be talking about new infrastructure being added to the grid to provide that new supply.

You can't just use, you can't just take from the existing stock of green supply and then count that as green. And this is one of the things that we've seen, like, I don't, Amazon made the news, I think a few weeks, a while, because they basically acquired a data center from a company called Talon, I believe, where they're right next to a nuclear power station, right?

So this, that you, there are some people saying, "oh, this is great, isn't it good that Amazon's using a bunch of clean power," but then you've got to think about, well, okay, who was that clean power going to before? Was it going to the grid? Like, there's a whole discussion there about this. Yeah, so there's a whole set of things to be talking about and this is why this is such a kind of fraught area, basically.

Asim Hussain: I mean, but I think the way to bring it back to something that people understand is when we talk about carbon offsets, I think now it's more understood that it's kind of like you, you have carbon removal offset. So you can plant a tree. Right? And then you planted the tree, that tree will grow, and there's issues there.

That tree will grow and also suck carbon from the atmosphere. And you can say that's a carbon credit of planting a tree. Or, there's carbon avoidance offsets. And there's many various, and there's actually very good variants of carbon avoidance offsets. But there is a variant of a carbon avoidance offset where I've got a tree.

And you pay me not to cut it down. And so where's the additionality? If I'm actually planting a tree, it's happening. I'm planting a tree. I'm adding additional kind of capacity in carbon removal. And in the renewable energy markets it's exactly the same. You can have renewable energy. Which if you buy means a renewable power plant is going to get built and you can have renewable energy which is just kind of sold and if you buy it or you don't buy it there's no change nothing's going to happen there's no more new renewable plants going to get built. Only one of them has that additionality component and so, therefore, only one of them should really be used in any kind of renewable energy claims.

But both of them are allowed in terms of renewable energy claims. So in terms of what this article is talking about, when they're saying "tech companies obscuring AI's real carbon footprint," they're actually talking about companies using what's called those unbundled RECs, which is those RECs which do not have that additionality component.

And then use buying them and then saying, "well, that's mitigating my environmental impact." And what the article is talking about is really, you should be looking at higher quality, Renewable Energy Credits, ones with more additionality components to it. And I think that's really interesting. There's actually also really, Olivier Corradi from, don't know if I'm pronouncing the second name correctly.

Electricity maps. Yeah, he, when I was talking about, he shared a really interesting article he'd written a year ago, which I thought was interesting as well. I'll share that here if you've got it. Yeah. He's actually advocating for like a more nuanced approach to looking at renewable energy in that there's additionality, then there's additionality, and then there's additionality.

There's like different levels of additionality. There's like, "this definitely 100 percent would never have been built unless you bought this renewable energy credit." And there's other ones like, "we may not have been able to build it, but we had some funding from here and there." So there's kind of different levels of additionality here as well, which I thought was really interesting also.

I'd never thought of additionality more than just a binary yes/no. And he was saying it's actually more of a score for a renewable energy credit. But

Chris Adams: Yeah, there's totally a continuum there. So the thing I might share for people who are looking for something actionable to work with here is basically the numbers that you often see reported by technology firms. There's all this, there's all this nuance hidden behind it. And there's one project called the Real-Time Cloud Project inside the Green Software Foundation, which essentially is a data set of the largest three providers.

So that's Amazon, Google, and Microsoft. And they've got the figures shown in both the kind of location-based figure, which is the closest thing you might think to, like, the physical location, the physical impact on the ground. They also talk about some of the market-based figures, which is what lots of firms like to use, like market-based on an annual basis.

But they also provide a few other details and a few other ways of talking about it, because some firms are now talking about hourly, basically hourly green energy versus annual green energy, with the idea being that you, it's a way to try and avoid making claims about saying, "I'm running a data center at night with certificates coming from a solar farm." This is inherently a little bit silly. So they address that stuff. So there's a, there's now, I think it's in the final stage of what's referred to as consistency review, where every member in the GSF is able to just say, "Hey, I object to this, or I'm not sure about this." And then, yeah, there'll be an open data data set for every single region from the three largest providers, which make up more than two thirds of the entire cloud market.

So you have some meaningful numbers that have come from the actual big providers themselves that you can actually, that we can work with.

Asim Hussain: And I think, like, I think basically my, I think the point I'm generally raising out, out there with another kind of, one of the reasons this article was very interesting to me, and especially the work that Realtime Cloud is actually interesting. Three, four years ago, everybody treated all carbon offsets the same.

They didn't realize there was nuance between them. Now that's changed. Everybody needs to now pay attention to the same thing in terms of renewable energy. If you do not pay attention to the fact that there is a lot of variability in a lot of this stuff, it's all going to get tarnished with the same brush in the future and any renewable energy claim is not going to be trusted.

So I was, I'm kind of a guiding and advising organizations to pay very close attention to kind of the type of renewable energy that you're buying. And be aware that because of podcasts like ourselves, there's generally, it's a Bloomberg article talking about this right now. It is now becoming very aware in the minds of a lot of people who care about this space, who listen to our podcasts, who are paying attention, that there is nuance here.

They're paying attention. And so as an organization, you need to pay attention to this as well.

Chris Adams: Cool. Asim, I'm just going to add this one thing because I realised I should have mentioned this. So I work in an organisation where we do track some of this stuff. We track the transition of the internet away from fossil fuels to greener energy. And, I've shared a link for the show notes. Because we're basically reviewing our own evidence that we accept for green hosting.

And we've linked to a couple of papers. And specific reports, which dive into this a bit more, which have kind of also prompted us to start looking at this. So, organizations like the Science Based Targets Initiative, we mentioned there. We talk about some of the other things that we, some of the nuances around RECs.

And yeah, this is, this will be something we'll be doing. So we're going to be essentially figuring out how to do this ourselves in the open over the next few months. So, Yeah, I guess it applies to small firms as well as large firms.

Asim Hussain: Yeah, yes, absolutely.

Chris Adams: All right, should we move to the next story? So, this is a story.

Researchers analysed 1 500 climate policies to find what works. And these are the lessons for Australia. I think this is the link you shared with me, Asim. There's a very kind of Australian centric kind of point of view, which, as someone born in a small mining town in Australia called Prospect, because what else would you name a mining town other than Prospect, because it's full... you? Yeah,

Asim Hussain: I didn't know you were born in Australia.

Chris Adams: Ah yeah, born in Australia, small mining town.

Yeah, I was literally born in a mining town called Prospect, and it's near One Tree Hill. Can you guess how many trees are on that hill? And it's next to Dry Creek. Can you guess the conditions of that river? Yeah, it's descriptive rather than creative, is the term I've heard people in Germany who do similar things talk about, actually.

Okay, so you shared this story, maybe you can introduce this one here, because I think it's quite relevant in this discussion, what we were just talking about in the previous two stories, actually.

Asim Hussain: Hmm. So I think it was just really interesting. It was an analysis of like 1500 climate policies and just really looking at what worked and what didn't work. And I thought it was interesting because we spoke a lot about, we've spoken a lot about things like carbon levies and things like that on this podcast.

But what I found interesting about this article was they, again, brought nuance to the discussion and saying, "actually there's different, different policies seem to work for different types of organizations and also combinations of policies seem to work better than individual policies." So a couple of interesting ones.

So one of the highlights I got, so some of the, in developed economies, some of the most successful cases were the results of two or more policies working together. So that could be like a ban or something, as well as like a carbon tax, kind of pulling those things together. Oh, for instance, like a great example they had here was like, for instance, example, a fuel efficiency mandate for vehicles combined with subsidies for developing like charging stations and things like that. So then you've kind of got the pressure on both sides. And another thing that was really interesting was cause we spoke about kind of carbon levies and pricing was particularly effective policy in sectors dominated by profit orientated companies, such as electricity and industry.

So I just think it was really interesting to kind of think through it from that perspective.

Chris Adams: So there's a really nice example, there's a few really good examples. Good concrete examples of this to make this, like, something you can, like, get your hands around. So in America, right, we've seen the Inflation Reduction Act. So that, in many ways, are kind of it's all carrot, no stick. So the idea is that there's massive amounts of subsidies for building out, like, for, like, EVs or building out new, kind of, battery gigafactories, all this stuff like that, or things which are essentially make deployment of renewables much, much more profitable than they otherwise would be, because they're gonna, because you have all these kind of subsidies saying, well, we're going to give you a production tax levy.

So for every unit of energy you produce, you'll be able to get, you'll be, you'll get a credit that you can actually apply. So your project over the entire length of it will end up being slightly more profitable. That, there's stuff like that, that you see, which is coming from one end. But we've also seen, In America, the EPA, the Environmental Protection Agency, they've got, they've now come in with a stick, or they're coming in with a stick now, to basically say, well, we're going to have to regulate carbon emissions.

And this now means that it's going to be all new kind of coal fired power stations or gas fired power stations, there'll be all these restrictions on how you should, how you can use them. And this is particular, the reason I raise this in America specifically is we were talking a little bit about AI before, right?

Now these regulations, I don't think that many technologists are aware of right now. They basically say if you're going to run a gas-fired power station, you need to fit loads of carbon capture and storage onto it, which is, broadly speaking, if it does work, it's not something that's really used in large amounts right now.

And what you currently see right now is you see lots of utility companies basically saying, "Oh, the only way we can possibly meet demand for AI is to build all this gas right now." And the problem with that is that ends up locking in all kinds of emissions. Because once you've built something, you have this incentive to kind of try and get your return back on building this in the first place.

And this feels like," I don't think people have realized just how much of a stick this is going to be, because as far as I can tell, all the laws from the EPA basically say, look, you can't build gas like this, and you can't actually do this." So we're going to have, we've got like this case of massive build out of AI coming up against all these regulatory forces as well.

And it's going to be quite a significant fight in the next 6 to 12 months, I think, because yeah, this is, we've now had the honeymoon period of all carrot. Like you said in this piece, and now we're coming up to the stick, which is the other part, to kind of make sure that you can, make sure the significant part of the US grid is going to be decarbonized by, I think it's the mid 2030s, basically, is what they're doing, that they're aiming for with this.

But we have the same thing in the UK as well, like, UK right now, we've got a target for, the UK has agreed to try and decarbonize the grid entirely by 2030, which is great for us as an organization because we, we want a fossil free internet by 2030. So we're like, "oh, thank God the UK is doing this." The UK government, one of the big kind of manifesto policies from Labour coming in, who've just won the election is "we're going to have a clean grid, entirely clean grid by 2030."

So five years, basically five years time, they're going to get rid of all the fossil, almost all the fossil gas generation, right?

Asim Hussain: How are they going to do that? 

Chris Adams: That's what we'll find out. But the thing I found out when I spoke to some people who, basically, this is actually all based on some modeling using a piece of software that we interviewed a chap called Iegor Riepin, he was talking about this in one of the episodes, we'll share a link.

That software was, basically, these kind of things were put together by some analysts on our laptop saying, well, this is what you can do. There's a report from Ember Climate where they, the report is called Escape from Gas, I think, or A Path Out of Gas. And this was one of the things that was written in 2022, when gas was super expensive, to say, "well, this is one thing you could plausibly do for this."

And yeah, when, the thing about policy, people reach for what's there. This turned into one of the things that one of the parties led on, and now we're going to see if we do see a fossil free grid and fossil free internet in the UK by 2030. Because, yeah, it's fascinating. I'm so, this is the most exciting, most excited I've been about UK politics in a very long time.

Asim Hussain: I don't know. I might dampen it for you. I'm just not, I'm just not very, I'm just, there's a lot of manifestos that come out from governance when they join and there's a lot of disappointment in the years later when they, when it doesn't manifest, when their manifesto doesn't manifest.

Chris Adams: So this is the final thing that might come in, might be relevant. So the modeling that was used for this as the basis to say, "yeah, we can do this." This one thing that ends up being, so I'll share a post to it, which I end up doing a bit of research and speaking to some of the people about this. It's actually very conservative, more conservative than the National Grid's own

estimates about, specifically in our industry, demand size and batteries. So, these are the two big things that we're likely to see a massive increase in.

Asim Hussain: That's what gas is used for more like this is specifically to get rid of gas.

Right. 

Chris Adams: yeah, so the,

Asim Hussain: peaker plants and then therefore you can do a little demand, demand responsible.

Chris Adams: Partly that, the thing that they said is like, the, their plans basically are relatively conservative about the ability for demand side of reduction, making your, you know, Carbon Aware in stuff like that, right? And there is another thing that we've seen is that the UK government is actually being quite gung ho about deployment or deploying all these new data centers.

So I'm kind of thinking, is there a chance to actually say, "well, okay, if you're going to have this deployment of all these data centers, and you know that one thing you're going to need to have is a much more responsive grid, is there a path for all this kind of carbon aware infrastructure to actually serve some of the roles that you wouldn't have to typically rely on peaker gas plants to actually fit, to like kind of fit?"

There's a bunch of stuff there and I think we'll learn basically because, yeah, this has been a really ambitious goal and you've also got this other idea to like bring in something which, can be quite flexible, but only if you incentivize infrastructure to be flexible, because for the most part, we don't see an economic incentive passed down to the consumers of infrastructure to be using this right now.

So, yeah, maybe this is a help of one piece. Yeah.

Asim Hussain: Yeah, I mean, I think there's been, there was some really good work done kind of several years ago, and this will be really good because one of the things I've seen is that the, all the positive moves I saw kind of three years ago regarding new data center rollouts, hydrogen fuel cells, kind of building kind of a much more advanced data center seems to have gone back a little bit.

And yeah, You're right, I think data centers could lead the way in terms of demand response. I'm not even talking about compute demand response. You can just take batteries, you can fill data centers with batteries and then they can store and then they can do their own sucking from the grid when it's clean and powering their own infrastructure when it's dirty.

You know, there's, there's other solutions, which doesn't even necessarily need kind of a software,

Chris Adams: Yeah, exactly. I mean, this is one thing that we've seen in Ireland. There's precedent in Ireland where people have said, "if you're going to be connected to the grid, you need to be prepared to be kind of complementary or sensitive to the needs of the grid for this." So, I think there's actually room for this, and it will be really nice.

I think that this feels like, given such an ambitious target, it does feel like a role where you could actually tell a good story about green Software, and be part of the solution as opposed to part of the problem, because a lot of the discussions around like rolling out of digital infrastructure is basically saying we can't possibly meet this demand.

But if we accept that demand is dynamic, then there is a chance to actually fit this in, and that feels like definitely worth going for, particularly to kind of maintain this kind of social license for operation, particularly for technology firms.

Asim Hussain: And I think a lot of what you've just said over the last couple of minutes runs very counter to what we were saying before about, I mean, everything you just described, this is all related to that whole idea of additionality. It's all about how do we transition, truly transition the grid to be fossil free?

And you need solutions like this. Not necessarily buying unbundled RECs, but you need to actually, like, think through, "well, how do I how do I be a better citizen in the grid infrastructure, do demand response, be sensitive, not demand energy when everybody needs it and therefore we have to spin up a gas power plant or something like that."

So these are the kind of things you need to actually transition the grid.

Chris Adams: Asim, I think we might have fallen down a bit of a grid rabbit hole,

Asim Hussain: Yeah, we've done it again, haven't we?

Chris Adams: Yeah, so, we're gonna have to move on, I think we've come up to time, but Asim, it's really nice to see you again, I'm glad you had a nice holiday, and I guess we've got a bunch of new things to do this quarter, right? With various projects we have inside the Green Software Foundation, and in the other member organizations related to it.

Alright dude, it's Friday, so have a lovely weekend, and for those listening, we'll put all the links to everything we've discussed in there, and if there's something you didn't see, Please do let us know, and we'll make a point of adding it. Alright, thanks a lot folks.

Asim Hussain: Thanks, Chris. Bye.

Chris Adams: See you around soon. Bye! Hey everyone, thanks for listening!

Just a reminder to follow Environment Variables on Apple Podcasts, Spotify, Google Podcasts, or wherever you get your podcasts. And please, do leave a rating and review if you like what we're doing. It helps other people discover the show, and of course, we'd love to have more listeners. To find out more about the Green Software Foundation, please visit greensoftware.foundation. That's greensoftware.foundation in any browser. Thanks again and see you in the next episode.

Host Chris Adams is joined by special guest Philipp Wiesner, a research associate and PhD student at TU Berlin, to discuss how computing systems can better align energy consumption with clean energy availability. Contributing to Project Vessim, Philipp explains how researchers are now able to model different energy consumption scenarios, from solar and wind power integration to the complexities of modern grids despite the scarcity of available testing environments. They discuss federated learning and its role in carbon-aware designs, along with challenges in tracking real energy savings. Tune in to learn about the future of carbon-aware computing and the tools being developed to help software become more sustainable.

Learn more about our people:

• Chris Adams: LinkedIn | GitHub | Website
• Philipp Wiesner: LinkedIn |GitHub | Portfolio

Find out more about the GSF:

• The Green Software Foundation Website 
• Sign up to the Green Software Foundation Newsletter

News:

• Let's Wait Awhile: How Temporal Workload Shifting Can Reduce Carbon Emissions in the Cloud [03:26]
• FedZero: Leveraging Renewable Excess Energy in Federated Learning | Proceedings of the 15th ACM International Conference on Future and Sustainable Energy Systems [11:56] 
• Vessim: A Testbed for Carbon-Aware Applications and Systems 
• Cucumber: Renewable-Aware Admission Control for Delay-Tolerant Cloud and Edge Workloads 

Events:

• 1st International Workshop on Low Carbon Computing |  (December 3) [41:26]
• Code Green: The Intersection of Software Engineering and Sustainability | (September 12 at 11:00 am AEST · Virtual) [44:41]
• Doing for Sustainability, What Open Source Did for Software | (September 16 at 11:20 CEST · Vienna) [45:02]
• Engineering a Greener Future for IT | (September 18 at 5:00 pm BST · London) [45:24]
• Collaborating On Digital Sustainability | (September 18 at 6:00 pm BST · Brighton) [45:33]
• Green IO London | (September 19th - 9:00-18:00 - London) [45:45]

Resources:

• Environment Variables | Episode 9 w/ Philippe Wiesner [02:48]
• Vessim: A Testbed for Carbon-Aware Applications and Systems [04:42]
• Towards More Carbon-Efficient Federated Learning (Flower Monthly 2024-04) [13:18]  
• Vessim | GitHub [24:02]
• FedZero | GitHub 
• Let's Wait Awhile | GitHub 
• Kepler [26:08]
• GitHub - green-coding-solutions/cloud-energy: Cloud Energy is an XGBoost & linear model based on the energy data from the SPECPower database for the cloud to estimate wattage consumption of server by just a few input variables [26:14]
• [2210.04951] Ecovisor: A Virtual Energy System for Carbon-Efficient Applications [29:14]
• The Trouble with European Green Electricity Certificates | Industry Decarbonization Newsletter [40:07]
• Software‐in‐the‐loop simulation for developing and testing carbon‐aware applications - Wiesner - 2023 

If you enjoyed this episode then please either:

• Follow, rate, and review on Apple Podcasts
• Follow and rate on Spotify
• Watch our videos on The Green Software Foundation YouTube Channel!

Connect with us on Twitter, Github and LinkedIn!

TRANSCRIPT BELOW:

Philipp Wiesner: Optimizing energy usage for when the sun is shining and the wind is blowing, of course it makes sense. But the big problem is that these green energy certificates, these guarantees of origin certificates, are traded independently of the physical flow of energy.

Chris Adams: Hello, and welcome to Environment Variables, brought to you by the Green Software Foundation. In each episode, we discuss the latest news and events surrounding green software. On our show, you can expect candid conversations with top experts in their field who have a passion for how to reduce the greenhouse gas emissions of software.

I'm your host, Chris Adams. Hello, and welcome to Environment Variables, where we bring you the latest news and insights from the world of sustainable software development. I'm your host, Chris Adams. Today, we're diving into the world of carbon-aware computing, how computing systems can align the energy consumption with the availability of clean energy.

Natural cycles govern all kinds of things we do. We mostly go to bed at night, and the food we eat is often influenced by what season we're in, and so on. This also shows up in how energy is available to us. You can see how day follows night in the output from a solar panel. And broadly speaking, it tends to be windier around winter than it is in the summer.

Even when we burn fossil fuels, we're basically digging up energy stored millions of years ago. As we move away from fossil fuels though, where we get to use the sunlight that was conveniently stored for us, we move towards a system based around harvesting energy, like we harvest food. The way we consume energy needs to adapt as well.

And that's the idea behind carbon-aware computing. But how do we know if our carbon-aware designs are actually better for the environment? With us today is Philipp Wiesner, a research associate and fourth year PhD student at TU Berlin, Technical University Berlin, whose work focuses on carbon-aware computing.

Philipp is deeply involved in research that connects the dots between computing power and renewable energy, and today will unpack the work he's been doing, including his work on Project Vessim, a testbed for carbon-aware computing. So let's dive in. Philipp, thanks so much for coming back, and for those who haven't heard the last episode when you joined us in 2022, can I give you a bit of space to introduce yourself?

Philipp Wiesner: Yeah, so hey Chris, and thanks a lot for having me back. I'm really excited. Yeah, so I'm Philipp, I'm a PhD student, as you just said, and a research associate at TU Berlin. I'm in the research group for distributed and operating systems. And I've been doing research on carbon-aware or in general, sustainable computing since the start of my PhD or my master's thesis in 2020.

And yeah, at the start of my PhD, I was mainly looking at the basics. So that's the first podcast we had two years ago. The field was still very young and we were looking at the potential of delaying the execution of workloads to make use of cleaner energy. Later, we also looked at migrating workloads from this to different data centers and so on.

And it was all based on carbon intensity. And then from then on later, we moved on from only optimizing for carbon intensity to really better understanding how electric grids work and rather optimizing for curtailed energy, excess energy. So, for green energy that would otherwise go to waste.

Chris Adams: Cool. Thank you. Okay. So you mentioned we spoke in 2022, two years ago, and the field has moved. And while we'll share a link to that podcast, where we go into a lot of detail, can I just give you a bit of time to just provide a brief summary of the paper that you spoke about then? I think it was called Let's Wait Awhile.

That was one of the projects that you're working on, and that was one thing related to the cloud. If we can talk about that'll be a nice scene setter. Then we can talk about the new work that we, honestly, I'm really excited about in all kind of nerdy glee. So yeah, Wait Awhile. What's the deal with that one?

Philipp Wiesner: Yeah, sure. So the idea in Let's Wait a While was just that in different regions of the world and at different times we have different carbon intensities. So carbon intensity is a metric that describes how dirty or how clean is the energy that you draw from the public grid in terms of emissions per kilowatt hour.

And at the time we were looking just, if you have a single data center and you might have some flexible workloads. So workloads where you do not really care if they happen now or tonight or tomorrow morning, how much potential do we have in shifting them, delaying them, computing them a bit earlier, maybe dissecting them in different parts and computing these times in different parts and so on?

And it was like a bit of a, yeah, it was a basic work at the time.

Chris Adams: And there were some, I think, when I first read about it, there were some quite, the savings there were around the low tens of percent for, compared to what the other ones would be. It was something like that, I think, from there. So that was one thing that caught my attention while we ended up speaking about it last time.

This new work that you presented at Hot Carbon, Vessim, as I understand it, this is something which is more general and applicable to all kinds of grids rather than being focused on one or two. Can you tell me a little bit about why you worked on this and what the general idea for creating a testbed for hardware computing was in the first place?

Philipp Wiesner: Yeah, sure. So the thing is that in carbon-aware computing, it's really common to simulate. The thing is that we're often talking about highly distributed systems, often globally distributed systems. And also these approaches often only take effect if you like simulate them or like run them for weeks or months because we need different seasons.

We need different, yeah. So the thing is that in Let's Wait a While and also in basically pretty much all the papers I wrote so far, we have to simulate systems. And in pretty much all the literature so far, people simulate systems because it's just extremely expensive to actually build and operate such systems.

Only big players can do it. And also results are not reproducible. So no one can really check if the stuff that you're reporting is correct.

Chris Adams: So let me just check. So basically, there is no way to... essentially there's no counterfactual, there's no way to see if my, like, I might say "yeah I've made the savings" but without showing your workings it's quite difficult to stand by them or even challenge them actually, that's one of the things that seems to be a bit complicated with that, right?

Philipp Wiesner: Exactly, a general problem in research, to make stuff reproducible. But here we have the big problem that everyone is implementing their own simulation. So everyone is doing it by themselves and just iterating over a big time series for carbon intensity is maybe not that hard. But the moment you have more complex systems and you're working with forecasts and so on, there's a lot of things you can do wrong.

And it's quite a heavy development effort often to do things right. So, just over the years we collected a lot of experience and eventually came up with the simulator. So, Vessim is basically a big co simulation suite where you can plug and play your own simulators for renewable power generation, carbon intensity of the grid, energy storages, and so on.

And you can connect them, you can connect existing simulators, and then test your real system, real applications, as if they were interacting with this simulated energy system.

Chris Adams: Ah, I see, so let me just check if I'm So one of the things you mentioned before is that we tend to look at the grid as like, just one thing that might be changing. And even our interactions are influencing the grid as well, so there's, that's one thing you need to take into account. But you also mentioned just then, that while most of us think of electricity as just power coming out of the wall, there's actually, it's more helpful to think about it in terms of, well, there's maybe on site clean energy, which is maybe very, clean, but isn't always available, and there's maybe on site batteries where the carbon intensities might be influenced by what has gone into it.

Like if it's clean energy that's gone in, it's going to be very clean. If it's dirty grid energy, it's going to be quite dirty. And then there's the grid. So there's multiple different, kind of, ways energy might go into the system that you use for your computing, for example, and you need to account for some of those things.

And then in addition to that, different parts of the world might have totally different grids. So Poland, for example, right now, loads and loads of coal, but comparing that to Iceland, which is, or France, which is like full of nukes, it's going to be very different carbon intensity. So you can't make the same assumptions, and you'll need to essentially do something to recreate the conditions that you might be in so that you can make somewhat more robust and intelligible statements about the savings you might achieve by making any kind of carbon-aware designs.

Philipp Wiesner: Yes, exactly. So yeah, as you said, like most of the existing research is looking at carbon intensity is looking at the global grid. And the problem with this metric is in general, that it's oversimplifying how complex grids actually are. So energy markets are really complex, grids are really complex. So if we only optimize for carbon intensity, which is a relevant metric in the future, we will run into several problems.

Because it's just hiding a lot of underlying complexities. And, as you said, we can maybe avoid these complexities if we're looking more local at microgrids. So if we assume, I don't know, a data center provider is putting solar panels on their roof or like having a direct power purchase agreement with some close by wind park, then we have a lot more direct connection of who's producing and who's consuming, and Vessim is exactly designed for this, to simulate such systems so that eventually you can optimize really for the energy that you're producing or storing.

Chris Adams: I see, okay, so you're still looking at this and we're looking at more than just a grid, basically. We're saying, we're focusing on the carbon intensity because the power can be coming from different places, like the one I'm immediately thinking of, actually. So in the news, Elon Musk's talking about this new data center where he's, where they bought 100 000 Nvidia H100 GPUs using 100 megawatts of power.

And that's so much power that they can't get that from the grid. So what they've done is, it's not clean energy they've used. They've used, say, like, a bunch of mobile gas generators, plugged into the data center that's come in. So we spoke about on site clean energy, but there's actually all kinds of energy that might be going in, and we're now seeing something like this in this kind of world of centralized AI, where you do see organizations looking to things other than just the grid to power things.

So we might need to take into account a whole basically the system seems to be getting more dynamic and being able to simulate this seems to be one of the ideas that Vessim is intended to kind of support so that we can essentially show our working and understand what kind of changes we might be making, or under these conditions, this is what a carbon-aware design might achieve, for example.

Philipp Wiesner: Yeah, exactly. Yeah. As you're saying, it's unfortunately becoming more and more popular to again, build gas turbines or even small nuclear power plants right into data centers to like come up with this insane amounts of energy that they need by now. We of course, like rather focusing on renewable sources because they're also a lot more difficult to schedule.

But yeah, this is one of the use cases of Wessim. So you can like explore how does it affect my data center if I cover the roof with solar panels, build a wind park nearby. How does it affect my data center if I invest into a lot of energy storage? How can I optimize my data center to make best use of these local resources?

And then there's also a lot of like systems questions, like how do I optimize? How do I make applications aware of locally produced energy? How do I make them aware of the state of charge of local batteries? What are they allowed to know? Like there's also security questions involved. And all of this we want to make explorable, so to say,

by providing this very general purpose, extendable simulation framework.

Chris Adams: Okay, cool. That sounds, okay, this sounds lots and lots of fun, and I remember getting quite excited when I read the paper from Hot Carbon. So you've shared this, but as I understand it, some of this has actually been used for some production work, or this isn't just like a wacky paper, there's actually people, there are examples of this being used.

I know that your paper spoke about the use of "Okay, here's how we've been running this on a kind of test system in a Raspberry Pi. Here's how it works with AWS cloud instances." So there are various examples of it being used. But there is a new project that you told me about beforehand called FedZero.

Can you tell me a little bit more about that? Because that seems to be applying some of these ideas in the real world. So yeah, maybe I'll just give you the chance to plug that one, because yeah, it was a fun read.

Philipp Wiesner: No, sure. Thanks a lot. Yeah, exactly. So we're trying to, deploy Vessim in a lot of different projects now. One of the main ideas behind Vessim is also to be have like hardware and software in-the-loop integration. So you can also plug in real devices, real energy generators, and so on.

In FedZero, we didn't do that. FedZero is also fully simulated. But it's a new, it's a scheduler basically we proposed for federated learning systems. This is,

federated learning is some type of distributed machine learning training. So you have a bunch of different clients and they locally train a machine learning model.

So they locally consume energy and the scheduler basically decides when, who is training. So it's like distributed batch processing in the end. And the entire evaluation of the system was done with Vessim. We have, for example, a lot of data sets that are coming with Vessim. They've been provided, for example, by Solcast.

It's a big solar data provider. And the entire idea behind the system is to only use excess energy. So only use curtailed energy and spare computational resources to perform a distributed machine learning training to make it effectively (an exclamation mark) zero carbon! Because you don't use any energy that would have been used by anyone else, basically.

Chris Adams: Oh, I see. Okay, so there's two things that I'd like to unpack there. So could you maybe just talk a little bit about the relationship to, I think I was confusing this with Flower, perhaps. So Flower is one of the wider projects which has something to do with federated machine learning. you maybe just talk on that because I think, for people who have listened to this before, they might not be aware of Flower or the even idea what federated machine learning might be. Because I just spoke about a centralized massive data center, but that's not the only way you can do machine learning and AI training by the sounds of things.

Philipp Wiesner: No, sure. We can also go a bit more into detail. So in federated learning, we assume that our data is distributed. So it's a different clients, a very common example are hospitals that, for example, want to train a common model for skin cancer detection whatsoever, but they cannot share their private patient data.

So we need to somehow train on their data without them releasing this data. And what's happening and Flower is one of the frameworks that lets you do this, that lets you perform federated learning as we distribute the model to the different hospitals, they locally train on the data and send it back.

And we try to schedule this local training in a way that it's only using green energy. Yeah, that's the main idea.

Chris Adams: Ah, I see. So, these would be devices which might already exist, for example, you don't have to have the same kind of emphasis on having a bunch of new stuff that you're using all the time to kind of sweat the asset and make the most of it, so you could be using maybe older technology or a more kind of mixed set of computing, rather than 100 000 H100s, for example, in one straight place, for example.

Philipp Wiesner: Yes, that's true. That's true in one sense. On the other hand, we always have to see what we're optimizing for. And if you're optimizing for something like curtailed energy, like for example, in the north of Germany, we have a lot of wind parks. But the German grid is not always strong enough to get all this energy to the south where like the big consumers are, the big industries.

So often times we have an excess of energy in the north. Now if you have hundreds of kilowatt hours of excess energy in the north, it doesn't get you anywhere to turn on a bunch of iPhones or old computers to make use of this energy. So usually with our approaches we target big consumers. We target the A100s and H100s to, because they actually make a difference.

Like turning them on or off actually makes a difference to the grid. A few thousand iPhones, not so much.

Chris Adams: Okay, so you still would be using some of this kind of relatively powerful hardware, if it represents a significant power draw on the grid, for example. And, is there anything special about where these might be placed? Because I've, seen examples and I've seen people talking about the idea of like literally putting this stuff directly into, say, the towers of wind turbines.

I know this is one thing that Windnode in Germany were doing, for example, because that's basically, where the power's not being used, it's one of the most fastest ways to actually

Philipp Wiesner: yeah, clearly, yeah, that's a super important topic, like placement of data centers. Just what I said, of course, you like place these data centers again, and if we talk about Germany in like Frankfurt region or in the south of Germany, it doesn't get you anywhere because you still need to transfer the energy there.

So you would actually need to build data centers very close to where energy is produced. However, like the big problem that we're facing right now with all this LLM and GPU stuff. is that there's actually not a lot of flexibility. So, FedZero for example assumes that we have some spare resources and some excess energy, and maybe the excess energy is fair, we do have that but we do not have spare resources.

Like, whenever I keep telling people since the last six months about this project they're like, "no, there are no spare resources." Like, GPUs are used all the time, we have a GPU shortage so, carbon-aware computing on GPUs is not really working at the moment.

Chris Adams: Yeah, I'm glad you mentioned this because this was one thing I've always been struggling to make sense of because when you've got a graphics card which is like 40k and you've got only a certain amount of years to use it, for example, the cost of having it off is actually quite high. So therefore, being able to simulate what the savings might be from some carbon air computing might, would actually be meaningful.

Or, if you're not going to switch off the power, then you might need to find another way to, like, not draw power from the grid. So that kind of implies either an increased amount of storage that you might need, so that rather than switching things off, you just reduce the power you're drawing from the grid by pulling it from somewhere else, for example.

So that's another kind of way you might flex your system to... still reduce the emissions, because it may be the case that you're able to load up on clean energy when there is an excess in other places, whilst keeping that thing still running, for example.

Philipp Wiesner: Yeah, exactly. No, you're completely right. And that was different, a few years ago when you, like, were mostly talking about cloud computing on CPUs, when we were talking about carbon-aware computing, because there you usually have a big buffer, because cloud providers want to have this image of infinite scalability, so they usually have spare resources at hand.

So, yeah, like this entire load shifting stuff was somewhat reasonable. Now with GPUs, I mean, you're right, you would need to plug them off, but no one's doing that. Like, if you're buying a GPU, you're using it.

Chris Adams: Using as much as you possibly can to make it, to make the per hour cost as low as possible, because you're, I see, okay, so, this is one thing I'm really glad you allowed us to dive into this because one of the things that I would initially think about, like there's some trade offs you might be making, and I might, before I spoke to you, one of my assumptions was that by using a federated model, you might have a lower embodied carbon because you're able to use a wider set of computing, which might already exist, for example, but it sounds like, because some of these chips are, they use more power, but per instruction, they're much, much more efficient per instruction. So, you're probably not going to save that much from the total embodied carbon of your fleet, for example, by reusing existing stuff in some cases. Is that what it would be like?

Philipp Wiesner: Yeah, there's different studies on whether highly distributed training or computation in general is actually more or less efficient. It often really depends, like usually the centralized, like you won't get anything more efficient than an H100. Energy per instructions. It's super, super efficient, but then of course, if you put them all in one place, you need to cool more and so on.

So often it can be more efficient to distribute it. there's no clear answer to that. I would say it really depends on the system, whether federated distributed system is more energy efficient or a local one, but of course in distributed system, you might have more flexibility for scheduling and waiting and yeah.

Chris Adams: Ah, I see. And I'm just going to move on to another question in a second, but while we talk about that, I just wanted to speak about maybe this idea of more distributed federated learning. A lot of the time you see people talking about saying, "I need a cluster of machines in one place because they need to talk to each other."

And the closer they are, the lower the latency, the faster the responses might be. And. When you're using a more distributed version, a distributed model, if you're not doing, like, even if you're not thinking about machine learning, you still need to think about latency, how long your response comes back.

How, like, what are the trade offs that you have there when you're looking at a kind of federated model, like with Flower, or providing some of these kinds of more distributed, carbon-aware style approaches, for example? Like, how do you address that?

Philipp Wiesner: So in general, in federated learning, we differentiate between cross device and cross silo federated learning. And cross device is what you were mostly referring to as like you have a lot of iPhones, a very distributed small system. They're often connected via mobile network. So latencies are usually quite high.

And there you really need to consider this latency trade off. In cross silo federated learning, it's a bit less important because you assume that each of these clients is, as I said, like a small data center or a big data center, they are connected via fiber. So, latencies are usually very low and not super relevant compared to how long a training takes.

So like sending models back and forth, model weights is not really relevant compared to how much time you need to compute actually. 

Chris Adams: Ah, thanks for explaining that, because there are various companies who are talking about having a distributed model, or saying, you need to do, you need to have, maybe, infrastructure that is complementary to the grid. And I was never able how to, how you get around the issue of speed of light, right?

Because they're going to be very, different if, even if it's just opposite ends of the country, that's going to, depending on the size of your country, that's going to be somewhat meaningful there. Okay. All right then, so we spoke a little bit about the use of Vessim, we spoke about why it exists, and like kind of this idea that you can simulate various kind of configurations of grids or configurations of power coming into the system.

I imagine there'll be some people who are listening thinking, "this sounds cool, how do I actually use this on a project when I'm designing something in a kind of carbon-aware way, or how do I actually, what am I looking for?" Could you maybe talk me through? The actual process of using Vessim specifically.

We can start with something simple like a machine learning job first and then we might talk about some other configurations like perhaps a CMS because that's what lots of people might be working on for example.

Philipp Wiesner: I mean, the entire idea of Vessim is that you do not really notice that it's a simulator. It should behave as like a proper energy system. So you can, on the one hand, you have your normal applications. It doesn't matter if it's a machine learning job or if it's a CMS or if it's anything else. You just have your applications that are running.

And then all you need to do is you somehow need to get their power usage into the simulation. You can either do that by modeling. You can, I don't know, look at the CPU usage of your application and somehow estimate how much power it would use. Or you actually let it run on a machine, plug a proper power meter, and then periodically feed this information into the simulation.

And the simulation itself you configure as you wish. So we have a lot of READMEs. You can plug and play custom simulators. We have some implemented, we're right now implementing the ones from the National Renewable Energy Laboratory. They have lots of simulators on everything from solar to wind to geothermal and whatsoever.

So you can plug your microgrid together. Also provided with energy storage. We are releasing Vessim 0.9 this week, I think, which then comes with a few more energy storage solutions. So you can really plug this together as you wish. And then you can write your own API basically to access this data center. The new applications, as in normal carbon-aware computing, can just query a REST API or something and really ask "what's the state of charge of my battery, how much energy is produced right now, what's the forecast for carbon intensity," all of this.

Chris Adams: Okay so I'm going to break that down a bit because there's a couple of quite interesting concepts that you spoke about. So first of all I feel a bit embarrassed by not mentioning this, we just assumed it, of course, was open source, but yes, this is entirely open source, anyone can access it, if they search for Vessim, V E S S I M, they'll likely find the project, because there aren't that many other things called Vessim right now, is that the case?

Philipp Wiesner: I hope so, because we were searching for a searchable name. 

Chris Adams: We'll also share links to the actual project on GitHub as well. It's written primarily in Python as I understand it, so if you know Python or know someone who works in Python, and given that Python's basically the most popular programming language on Earth, or by various metrics, is one of them, then the chances are fairly high that you can start doing this.

Are there, like, presets for different grids? So in the same way that you might have a preset for, I'm viewing a website on a mobile phone, or I'm viewing a website on a desktop, you said a little bit about NREL, the National Renewable Energy Laboratory. They've said, "well, this is what the, this is how a geothermal plant might work, or this is how a wind turbine might work and this is solar panel might work."

And that gives you an idea of saying, "well, this is somewhat similar to my grid." And then there's kind of, are there some presets or things so people can come up with some approximation of what it might be in their part of the world?

Philipp Wiesner: Not yet publicly. We're working on this and we will publish it soon. But right now there's no presets, but Vessim comes with some data sets that were provided. One was provided by Watttime on carbon intensity. One was provided by Solcast on solar data. And you're right, Vessim itself is in Python, but we're like building on a library called Mosaik.

Been developed in Oldenburg, also in Germany, and this is basically the entire co-simulation system behind Vessim. And they allow you to basically plug and play any type of simulator that can be written in Java, or it can also be a real system. You can also just take a real battery and measure that and just feed this information.

So it's really very flexible in that sense.

Chris Adams: Ah, cool, I'm glad you mentioned about the power use thing, because that's the thing I was going to speak to you about, because you said that you can draw power from a physical system, so you might time series of power usage, but one thing that we've spoken about previously with other guests on this podcast is that if you know what kind of compute usage, how much of your compute of a server is, you can make some I'm a bit worried about using the word inference when we've just spoken about AI.

But like, you can, there'll be some idea of what the likely power might be, just because machines tend to use more power at 100 percent utilization versus maybe 50 percent utilization. There's various curves and things that give you some numbers. And I believe there's an organization, Green Coding Solutions, who share some of this.

But also, I think this is under the hood what, other projects like Kepler also have some reference to. So, there are different ways that exist right now to give you some numbers. Even if you don't have something plugged into a system, for example. Maybe, you don't have access to, a power meter like you're deploying in a cloud, for example.

Philipp Wiesner: Exactly. Yeah. And this is cool that you mentioned that because this is one of the, it's a master thesis that we are running right now with one of the students who has also been developing Vessim for the last two years. And we are also using Kepler and trying to really calibrate Vessim for real systems.

And of course, these calibrations can be published. So we can really see, because there's this big difference between software-based power metering and like actual hardware power metering. In software-based power metering that's like included in Intel processors and also NVIDIA graphics cards, they just estimate how much power do I draw right now.

But it comes with a certain error. And what we're doing is like we're benchmarking these machines, using the integrated software-based power metering that actually everyone has access to but also like plug a real power meter and see how big is the difference under different conditions and so on and try to, yeah, make Vessim a bit easier to configure for bigger systems. Because this is the entire idea, maybe you have access to one GPU that you can benchmark but then you want to explore, "okay, how does my system behave if I have 200 GPUs?" And you can scale up in the simulation.

Chris Adams: Okay, so when we, so when someone is starting with this, like, we're going to talk about actually running the software, but we're just setting the scene to figure out, okay, what does the setup look like? Because I think most developers might be familiar with the idea of running a test to see if it works as intended.

There's this whole new world outside of your computer system that we need to talk about that I think you might need to kind of, that people need to understand before they start thinking about what, how the system is supposed to be designed. So we spoke a little bit about the fact that Vessim is using a existing tool called Mosaik, which designed for simulating grids.

So someone's done all that work already to figure out how to simulate a grid and you're using, you're relying on some of that. So there's a bunch of work that's gone into that already that you can draw upon. And we spoke a little bit about, there are ways that you can get readings on power, so even if you don't have something plugged in, you can still get some maybe low precision but still useful numbers to help inform some of your, like, design choices.

And then, now you've got that part there. You said a little bit about maybe things being in the, software being in-the-loop here, and I think while I've spoken excitedly about the notion of ecovisors and things on previous podcasts, I think that's what you're referring to here, like if the software itself knows how much, how full a battery is, or how green the power is, it might change in certain ways or it might respond to certain things.

Could you talk a little bit about that part? Because I think the software in-the-loop part is the thing that most people who've been taking an existing project and maybe trying to extend it to be carbon-aware. That's where they might be spending their most times either thinking about how to test something or figuring out how to make and adapt their existing system so they can start coming up with some metrics.

Philipp Wiesner: Yeah, It's interesting that you mentioned Ecovisor because it was a big inspiration for this project. Because I think they spent tens of thousands of euros on this one solar array simulator and a testbed. And we wanted to, like, explore similar ideas, but we didn't have the funding at the time to just buy such a machine.

So I had to write a simulator, more or less. So this is how it all started. Sorry, what was the question?

Chris Adams: in-the-loop, software in-the-loop,

Philipp Wiesner: Ah, software-in-the-loop, sure. Yeah, 

Chris Adams: We've simulated all this stuff, we know roughly how much power we're using, so the system, the software now has the data going into it. So what happens next? That's the thing.

Philipp Wiesner: Yes. So the idea is we need to run the simulation in real time, right? In wall clock time. So one second of simulated time needs to be the same as one second of real time. And if we manage this, then we can simply from the simulation, host any kind of API. We can just spin up a REST API that gives us access to the system.

And in Vessim, users can define these APIs themselves. So in Ecovisor, for example, they already came up with an API. I don't know, "/battery" and then you can query the battery state of charge, but something like this. Vessim lets you explore this. How do we abstract the complicated energy system and how do we like expose relevant information for the computing system?

And then you can just run a real normal application running on your computer, running in a docker container whatsoever. And it can send requests to this energy system API. And best case, if we did everything correctly, it behaves exactly as if you were doing a request to an actual energy system. So your application doesn't know that it's working with the simulation.

Chris Adams: Okay, so I'll try and make this a bit concrete. So when I'm working on a project, in an existing kind of carbon-aware system, I might make an API call to like, I might ask electricity maps, "Hey, what's the local grid intensity where I am right now?" For example, and like, it's a little bit like that kind of request I'd be making, but I might be making it to Vessim to say, "what's the grid intensity?

What's the state of my charge or stuff like that?"

Philipp Wiesner: Exactly. You just write a normal application that is running on a system, just that the system is power metered, so Vessim can simulate how it would be behaving if it was plugged to batteries, for example, and then it can talk to this simulated energy system.

Chris Adams: Gotcha. Okay. And we're using Vessim as just the kind of, simulator to share this stuff and it may be that in the real world, like I might be, speaking to the electricity maps API or something like that.

Philipp Wiesner: Absolutely, would be, yeah, absolutely.

Chris Adams: Okay.

Philipp Wiesner: A plug and play thingy, yeah.

Chris Adams: All right, that part, I think I understand. So then at that point, I'm then able to basically say, well, let's imagine electricity maps has given me this figure for carbon intensity, or let's imagine my, the state of my battery is high rather than low, I then might make some changes and say, well, "When the energy is low, don't do this stuff, or just delay it for maybe half an hour to check it again."

So that's the kind of stuff. So maybe you could talk a little bit about, now that we're at that point, how does that manifest into me seeing if that has actually reduced carbon emissions over a period of time, for example. Maybe we could talk about that part, because I think that's the final thing, now that we've set up our system, we've connected to, we've figured out how to get data coming into my system to tell me what the carbon intensity might be, or what batteries might be, or what the power from a wind farm might be. So yeah, maybe you can talk about that. So over a given time, has this actually helped, for example?

Philipp Wiesner: Yeah, And this is actually what makes simulations so powerful, because you can just execute a bunch of different scenarios. You can execute your scenario one time without any carbon-aware strategies. You can execute it a second time with the carbon-aware strategies. You can simulate as if you were executing it in summer or winter, or you can simulate power outages and whatsoever.

Because it's all really fast and really cheap to do. And then of course Vessim is recording all these metrics. Like for every second it's recording how much power has been drawn, produced, how much emissions were there. So afterwards you can analyze these results and compare the different runs.

Chris Adams: Ah, okay. I think I'm understanding this now and I

kind of figure out how I might build some of this now if I was to be building maybe a batch job, for example. I can kind of maybe imagine how this might work for maybe a content management system or like a CMS or WordPress or something like that. So let's say I would do, would I be looking at like, this would be my traffic the last month, run it through and then compare it to what the carbon intensity might have been.

Is that how I might do this?

for 

Philipp Wiesner: Yeah, absolutely. Absolutely. And you can also use it in testing pipelines, for example, in continuous integration pipelines. You can just simulate a bunch of different scenarios for your carbon-aware application. Some of them might be power outages or something like really edge cases that you do not want to happen, but you can really test cheaply that your behavior, that your system or your application would behave as expected

Chris Adams: Ah, 

Philipp Wiesner: In a reproducible manner.

Chris Adams: So you've raised something quite interesting for me here because you, when you spoke about the idea of like things, power outages or stuff like that, this kind of implies that if I've got a system here, I could see to what degree I might make a change to the software, or I might just say, well, given I have control over my infrastructure, I'm running a data center, what if I was to just put a battery inside that data center?

How would that impact the carbon footprint of using this system every month, for example? Would that, could that be something that I could simulate with that?

Philipp Wiesner: Yes, absolutely. Absolutely. And if you have a model that is able to, like a battery model, a simulator that is able to simulate battery aging, for example, you can also simulate what's happening over the course of the next three years with my batteries and so on.

Chris Adams: Ah, okay, so it's actually two way So therefore, I can look at, I can, look at, say, making interventions at my code level, and I can interventions at the kind of deployment scenario level as well, actually. Okay, that is something I hadn't actually realised before then. Okay, so we've dived into this quite a bit and I hope there are at least some people following us now.

Can we just take a step back to say because we have just been assuming that carbon emissions or carbon intensity is absolutely the figure we should be looking for and that is our North Star and you mentioned before that maybe that isn't or there's a bit more to it than that because we can't just look at this number, see if it's going down, and say, "Oh yeah, we're making progress."

Maybe you could outline on some of that, because it's a lot of nuance here that people often miss, I think. 

Philipp Wiesner: Ah Yeah, sure sure sure. So carbon intensity is a metric that defines how clean or dirty the energy in the public grid is. That means if it's low, you can use it if it's high, you should maybe rather not. And the problem here is that it's hiding a lot of the underlying complexities of how grids work and how energy markets work.

So for example, one big problem that we have in energy markets is, like in carbon accounting in general, is this double spending of guarantees of origin certificates. So for example, you can easily buy certificates produced in Iceland or something. They produce almost all the energy through renewable sources, hydropower, geothermal power.

You can buy their energy and then tell everyone you're 100 percent renewable energy. But at the same time, of course, any data center that is located in Iceland will tell you that they're 100 percent renewable, which is also true. and in fact, there's not even a power line between Iceland and mainland Europe, and still you can buy their certificates.

so we have a lot of double accounting and of course, everyone is coming up with new methodologies of what's good and what's the right thing to do. And of course, everyone's also just doing what's in their favor. And this is the big problem with carbon intensity. So of course you can say now "carbon intensity in Germany is really low.

I should consume energy." But unless you have a green energy contract, you're for sure only getting the dirty fraction, because all the green energy is already sold.

Chris Adams: So let me just check if I can stop you there for a second. So what you spoke about just then was there are kind of two approaches or like there's one one approach of looking at how kind of green power is might be the existing kind of system that you see in most developed energy markets right now where there's like a market-based system where you essentially can purchase renewable energy certificates and apply that to kind of brown energy to then say it's green.

And there may be reasons why we have ended up with this scenario. As you said, there are two problems there when we talk about, say, the idea of buying green power certificates from Iceland and then applying them to dirty power in, say, Poland or Germany to say it's green. There's like a, there's one issue which is that there's literally no way to connect, there's no connection, like we definitely are not using geothermal that's generated in Germany to get there. There's no physical connection so it's not deliverable. And the other issue is that if Iceland was to do that, they would then have to stop saying their power is green because they've already sold on their greenness. And in many cases they're not doing that. They're saying, "oh, we're still green even though we've solved the greenness."

So, there is a kind of problem about the, yes, let's assume that the market, the concept was conceptually sound. There is a whole issue about people kind of cheating a bit and saying no, it's still green when they've sold the greenness on it already, basically.

Yes, 

Philipp Wiesner: and exactly in this sense, optimizing for carbon intensity is cheating, because you're doing exactly this. You're like saying, "I'm using low carbon energy," but all the green, as you said, has already been sold. At the same time, it makes sense. The metric also makes sense. Optimizing energy usage for when the sun is shining and the wind is blowing, of course it makes sense.

But the big problem is that these green energy certificates, these guarantees of origin certificates, are traded independently of the physical flow of energy.

Chris Adams: Ah, okay. Okay, so there is, there's a whole discussion about this, like, where we work at the Green Web Foundation, we're kind of borrowing some of the ideas we've seen in the hydrogen sector, where they talk about if you want to say energy is green, it needs to be, I believe it is, timely, deliverable and additional.

So you need to be new power, like the, you mentioned this example of nuclear being used in other data centers, where essentially you're reallocating power that would have been going into the grid. So everyone has slightly greener power, but you say, no, that all has to go over the data center. Then that has the impact of essentially making the power that everyone else is using be seen as slightly less clean because of this accounting system, and that's one of the unfortunate artifacts of that. So that's one thing you mentioned about that. And I think the system that you might be talking about was the idea of like a location-based system, which is much more about like the physical environmental footprint of the power you're using rather than, say, the market-based approach, right?

Philipp Wiesner: Exactly. So we need a system that is somehow time based and location-based, a certificate system, then this all would make a lot more sense. But even then, grids are very complex and it's too easy to boil it all down to a single number and say at Germany, for an EPM we have this number of carbon intensity.

It would maybe make more sense if we, like, break down everything into smaller pieces, again, go more local. I have, for sure, Northern Germany separated from the South and stuff like this. But still, it makes sense, and this is what, for example, Vessim is good for, to really think about and understand how is energy actually flowing, how is the energy system actually working, because it's a bit more complex.

Chris Adams: Ah, I see. Okay, so for people who want to go down this weird accounting rabbit hole that we just kind of skirted around, I will share a link to a journalist called Hanno Böck who's dived into the world of Iceland, and in particular about this whole idea of selling green energy twice like this, because if you are looking into this, it's, as we end up with a much more decarbonised grid, this is going to come up again and again.

And right now, we do see the systems being used, like the GHG protocol, the people who actually define these standards. They're going through a process of reform and basically saying, "well, these are the changes we need to make and this is why this is possibly not fit for the purposes that it was 10 years ago," for example.

So there is a chance to make interventions and talk about this, if you want to look into this as well as looking into the other things we spoke about with Vessim. Okay, so we've spoken about this idea of setting up a kind of testbed that we've established that this is kind of open source, that anyone can pick up and start playing with right now, like the code is open source on GitHub, and as I can see, documented well with like, videos and stuff like that.

We're just coming to time, so I just want to give a bit of space for some of the future events that are coming up, because I understand that, well, part of your job, as you told me, is like, I'm the publicity chair for this all about low carbon computing, so I should probably give you a bit of space to talk about that as well.

Philipp Wiesner: Yeah, that's super nice. Thank you to just take two minutes. So we're organizing this workshop. It's the first workshop for low carbon computing, which will take place on the 3rd of December in Glasgow, and It's mainly an initiative, a local initiative from some folks at the University of Glasgow, but it's an international conference, it's an international workshop I mean.

And we expect submissions from all over the world, but mainly probably the European community, I guess. And yeah, welcome everyone to submit your work. So we have a very broad scope on really anything related to sustainable computing, from energy efficiency, carbon efficiency, embodied carbon and so on, but also topics that are often a bit underrepresented in computer science like circular economy, sufficiency, demand reduction, and all of this. exactly. So this is all very welcome. And for regular talks, all you need to submit is like a two page extended abstract by the 24th of September. Although we're like currently discussing maybe a tiny deadline extension and all accepted authors, all accepted papers will afterwards be invited to submit like a full workshop paper of up to eight pages if you're interested, which will be published in our post proceeding.

So yeah, feel free to have a look. I think we will also put a link in the

Chris Adams: a link in the show notes, for sure, absolutely. So if I understand this correctly, is this a little bit like, possibly, a kind of like a sister conference to Hot Carbon, but where it's a little bit colder, and you're more likely to kind of find a burrito served at Legal Jack's.

Is the name of a restaurant in Glasgow that eaten at before.

Philipp Wiesner: Yes, I guess like it's the first version. Let's see where it's going. But maybe we can, if it's going well, we can maybe establish it as like a European alternative to Hot Carbon or something, which will, I guess, take place in the US in the upcoming years. 

Chris Adams: I see. And there's one thing I wanted to just check because when I looked at the kind of call for papers, the set of topics, it's a little bit more wide than just like let's focus on efficiency. There seems to be a kind of systemic angle that I hadn't seen quite in the same way, for example. So there are discussions about like you said,

okay, looking at the demand side as well as just the supply side, for example. And like, I think I've even seen the word frugal computing used inside here. Like, okay, this is almost like somewhat possibly like the limits conference or maybe some of those ideas somewhat comparable to inside this. 

Philipp Wiesner: Yeah,

guess, yeah, I guess it's somewhere in between that, in between Hot Carbon and limits, maybe.

Chris Adams: cool. All right. Well, I am now thinking it might be nice to figure out a way to get a series of trains to get up to Glasgow, because, I really like that city.

Philipp Wiesner: Yeah, 18 the last times.

Chris Adams: Okay, cool. So that's one thing that's taking place. So the deadline was the 24th of September. Easy to apply to, there is, and that'll be taking place in December.

Philipp Wiesner: Just one last thing, it's a hybrid event. It's a hybrid event, so anyone is welcome to participate, and also authors are of course welcome to present remotely.

Chris Adams: Okay, so I don't need to get on that if I really don't want to, because, well, I might want to see people, but it takes quite a long time

Philipp Wiesner: only a one day event, so you need to consider if it's worth two days of travel.

Chris Adams: Alright, thanks for that Philipp, I appreciate that and we'll share a link to that so people know when the CFP is open how, well actually when it closes more specifically because I think it is open right now. Okay, cool. So I'll just touch on some other events that we have inside here and then I think we'll wrap up if that's okay with you.

So these are some of the other events that our producer Chris Skipper has shared. So there's an event called CodeGreen, the intersection of software engineering and sustainability. This is on September the 12th at 11am AEST. This has Sara Bergman from Microsoft Norway, and one of the, one of the authors of the book, Building Green Software, and also a regular on our podcast.

She's presenting, speaking with Michelle Sandford. And there's also an event in Vienna, in Austria. This has Asim Hussain, our Executive Director of the Green Software Foundation. He's talking about the Impact Framework, this radically transparent tool for sharing the environmental impact of your tooling, and that's on September the 16th, and I think that is in person, but then maybe we'll share a link for people to see about if they can get involved remotely.

There's also another event in the UK, Engineering a Greener Future for IT, that's on September the 18th at 5pm British Standard Time in London. And there is also another event in Brighton as well, so the UK is quite heavily represented right now, Collaborating on Digital Sustainability, that's September the 18th.

I'm going to use this platform while I still can. I'll share a link to Green IO London as well, which is the conference that I'll be keynoting at with a number of luminaries. We'll share a link to that because I, it's very front of my mind right now and I'll be there for the day. So if you listen to this and you fancy coming along, we'll share a link where you can get a free ticket to join at the conference.

And yeah, that's us. I think we've come to the time. We'll just actually make sure that all the things we've discussed, like I mentioned before, we always try to get as many of these show notes full of links as possible, because we cover a lot of ground on these free wheeling conversations, so that will be visible.

The thing you might need to do though, if you are listening via Spotify or the Apple platform, you might need to look up podcast.greensoftware.foundation, then follow the link to this podcast to see all the links, because I don't think they're made available if you just are listening to it in Spotify, for example.

And we do put quite a lot of time into doing that. And Philipp, I just want to say thank you so much for helping unpack what can be an extremely complicated topic and putting the work in to make a open source project that makes it easy for people to figure out what to do with carbon-aware computing and hopefully guide themselves to slightly more sustainable software.

Philipp Wiesner: Yeah, no, thanks a lot for having me. Yeah, had a lot of fun. It was a pleasure. Thank you.

Chris Adams: Okay, final question I should ask, if people do want to find out about you, where should people look? Are you on LinkedIn? Are you still using X slash Twitter slash what's the best way to follow your work these days?

Philipp Wiesner: Probably just Google me, I guess there's maybe one or 2, 3, 4 Philipp Wiesners that are more important than me, but if you just look at TU Berlin or Philipp Wiesner carbon-aware where I should pop up and then you can see.

Chris Adams: Cool, well your name will be in the show notes and the show title, so if anyone is curious about how to spell Philipp Wiesner and not sure how many P's they should be using, then that'll be inside there. All right, Philipp, really lovely chatting to you, mate, and thank you so much for coming along, and yeah, have a lovely week.

Philipp Wiesner: Thank you. You too. Thanks a lot.

Chris Adams: Cheers, bye!

Philipp Wiesner: Bye.

Chris Adams: Hey everyone, thanks for listening! Just a reminder to follow Environment Variables on Apple Podcasts, Spotify, Google Podcasts, or wherever you get your podcasts. And please, do leave a rating and review if you like what we're doing. It helps other people discover the show, and of course, we'd love to have more listeners.

To find out more about the Green Software Foundation, please visit greensoftware.foundation. That's greensoftware.foundation in any browser. Thanks again, and see you in the next episode!

This week we are joined by two PhD researchers, Silke Kaiser and Chiara Fusar Bassini, from the Hertie School in Berlin. With host Chris Adams they discuss their use of data science and machine learning and how they are using them to tackle some of today’s most pressing environmental challenges. Silke shares insights into her research on predicting cycling traffic in cities to better inform urban planning and promote sustainable transport, while Chiara discusses her work on analyzing European energy data to support the renewable energy transition. Together, they explore the intersection of technology, data, and policy, highlighting the importance of data-driven decision-making in achieving sustainability goals.

Learn more about our people:

• Chris Adams: LinkedIn | GitHub | Website
• Silke Kaiser: LinkedIn | Website
• Chiara Fusar Bassini: LinkedIn

Find out more about the GSF:

• The Green Software Foundation Website 
• Sign up to the Green Software Foundation Newsletter

News:

• From counting stations to to city wide estimates: data driven bicycle volume extrapolation | Silke Kaiser [09:46] 
• Pedalling Towards a Greener Future: The Impact of Cycling and Active Transport on Climate Change and Public Health - Catalyse [12:46]
• Chapter 10: Transport | IPCC [14:10]
• Estimating Coal Power Plant Operation From Satellite Images with Computer Vision [24:11]
• Does the EU AI Act really call for tracking inference as well as training in AI models? [38:27]
• What is the methodology used to measure the carbon footprint of training Llama 3.1? [41:12]
• Climate policies that achieved major emission reductions: Global evidence from two decades | Science [44:27]
• Microsoft employees spent years fighting the tech giant's oil ties. Now, they’re speaking out. | Grist [46:51]
• A review of the ENTSO-E Transparency Platform 

Resources:

• How does AI and ML Impact Climate Change? | EV Ep 5 [05:38]
• French Revolution: Cyclists Now Outnumber Motorists In Paris [07:25]
• Berlin’s Efforts to Reduce Driving Stalled by German Car Culture [07:42]
• Emissions of Carbon Dioxide in the Transportation Sector | Congressional Budget Office [16:47]
• Electricity Maps [21:59]
• Machine Learning for Sustainable Energy Systems | Annual Reviews [30:41] 
• CodeCarbon [40:49]
• Light bulbs have energy ratings — so why can’t AI chatbots? [42:16]
• The Chronic Potentialitis of Digital Enablement | Vlad Coraoma [46:10]
• Blackout (Elsberg novel) - Wikipedia
• International Energy Agency - an overview | ScienceDirect Topics
• Taxation - an overview | ScienceDirect Topics
• How fast is Germany cutting its greenhouse gas emissions? – DW – 07/10/2023
• Tackling Climate Change with Machine Learning | ACM Computing Surveys 

If you enjoyed this episode then please either:

• Follow, rate, and review on Apple Podcasts
• Follow and rate on Spotify
• Watch our videos on The Green Software Foundation YouTube Channel!

Connect with us on Twitter, Github and LinkedIn!

TRANSCRIPT BELOW:

Silke Kaiser: I like the term of fighting fire with fire. You know, you're trying to make it better, but you're making it maybe even worse. But I think if we make some smart choices along the way, I rather like to compare it to the idea of fighting a forest fire with a controlled burn. What I'm trying to say is that there are different approaches that we can actually also reduce the emissions caused by AI.

Chris Adams: Hello, and welcome to Environment Variables, brought to you by the Green Software Foundation. In each episode, we discuss the latest news and events surrounding green software. On our show, you can expect candid conversations with top experts in their field who have a passion for how to reduce the greenhouse gas emissions of software. I'm your host Chris Adams. Hello and welcome to Environment Variables, where we bring you the latest updates and news in the world of sustainable software development. I'm your host Chris Adams. It's important to understand that when we talk about sustainability and technology, it's easy to mix up sustainability of software with sustainability through software. Sustainability of software development is about understanding the direct impacts of technology and doing as much as we can to reduce it without delivering a worse experience for people using the software. Here, we care about the impact of code, like making it more efficient or making sure the energy we use is cleaner and coming from the cleanest possible sources. So when we talk about green software or green IT, this is what we're talking about. Sustainability through software development, this occurs through the application of software to solve a specific sustainability problem or provide us with insights that we didn't have previously to help us meet some of our sustainability goals. To make this really concrete, you can use green software and talk all day long about the sustainability of software whilst helping people drill for oil and gas. Now you can do that, but it's really not a good idea if you wanna hit any kind of societal climate goals. And if you're listening to this podcast, I think you probably don't wanna do that either. So while we usually cover the sustainability of software. It can also be helpful to look up from our keyboards sometimes to talk about the effects of software, the effects it can have for helping us reach our climate goals. So in this episode, we'll be diving deep into the work of PhD candidates who are pushing the boundaries of what's possible in sustainability ad software. In this episode, we're joined by two researchers from Berlin Institutions, the Berlin School of Economics and the Berlin Hertie School of Governance. So what insights can we gain from their research? How are they using technology to address some of the challenges around sustainability today? Let's find out. So, first of all, Silke, can I just give you folks a bit of time to introduce yourselves? I'll hand over to you, Silke, and then hand over to you, Chiara, to introduce and give you the floor. So, yeah, Silke, thank you very much for joining. The floor is yours.

Silke Kaiser: Thank you very much, Chris. I'm Silke Kaiser. I'm a PhD researcher at the Berlin School of Economics and at the Hertie School Berlin. I'm excited to be here today. My research focuses on the analysis of sustainable transport data, with particular emphasis on cycling data. I utilize various tools from machine learning, data science, and spatial statistics to explore this field.

Maybe more on a personal note, outside of academia, I would say I'm a bit of an auto enthusiast. Just earlier this week, I came back from a vacation in France, but I'm happy to be back and to join you for this episode today.

Chris Adams: I'll be Back just in time for the weather to be nice in Berlin, right?

Silke Kaiser: Exactly.

Chris Adams: Cool. Thank you for that, Silke. And, Chiara, can I do the same to give you some space to introduce yourself as well?

Chiara Fusar Bassini: Yes, thank you for that. My name is Chiara Fusar Bassini. I'm a PhD researcher at Hertie school in Berlin. I'm very excited to be here for my very first podcast. My own research focuses on the analysis of European energy markets. And I use data science and machine learning to analyze time series of dispatch of single power plants and try to look at how power plant dispatch has changed in the context of an evolving energy system. Beside being an academic, I'm also a rather lousy actor in an amateur theater group here in Berlin.

Chris Adams: Cool, thank you. So you said you're working in a theatre. Are there any particular roles you play or anything like that? Because I think you may be the first actor who's come onto the podcast actually, Kiara. Chiara.

Chiara Fusar Bassini: Well, I have a tendency to take on the roles of either mad people or police men or like with gender changes. So anything in between is business, usually like mad people it has been, like the latest role I've had was a police officer. And before that I was a mentally ill person.

Chris Adams: Wow. I did not, I was not expecting that. Okay. All right. Well, welcome onto the show. And I guess that maybe we'll see some, productions in future. Folks, if you're new to the podcast, my name is Chris Adams. I am the executive director of the Green Web Foundation. That's not the same as the Green Software Foundation.

It's one of the members of the Green Software Foundation. I'm also one of the chairs of the Policy Working Group and also one of the hosts of this podcast. Okay, so before we talk in depth about your research, I just want to check for people who are also listening to this, you've been doing this research under the supervision of Professor Lynn Kaack, I believe, is that the case?

Chiara Fusar Bassini: Precisely. We've been working with her for a few years now, each of us.

Chris Adams: Brilliant. Okay, so what we've, the reason I'm sharing this for listeners is that we did an episode 5, where we spoke to Lynn and another person, Will, oh his name has changed, I think it's Will Alpine now, talking all about climate change and AI, two years ago. So if you enjoyed this, I would suggest looking at that to learn a little bit more.

And that might provide some extra context for this discussion. All right, then. Are you two folks sitting comfortably? Happy to go ahead with this?

Silke Kaiser: I think it's good to go and share insights on our research.

Chris Adams: Good stuff. Thank you. For anyone who is listening as well, the thing I'll just share is that we will share show notes with links to all the projects and papers that come up to this. So if any of this is interesting to you, then yes, we, you can continue your quest for more knowledge and insight outside of this podcast. All right. Silke. Let's start with you. Your research focuses on predicting cycling traffic in cities using data from bike sharing systems. And this is something I believe you worked with Lynn and another researcher, Nadja Klein, on from this. Could you maybe just explain a little bit about how this actually helps people when they're trying to design how people move around in cities like, say, Berlin or Paris or things like that?

Silke Kaiser: Yes, I'd love to. So what we generally see when we think about transport in cities is that public space in cities is limited. Whether it be in Europe, the USA, or any other place. Generally, then when we think about how we want to redistribute the space among different mode shares in city, we see that there often tends to be a heated debate.

And especially as we work towards promoting more sustainable modes of transport and therefore reducing the CO2 footprint of our cities, conflicts often arise. And the question then remains, how do we actually want to prioritize these different modes of transport and allocate the space and also financial resources among them?

So, for example, take Paris as an example. I've lived in the city for several years during my studies. And what you can see in the city that in the past few years, they made a lot of changes to prioritize cyclists, which has improved the uptake of cycling, but which has also led to quite heated debates.

The same we can see here in Berlin, the city that we're currently both located in, is that we had a re-election last year here in February, and a lot about the debate actually hinged in part on the choice between prioritizing cycling and individualized motorized transport. So what I do try to do in my research is actually to provide more data to this debate, because what I see as the main challenge in these kinds of debates is that we don't have accurate data on how much cycling traffic we actually have in cities.

Chris Adams: Ah, okay. So maybe just kind of dive into there. So it's basically, we don't have the data to really have a data informed discussion, basically. That's one of the things that is the challenge here. And for context, so the three of us live in berlin. We saw, basically, the new government and the new mayor come into power on a very kind of pro car platform, basically.

So this is what you're referring to, right?

Silke Kaiser: Precisely. So, answering maybe the first part of your question, so for example, in Berlin, we have only 40 locations where we count cyclists. I think in Paris, it's around 53, in New York, it's 41. And then in Berlin, for example, we have around eight times more locations where we count motorized traffic. So we just have much more data and much more information on motorized traffic than we do have on cycling.

And then yes, in Berlin, actually, there was quite a bit of a heated debate pretty much between let's say the inner city, which was more pro cycling and in the suburbs, which were more likely pro cars. The government switched from a green to a more conservative government, which actually decided to suspend actually just this month quite some, projects, long distance commuter paths, and both bicycle parking houses in cities.

But that's just more really on the political side of the debate. And then what I really see as the main challenge is that just this data and the information is missing on where we actually would need the infrastructure most, given how much we want to prioritize cyclists.

Chris Adams: Okay, so maybe I should ask, where is the data coming from then, for this?

Silke Kaiser: So, we do have these 40 counting stations in Berlin, that's the case study that I'm looking at, and what we then figured is, well, we don't have that much precise data on cycling, but we do have an abundance of other data on cycling. So, for example, we do have, as you mentioned, the bike sharing data. We do have as well data from Strava.

That's an app popular to record yourself while doing sports. We do have data on infrastructure, we do have data on weather, we do have data on socio economic factors. And we figured, well, why don't we use all this available data to actually extrapolate from these few isolated locations to actually obtain city wide estimates.

And that's what I did in this research with Lynn Kaack and Nadja Klein. So what, precisely we do is that we train various machine learning algorithms to use all this kind of data in combination with the cycling counting station data that we have to obtain citywide estimates. And what we actually found is that only using this data is a bit tricky.

It provides us estimates for completely new locations 32%, which is you know, rather good in comparison to having no data at all, but still 32 percent is an error I take seriously. And what we then simulated, continuing with this research was, what if we make some sample counts for new locations? So for example, if I want to estimate the traffic in front of your house, Chris, the cycling volume in front of your house, we figured that if we would maybe put you, someone else, on an automated machine there to count the cyclists, and we were to count the traffic for 10 days.

And then combine it with our models, we're able to get estimates with an error of only 17%, so really rather low for complete new locations. And this gives us a good estimate of how much cycling traffic we actually have in every single street of a city.

Chris Adams: Ah, I see. Okay, so you're using the machine learning model, basically, to make the extrapolations when there's, there might not be so much data, give you more accurate so you can say, "well, I'm more confident that this many people are trying to get around using an active, non gas burning form of transportation," for example, right?

Silke Kaiser: Precisely,

Chris Adams: Ah, okay. All right. Thank you for clearing that up. And I understand. And just, we'll come back to this a little bit later, but you said you're using ML. not the same as generative AI or something like that. That's a different, there's, a whole flavor of different things you might be using there, right?

Silke Kaiser: precisely. I mean, there's many different models out there. In this paper, we used rather simple machine learning algorithms, nothing comparable to maybe what most people think of when they think about chat GPT or whichever generative AI you might think about. Those are really rather simple models making usage of the data we have.

And those models So, I just tend to have, I sometimes, you know, sometimes I think about, I mean, you can target these problems with very complicated algorithms, but sometimes using rather simple algorithms might just be sufficient. And that's what we actually found in this research.

Chris Adams: Okay, alright, thanks. When I was doing a bit of research, I realized that you also wrote a piece around, I think in a publication called Catalyse, where you're talking about active transport and why it's important for greenhouse gas emissions, because you've spoken about like mode shift, which I assume means basically moving from a Being in a car to moving and maybe a active transport, which people use like scooters, bikes, stuff like that.

That's what I think you're referring to there. Could you maybe talk a little bit about how this research actually contributes to like the adoption of cycling as one of the potential modes, because you spoke about it in Paris and I went to Paris and was being pretty transformational when I was there compared to being a few years back.

And I also bike around Berlin too. So I have a vested interest in learning this. Maybe you could like lay out why that's actually benefiting and how active transport helps. Basically, us meet our climate goals.

Silke Kaiser: Absolutely. So, when I, Referring to other research that I've read, research that I haven't done myself, but it is out there and it's been cited a lot, is that we do see that cycling has numerous benefits. It benefits your individual health. If you cycle, it's good for your physical health and then all the sicknesses or illnesses related to insufficient physical health.

We can also see that if you cycle, it's also good for me because then generally we see a reduction in noise and air pollution in cities. So it really benefits the public health, the broader public. And then yes, absolutely. I mean, I did read the IPCC report, which is a report on climate change and it comes out, the last one came out in 2023.

And what they found is that actually 15 percent of net global greenhouse gas emissions are related to transport. This of course includes all kinds of transport, but also, one of them is urban transport. And then switching, there are many levers how to tackle this, right? And within general, as in climate change, there's no one solution fits all, but switching from motorized traffic to cycling is one of those means to actually reduce those greenhouse gas emissions.

And coming a bit back to my research as well, what we find also in research and science is that all you can think about, you know, talk to your friends and family and gather some anecdotal evidence, you'll probably find that one of the biggest deterrents that keeps people from cycling is that they actually are afraid because there's not enough cycling infrastructure.

They're afraid of accidents. And that's a relevant fear. We do see many accidents in cities right now, mainly between motorized traffic and cycling, but also all kinds of other accidents. And what we do can do in cities is, to actually promote cycling, is to build more attractive infrastructure for cyclists.

This can include bicycle lanes, a better design of roundabouts. And all this attracts people to cycle, but, it actually also reduces the risk of serious and fatal accidents. So what I really try to do with my research is that again, if we have these heated debates in cities, how we want to distribute space among cyclists, cars, delivery trucks, etc.

I'm trying to provide data on where we actually have how many cyclists in which street, so that when policy makers or transport planners come around, they can use my data and actually make fact driven decisions when and where infrastructure benefits the most, the greatest number of people. And that's how I hope my research can contribute 

Chris Adams: Ah, I see. Okay. So that's really helpful. And I think there are some kind of comparisons I can make, which make to maybe help me understand if, and some of the listeners. So you know how a couple of years ago, in the middle of the pandemic and COVID, one way to reduce the number of COVID cases was just to reduce the number of people taking tests, right? You know, that's not necessarily the best way to solve that, know, and it feels like

we've got a similar situation. We've got a data asymmetry problem here it looks like you're doing some work to address for that. I mean, Also, you've spoken about, as I understand it, there are various parts of, like, our economy which are easier to decarbonize than other ones.

Like if, for example, in Germany and in America, transport's the biggest, one of the places where we've seen not so much progress on reduction, on carbon emission reductions compared to things like the energy sector and stuff, which is decarbonizing relatively quickly. So this is what some of this is a reference to. Okay, so what we'll do is we'll share some links to Catalyst, I'm sorry, Catalyse, the paper there, and also some of the papers that you have. So we spoke about Paris, and we've spoke about Berlin, where we both live. Are there any other places you would point people to as examples of, okay, this is what good might look like, and this is one place which actually has quite good data to show where you've actually seen quite effective policymaking to kind of change the environment to make it easier to cycle? Because, yeah, not everyone wants to become a MAMIL, like middle aged man in Lycra wearing the helmets and everything like that.

Silke Kaiser: So, I mean, there are definitely some cities that you know that are popular for cycling, for example, just earlier this spring, I did a research day for some months in Copenhagen. And obviously Copenhagen is a bit of a dream for cyclists, right? I'm not the first one to mention this. And then there are other cities, Amsterdam, you name it, but generally I do have to admit that in my research, I haven't really come across cities that do have much better data.

I would say it's a grasping problem across different cities that data is missing. Copenhagen and Amsterdam have taken political decisions to prioritize cycling, but I do have to admit that I didn't, I haven't come so much across that they've made this as a data-driven decision, but this was more of a political decision.

Chris Adams: Ah, okay. And there's one thing I'll just ask before, Chiara, I'm, okay, I am a closet energy, well, not very closet energy nerd. I'm totally gonna, looking forward to talk about that. But Silke, I was just going to ask you, so you mentioned use of Strava and you mentioned the use of, Okay. It's useful to have these new sources of data, but there's also a question about the provenance of that data and like the circumstances under which it's collected.

So for example, we've seen Strava used in lots of other places and if you're using Strava, you tend to be a bit richer, a bit younger, a bit healthier than most people. Maybe you could talk a little bit about some of that, because there are various sources available to inform these policy sessions, and like, Strava is one.

But like, where else, like, assuming you had, you were suddenly queen of the world, where would you wish you could get some of the data from to kind of inform this in future?

Silke Kaiser: So you're absolutely right, Strava definitely is quite biased. It's the data, for example, for Berlin, I definitely know that they're all male, they're mainly male, young, and they do tend to do very sporty biking in comparison, for example, what I probably do to commute work. So it is true that some of the data we use is biased and we're trying to balance this off with the other data sources that we're having.

We're also taking socioeconomic factors into account because obviously we do not want to have, infrastructure is meant to be there for everyone and not for privileged or less privileged people. It's, meant to be equal for everyone. But then obviously I thought about a lot, well, how could we actually improve the data availability in cities?

And I definitely see two levers that we have. Well, first of all, we can place more cycling counting stations. That is a bit challenging because, for example, we have so many kilometers of roads and it's hard to track them all. There are cheaper options than the ones that we're currently employing. So this might be our one option.

Some of them are then using cameras, for example, that are just much cheaper to put them out there. And then the other question, and that's something I'm also looking forward to, to answer this question is because I'll be looking at this in my future research, is that actually how can we place the sensors that we have better across a city? Because currently look it up again for your city.

You'll probably find a similar image is that we do tend to place these censuses as very busy and scenic roads. And the question is actually, can we maybe place them at more diverse spots within a city? And if yes, how can we choose those streets to place those sensors at to actually get a more comprehensive image of the cycling traffic and then also of all kinds of socioeconomic areas and a more equal data image.

Chris Adams: Cool. Thank you for that, Silke. All right. We'll come back to you a little bit later about some of the specific techniques that we were using, because we spoke a little bit about ML and there's a lot more we might dive into there. Chiara, if it's okay, can I ask you a little bit about your research analyzing European energy data?

Because you didn't hint a little bit about how this can affect renewable energy transition, and one of the, one of the things that Germany has in particular is a target to have 80 percent of the grid running on renewables by in, wow, in five years. So that's not much time, and we've also spoken about on the grid, we've spoken about things like time shifting and location shifting as a kind of carbon-aware software, particularly in changing how data centers, like, fit into the grid, I suppose, or the energy they use. Can you maybe talk a little bit about some of the challenges you've found actually working with this data hands on? Because we At best, most of us developers, we might use it in a really nice, pretty fashion from electricity map or Watttime or in an SDK, but it sounds like you're pretty much at the front end having to figure out how the sausage gets made. So yeah, if I can ask you, maybe you could tell us a little bit more about how this data comes about and what are some of the challenges.

Chiara Fusar Bassini: I mean, if you've been using Electricity Maps, you probably have been using an application that in the back uses NSOE data. So you, in Europe, we are rather lucky because. There have been two regulation that have been released in 2011 and 2013, which forced in a way transmission operators to publish a variety of a time series of energy data from the grid and from the markets in an effort to increase transparency. And we have a number of data in a central, on a central repository, which is called the Transparency Platform. We have load data, we have generation data, we have transmission data about the grids, we have balancing data, balancing markets, but also a lot of information on individual power plants. This data is overall extremely useful, but unfortunately it's not Always accurate and it's not always complete and not all the data is not always published in a timely manner.

That very much depends on the type of data, the country itself. We are still very much better off than other markets where there's no data at all. But it's still an issue of like how good and like what the data quality actually is. Because you mentioned time and location shifting. To do time and location shifting, most likely you will be working with aggregated data.

For example, load data, load data, load forecast data. And. One could analyze, for example, a load to decide whether to shift more energy consuming activities at night or at moment where there are off peak time windows. And on the other hand, one could look at aggregated renewable generation data to try to relocate some more of energy consuming activity to time of the day where the grid is actually greener and there are a lot, there's a lot of interest in academia, but also in, in the industry sector to provide us information, to have an estimate of carbon intensity. there are a number of startups out there. You mentioned Electricity Maps, but also academics have come up with top, bottom up, top down and bottom up approaches to compute really at hourly or quarter hourly level, these carbon intensity estimates. The trick here is that you are working with, the aggregated data and aggregate the quality of aggregated data and the timeliness of aggregated data is rather high, the situation is a little bit different when you move to a more geographically granular, like a higher geographical granularity.

Chris Adams: Okay, so from Germany going to like Berlin or Germany going to another part like Frankfurt, for example, something like that. Yeah? 

Chiara Fusar Bassini: Rather, when you're looking from, the aggregated generation to the generation of individual power plants, because in that sense, you might be interested to know which power plants are actually generating right now. And you might be interested to know which areas are generating more solar, for example, which areas are generating more wind. Unfortunately, we don't have data on all power plants. Which would be rather impossible in terms of like amount and extent of the data, but we have only data for power plants that are at least 100 megawatt. It's mostly conventional power plants. So, for example, we have no individual information or very little information on wind farms, for example, because Some of them are not big enough to qualify for this criteria. And also this data get published with a significant delay of four days so that you can't really use it to

do anything operational. It's also not conceived for that. And we can or we cannot use the data to properly, like we cannot do it for, we use it more for analysis than for forecasting, but nonetheless, we can use this data to understand a lot on individual power plant data, why they decide to dispatch on how they are dispatched, and especially in the context of conventional power plants, how their dispatch has changed over time because of political reasons, but also because of the increase in cycling of fossil power plants, because they have to adapt to the renewable energy generation, to more renewable energy generation.

Chris Adams: Can I just quickly stop? I just want to check I understand some of the terms you've used for listeners who might not be familiar with load, cycling, some of these things here. So when you talk about loads, you talk about energy, like basically that's power draw, what people are trying to draw from the grid.

You mentioned that. And then you also mentioned, I think, like cycling. So that's like basically scaling down a power station in response to there being loads of wind on the grid or stuff like that. Maybe is that about right?

Chiara Fusar Bassini: Yes. Thanks for asking. So to clarify, so load means the demand and in the past, I mean, demand, especially from, industry, but also from household, it has been rather predictable. And the way we faced demand, or we satisfied demand, because well, in an energy system, demand and supply need to be equalized at any time. In the past, the most of the baseload, so the main bulk of consumption, have been satisfied using traditional fossil fuel technologies, so called dispatchable, because you can, decide when and how to switch them on and off.

Chris Adams: Ah, okay.

Chiara Fusar Bassini: But the thing is, as more and more renewables enter the grid, they cannot be dispatched whenever, they can only be dispatched, or they can't choose when to dispatch a

Chris Adams: Yeah, control the sun and the wind. We can just respond. Yeah.

Chiara Fusar Bassini: They just respond to external weather factors, right? But that also implies that we still have these conventional power plants that are

dispatchable, but we now have to operate them with increased flexibility. So they have to be able to ramp up and ramp down as the load is more and more satisfied by renewable energy sources.

Sources when they are there. And we always make the assumption that say conventional power plants are a hundred percent flexible, but that's not actually the case. For example, some power plants, when they are turned on, they have to generate a minimal capacity. And if the demand for that capacity is not there, that might be an issue, or they might have some minimal times to be switched on and switch off.

So there, there is a plenty of interesting question that arise from the increase of renewables. Like how will conventional power plants cope with more renewables in the grid?

Chris Adams: Ah, I see. Okay. So one thing you're saying, she's like, yes, it's not like a computer. You can't turn it on straight away, like in milliseconds. And so that's one thing you mentioned and the fact they need to do that more is another issue. And if I understand it, what you described was quite a physical process.

It's like, we're not using bits, we're using atoms, like burning coal, things are expanding and contracting. Like, is there a risk that, you know, the, a big power plant could be damaged some more, or does that introduce any wear and tear when people need to scale something back? Because I can imagine someone saying, "hey, you're making me change how I do things, and therefore you're introducing some risk into this.

That's not what this was designed for in the first place."

Chiara Fusar Bassini: Yes, that's actually a very interesting question. What I mentioned, cycling, meaning that you operate conventional power plants more flexibly, has some consequences on the lifetime of power plants, especially if you keep on turning it on and turning it off. There are some wear and tear indeed for thermal power plants, wear and tear consequences, some of some power plants may not even be able to do so because they have some agreements with O&M managers that tell them, you know, "you can do that, but then you'll have to pay more because you will have, we'll have to do more main maintenance." And also, there are a number of obstacles that arise, especially for older power plants that have not been conceived with this flexibility option in mind, but rather to satisfy baseload.

Chris Adams: Ah, okay, thanks for clarifying that. And you're essentially doing some of the research to see how you might predict some of this better to either reduce or basically accommodate some of these changes that we might have when we've got a much more dynamic grid that is influenced by the sun shining and the wind and all the things like that, right? So maybe if we can talk a little bit about some of the techniques being used to track this and reduce the amount of, maybe, reserved capacity that needs to be done, or reduce the amount of wear and tear that might be imposed on the kind of entire system full of all different power generation. You said you spoke a little bit about using machine learning, and We spoke to Silke.

Silke mentioned that she's using some ML models, which are not like generative AI. That's a very, it's a different kind of AI. Could you maybe talk a little bit about like how that gets used, because you hinted at it, and like what some of the barriers are for using some of that, because it, that sounded quite enticing and interesting.

Chiara Fusar Bassini: Yes, before, before that, I might, I want to add on this. So there is some parallel research being done, especially like at engineering department of a lot of engineers trying to use machine learning to efficiently operate conventional power plants to reduce this wear and tear of wear and tear problems.

And in general, like damages from cycling while still satisfying a change in demand. What I'm doing is and Rather different analysis of historical usage of power plants. So to see how power plants are, have actually been operated so far in the markets, how they're, how they, how flexible they actually are.

Because sometimes we assume that they're 100%, again, we assume how they're 100 percent flexible, but how flexible are power plants that we already have in the grid? And also how available are there in cases of outages, for example, how, like. What's the percentage of time in the year that they actually could provide electricity, for example? And in terms of techniques, well, it's a lot of time series data, so most time series apt methods can be used here. It very much depends on the ultimate task, but one of the major obstacles I encounter is that this high granularity data is by far not as good as the aggregated data, especially, for example, an availability of power plants that has to be reported in a rather accurate way, but then is not one to one translatable to time series format because it's published as market messages, meaning that the data that we have is not in a format that makes it directly usable for a researcher. So there there are a number of obstacles that are really determined by the data quality rather than by the task itself.

Chris Adams: Ah, okay, so it comes down to the data a lot of the time then, basically, yeah?

Chiara Fusar Bassini: Again, like as Silke said, sometimes it's really just a matter of the data that you have, like the research that you can do is going to be determined by the quality of the data that you have.

Chris Adams: Okay, we'll touch on that a little bit later, but I guess the, that does make me think about, particularly in Germany and countries where we've seen very rapid changes. Like, Germany, there's, you know, there's a massive craze of balcony solar, for example, or we've seen loads of battery coming onto the grid, or even Pakistan. We've seen, like, a third of the power, the new power introduced this year, was come from rooftop solar, and each one of those is individually less than 100 megawatts. That's an enormous chunk of power. So there's all this new stuff that we're not, don't necessarily have access to the data for to actually figure out, okay, how will the grid work and how can we make sensible predictions on this? That's useful to know. Brilliant. 

Okay. So we're speaking a little bit about the upsides and how, where some of the potential might be. We do speak about Green Software, about reducing the environmental impact of some of this, and obviously when we're doing some of this work, I've asked a little bit about the kind of models you might be using, partly because there's a question, whenever we start using technology to help us meet climate goals, it's when some of that energy is still coming from burning fossil fuels, for example, there's trade offs to be made. Does anyone want to go first, talking about how we think about these trade offs? Because as practitioners, I imagine you're at the coalface, but you're also working with some of the people who think about this every single day. And like, if you're working with Lynn, and like, Lynn was one of the founders of Climate Change AI, I reckon she probably has some reckons and you've probably had some conversations about this, right?

Silke Kaiser: Absolutely. Actually, just, I think just the last group meeting we had, we just discussed about precisely this topic, because it obviously, it is a question that keeps coming, a question that we do want to answer. And it is also like, it's in our minds, right? Because if we want to do something positive for the climate, and then actually, the net result might be negative, because our models consume that much energy.

This definitely is a topic that we think about a lot, I would say. I see Chiara nodding. I think she's agreeing with me, but, and I can see that maybe to the outside world, often this can seem a bit like, I like the term of fighting fire with fire, you know, you, you're trying to make it better, but you're making it maybe even worse, but I think if we make some smart choices along the way, I rather like to compare it to the idea of fighting a forest fire with a controlled burn. So right, that we do try, for example, in, in the models that I was employing, I did partially check how much, how big the energy usage was.

I was using simpler models, as I mentioned earlier. So the energy consumption wasn't that high, but I think it's good for us and for everyone out there using, similar models. To track your energy consumption and there are very nice packages and libraries out there, tools, all kinds of things, open source, freely available that are very good in, in managing or in measuring the energy consumption you have.

And then of course there are a whole bunch of other approaches that you can take. Right? I mean, you mentioned it's an issue if it comes from fossil fueled energy, but obviously you know, you can think about, I know that there are a lot of like. Service and data science centers, for example, out there in Iceland, where you tend to have more natural cooling, where a lot of the energy being produced is renewable.

I'm not saying that at all perfect, but what I'm trying to say is that there are different approaches that we can actually also reduce the emissions caused by AI. 

Chris Adams: Ok, so there was one thing about the actual technique, like, AI is not a monolith. There's all different approaches within this, in some ways, not particularly helpful term, like the use of relatively small machine learning models, which are relatively simple, that's going to have a totally different footprint to the model used to generate SOAR, like a video or something like that.

And that's something that we probably would benefit from having a better kind of intuition off as practitioners, for example, and you spoke a little about the carbon intensity. So there's two, two things that you have there. And you mentioned some software that you have. And you said that, Chiara, if I can kind of give the floor to you, because I think you mentioned you, you've spoken about some of this before about, yeah, there are some tools and I use them as well. Can I ask you a little bit about when you've been thinking, I mean, how do you think about these trade offs? Or is it a trade off? Or is there another set of dimensions you might be thinking on rather than like forest fires and controlled burns, for example?

Chiara Fusar Bassini: Yeah, I think, there are two things that need to be thought through when using AI. Number one is like, how do you develop your model, and then what do you use your model for? So how do you develop your model? That's similar to what Silke said, for example, doing emission emission tracking while developing the code and while training the code. And at the moment, I think AI is missing some embedded indicator of the social environmental cost of the training. So. We kind of think of performance metrics such as accuracy, such as like classic cross entropy losses and so on, and we think only about precision. But sometimes we need to be a little bit more critical of whether an increase of accuracy of 0,1 percent is worth an increase in the training time of two hours or an increase in the size of the model of 25%. These are like actual numbers and scientists have coined for that the term green AI, meaning Okay, can you know, can we, in a way, embed this measure of the size of the model within the loss that we are trying to minimize in the training of our model? There is another, a good example, for example, is the Bloom model that is an alternative large language model to GPT. It is similar in size, but it required Like the CO2 emissions of the model are 20 times lower than GPT 3. And this has been made possible by, first of all, in smart usage of the training and also tracking of the carbon intensity of the grid. It system was trained, the model was trained mainly in France, which is, which has runs predominantly on nuclear power. So in like carb, much more carbon neutral system. So there are a number of things that one can consider while training their model. But also another thing that is very important, and I think that we sometimes don't really think through, is what are we using AI for? And in that sense, there is currently no standard assessment in place. Like, is this application really worth using AI? AI is, by its nature, ethically neutral. It can be used for anything from targeted advertisement that will have probably a negative impact on environment to detecting wildfires. So very positive impact. I think policymakers in that sense can make a great deal to really make a difference and start, for example, by providing a classification of which user cases are positive for environment and which are negative. It sounds, it may sound like science fiction, but it has already been done in the European AI Act in looking at the perspective of risks, like which application have a high risk and hence should be more controlled and which other have lower risk. And I think a similar classification would be also very useful for environmental purposes.

Chris Adams: I'm really glad you mentioned that because I ended up reading through the AI act for research recently. And the idea of the risk that is, you're right, there doesn't seem to that much be that much reference to the use of AI for, let's say, you know, increasing the extraction of fossil fuels, right, versus that.

That's, there doesn't seem to be much to mention about that, but there is some information about the transparency around training. And now that we've looked at it a bit closer, so within the Green Software Foundation, there is a group called the Real Time Carbon Group. We've been looking into some of the specific implications of this, and it looks like the AI Act, it also, it looks like it's probably going to suggest not just understanding the training, but also the cost of inference, like the use of the model rather than just the training of the model. If I can just quickly, you've, you, mentioned there are tools out there, and Silke also mentioned there are tools out there. If I did want to measure some of this, and if I did think there was some legislation coming for this, what tools are there available for me to measure the direct impact? So at least I know what the trade off might be.

So we understand that the carbon footprint of decarbonizing transport, like Silke mentioned, that's going to be, you know, positive, but quite, but there's ways of calculating that, but for us as practitioners, are there any software or any tools you might recommend that are kind of common in the field now? Either goes. I'm happy to, whoever's more comfortable talking about this.

Chiara Fusar Bassini: I'm thinking CodeCarbon is more probably a standard used by many scientists. I know there are more applications that might have a higher granularity, but I guess that's a

Chris Adams: That's the one that you folks have used, right? Okay, I hear CodeCarbon used a lot, and I, as I understand it, that's the one that's been used for the Bloom model when they wrote a paper about that. That's what I'm not sure Facebook have actually explained this because when I was looking at LLlama's model, so AI models have model cards, which basically, which I think, various responsible practitioners now say, "this model took this much carbon, or they had this much energy gone in to kind of create it," for example, if you go to the existing Llama 3.1 model card on Hugging Face, and you try to follow a link to the actual methodology, It's not actually explaining how it works. there's now a bug. I filed a bug to ask out, ask, well, how did you work these figures out? Because these feel like it's quite important, especially because when you look at the numbers, it's significantly larger than Bloom, basically. That's, so, so what you're referring to is CodeCarbon. That's one tool that people can use that will give you some idea that is in use in a few places already that's relatively safe to start off with. Great. Okay. And we spoke a little bit about some tools. So if someone is, wants to take their first steps, they might look at this.

And there are various projects I'm aware of to make it a bit easier to understand the impact of one versus another. I believe there's one Energy Star AI or Energy Star, AI Energy Star or something like that. There's one person who I've spoken to who's involved with it. Boris. I'm so sorry I can't pronounce your surname, but I do know you're the AI lead, the AI sustainability lead at Salesforce. Boris G is one of the people who's been writing about this. He's not the only author, but he's the person I know, and we'll share a link to that as well, because that's the first thing I've seen of a useful, like, A kind of nice idea to give you an idea of what the inference, the usage as well as the training might actually be. If you were to look at this, we've still got this issue of data or having access to data like, and Silke, I ask you, if you were like queen of everywhere for a moment, how would I change it for here, right? Let's say that you want to be responsible AI practitioners, like what are the things that we need to see in the next, in the coming years to make it possible to be like responsible practitioners so that when we do use AI, we're using it in the kind of greenest possible fashion. Silke, I asked you first about Queen of Everywhere, so maybe you go first and I'll hand over to Chiara.

Silke Kaiser: Well, that's a very good question. I definitely say, as in general, with all kinds of, you see, in more technical approaches, we do need reproducibility and traceability of what we do in our research. I mean, just as you mentioned with the Llama, I think it's important that other people are also able to understand what we did, what was the energy consumption of what did, how can they, how can we check the things that we've done and, see if, we did it properly, if it took a right approach?

And then obviously, I think this is a bit less related to, the topic that I'm or that Chiara was working on, but also in the longterm, we do need to think about ethical concerns coming down to this. And then again, I think just, really, transparency. So I really think that transparency is a good way to address this.

What take do you have on this Chiara? I think one of my major takes also from what Silke mentioned, and I'm really glad you mentioned, is the fact that when we were talking about policy making is that very often policy making is not data-driven. One problem is that we don't have the data and it can be addressed partly by regulators asking for those data, right? But another issue is also that we don't really do data-driven assessment of the policy that we implement then. And I came across very recently a paper that tries to systematically evaluate policies.

And having been implemented in the last 25 years, this very recent paper has been published like a few days ago.

And I thought it was very interesting to well, once again, the results of the paper is there's no one size fits all and some countries depending on their level of development might need different policies. And we have to keep that in mind that we can't use the same policies for a developed country whose energy consumption, for example, is no longer linearly dependent on its GDP from a different, from a developed country or a developing country that has very different issues. But I think in general, this approach of doing data-driven policymaking and science-driven policymaking is something that would really, we would really need in this space.

Chiara Fusar Bassini: I think that's something I can really agree on. I often feel that as a scientist, we feel like we're trying to really produce clear results, objective results. And then often we feel there's maybe a bit of a lack between the two. The research that we do and how much this is sometimes uptaken, by policy.

And obviously we hope that because we do really put so much effort into this and always try to be objective. We hope that this will eventually be more used more and more in the policy sector.

Chris Adams: You've touched on a really interesting point, and I can think of some examples that just occurred to me. So, we had an interview with, oh, Vlad Kor, his first name is Vlad, I'm gonna mispronounce his surname, but we spoke a little about, all about the rebound effect, and Vlad Coraoma, that's it, Vlad Coraoma had this lovely post actually on LinkedIn talking about the curse of potentialities, potential itis, which is basically talking about, we have all these kinds of really exciting projects, but whether people follow through to check whether the actual gains materialised, or the benefits materialised, there's much less effort put into that.

And we've been seeing, like you said, Chiara, from the last 25, we've been seeing predictions for things that would happen in 2020 or 2030. And 2020 is in the past now, we can check if this is, if these actually delivered, but a lot of the time we do not see that. And in our field, specifically as kind of cloud providers, or people who might be consuming services, there's some, there's a really, I'm thinking of a really good example. Microsoft has a whole thing about pushing for AI and everything like that. And we know that, as you mentioned, AI can be used for good, and can be used for bad, or used for Climate aware things, which are really helpful and things are not so good. And we've even seen like people who are workers really pushing for this. I'll share a link to an article in Grist where, written by Maddie Stone, where she talks about some of the sustainable connected community inside Microsoft, speaking to some of the management there. There's a guy called Darrell Willis. He's the vice president of energy. And they spoke and said, "hi, we are pushing for," you know, "can we please have a conversation about what we're using AI for inside our company, because we're one of the largest companies in the world and we're one of the leaders in various industries," right?

And there was a commitment to say, we're going to produce, as the management said, "we're going to start releasing information about, okay, how much of our use of AI is coming from the fossil part of the industry versus the renewable part of the industry?" And this feels like a really important data point if we're going to be looking at tens of billions of dollars used on AI.

I mean We know that it's an accelerant. If it's an accelerant of fossil fuel extraction and burning, that's a very different story to using tens of billions of dollars for renewable energy, for example. And if we've seen commitments at a management level, then it would be nice to see these. As we understand, these commitments were made, these were shared inside the team, but we don't have this, and we'll share a link to the specific terms, because actually, I'll just share the quote with you, because I think it's one thing that, if you're an employee of a cloud firm, or a customer of a cloud firm, it's the kind of thing you might want to know about, so on the call, "Darrell Willis, committed to providing employees with updates on net zero requirements as Microsoft continued to implement these energy principles. Committed to providing a breakdown of energy divisions revenue across six different sectors from oil, gas extraction, to all zero, low to zero carbon energy. So sharing this information internally." Now this feels like a thing that employees probably should be aware of or asking for. Also feels like something that if you're an investor of Microsoft or a customer, you might want to know.

Because there's an impact inside your supply chain thinking about this. And if you're choosing one provider because they have really strong GSG credentials, this may make you view it somewhat differently. We'll share the links because it seems to be the best concrete example I can think of at significant scale that we might be talking about. And I'll get down on my soapbox because that's just the thing that really leapt out when I, when you spoke about that. So we coming up to time, and we've spoken about the different uses of AI, sustainability of software, as well as some of the Things you might want to use or be aware of as a practitioner. If people do want to find out about the work that the two of you are doing, where should people be looking? So Silke, if people are interested in your work, is there a LinkedIn page or is there a website that you direct people's attention to?

Silke Kaiser: I normally try to direct people to my personal webpage, which is silkekeiser.github.io. Or you also, you can also find me on X or on LinkedIn. And I'm always happy to share news on my research as well as the articles that are out there. And I'd be happy to, if people were to look at those pieces of information.

Chris Adams: Cool, thank you. Alright, and Chiara, if I just hand over for you?

Chiara Fusar Bassini: I've seen Silke's website and you guys should really see it.

It's a very nice animation. I don't have myself a website, but I'm very active on LinkedIn. You can find me under Chiara Fosar Fusar Bassini.

Chris Adams: Chiara F U S A R, we'll put it in the link, we'll add it in show notes. So, Chiara Fusar Bassini. Brilliant. Thank you, folks. This has been lots and lots of fun. I've learned a lot from this, and this has been a really nice chat. Hopefully, we'll cross paths sometime in Berlin, but otherwise, thanks again for coming on, and have a lovely week.

Silke Kaiser: Thank you very much for having us.

Chris Adams: Ta ra! Hey everyone, thanks for listening! Just a reminder to follow Environment Variables on Apple Podcasts, Spotify, Google Podcasts, or wherever you get your podcasts. And please, do leave a rating and review if you like what we're doing. It helps other people discover the show, and of course, we'd love to have more listeners. To find out more about the Green Software Foundation, please visit greensoftware.foundation. That's greensoftware.foundation in any browser. Thanks again and see you in the next episode.

TWiGS host Chris Adams is joined once again by Gaël Duez to discuss the latest news in green software around AI. They discuss insights from recent reports by Google, Meta, and Amazon, as well as looking at the implementation of the GSF’s Software Carbon Intensity metric. Similarly, the conversation touches on the distribution of renewable energies and the use of different means of measuring carbon in reporting, and how this can affect the behavior of consumers and organizations alike. Tune in for an enlightening discussion on the latest in green software.

Learn more about our people:

• Chris Adams: LinkedIn | GitHub | Website
• Gaël Duez: LinkedIn | Website

Find out more about the GSF:

• The Green Software Foundation Website 
• Sign up to the Green Software Foundation Newsletter

News:

• Green Coding AI [05:02]
• Sustainability: How Did Amazon, Azure, Google Perform in 2023? - The New Stack [19:33]
• Big Tech’s bid to rewrite the rules on net zero [34:20]
• Can AWS really reduce the carbon footprint of AI workloads by 99% or is it 'greenwashing'? [43:36]
• Are the CO2 emission reductions promised by cloud providers realistic ? | Boavizta [44:44]
• Mark Butcher on LinkedIn: #sustainability #scope2 | 58 comments [45:09]
• AWS, Digital Realty, Google, Meta, Microsoft, and Schneider Electric call for greater supplier transparency on Scope 3 emissions 

Events:

• Green IO London 2024 [1:01:15]
• CNCF Cloud Native Sustainability Week 2024 [1:05:48]

Resources:

• The Week in Green Software: Complex Carbon Accounting with Gaël Duez [03:54]
• EcoLogits tracks the energy consumption and environmental footprint of using generative AI models through APIs. [06:43]
• https://hotcarbon.org/assets/2024/pdf/hotcarbon24-final109.pdf [11:02]
• Making an image with generative AI uses as much energy as charging your phone | MIT Technology Review [15:21]
• Energy Star Ratings for AI Models [15:53]
• http://CodeCarbon.io [17:02]
• https://github.com/mrchrisadams/djangocon-eu-2024-workshop/issues/1 
• Marketplace · GitHub [18:40]
• Why have Irish energy companies been told to drop 'misleading' 100% renewable claims? | Euronews [30:24]
• RE100 TECHNICAL CRITERIA [31:04]
• https://github.com/Green-Software-Foundation/real-time-cloud/issues/59 [31:42]
• ISO/IEC 21031:2024 - Information technology — Software Carbon Intensity (SCI) specification 
• Asim Hussain on LinkedIn: Hey, so according to a new carbon offset standard (CCP), 32% of the… | 27 comments [50:41]
• Making sense of the energy reporting deadlines for datacentres in Europe 
• The bigger picture: A data centre revolution - Green Web Foundation [1:00:20] 
• Green Coding 

If you enjoyed this episode then please either:

• Follow, rate, and review on Apple Podcasts
• Follow and rate on Spotify
• Watch our videos on The Green Software Foundation YouTube Channel!

Connect with us on Twitter, Github and LinkedIn!

TRANSCRIPT BELOW:

Gaël Duez: We cannot fully rely on companies agreeing on how they should measure their own environmental impact, even if they are well meaning, with tons of great people trying to do the right things, etc. It's not a black and white world out there, but there is a question at some point of financial pressure, shareholder pressures in many of these companies.

They're just stronger than the entire stakeholder's pressure.

Chris Adams: Hello, and welcome to Environment Variables, brought to you by the Green Software Foundation. In each episode, we discuss the latest news and events surrounding green software. On our show, you can expect candid conversations with top experts in their field who have a passion for how to reduce the greenhouse gas emissions of software. I'm your host, Chris Adams. Hello, and welcome to Environment Variables, where we bring you the latest news and updates from the world of sustainable software development. I'm your host, Chris Adams. Round about this time last year, Gaël Duez, the voice of the Green IO podcast, came on to Environment Variables, This Week in Green Software to talk tech, sustainability, and moving from France to live on La Reunion Island.

And the ups and downs of consulting remotely on digital sustainability from a small island off the coast of East Africa. It was a fun chat. And when I asked him if he'd be up for coming on again to review a few of the latest stories around green software, he basically said, "yes, Chris, but I can't do Friday because I promised my daughter we'd pop out to go whale watching." As one does on a tropical island, I guess. So here we are, recording on a Monday morning instead. Gaël, thanks so much for coming back on. Can I give you a bit of space to introduce yourself and what you're up to these days? Also, how are the whales? 

Gaël Duez: Oh Thanks, Chris. It's great to be back again on Environment Variables. And the whales were there, like several made quite a show, including a mother and her little newborn. Well, little meaning four tons. So it's always impressive to see there's a 20 tons mammals jump in the air like this and cherry on top of the cake, actually, we saw a lot of dolphins and two turtles.

So it was a perfect trip. 

Chris Adams: Wow, I'm jealous. Okay. Yeah? 

Gaël Duez: Actually, to connect a bit more on our sustainability topic, I'm also relieved that the best practices for whale watching are more and more enforced, such as, minimum distance to approach them, turning off the engine when they come in, or direction, et cetera.

And, I also say, on top of the pleasure of, watching them and being on the boat, etc. It's a very positive sign that we can enjoy nature without destroying it. And that's pretty cool.

Chris Adams: Cool. So it sounds like there might be, hopefully, when your daughter's taking, maybe some of her kids for whale watching, there's a chance to continue that, by the sounds of things.

Gaël Duez: Yeah, I'd like to focus on whale and dolphins rather than coral reefs, which are something that really is puzzling me, but yeah, we might, hopefully. And to answer your other question, when I'm not whale watching in Reunion, I try to be useful in the tech community, by advocating for more sustainable ways of designing, coding, hosting, and even considering the use of technology itself. And my main tools remain the Green IO podcast, as you mentioned it, as well as the Green IO conferences, which I started to organize in several cities, but I'm sure we'll get back to this point later. And besides, as you already mentioned, besides as volunteering activities, I do public speaking and consulting on systemic strategy for mostly for tech companies, for both paying the bills, but also to have an impact.

Chris Adams: Cool. Thank you for that. I will also share a link to the previous episode if you're for people who are listening, so they can get an idea of some of the other things that Gaël has been working on and been discussing previously. If you're new to this podcast, I should probably introduce myself as well. As I mentioned, my name is Chris Adams. I am the executive director of the Green Web Foundation, which is a Dutch nonprofit based around reaching a fossil free internet by 2030. And I also work as one of the policy chairs in the green Software Foundation, particularly in the policy working group. Okay, and before we dive into this podcast fully, just a quick reminder, everything we refer to, every story, we'll share in the show notes and there will be a transcript as well for you to kind of search through and look into later. Okay, so as per usual with this week in Green Software, We run through some of the latest stories or projects that have caught our eyes over the last few weeks. And Gaël, I should probably ask you, are you sitting comfortably, ready to go?

Gaël Duez: Perfect. Everything is fine.

Chris Adams: Brilliant. Okay. Let's start with the first story then. So this is the first one that kind of came up on my radar. There's a website called Green Coding AI. And we've spoken about the environmental impact of AI on this show multiple times.

We've also spoken about there is different models that you can actually use to maybe ask the same question and get something back. This one is actually a project from, the Berlin based group, Green Coding Solutions. And what they've actually done is put together a service, running on their own hardware, where you can basically try various, models to prompt them to, like, try out, say, Llama 3, one of the big versions, maybe a smaller version of some of these to see, to ask a question and see what kind of responses get back. And one of the things that's particularly interesting in my point, from my perspective is A, the transparency, but also it gives you a software carbon intensity score for every single inference when you ask a question. So you can start comparing the utility of a large model versus a small model. Gaël, I think you might've had a chance to play with some of this already.

When you looked at it, what, were there any things that sprung to mind for you when you first had a bit of a kind of kick the tires and mess around with it at all?

Gaël Duez: Well, what I was pleasantly surprised with is, as you mentioned, the use of the software carbon intensity. I, as far as I know, this is the first time that I've seen it being used for this kind of tools. I know that what is very interesting is that we can see now, not blossoming yet, but several initiatives have popped up with that kind of approach and I need to shout out and at least I need to give a big kudos to Arne and his team because once again it's very thoroughly done for what they've done in Berlin.

What I also like is the work that has been done with another initiative, and I hope that they will start discussing, with each other, which is, it used to be a data for good, project, but now it's a, an, an association on their own called Gen AI Impact. Yeah. So the association is called Gen AI Impact and they created this tool Ecologits, which is kind of just kind of the same approach.

With a very strong focus on transparency, how they calculate everything and all the code is accessible, et cetera. And there is also obviously Code Carbon by Jürgen Fais, which was kind of the trailblazer, if I recall. And I think it says something positive about the trend in the AI industry, that they need to tool up to Assess more and more precisely what is the carbon impact and hopefully soon the water impact.

And that's still, I know that it's not feasible at the moment because of the lack of transparency around water, but we will discuss water later in this episode. I'm pretty sure regarding the other news that, that we have, but it, is very important. I cannot count how many times I've got potential customers or heads of IT telling me what are the tools we can use to measure the footprint of AI.

And Also to use them to push back a little on the AI hype. everything should be AI'ed. I don't know about it. With a pinch of AI everywhere, but it comes with a cost. Now, going back to your question. So I really love the multiplicity of tools that we see at the moment. And now going back to more specifically in Green Coding AI, I think the way they, try to cover all the angles when it comes to energy, having both the CPU joules, the GPU joules, but also the temperature.

It's very interesting because the temperature, it connects a lot with a discussion I had with Professor Lee. One of the big experts on water cooling, especially in data center in tropical island or in tropical climate, sorry, not island. I'm the one on a tropical island, but there are a lot of tropical climate without tropical island, starting with Virginia in US that we tend to forget.

And he was telling me how important it is to understand that there are thresholds when you use a chip. And there's a threshold that impact massively the energy consumption. So it's not the same to have a chip being used around 25 degrees Celsius, for instance, and 30 degrees Celsius. It's just a linear progression in energy consumption versus, temperature.

And I was really delighted to see the temperature being put as a metrics, because that's where we are getting more and more in details and understanding that it's not that easy, it's not that linear, and we need to investigate things in a more systemic view. And the temperature for many people operating data centers, is absolutely pivotal when it comes to anticipating energy consumption.

So I think, yeah, I was very pleased with this CPU temperature metric that they added. I have no idea how they made to calculate it. So that's something I'm going to investigate, but I like the idea of putting this kind of metrics on the table as well. 

Chris Adams: So, thank you for that, Gaël. So, what I can share with you, there's a chap, one of the chaps working at Green Coding Solutions, Didi Hoffman, was one of the people doing a significant amount of the work on this, and I've used some of their tooling, they've made open source before, and they basically pull some of the temperature figures from the, from basically the CPU itself, I mean from the system that they use, because for example, most machines can tell you how hot or cold they are, for example.

And that's for pulling some of those numbers actually from. The thing that I might share with you that I think is interesting here is, normally, Green Coding Solutions, they tend to open source as much of the work as possible. And they have one of the nicer tools out there for looking at the environmental impact of pretty much any kind of service, for example.

And the thing that So the, I did actually ask Diddy about this and said, "Diddy, this looks really cool. Are you planning on open sourcing?" He says, "yes. The only thing that has stopped me open sourcing it so far is tidying up because it is a bit of a mess." But the thing we can point you to in the meantime is a paper that he worked with, I think Professor Verena Majuntke from HTW Berlin. They submitted a paper specifically for, to Hot Carbon talking about, essentially being able to optimise different models for optimize the use of AI services specifically for particular tasks. And that paper is really interesting in my view, because they say, well, given these particular tasks, some tasks are amenable to running on small models.

And some tasks are better for large models, for example. And they were basically demonstrating how you can pretty much use a system, which looks at the prompt and then will kind of route it to the appropriately sized model to reduce the environmental impact of this. But one of the things they said was that they can look at the energy used when making an inference like, asking for a response back, but it's still a massive challenge to get any numbers from the embodied energy in making some software. And this is one of the key things that we might, maybe wasn't so obvious when we first spoke about this, that right now there are kind of three broad life cycle stages in an AI project. There's a training part where you create a model. There's an inference part where you have the use of the model. And then there's the kind of embodied carbon around this. And this is one of the first projects we've seen that really makes the figures for the inference part quite visible and quite easy to work at, and one of the key differences between this and the Ecologits project that you just mentioned before, was that this is actually running on the software itself, so they have access to the hardware, they're running the hardware themselves, and they, actually document how they do this. The, when I spoke to Samuel Rince at GenAI, I asked him, "how does your thing work?" Because that's really cool, I really would love to see some figures for the inference prices. Well, what we do is we have to make some educated guesses and estimates based on how big a model might be, how much memory might be allocated for it, and how long it's run and how big the response is.

So they're essentially annotating responses that come back from Open AI or from Mistral or from Claude or anything like that to give you some numbers for this. And this kind of does beg the question, if we see green AI instrumenting their own physical hardware, and if we know that the best we can see elsewhere is us having to instrument things ourselves. Based on guesses, why is it such a challenge to find services that provide these numbers as part of how they work? This is still a challenge in 2024. And I think these projects here demonstrate that yes, there is demand and there's also ways to do it. So if these aren't being exposed, they really should, because it becomes much easier to be a responsible technologist if you have access to the figures about the environmental impact of what you're using AI, for example. So yeah, those are the things I'd share, and I'll add some links to that to follow on from this.

Gaël Duez: I wasn't aware that they were using their own materials, and that, that's a massive change indeed, because no more educated guess, as you said, and the ability to measure yourself. Of course, it means that they create their own assumption when it comes to the hardware setup, but Yeah, I really love the idea.

And I have a question for you because I wasn't able to unpack yet all the publications from Hot Carbon. So how was this article aligned or not aligned with the last one from Sasha Luchoni when she was actually testing? So once again, on the inference side, because we know that this is where most of the impact comes from, when she was categorizing the use of a different LLM

in front of different use case and, she was, for instance raising the alarm that for speech recognition text or other, I would say, basic AI solution, using large LLM was a complete waste. And that's something that really struck me that how important it is to pick the right model for the right task and not using LLM, especially GPT 3 or GPT 4 for everything.

So I was wondering, do you know how well that was aligned or not? Their own findings.

Chris Adams: Okay, so it's been a couple of months since I read the paper, or about one month, because the actual paper was released, I think, on the 11th of July or

something. The key thing was this paper from Verena and Didi, the people at the University HTW Berlin,

and, Green Coding Solutions. Their one, while they do talk a little bit about classification of tasks, they don't go into the same detail that the paper from Sacha Luchoni, which was, that was the one that we saw in MIT Technology Review, which very much said, generating an image has the same kind of energy demand as charging a phone, for example.

So they don't provide that same kind of breakdown because their primary focus was working with text. However, there is something we should share a link to, which is some recent work in April where the, where Sasha Luchoni and other people, other luminaries at Hugging Face spent some time working on what you might refer to as like Energy Star for AI.

This idea that for different tasks, you do have models and so you can start making some more informed choices about your choice of model when you're doing some of this stuff. So we should share links to some of these things because

there is quite a lot of work taking place that I think not everyone is aware of right now and it's a really useful kind of jump off point for this.

So yeah, useful questions Gaël, thank you.

Gaël Duez: It connects well with what you've mentioned, that there's a rating should be provided by the one making these models or operating these models. And it's great to see all those initiatives, but eventually, and I think this is a nice connection to the next stories, but when you operate something, you need to be transparent about the environmental footprint of what you do. 

Chris Adams: So this is actually, you're right, that's a very good point, Gaël, and maybe just the final thing I'll share with this before we move on to the next story is that for training, we've seen pretty much the industry start to settle on the use of CodeCarbon, which is an open source Python project that you can use to essentially work out the energy use for any piece of Python code, but specifically it's used for quantifying the energy usage for training.

So, this is what's used for some of the papers by, from Hugging Face. I believe this was also the tool that was used for the most recent information from Meta when they shared the carbon footprint of Llama 3, their big model as well. So you do have some existing tools that are out there. One thing I've actually tried looking at recently was I realized that you can use these tools and there are tools like GitHub and GitHub Spaces which allow you to run a virtual machine, leave and run, like some inference locally, for example, to try and like test something out, for example. And, what I found was that I was preparing a workshop for, to deliver at DjangoCon in June earlier on this year to help people, like, figure out, okay, what's the environment impact of maybe using an AI service or figure out what some of the kind of service side environmental footprints might be. I found that, there's a competitor to GitHub spaces called Gitpod. They use a slightly more up-to-date version of Ubuntu, which basically means that if you're running a virtual machine inside this, you can actually use Code Carbon and get numbers back. But when I try to use this with GitHub spaces, because they're using a slightly older version of the underlying operating Linux that's used for this, you can't get the same numbers back. And I think this is important or worth being aware of because there's a recent release from GitHub, I think. I'm not sure if it's totally available for everyone yet, but there are now some tools specifically for using inference in GitHub spaces specifically. So it'll be really lovely, and I know I'm kind of nerd sniping the GitHub team here, if they could expose some of these numbers, because the tooling totally exists now, and the bar is so low that even just having something like CodeCarbon, which is useful, but has some, when you look into the details, has a few kind of compromises and few issues. That would still be massively more useful than what we have available to right now. So yeah, there's there's definitely worth, there's definitely tools out there that organizations can use to make it easier for consumers like yourself or me to use these in a more kind of responsible fashion. All right. Thanks, Gaël. Should we look at the next story now?

Gaël Duez: Let's do it.

Chris Adams: All right. The next one. This is a story from a former VP of Sustainability Architecture, I believe, at Amazon, Adrian Cockcroft. he's written a piece for The New Stack and the title is, How did Amazon, Azure and Google perform in 2023 Sustainability? So this is a piece by Adrian Cockcroft where he's basically read through the three sustainability reports. And as someone who actually does have a significant amount of context working on the platform side as well as, since leaving Amazon looking at the tools out there and trying to collate a kind of like useful data set with the Green Software Foundation Real Time Cloud framework, he's been able to say well these things which are easy to use this is where some of the data is helpful this is where there are real challenges and he's it's really useful to get to have someone who, in my view is very much seen as like a kind of real kind of trusted message of saying look,

these bits are okay, this is where the bar is really low and we probably should be expecting quite a bit more given the amount of resources available to people. And yeah, I wanted to just check, is there anything that caught your eye or that really leapt out at you when you were looking at this? I really like this piece and I'm really glad it's actually out in the public domain.

Gaël Duez: Yeah, I love Adrian's work. I love his optimism. I'm not, I would be a bit more cautious is when he, well, especially regarding Amazon, but I guess we're all biased at some point. And what caught my eyes was we discuss a lot numbers that are not that easily assessed and separated. The very big first issue that I had is when we talk about Amazon, it's not the main, the same thing that when we talk about Google or Meta, because, or Microsoft, because there is this big on premise brick and mortar, as we used to say, chunk of Amazon's carbon footprint.

And what is strictly related to AWS should be extracted. And that's not the case with all the numbers.

Chris Adams: You're referring to AWS and

Amazon the retailer, like there's

been separate business, that's what you're talking about here. fact they're not breaking down makes it harder to understand, right?

Gaël Duez: It's, very hard because it's a bit like if we were discussing the, Fusabi report of Google plus Walmart or Sainsbury's or whatever, or Carrefour, and, I'm always very concerned about how globalized the data are.

And I've, got some cave hats with what Adrian said. I won't challenge the numbers or his analysis because I think it was well done and the trend are there. Like AWS, Amazon is doing a bit better and Google and Microsoft are slipping up as the, as he mentioned, but, focusing a bit more on Amazon, for instance, there are a few things that I'm a bit concerned about.

So the first one is that they gather everything in this big Amazon and what actually as techies, we would like to understand better is AWS on its own. I think it's big enough to have its own sustainability report. The second is that they continuously provide numbers on the market-based approach, especially for energy. And I think that there are now countless examples where it's not really how Sustainably is done. Sustainably is as much a global matter than a local matter.

And I'd like just to take the example of Ireland. So if you run as many European techies, your instance on AWS, there are a great deal of chance that by default, you will be using the Ireland region. And when you log in the dashboard or your Sustainably dashboard, exactly as Adrian mentioned, you will see that everything is fine.

You're zero percent, you're carbon neutral, everything has been offsetted, and ciao, bye bye, well done, you can, do business as usual. Now the reality of the Ireland electricity grid is that one year ago, in 2023, the amount of electricity consumed by data center equals residential urban residentials.

It means that every houses, every buildings in Ireland consumes now less, a bit less electricity than data centers. So it has put a tremendous pressure on the electricity grid and the Irish electricity grid is not the cleanest or sorry, the lowest carbon on Earth at all. So, technically speaking, when we add resources, when we add instances on AWS Irish region, we are adding pressure and pushing the Ireland electricity providers to emit, to produce more energy, which is kind of high intensive energy.

And now you've got this market based approach, which has is it's prone. I'm not like, it's not black and white here, but saying, okay, but we invested energy elsewhere. And we show it either by a, power purchasing agreement or AAC. And so that's all good on the market because everything shall be offset.

But the local realities matters and that's even more true for water. But let's put that aside for the moment. So as long as they don't at least try to localize a bit more the carbon emission and the related energy carbon emission. I think it will be always very hard to say, okay, the trend is okay, the trend is not okay.

So that's my first issue. My second issue is that, and I think it was our dear friend from SDIA, Mike Schultz, who once said, one of the most precious resource on earth today is renewable energy. Because of course it's growing, but we don't have that much. And we should always question how much we allocate to which use.

And by having this 100 percent focused on offset or net zero approach, that is the one from Amazon, Google, etc. We cannot leave the elephant in the room, which is, but what are the absolute numbers? And when the absolute number are getting higher and higher, almost from a logarithmic perspective, it's almost exponential, not fully, but almost, we should question ourselves, but where is the limit?

Because we do know that in systemic and in environmental ecology, there are, there is always a limit to how much a system can grow. So, that's my two big issues. It's not localized enough. And it doesn't talk about absolute values. It only talks about the potential of things being offset or being carbon neutral.

And we need to think more about when we slow down or even we reduce our energy consumption. That's not on the table at all. So yes, of course, there are a lot of progress being made. They buy a lot of renewable energy, but is it the best use what we can do about renewable energy? And what are the trends? I know I could speak for ages about it, but sorry, Chris, and I didn't even mention water.

Chris Adams: So just, so let's just check if I understand the key things you're referring to there. So one of them is, just to be clear, we're talking about the Republic of Ireland, as in the island of Ireland here inside Western Europe.

That's we're talking about here. And if I understand what you're referring to here, there is one of the big things, big parts of this story this year is that Amazon has made a big song and dance about saying "yes, we are now 100 percent renewable powered for all of our infrastructure." And what you're, what it sounds like you're saying is that The physical reality in Ireland doesn't necessarily match this claim because it may be that the kind of the way people are substantiating this green claim is that they're basing this on credits like renewable energy credits and while these may be kind of considered kosher or like considered like legitimate in like maybe a trade electricity trading market kind of sense, the fact that we don't see the actual location based figures for these data centers brings up all kinds of questions. And also, the, there is also questions about, are renewable energy credits the correct way to actually basically back up any claims around the use of green energy, particularly when we know that the underlying grid, there may be more power being used than it actually, than renewable energy is actually generated in Ireland itself for this, right?

Gaël Duez: Absolutely. And then to, give them credits, they use less and less, renewable energy credits, which are highly questionable tools, and they use more and more PPA, which are Purchasing Power Agreement, where actually they commit to add new renewable energy via partnership, long term partnership with electricity producer.

So it actually increases the amount of renewable energy available for everyone on the grid. So I'm not saying that everything is bad or everything is great. My question is if you, for instance, just staying within Europe, invest in northern Germany in a wind farm to produce that amount of gigawatt of renewable energy, that's great.

That's necessarily, that's something that is very useful for the German market and for German users, but it will not offset the fact that there are still gas and even, correct me if I'm wrong, coal based port plant in Ireland, and that the use, the rise of energy use in this part of the world will emit more greenhouse gases.

So, once again, it's the incredible ability of humankind to tell itself stories, which has made us what we are today, has also a dark side, which is it's not because we decide that we create a fancy story called the market or the energy market, et cetera, that it is completely disconnected of the physical reality of thing, as you mentioned.

And the physical reality of thing is, it's great to add more renewable energy to the grid every day. Anywhere on earth, because anywhere on earth, we need more renewable energy, but it cannot really offset or compensate the fact that if we put some stress on electricity grid somewhere, it will add the emission of greenhouse gases and, eventually everywhere around the world, because I think it's any like carbon molecule that take 15 days to do a round trip.

So, it's a global challenge.

Chris Adams: If I understand what you're saying, basically, the instruments being used do not fully capture the physical realities of what's taking place. And while there may be progress, we probably need more progress in order to actually face the challenges that are being kind of set out by the actual, the real science that we're seeing. I'll share a couple of little points around Ireland specifically before we move on to this. So Ireland is actually one of the few countries where Green and IT claims around green energy have actually been challenged by the Advertising Standards Agency, specifically saying if you're a green energy firm and you're saying you're using green energy. We've, there have basically been cases where the Advertising Standards Agency in Ireland has said, you can't make these claims in Ireland if you're using just renewable energy credits as the basis for making this claim. So that's one thing we've seen. And that has interesting implications for technology firms that are using these green energies if they're substantiating their kind of claims around green energy by using these certificates.

If you've already had a ruling saying, "nah you're not allowed to do that." The other thing that surprised me, when I was looking into this, because the Renewable Energy 100 is a ranking of the top of a large number of firms who are significant investors in renewable energy. They actually don't accept the use of these kinds of renewable energy credits if they're not physically deliverable.

And one of the challenges you see in Ireland is that there's a limited amount of capacity to move the kind of like green energy that might be generated elsewhere in the world to there for this. So that's one of the challenges that you see. And we'll share links to both of those two things because for people who are kind of wonkish and want to get down to some of the bottom of this, they're really, I think they provide some interesting background to this.

We'll also share a link to the real time, to the Green Software Foundation Real Time Cloud dataset where there's been a bunch of work into trying to find some location based figures for this stuff so you can come up with some more accurate numbers than what we're seeing here. And I think, okay, I'll leave the last word with you, then we'll move on to the next story.

Go for it.

Gaël Duez: There are two things that I'd like to give credits to Adrian in his article. It's like stressing how much there are two sides of the story. And there's reports that focus a lot on sustainably of the cloud. And that's definitely what Amazon, Google, et cetera, are trying to do. But there is also this question of sustainably in the cloud, which is how as a user I can do or not a better job mitigating, reducing my carbon emissions, my water consumption, et cetera.

And he's right to say that not significant, no significant progress has been made on Amazon side and on AWS side story. And they are still infant phase at Google and even at Microsoft when it comes to transparency. And as a CTO, as an software engineer. And when you look at these dashboards and you see that everything is fine, everything is offsetted, you've reached carbon neutrality, it doesn't empower you to do the right things, which is optimizing, reducing your carbon emissions, your water consumption, etc. So that part, empowering consumers is still lagging of what we should expect from these tech behemoths. And my last comment is that I was very pleased that he mentioned and he reviewed, thoroughly the water consumption because for water and that my message about global versus local, it doesn't really matter.

It doesn't really make any sense to analyze the water consumption in terms of global consumption. It's water is a local matter. And it's really region per region, even data center per data centers. How much water comes in? How much water comes out? And in which state? Is it reusable, not reusable? Is it a closed loop or not?

In most of the data centers, including the one from the hyperscalers, are far from a closed loop. I know that Google has experienced once and they told quite a lot about it and it makes total sense. But we need more. precise and localized information on water. And that's a massive challenge as well. We focus a lot on carbon, but water is the next big issue that we need to pay attention to.

Chris Adams: Alright, water, that's the next horizon. I'm going to park that because we'll come back to it a little bit later. The next story is actually from the Financial Times. This is talking about Big Tech's bid to rewrite the rules on net zero. Now, at the time of this going out, it may be that the really nice looking piece may be hard for people to see, but no, the link does seem to work actually still, thankfully. The Financial Times has a really interesting piece, basically talking about the large technology firms that we often see coming up again and again. And this is a bit of a deep dive into some of the things you just referred to about like location based carbon footprints for electricity, because that's one of the key drivers of emissions for our use of digital services, and the market based approaches. And this pretty much dives deeply into something of a bun fight that's taking place where you have two kind of schools of thought where there's one set of companies like, to an extent, Microsoft and Google are pushing for this notion of 24/7 renewable energy and are having a quite kind of tight accounting process. And then you have another approach being largely put forward by Meta and Amazon talking about their kind of emissions first approach saying, no, what we should be looking at is decarbonizing the entire grid, not so much looking at our carbon footprint. And there's a couple of things that are really interesting inside this.

There's a few nice interactive graphics for you to see how people make green claims around energy usage. But one thing that I think is actually really stark is this set of charts showing the difference when you try looking at these figures. So, if you were to look at, say, the carbon footprints from, say, Microsoft, you can see, like, from 2018 to, like, now, you've got a figure of maybe, you see one chart showing the market based footprint, which is, pretty close to zero for Microsoft and close to zero for Meta and likewise for Apple. And then you see the location based figure for Microsoft. It's something in the region of like 8 million tons or zero tons, for example. And likewise with Meta, you're seeing 4 million tons versus zero tons. And Amazon's got the same issue where you're looking at like 15 million tons of location based carbon footprint from using electricity versus 3 million tons from using this.

So you, this really gives an idea of how these two different perspectives end up changing how you might report on this and how you might think about the environmental impact of using some of these tools. And like, to an extent, there is, there are reasons why you have a market based approach because, these come out of the fact that people who are inside large firms are looking for ways to be recognized for the investments they're making so they can justify this internally.

So there is a role that some of these play, but it often, it obviously gets quite a bit more complicated than that, especially because this is the year that the Greenhouse Gas Protocol, the kind of gold standard for reporting, is currently being overhauled to rethink how you should report this stuff and how you should be allowed to talk about energy being green or not green in this context. So Gaël, is there anything that kind of leapt out at you when you had looked through this? Because I would love more people to see this. I think it's a really fascinating story.

Gaël Duez: I think I've already commented it in advance when I was referring to the struggle between market based and local based approach. And once again, I think we should stress how important it is to understand that the way we build things in our mind and in our society as humans is one thing, and the physical reality of the world is another thing.

And when you add energy on a grid, wherever, et cetera, you have no clue on how it will be used, even if it will be used, because when you create PPA, it's potential energy to be used. You create new capacities, whether those capacities will be used or not remains a challenge. Obviously, they will be used, but not necessarily 100%, etc.

And I think the right approach is clean up your own mess. Everyone should start with this. So, I'm fine with having part of the sustainability report explaining what has been done and what could be the approach of market based, but the truth is local based approach. And when you see these figures, they're actually very consistent.

Yes, they're increasing massively their investments in data centers to fuel the AI boom. Their entire business model is based on infinite growth. The numbers go up, that's pretty, pretty logical. And what I've just kind of, when I read this piece of news, it also connected a lot with the crisis at the SBTI, the Science Based Target Initiative, that happened this year, when there was very strong push to allow more offset techniques to be recognized as science based, 

Chris Adams: Ah, you're referring to the Scope 3 thing. The push people being able to use offsets in their supply not just electricity, as a way to kind of decarbonize that without having to necessarily make some of the changes to like reform the supply chain. Is that what you're referring to here?

Gaël Duez: Absolutely. Thanks for making it much clearer than I was about to say. So, I think the struggle is everywhere because we see that the low hanging fruits, most of them has been already taken care of in this big corporation and they're entering the bumpy road where you've got harder choice to do. And when you face this kind of choices, well, either you do the right things and you go back to the physical reality of our world, or you try to change a bit the narrative or change a bit the rules, and I think this is exactly what we've been seeing at the science based target initiative where, some companies were obviously not able at all to meet the decarbonizing plan that they proposed just a few years ago, and they were trying to change a bit the rules.

And that should really question ourselves when it comes to transparency and acknowledge that even the most well intentioned CEO, the most well intentioned C suite, they cannot really do the right things without a bit of external help, whether it comes from pressure from activists or governments or UN, you name it, but we cannot fully rely on companies agreeing on how they should measure their own environmental impact.

Even if they are well meaning with tons of great people trying to do the right things, et cetera. It's not a black and white world out there. But there is a question at some point of financial pressure, shareholder pressures in many of these companies. They're just stronger than the entire stakeholders pressure.

Chris Adams: I think I know what you're referring to here, and what I'll do is we'll share a link to some of the stories about the SBTI, the Science Based Targets Initiative, the whole kind of bun fight that took place there, that ended up resulting in the then CEO stepping down from the organization, partly because there was so much pushback against this idea that you might be able to use offsets for looking at your supply chain emissions is in this same fashion. So we'll definitely add a link to that, thanks for that Gaël. The other thing that we maybe might maybe you must referring to inside this is there are some really nice quotes. Is that what you're talking about here does actually there's some relevance back to the Financial Times piece here, about when you look at the decarbonizing energy in this particular space. One of the approaches being used, which is described as Emissions First, for example, this is the one that's being largely put forward by Meta and Amazon. And one of the arguments for this is that we should be able to be optimized for absolute impact, not necessarily our own carbon footprint.

So given a hundred euros or a hundred dollars, we should be able to deploy that in the most effective place globally, wherever it might be. So one of the arguments being made is that we, if we want a data center in say, North Virginia, we should be allowed to basically purchase the right to claim the emission reductions in somewhere like India, which has a very coal heavy grid, and then kind of count that against our own emissions. And on one level, yes, you are reducing the environmental impact, you can make the argument, but it also means that some of the cheapest possible emissions in India are now being bought by one of the richest, some of the richest companies in the world. And this is very similar to the dynamic you were referring to with offsets, where, essentially, if you have a kind of, climbing scale in terms of how difficult it will be to reduce emissions over time. If you grab all the low hanging fruit, that raises all kinds of equity issues about what's left over for people who don't necessarily have the same resources available to them as Facebook or Meta or Amazon, for example, when looking at this. So, this is one of the challenges you do, find. But I'd really urge people to look over the FT piece and what we can do is we'll share a link to an archive link for this if the one, if it's behind a payroll, because it's a really fascinating piece and it's really worth looking into. Okay, Asim, sorry, Gaël, should we move on to the next story here?

Gaël Duez: Yes. I'm honored that you thought was... I 

Chris Adams: Bit of a Monday morning moment here. Okay, so this is actually a story speaking about AWS again, by moving an AI workload to AWS, you can reduce the carbon footprint by up to 99%. That's the argument being laid out inside this. And this is a piece from The Stack, which is not the same as The New Stack, who covered our previous story.

It's another organization. This one basically takes apart some of these questions, or at least dissects some of these claims and say, well, what's the basis for this? Gaël, I wanted to kind of give you a bit of space to talk about this because I know that you've read this and I know that this is something that is... and you often advise firms who are actually trying to figure out how do I reduce the environmental impact of the services I'm using. So when you read this, what leaps out at you and what kind of things are the kind of most salient points would you say?

Gaël Duez: Well, I think my main reaction was, "oh no, yet another one."

Okay. So there is the SCI piece, which is interesting, but let's go back to basics about it. And I encourage listeners to have a look at two great resources. The first one is an article from the Boavizta Association investigating the claims made by several hyperscalers that when you migrate from your services, from a traditionally hosted perspective to a hyperscaler perspective, you save 60, 70, 80, 90 percent of carbon emissions.

And the second source is actually a man, and I was very pleased to see him being quoted and interviewed in the article, which is, he's Mark Butcher. He has been very vocal about hyperscaler claims, how they measure things, the scope three not being that well taken into consideration. Mark works at Positive Cloud and he works with a lot of clients across the UK on these topics.

And why I'm mentioning this too is that hyperscalers, when they say, basically "move your instances more, move your computing power to our facilities, because we are so much more efficient than the other data centers or the other hosting solutions," they might be right, but the others are a role, a world on their own.

And this is really the question of the baseline. Yes. If I run two or three servers in my office building, in a small room with a very old air conditioning, yes, I might reach a PUE of 2, 2.2, maybe 3, whatever, et cetera. But most of the clients, they don't start with these baselines. They start with servers and instances being managed in already quite professional and by seasoned providers, okay, and I won't provide any name here, who have already reached a decent PUE and the gain is much smaller.

So that's. On, from which baseline do you start?

Chris Adams: I see, like the example given here is like a really inefficient thing. If you compare really an inefficient setup compared

to this idealized, going to give you a disproportionately large saving. That's

Gaël Duez: And this is when you study the case studies, which have been provided by hyperscalers. it's literally my grandma managing a data center. Worst practices possible that are accounted as a baseline. And this is not the truth. Many data centers already reach pretty decent or, pretty good actually,

power efficiency. So where you start from matters. And that's my, point number one, my point number two, and we go all the way back to this local versus market based approach. But I'm sorry to say I'm, I, feel a bit like I'm, rambling here. But once again, if, even if, okay, I run the worst possible data center on Earth, I've got

maybe a PUE of two, maybe three, et cetera, but I run it in a very low carbon place, say France, if you account, for, the nuclear, a share of energy, say Scotland, for instance, and I recall Mark wrote something about it, Mark Butcher wrote something about comparing the energy intensity of North Scotland and Ireland.

And once again, just by migrating your instances there. You might, let's say, divide by two, your energy consumption, thanks to better energy efficiency by AWS, because they're very active in Ireland, but you can do the same math with Google or Azure, but then you start operating in a country which is, and that's the case between, for instance, Northern Scotland and Ireland, six times more energy carbon intensive when it comes to electricity production.

Chris Adams: So let me check if I understand, well, I think with the point you're getting at, you're basically saying, yes, you may, the infrastructure may be more efficient, but if the local energy is dirtier, it doesn't necessarily matter that it's six times more efficient. If, say, the infrastructure is twice as efficient, but the energy is six times worse, then it's still, you're not coming out ahead. That's what I think you seem to be making the argument there

Gaël Duez: I actually, I was actually, I was not the one making the argument. Mark did it and 

Chris Adams: but that's, that's you're saying. Okay. 

Gaël Duez: but that's exactly what I'm saying. And so this question of where do we start from the baseline is super important. Then there is a specificity of the AI itself. And we should always remember that most of the time by AI people, and I think this article is a bit misleading here as well, they think generative AI, which is part of AI on its own.

We start from so high. It's still very infant. The LLM are not that old. And of course you can decrease very significantly the energy consumption and the emissions of your AI model, because we are, we've just started to do so, but what will be the trend for, I would say, everyone rather than AWS is an open question.

Obviously, they're doing a better job reducing and optimizing everything, but it's fair to assume that other actors as well are doing the same. So I'm always very concerned when we take a very new algorithm, I would say, or a new part of the AI industry and say, "Oh, look, we're going to reduce by that, that much."

But of course, it's like with cryptocurrency. They've started at such an inefficient way that they made a great progress. Still, they consume a lot of energy. So you see that's my point. So I will be always super cautious with this kind of stuff. And then comes the good part of it, which is using the, as the software carbon intensity, but maybe Chris, you want to elaborate a bit more on 

Chris Adams: So, it's nice to see an international standard being used in reports like this, but in order for a standard to be used, you need to follow the standard. And one of the key things that the standard actually does is it, basically, when I've looked through this, I look through the report, I can't see any numbers for any of this.

So, it's like you have a car saying, well, this car is twice as efficient as this other car. If you don't have any numbers about, like, the miles per gallon or the equivalent like that, it's very hard to be able to trust that number, for example, or trust any of these claims here. So you have an issue about, well, there's a lack of underlying numbers. Also, the thing we see is that a significant part of this is based around the environmental impact of the energy itself. Again, we don't have the numbers for this, but in particular, one of the key things, one of the key claims being used in the report was we're going to take into account these market based figures here.

Now, the Software Carbon Intensity specs explicitly says we don't use these inside this. So, you've essentially got people using this term, Software Carbon Intensity, without actually following any particular nuances of this. And this, It makes it very difficult for me to recommend this report for anyone else because it essentially is going against how this is intended to be used. And the firm that was working with Amazon, they are very involved inside the Green Software Foundation. This really needs to be a thing that we can't do if we want to see this to be adopted and respected because this essentially, in my view, undermines a huge amount of work that's gone into developing a standard here because this makes me trust the Software Carbon Intensity less after reading this report and seeing people cite it, because it's being used incorrectly. So that's one of the things I would actually raise and something that does need to be addressed. Like, there are mechanisms that the GSF has to say, please do not use it in this way. It's misleading and it undermines some of the work we have. And I think that's something that will need to happen because, yeah, I cannot recommend anyone looking at this report or even recommend using or referring to this standard like this because it's an incorrect use of the standard. So yeah, that's my take on it. 

I'll move on from this because that was a bit of a downer, but it's really important if you want people to trust this in my view.

Gaël Duez: And I think also that there is a way to protect the SCI and the Green Software Foundation tools, which is using the Impact Framework manifest, 

Chris Adams: Yeah, I'm glad you mentioned this, Yeah.

So there is a mechanism. The whole thing around the Impact Framework was specifically set up to say, "hi, you're going to make a claim? Make this transparent. Show you're working inside this." And that's like, there's a huge amount of work that's gone into providing this, and if you're not going to share any of the numbers or share the basis, and there are now lots and lots of really helpful case studies demonstrating how to do this, like we'll share links to this to show this is, these are the correct ways to use this. When you have it being used in a way which is so unhelpful, it's, really problematic, and you can see why people are going to struggle, and why, you can see why people end up essentially dismissing so much of these, efforts as greenwash, when people aren't sharing the underlying numbers for this stuff.

So yeah, that's like the framing I would take, and I would really like to see this addressed, because it's going to be, it's going to be a real challenge going forward, in my view.

Gaël Duez: Transparency, transparency, that's all that matters with this kind of claims.

Chris Adams: So, this probably takes us to the last story I think we have time for. This is one actually, this isn't so much a story, but more of a kind of discussion about some of this. So we've been talking about how it's real, really difficult to actually get some numbers from this, and how the way people report this is also a real challenge.

The thing I kind of want to share with you, and I realize I won't have time to talk about this while I, this is partly what I'll be talking about at some events in September, is the fact that we have, we Some regulation, which is forcing some of this. So in Europe, all across Europe, there is a law called the energy efficiency directive, and basically any data center that uses more than 500 kilowatts of power, which is, that's not a small data.

It's not a tiny data center, but pretty much every single hyperscale you imagine would have to do this. There is now a law, which basically says, every data center and any organization operating a data center has to make publicly available, Information like the name of the data center, who runs it, how large it is, how much power is used. It talks about the amount of energy used, the water used, all this stuff. So we do actually have laws which are kind of forcing some of this now. The, there is one caveat in that where companies consider this information to be a trade secret, they don't necessarily need to publish this information into the public domain, but where companies are not saying this is a trade secret, we now, as technology professionals, can ask and say, this information should be in the public domain if you're in, if you're in Europe. And the thing that I might share as a follow on from this is that for companies that are not sharing this information, they are now mandated to report to a centralized database with the idea that some of this information will be shared in an aggregated format.

So for the first time, we can actually get some meaningful numbers that come out of this. So, companies that are prepared to be transparent, you can ask for this stuff. Companies that are then saying, "we're not going to make this transparent because it's a trade secret." There is still a mechanism by which they will need to report so that we can finally have some data informed policy around this. Because one thing that's come up again and again in this discussion has been that we don't have access to this information. And there is so much pressure or there's so much, there's so many incentives to construct a message which makes you look good that it becomes very difficult to trust a lot of the statements around green software that come from lots and lots of large firms right now.

So yeah, this is, I'll share a link to the issue where this has been discussed inside the real time cloud working group inside the Green Software Foundation. But Gaël, I wanted to just check, as someone who's not in Europe, what's your take on this? Is there anything that kind of caught your eye when you were looking at this?

Gaël Duez: Well, the first thing is that it is required to disclose numbers in both relative, but also absolute numbers. And that's very important. It's not only PUE. It's also how much energy did you consume overall? And that really connects well with what we've discussed before. I also believe that this is something that is

pivotal for country with weakest electricity grid to consider. It's always claimed that for instance, Kenya, my neighboring country of La Reunion island, has a very strong policy of attracting data centers. They want to become a computing power. 

Chris Adams: Iceland of Africa, because have more geothermal than

Gaël Duez: Absolutely. And I reckon that this kind of disclosure will also have them a lot anticipate what is needed for them to prepare the electricity grid to this kind of increase in electricity consumption caused by data centers.

And also making sure that what has been dealt and agreed when they do this big hyperscale deal are actually what is provided, that the energy is there, that the water is there, and I believe that local populations, which are often caught in between, like, "oh, it will create a lot of jobs, but when you do the mass, not that much," so it's not that an obvious investment to say, okay, we will welcome a lot of data centers in our country.

It might be, but it's not like a big investment. A clear win or the case, they will have the ability to scrutinize, how, what are the impacts and environmental impacts. And I must admit that, if you look a bit at the history, in 1982 in Europe, a European directive created the Seveso listing, and the Seveso made it compulsory for every state in Europe to list what are the facilities, industrial, agricultural, mostly industrial, which can create significant environmental risk.

And we were talking about chemical industry, et cetera, et cetera. And to some extent, data centers, they impact a lot of the environment. It's just that they will not blow away like a chemical industry, but on the long run, they've got a lot of impact on their environment and it makes a lot of sense to, at least for the bigger now, the biggest data centers, to be able to provide

environmental information, in a comprehensive way, a comparable way, and to make sure that we monitor the environmental impacts of these big facilities. we're talking about facilities that are built on hectares of land. There's not like the small, tiny warehouse that we might still think of.

They're like massive industrial facilities. So having open and transparent reporting seems to be quite straightforward.

Chris Adams: So maybe we might see some of these ideas adopted in other parts of the world, especially because on the underlying data for this. We'll share a link to this. We've been doing some research ourselves in the Green Web Foundation. A lot of these data points are based on the EU code of conduct, which is a public document for people. So it may be the case that you might see some of these data points being reported in other parts of the world as well to set a precedent. So they actually have the data to make data informed decisions about how, about the role digitalization plays in society and

how the impacts are actually shared around this.

Thank you for the link about Seveso as well. I didn't know about the Seveso directive. That's totally new to me. What I might share is a link to some of the work that the Green Web Foundation has been doing. We have a fellowship and one of our fellows, Samantha Nidwalana, she's based in South Africa and the Netherlands, and she spent a bunch of time looking through, trying to get some numbers in South Africa for, to basically explore okay, what's the environmental impact of data centers inside this?

And we'll share some links to her challenges in this because she's been trying to find these numbers and it's been a real, it's been a real challenge in many cases actually. And it does give you some idea about like where some of this might go, but also hopefully stories like this and seeing some of these laws being passed might help set a precedent so that we can have more transparency in other parts of the world as well. Okay, I realize we're coming up to time. Let's do a quick just run through. We've got some events coming up. I know that I'll be catching a train in a few weeks time to go to London, and I think I'll see you in person for the first time for quite some time actually. This is an event called Green IO Conference. Maybe you could just briefly touch on that before we move on to the next set of events and then close out for the day.

Gaël Duez: Yeah. Well, absolutely. I will be pleased to see you again for the second time. The first time was in Berlin, if I recall, and I think it was for being interviewed on the GreenEye And what. I launched with my partners RPI Days last year, was a series of global conferences called Green IO, not very original, sorry about this.

And the idea was to have also on site events. We can see a lot of hybrid and a lot of online events when it comes to digital sustainability and they're great. We can give kudos to the Green Web Foundation, the Green Software Foundation, CNCF as well. They're doing a lot of these events. Now, what I also realized is that if you look at other specialties in our IT industry, let's take cybersecurity, for instance, accessibility, design, or you name it.

When you work in a city with a significant enough workforce in cybersecurity, for instance, let's say Barcelona, Berlin, Paris, New York, Singapore, Beijing, you have at least 2, 3, 4, sometimes 5 different on site conferences when you know that you will meet your peers. Today, if you are an IT sustainability specialist or green IT folks or whatever, like the dude who anyone reached out to because, "oh, it's about green, et cetera.

He's a person, she's a person we'd like to talk to." You've got basically nowhere to go except for one conference in Paris and one in Belgium now in Brussels this year, where you know that the topic will be a hundred percent your specialty. So don't get me wrong. We can give kudos to a lot of tech conferences like QCon, like even Reinvent AWS, you've got to sustainably track, that's fine, but having a 100 percent green IT focus or IT sustainably focus conference, where you know that basically this is the place to be to meet all your peers, most of the cities around, they don't have it. So that was really the idea of creating this Green IO.

Even since Singapore, London, and Paris this year, we are like 90, 90 percent sure that we will add New York and Munich next year also. And the idea is really, it's just It's a place that has been created for the local communities to do what they want with it. And so you, this year in London, it's a bit like go back to the basics and let's talk step by step.

What about low carbon infrastructure? What about design and eco design? What about beyond, understanding the organizational challenges and HR challenges that when you want to rule out more sustainable approach, but that's a one day conference. It's two days in Paris this year, but in London, it will be one day.

And we expect to have like several hundreds of people joining. And I know also that we will have a great keynote speaker that some of the listeners might known named Chris Adams. So I'm really delighted that you agreed to join Chris. Thanks a lot for this.

Chris Adams: Cool, thanks, so that's the event. It's the 19th of September, taking

place in Bishopsgate in London. And thank you, I am indeed keynoting in the morning, so I hope my trains are on time. But there's a number of people from the Green Software Foundation, for example. I can see Sara Hsu, who is one of the people leading the Green Software patterns project in the Green Software Foundation. There's some people from the W3C on working on web sustainability guidelines. There's also, I'll, share a link to Therese Gale who is working at Salesforce. She'll be talking a little bit about some of the experience, some of her experiences as well. So there's a number of people who

Gaël Duez: Mark Butcher will 

Chris Adams: and Mark, yes, the, yes, Mark Butcher of Positive Cloud.

He's been one of the people who's been really instrumental in some of the most recent work inside the UK government, put together a kind of Digital Sustainability Alliance. There's a bunch of things there. So I'm actually looking forward to this Gaël, and I want to just give people a heads up that it's taking place. Gaël, thanks so much for coming in for this and telling us the stories about Wales, And we're watching and all that sharing all your insights elsewhere. Gaël, if people do want to follow up and see what you're up to, what's the best way to find you for future work?

Gaël Duez: Well I would say on LinkedIn and I'm pretty easy to find Gaël Duez. There is not that many. And more generally, if you are interested in what we do in Green IO, it's greenio.tech. That's the website. You've got access to the podcast, the conferences you've got link to my own websites. But I think greenio.tech is the best place to start.

Chris Adams: Brilliant. Well, thank you for that. We'll share some links to all of those websites and to this event coming up. And, for people who aren't able to go to Green IO, the Cloud Native Computing Foundation, the TAG ENV essentially the green part of the Cloud Native Computing Foundation.

They have a series of remote events taking place in October. We'll show a link to that so that if you still want some, to get an events fixed, that's where to look. Alright, Gaël, thank you once again for making the time and I hope you have a lovely week. Okay. Take care of yourself, mate.

Gaël Duez: Thank you, that was great being there. Bye, have a nice week as well.

Chris Adams: Hey everyone, thanks for listening. Just a reminder to follow Environment Variables on Apple Podcasts, Spotify, Google Podcasts, or wherever you get your podcasts. And please, do leave a rating and review if you like what we're doing. It helps other people discover the show, and of course, we'd love to have more listeners. To find out more about the Green Software Foundation, please visit greensoftware.foundationon. That's greensoftware.foundation in any browser. Thanks again, and see you in the next episode! 

Host Chris Adams and guest Romain Jacob delve into the often-overlooked energy demands of networking infrastructure to discover A Greener Internet that Sleeps More. While AI and data centers usually dominate the conversation, networking still consumes significant power, comparable to the energy usage of entire countries. They discuss innovative practices to make the internet greener, such as putting networks to sleep during low usage periods and extending the life of hardware. Romain talks about his recent Hypnos paper, which won Best Paper at HotCarbon 2024. He shares his team’s award-winning research on how energy demand for networking kit powering the internet can be reduced by simply by powering down links when not in use.

Learn more about our people:

• Chris Adams: LinkedIn | GitHub | Website
• Romain Jacob: LinkedIn | Website

Find out more about the GSF:

• The Green Software Foundation Website 
• Sign up to the Green Software Foundation Newsletter

Resources:

• SCION Architecture [11:30]
• Environmental Impacts of Internet Technology (eimpact) [17:15]
• Why we should be intentional about the mental models we use for thinking when we think about digital sustainability | Chris Adams [18:30] 
• A Sleep Study for ISP Networks: Evaluating Link Sleeping on Real World Data | Romain Jacob, Lukas Röllin and Laurent Vanbever [18:59]
• Network energy use not directly proportional to data volume: The power model approach for more reliable network energy consumption calculations | David Mytton [38:55]
• Co2.js - The Issue | The Green Web Foundation [42:57]
• Rethinking Allocation in High-Baseload Systems: A Demand-Proportional Network Electricity Intensity Metric — University of Bristol | Daniel Schien [43:53] 
• Introducing Web Sustainability Guidelines | 2023 | Blog | W3C [49:31]
• Greening of Streaming [52:16]
• Network Power Zoo | ETH Zurich [54:46]

Other source material:   

• A Primer on Optimistic UI | Imhoff
• Response Time Limits: Article by Jakob Nielsen | NN Group
• Optimistic UI Patterns for Improved Perceived Performance | Simon Hearne
• Reducing the Energy Footprint of Cellular Networks with Delay-Tolerant Users | IEEE Journals & Magazine 

If you enjoyed this episode then please either:

• Follow, rate, and review on Apple Podcasts
• Follow and rate on Spotify
• Watch our videos on The Green Software Foundation YouTube Channel!
• Connect with us on Twitter, Github and LinkedIn!

TRANSCRIPT BELOW:
Romain Jacob: We used to consider that energy is cheap. Energy is there. We don't need to worry too much about it. So it's just simpler to plug the thing in, assume energy is there. You can draw power as much as you want, whenever you want, for as much as you want. And it's time to get away from that.

Chris Adams: Hello, and welcome to Environment Variables, brought to you by the Green Software Foundation. In each episode, we discuss the latest news and events surrounding green software. On our show, you can expect candid conversations with top experts in their field who have a passion for how to reduce the greenhouse gas emissions of software.

I'm your host, Chris Adams.

Hello, and welcome to another edition of Environment Variables, where we bring you the latest news and updates from the world of sustainable software development. I'm your host, Chris Adams. Back in episode 10 of this podcast in September 2022, we did a deep dive into the subject of green networking, because while a lot of the time people talk about the energy demands of AI and data centers, in 2024, in absolute terms, the amount of power consumed by networking was still larger.

Back then in 2022, the best figures, when we looked at this, came from the AI, which put the energy usage of data networks at around 250 terawatt hours per year. So that's about the same as all of Spain's energy usage in 2023, so that's not nothing. Now, it's a few years later, 2024, and the best figures from the same agency, the IEA, now give us a range of between 260 and 360 terawatt hours, which could be anything up to a jump of 50 percent in three years now.

Now because of much of this power is coming from fossil fuels, this is a real problem, climate wise. So what can we do about this? With me to explore this once again, is my friend Romain Jacob, who helped guide us through the subject in 2022, along with Dr. Yves Schouler at Intel at the time.

His team's recent research won the Best Paper Award at HotCarbon, the conference that has fast become a fixture on the green IT and digital sustainability circuit. So he seemed a good person to ask about this. Romain, thank you so much for joining me for this podcast. Can I give you the floor to introduce yourself before we revisit the world of green networks?

Romain Jacob: Chris, welcome. I'm very happy to be back on the podcast to talk a little bit more about this. Hello, I'm Romain, I'm a postdoctoral researcher at ETH Zurich in Switzerland. I've been working in sustainability for two to three years now, more or less full time, as much as full time research happens in academia. And yeah, I was, I had the pleasure to present some of our technical work at HotCarbon and I'm sure we're going to deep dive into a bit more in the podcast.

Chris Adams: Okay. Thank you, Romain. And for people who are new to this podcast, my name is Chris Adams. I am the executive director of the Green Web Foundation. We're a Dutch nonprofit focused on reaching a fossil free internet by 2030. I also work as one of the policy chairs inside the larger Green Software Foundation, which is why I'm on this podcast.

Alright, if you're also new to this podcast, every single project and paper that we mention today we'll be posting a link to in the show notes. So for people who are on a quest to learn more about reducing the environmental impact of software engineering, you can use these for your own practices and your own research.

Okay, Romain, are you sitting comfortably?

Romain Jacob: I am.

Chris Adams: Okay, great. Then I'll begin. Okay, Romain, when we last spoke, we covered a range of approaches that people are using right now to rein in the environmental impact of networking. And but before we spend too much time, I wanted to kind of see if you could help set the scene to help folks develop a mental model for thinking about, say, networking versus data centers, because I touched on this a little bit, but it might not be obvious to most people.

So maybe if you could just provide the high level, then we can touch on some of these Differences between the two and why you might care, or how you might think about these differently. So yeah, let's go from there.

Romain Jacob: Yes. Sounds good. It's true, you mentioned in your introduction that we sometimes get numbers that oppose or compare data center corporate footprint or usage and the ones from networks. But now, what does network really mean? It's not really clear what we mean by that. Because there is networking in data centers and, you know, the rest of networks are not completely detached from it.

But at a high level, you have a set of networks that are meant to provide internet access to individuals and to other networks. What this means is that you have companies that are specialized to just make your laptop, your phone, or other appliances you have, being able to talk through the internet. And typically when we refer to networks, without further details, this is the type of network you're talking about.

And data centers, on the other hand, are something that is in the scale of IT fairly recent, where we have this idea of if we centralize in, in one physical location, a lot of powerful resource machines that have a lot of compute, that have a lot of storage available, then we can use that as a remote computer.

And just offload tasks to those data centers and just only get the results back. In today's ecosystem, data centers are a very core element of the internet. The internet today would not really work without data centers. Or at least a lot of the applications we use over the internet only work thanks to data centers.

So from a networking perspective, in a data center, you also have, you need to exchange information and bits and packets between those different machines that live inside the data center, but the way the network looks like is very different from the cellular network that is providing mobile connectivity to your smartphone.

Chris Adams: Ah, I see.

Romain Jacob: So those are very different type of networks, and they have different, very different type of way they are designed, way they are operated, and how they are used. Typically, in a data center, you tend to have a quite high usage of the data center network because you have a lot of exchange and interaction between the different machines that live inside the data center, whereas in the networks that provide internet access, so the networks that are managed by entities we call ISP, for Internet Service Providers.

Chris Adams: Mm hmm.

Romain Jacob: Tends to be much less utilized. There is a lot more capacity. in those network that what is really demanded by the end user.

Chris Adams: Okay.

Romain Jacob: That's a fundamental difference between, between the two networks and something that we try to leverage in our research.

Chris Adams: Okay, so if I just check if I understand this, so you said that you might have networks which might be like, say the ISPs and things, they are individually not that high themselves, but because there's so many of them and because they're so diffuse, in aggregate, this can work out to be a very large figure, for example, and, and, that also speaks a little bit about, I guess, how you might power some of these.

So, like, when we think about a large hyperscale data center, that's something in the region of maybe, if you're looking at a large one, which is maybe the high tens to maybe low hundreds of megawatts, that's maybe thousands of homes. That's a lot of power in one place, whereas with a network, you don't have quite so much, but it's because it's distributed, you might have to have different approaches to managing that.

So, for example, you might and, you might have to take different strategies to either decarbonize that or deal with some of that, some of that load that you actually have. Okay, and I guess one of the questions I might have to ask you about this is then, when you have this, split between the maybe it's just worth talking a little bit about the different kinds of networks that you have here.

So for example, as I understand it, there's maybe an ISP I connect to, maybe my ISP, but then they need to connect to some other cable. And if I'm going across, maybe if I'm connecting to a server across in the Atlantic, then I'm going through some like backbone or something like that, maybe you could talk a little bit about the different layers there, and what some of the, and what the kind of, how much they might make, or if there's any differences in how those ones need to be powered, for example, how they're used?

Romain Jacob: Yeah, totally. So the, the name internet stands for an interconnection of networks. So the internet by name and in private space is a network of networks. Right? So when you connect to the internet. What it means is that you connect to another machine somewhere on the planet that has also access to the same global network.

But these global networks, the connection between those two endpoints has to go through, most of the time, several different networks. And so, typically, if we look at the internet infrastructure, there are several ways of representing this, but one division that we usually use is You have Core, IAP, so the one that kind of sit more in the middle and they provide transit for many, many different interconnections.

Then you have networks that are qualified to be more belonging to the metro area. So this is where it's getting closer to the user, but it's not yet the network that provides direct connectivity to, let's say, your phone or your laptop. And then you have the edge network. And the edge network is really there to provide what is called the last mile connectivity key to the end user.

And those categories exist and were proposed to Helsinki because this share of the network has different characteristics. The core tends to look a bit more like the data center. Like, it's more dense mesh, so there are more interconnections between the different points in that network. And the utilization tends to be higher and kind of constant because it's a global network.

Whereas, the closer you go to the end user, the more you're going to see filtration, because, for example, while users are awake during certain hours during the day, and this is where they tend to use their machine. You will see peak of usage during, you know, TV show primetimes in the evening, but much less at 5 a.m., where most people are deep and not using their phone or their laptops. And so, those networks look differently in terms of What they are used for and how they are built and designed, because we try to adapt the design for the particular use case.

Chris Adams: Okay, thank you for that. So it's a little bit, if you squint, it's a little bit like how you might have motorways and then A roads and B roads and maybe smaller roads, for example.

Romain Jacob: Very like that. It's very much like that.

Chris Adams: All right. Okay. So that's actually quite helpful. And when I think about other kind of systems, I think a little bit about say, like electricity networks, which have, you know, big fat transmission ones which send lots of things, but then you have like the smaller distribution networks which are, so it's somewhat comparable. Okay.

Romain Jacob: It very similar in principle.

Chris Adams: Okay, so this is helpful for developing a mental model about some of this. Alright, okay, so last episode when we spoke about the different techniques people spoke about, you, we spoke about things like carbon away networking, different protocol designs like I think SCION, which was one of the, projects proposed.

And you kind of coined this phrase that an internet of the future needs to grow old and sleep more. And I really, I found this kind of quite entertaining and it stuck with me. But for people who are new to this, maybe you could just unpack what you meant by that because not everyone has read the paper or seen the talk. And I think it's quite helpful for thinking about this subject in general, actually.

Romain Jacob: Yeah, sure. So, so two years ago in the first edition of the HotCarbon Workshop, we, we outlined this vision of what could be relevant to work on in sustainable networking area. And the two ideas that emerged were essentially captured by this growing old and fleeting more aspect. So what do we mean by that?

Growing old is essentially the idea that we tend to be using the hardware we buy not long enough. So if we take an end user perspective, we tend to change phones every couple years. Numbers are changing about this, but we can debate whether this is a good thing or a bad thing. In the networking area, so for the hardware that operators buy to make up the network, so devices that we call routers and switches, it tends to be a bit of the same thing.

Devices were changed, the standard used to be every three years. So in three years, the entire infrastructure would be renewed. So you would buy new hardware to get higher speed or better energy and so on. And there are various reasons for doing that. We can detail it if you're interested afterwards. But it has a very significant cost, financial cost, but also in terms of carbon cost.

Because one people need to understand is that every time you manufacture a product. Not just for networking, but for any product, there is a carbon footprint associated to it. This is where we typically refer to the embodied carbon footprint. And so this embodied carbon is a one time pay, but if you buy more often, well, you pay this price more often.

Now, there's a bit of a tricky thing, which is that you, you can argue that If I buy a new device, a new phone or a new router that is 10 times more energy efficient, then over time, I will then do more saving that would compensate for the embodied cost. The problem is that you, it's very hard to estimate how much you would save and how much is the embodied footprint.

It's a very firmware, a break even point where you're doing this upgrade, buying this new hardware, start paying off from a carbon perspective, but it's not necessarily clear ahead of time when that happened. Generally speaking though, what was pretty clear to us is that we could and we probably should be using the hardware longer.

And this is what we meant by the grow old idea.

Chris Adams: Ah, okay. Thanks for that.

Romain Jacob: Now, to describe more, this is kind of simpler. This is the idea that is very common in other fields like what we know as embedded systems or the Internet of Things. Think about devices that run on battery, to make things simple, that are more or less small but run on battery. And because they need to run on battery, for decades, engineers have been trying to optimize the energy efficiency of those devices.

And the most efficient way of doing this is essentially turning off everything you don't need when you don't need it. So if you think about your phone, your, the screen of your phone is off, I don't know, maybe 90 percent of the time. And this is to save the power drawn by your screen, which is by far the most expensive or power hungry element in a smartphone.

And we do this in order to save on power, on the average power and so on energy at the end of the day. And we are arguing that in the networking world, this is not done too much. And it should probably be done more. So now I need to be quite precise here when I talk about the networking world, I'm talking about the wired networking domain.

In the mobile domain, so in cellular communication that connect to your phone, or also in Wi Fi and so on, the idea of sleeping is already used quite a lot. But in the wired domain, it did not transfer too much. And so the reason why it did not transfer is because we used to come to the idea that energy is cheap. Energy is there. We don't need to worry too much about it. So, it's just simpler to plug the thing in, assume energy is there. You can draw power as much as you want, whenever you want, for as much as you want. And it's time to get away from that.

Chris Adams: Okay, alright, so if I just play that back to you to make sure I understand it correctly. So, the growing old part is essentially a reference to the embodied energy that goes into making various kinds of hardware. So, like, when we're looking at a laptop, around 80 percent of the carbon footprint, it comes from the manufacturing, compared to the running of this.

And, if I keep that laptop for a short period of time, It's a great proportion of the life cycle, lifetime emissions, for example. And we see the same thing in data centers as well. So, for example, Facebook and company, you know, some companies and hyperscalers, they might have had this three year period that you spoke about before, but in the 2010s, we saw figures anecdotally, but not published ever.

It was like 20 years ago. But sometimes these would go down to as much as as little as 18 months for some service because they wanted to get the maximum usage of kind of compute for the power they're using for example. So they had incentives to change like that. So that's what that part is a reference to.

And I think on the eImpact mailing list, where I've seen a lot of the discussion. I will share a link to this in the show notes of this really cool 3D chart showing how, where the break even points are that you mentioned about that. And the sleeping part seems to be this reference that, in many ways, networks are often designed for kind of maximum amount of usage, not necessarily what the average usage might be, similar to how, say, the electricity grid in America, for example, is designed currently designed for everyone to be using aircon at the same time, when normally it's maybe 40 percent utilization.

So there's all this kind of headroom, which doesn't need to be accounted for. And we, and it's a bit like service, you know, we have, as software engineers, we're taught generally to size for the maximum output because the loss of business is supposed to be worse than the cost of having that extra capacity.

But in 2024, there are new approaches that could be taken. And we do things like serverless and scaling things down. And these ideas are - they've been slower to be adopted in the networking field essentially, right?

Romain Jacob: Yeah, that's kind of the, that's kind of the idea.

Chris Adams: Brilliant! Okay, that is good. We'll share a link to the paper because it's quite a fun read and I really helped, it stuck with me ever since I saw you speaking about that.

Okay, so we've kind of set some of the scene so far. We've got some nice mental models for thinking about this. We referred to the energy and the embodied part and I guess the thing we didn't mention too much was that the growing old thing is going to be, you know, more of an issue over time because while we're getting better at decarbonising the electricity of the internet, we're not doing such a good job of decarbonising the extremely energy intensive process of making electronics right now. So we're only, this is only going to become more acute over time.

So, maybe I can allow you to just talk a little bit about this Hypnos paper, because as I understand it, it was an extension of some of this vision going forward, and I know that it wasn't, you weren't presenting yourself, but I know it was your team who were presenting it at HotCarbon, so maybe you can talk a little bit about that, and maybe say who was presenting, or some of those things there, because, yeah, I enjoyed reading this, it was quite fun, it, similar way, I enjoyed it as well, basically.

Romain Jacob: So Hypnos is a, a recent proposal that we've made to essentially try to quantify this sleeping principle. So in two and a half years ago, we said, okay, we could look at the embedded aspect, we could look at the operational aspect and how to improve them. And we thought back then that the one way was to just try to apply those principles of heaping to wired networks.

And so together with some math students from ETH, we started looking into this and say, okay, in theory, we know how to do this. Let's try. You know, let's try for real, let's take some hardware, let's design a prototype, protocol that would just put some things to sleep and see what happens. What was surprising to us was that the, the theory of how you would do the sleeping in a wired network and how much you would expect to save by doing that was old.

It was the first papers go back 2008 or so, so there's been a while and back then, people were saying, okay, assuming we have hardware that, that allows us to do everything we want, then we could implement seeding in this way and then we would save so much. So they knew that the proposals that were made, they were making back then were not readily applicable.

And so we felt like 15 years later, it's kind of interesting to see where are we today? Like how can we do things? And the key element, key there, was how quickly you can turn on something. You turn it off, you can always turn off something, you take some time, you save some power, okay? But then, eventually, if you need it back on, you want it to react quickly.

You know, I talked about the screen of your phone before. It's always off, and it's fine, because as soon as you press the button, and you touch with your finger, the screen lights up, right? It feels instant, right? So it needs to happen quickly to be usable. Except that in networks, it's not like that, it's, I mean, not, not, at least not today.

So if, if you think about a link that connects two routers, this was the first, the first thing that we started considering. Okay, let's put that to sleep. It's essentially the smallest unit in a network that you could put to sleep.

Right.

Chris Adams: A bit like a lane in a, like a multi-lane in a multi-lane car road.

Romain Jacob: Yeah. I would think about the road network, like, turning off a port or turning off a link in a network would be like. Cutting one road in your network. You know, like here in the city, you have many different ways to go to different end points and you would just say, okay, this street is closed. So you can't use it.

Chris Adams: Ah, I see. Okay.

Romain Jacob: That's kind of like the simplest thing one can do from an networking perspective, except that to turn the thing back on, to reopen the street would take multiple seconds.

Chris Adams: Mm.

Romain Jacob: And it doesn't sound like much, but in the networking area, multiple seconds is a lot of time because a lot of traffic can be sent during this time.

And. If you make things short and not too technical, it's way too long.

Chris Adams: So we're looking, you want milliseconds, which are like thousandth of a second, and if something, it takes two or three, it's two or 3000 times slower than you'd like it to be basically.

Romain Jacob: Exactly. So, without getting too nerdy and too technical, the problem is, we can't do what we're suggesting in the literature because we cannot sleep at short timescales as we were planned. And so we're like, okay, so is it over or can we still do something? And so what we were thinking is maybe we cannot sleep, you know, at millisecond time scales, but we can still leverage the fact that networks, some networks, are a lot more used during the day than during the night.

So we, we have a lot of patterns that are daily or hourly that we can leverage to say, okay, well, we have a predictable variation in the average use of the network. And so when we reach the value to declare night time. Then maybe there are some things we can share. And so we, we try to implement a protocol that we do do.

We say, okay, let's do the simplest thing possible and see how well it works. And Hypnos is essentially the outcome of that. So in essence, it's a very simple tentacle that looks at all of the roads, so all of the links in the network and how much they are used. And then we start turning off the, the unused one.

Until we reach some kind of like stopping condition that we say, okay, now it's enough. Like the rest we really need to keep it. At a high level, this is what we do. So one, one challenge was to get actual data to test it.

Chris Adams: Mm-Hmm.

Romain Jacob: Because if you stimulate a network and you stimulate the utilization of your network, you can make things as pretty or as, as ugly as you want, you know, depending on how you look at things.

And so it was, what was really missing from the literature was precise case study that says, okay, here is the data from a given ISP. Here's what the network looks like, and here's what the utilization looks like. In this network, what can we do? So, there has been a long, very long effort to actually get this data.

And then, the Hypnos paper is essentially say, okay, we have the protocol, we have the theory, now we have the data. Let's match the two things together and see where that takes us. And, we looked at two internet service providers that, that belong to the access part. So, those are, networks that are very close to the end user, where you would expect more of D&I fluctuation.

And we do see that. What we were a bit surprised to confirm is that those networks are effectively underutilized. You want to dare a guess what's the average utilization in those networks?

Chris Adams: I literally couldn't, I have no idea what the number might actually be to be honest.

Romain Jacob: Guess!

Chris Adams: Okay is it like Okay, so I said the national grid was about 40%. Is it like 40, 50%? Like, that's, like, not-

Romain Jacob: Four, four zero?

Chris Adams: Yeah, four zero is like what I, is what national, electricity grid is. So maybe it's like, something like that, maybe?

That's my guess.

Romain Jacob: Now you're an order of magnitude too high. So we are talking a couple of percent.

Chris Adams: Oh, wow! Okay, and the whole point about the internet is that if you don't have one route, you can still route other ways. So you've got all these under, you've got all these things which people are currently on that almost no one is using ever, basically, at like 2%. Okay, alright.

Romain Jacob: I need to modulate this, right? Okay. For the couple of networks that we got, we managed to get access to the data, right? So I'm not claiming this is the general number. I would love to know, if you have data, please let us know.

Chris Adams: Mm

Romain Jacob: But for the networks we could get access to, this is the type of numbers you would see. An average utilization of a couple of percent. And again, going back to what we were saying before, in a data center, things would be different.

Chris Adams: Mm hmm.

Romain Jacob: I actually don't know because I was there a little in data center networking, but I would expect things to be more in the 40 50 percent kind of like what you were mentioning before.

But in an end to end service provider network, the underutilization is extreme. There are various reasons for that, but it tends to be the case.

Chris Adams: That's really interesting, because when you look at data centers, so like, I can tell you about the service that I run, or that our organization runs, the Green Web Foundation, so we run a checking service that gets around between 5 and 10 million, like, checks every day, right? So that's maybe In the order of like 400 million per month, for example, something like that.

It's, a relatively high number, for example, and even when we have that, we've got around 50 percent, we, we did, we started working out the environmental impact of our own systems recently, and that's with us with utilization around 50 percent for our systems and in cloud typically you'll see cloud providers saying oh we're really good we're 30 or 40 percent like the highest I've seen is Facebook's most recent stuff about XFaaS and they say oh yeah we can achieve utilization of as high as 60 odd percent right but for lots of data centers the kind of old Data centers would have been in the low digits.

And you've had this whole wave of people saying, well, let's move to the cloud by making much better use of a smaller number one. So it sounds like the same kind of ideas of massive underutilization and therefore huge amounts of essentially hardware, you know, it seems like it's somewhat similar in the networking field as well.

And there's maybe scope for reductions in that field as well. Okay.

Romain Jacob: Exactly. And so, I want to make it clear, like, it's not happening this way because operators are idiots, right? It's just, there are a number of reasons why you have such underutilization. One what I would say is probably the main one from a decision point of view is that you want to provide high performance for number of connections in your network.

So to reach from point A to point B, you want to make sure that you want to have the lowest delay typically. And that requires to have a direct line. Do you have other concerns that are that things do fail in networks. Link failures happen and they can be quite drastic. You operate like a physical infrastructure in a country where people leave and work, you have incidents, fibers get cut, and those are things that take a long time to fix and so on.

So you want to have some resiliency in your network. So that if some part of the network goes down, you can still reroute the traffic the other way around and still have enough capacity to serve that traffic. So you have some names that are not used by default intentionally. So you get to a 2 percent or a couple of percent average.

But you don't want to be at 50 percent because if you are at 50 percent and something really goes down, then you may run into a situation where you don't have enough capacity left to run your business. So the point is, we have such a high underutilization and something that I don't think we explained so far is in network equipment, so routers and switches, you have very little proportionality.

What I mean by that is that the amount of power that is drawn by a router. It's essentially, from at a height, it's not exactly true, but at a high level.

Chris Adams: Hmm.

Romain Jacob: The amount of power drawn is almost independent or varies very little if you send no traffic at all or if you send at 100%.

Chris Adams: Okay.

Romain Jacob: So, what it means is that if you have a router that you use at 1 percent of its total capacity, you pay almost 100 percent of the power.

Chris Adams: It's almost like one person in that plane going back and forwards, for example. Like, if I'm going to fly, there's, you know, if I'm going to fly somewhere and I'm the only person, it's going to be the same footprint as if that plane was entirely full, for example.

Romain Jacob: Kind of, yes. It was the same kind of idea. And this is why for us, investigating this clipping was kind of interesting because we know we have such method underutilization, although I probably would not have guessed it was that low, and it wastes a lot. Third, we are essentially operating most of those links at the worst efficiency point possible. And so we try to remedy this and it goes one step in that direction.

Chris Adams: I see. And I think one thing that you mentioned before was this idea, you can power these things down and you know there's very, because there are alternative routes through the network at any time, it may be the case that even if you do have these things powered off in response to upticks in demand, just like with, say, national grids, people might. You know, switch on batteries to, or feed power into the grid from a battery or possibly a peak of gas plant.

You have, you still have the option of switching these route, these links back on when there is mass a, a, a big peak in power, for example.

Romain Jacob: Exactly. And it's the same with the kitting protocol we proposed, right? So, it still makes sense to have those redundant links deployed, you know, those fibers laid out. And then you may say, yeah, but we've paid all this effort to actually install this and it's there. Why should I not turn it on? Well, because it consumes energy whether you use it or not. That's for one. And second, It's good to have it in case you need it. But you can turn it off so that you save energy. If you can turn it up quickly, right, then it goes back to what I was saying before. The turning of quickly part is still problematic today.

Chris Adams: Hmm.

Romain Jacob: So orders of magnitude that the time it takes to actually do this would be in 10, 10 seconds, roughly a few seconds, let's say, up to more, a minute or so.

So then what, what do you have to wait? Yeah. The benefit you gain by turning links off in terms of energy versus the time you may have to wait until you go back to a good state in your network in case you have some failures in your network. And of course you need to multiply that risk by the likelihood of getting such link failures.

So if, let's say, if you have a doomsday event that, you know, will just kill the network error, but that happens one every hundred year. Maybe you can be fine having a day to day management policy that says, okay, to manage this doomsday event, we will need an hour long, but in all the rest of the time, it will be fine and we'll save energy every single day.

You know, you, you have weighed the pros and cons of a strategy in terms of performance and in terms of energy usage.

Chris Adams: And presumably, one thing that you've entioned is because you mentioned that you often have these regular, kind of, predictable cycles, like, most people don't, you know, fewer people use the internet when they're asleep than when they're awake, for example. Like, it sounds really silly, but like, yeah, you're going to see these predictable patterns.

Some of the work with the Hypnos paper was basically, essentially taking some of these things into account. So you can say, well, you need to have this buffer, but we don't have to have the buffer massive you don't need to have every single car in the world engine on idling just in case you need to use it you can turn off some of these car engines for it so that was the kind of idea behind this.

Okay neat so I've used this car model a few times but it suggests that actually think about how, the amount of energy usage and how it scales with how we use the Internet. It might not be the correct mental model. And I just want to kind of run this by you, because this is one thing that I've been thinking about recently is that a lot of us tend to instinctively reached to a kind of car and driving and burning fuel model, because that's how a lot of expo experience costs of energy a lot of the time, right?

But it feels almost like if you've got this thing, it may be a different model might be like, I don't know, like bike lanes where there's a matter of time that you need to build something. You might need to light, make sure a bike lane is well lit, for example, if you do use, but the amount of people using the bike lane that isn't the big driver of emissions in this, for example, maybe something like that.

Romain Jacob: Yeah, I was about to say, I think the analogy is not wrong per se, it's just a matter of the trade off between the, the infrastructure cost and the driving cost, let's say. So let's assume you, you're using a, a mean of transport, whatever that may be. That has a cost X per kilometer, but then you need light and you need, I don't know, cooling or if you're using something that works on under like, I don't know, superconductive environment, then you need extreme cooling. And so the cost for the environment gets very high. I think the superconducting thing is actually a, a, a pretty, a, a much, very much closer analogy to how the way network works.

And you need to spend a lot of power, or to draw a lot of power, just to get the infrastructure on. But once the infrastructure is on, once you get your superconductive environment, then traversing this environment is very cheap.

Chris Adams: Ah, okay.

Romain Jacob: And networks are a bit like that today, right? So turning the wires on costs a lot, but when it's on, sending the bits through the wire, it's pretty cheap.

Chris Adams: Ah, I see, and if I understand it, when I've spoke to other people who know more about networking than me, they've basically told me that at some levels, even when you're not sending any data, there is a signal being sent that basically says, I'm not sending data, I'm not sending data, I'm not sending data, just to make sure so that you've got that connection so that when you do send some data, there's a fast response time.

So, just because we aren't perceiving something doesn't mean there isn't energy use taking place for example. So there's maybe some leakiness in the models that we might instinctively just use or intuitively try applying when we're trying to figure out, okay, how do I make something more sustainable for example?

Romain Jacob: Yeah, this is very true. And it also, it gets a bit more detailed than that. It also depends on the type of physical layer you use for sending your information. In networks today, you have, I think, I guess we could differentiate between three main types of physical layer. One is the electrical communication, so you send an electrical signal through a power rail.

You have optical communication, so essentially using light that you modulate in some way. And then there's everything that is kind of wireless and radio wave communication. So I'll leave the wireless part out because I know less about it and it's a very complicated bee. But if you compare electrical to optical, things work kind of differently.

In the electrical environment, you, you can, you have essentially a physical connection between the two points that try to talk to each other. And so, when the physical connection is there, you may send messages as you were saying before, like, I'm not sending, I'm not sending, I'm not sending, but you can do this, for example, once it be, I don't know, 30 seconds or so.

It will be enough to, yeah, keep the connection alive. Whereas if you use optical, it's different. Because if you use optical communication, the line does not exist. The line between the two exists only because you have a laser that is sending some photons from one end to the other. So, where it's different is that a laser is an access component.

You need to, to, to send energy to create this link between the two ends.

Chris Adams: Mm.

Romain Jacob: And so, now, it's not that every 30 seconds you need to say, I'm not sending anything. It's like, all the time, you need to have this laser on so that the two endpoints know they are connected to, to each other.

Chris Adams: Ah, okay.

Romain Jacob: And it's actually one of the reasons why the early ideas about tweaking are not so much in use today is because they don't work nicely with optical communication.

And optical communications are the de facto standard in networks today, in the, in the core of the internet and in data centers as well. For reasons that we don't have time to detail, optical is the primary means of communication. And it is by design the laser needs to be on for the link for the communication to exist.

Whether you send data or not.

Chris Adams: Alright, okay, thank you for elucidating this part here. So it sounds like the models we might use a lot of the time, as lots of, when you're working with digital sustainability, it's very common to look at a kind of figure per gigabyte sent, for example, and like in some cases It's better than having nothing, for example, but there is a lot of extra nuance here.

And, there is, we have seen some new papers, I think there was one paper by David Mytton, who, that we'll share a link to, he's been speaking, he's, he shared one recently about the fact that, there are other approaches you might take, for example, for this. There, if you could, just brief, it'd be really nice to just touch on some of that, if we could, and then just, and then to add some extra nuance, realize that, like, It's not that there is no proportion, because there is something you need to do.

Maybe we could just talk a little bit about some of the things that David Mytton's been proposing as an alternative way to figure out a number here, because I'm mainly sharing this for developers who get access to these numbers, and they want to make a number go up or down, and it's useful to understand what goes into these models so that you are incentivizing the correct interventions essentially.

Romain Jacob: Yeah, of course. So that's actually a very important point, I think. You will often find if you look in the, on the web or anywhere. Figures are in energy per bit, or energy per X, or energy per web search, or energy per email sort of thing. Whatever we can think about whether computing such numbers make sense, what, what is very important to understand is that those numbers were derived in an attributional way.

That means that you take the total power cost of a system. And then you divide by the number of bits that were transmitted. If you take a network, you take the sum of the energy consumption of all the routers and all the links and all the calling and all of everything. And then you look at the total amount of traffic you've sent over your reporting interval, like a year.

And you take one, you divide by the other and ta da, you get energy per bit. That is interesting. That is interesting to get an idea of how much, how much energy you spend for the useful work you've done in that network. But it should not be interpreted as, this is the cost for a single bit, because if you do this, then, and that would be a different type of reasoning that we call consequential reasoning.

If you do this, you would then draw the conclusion, the wrong conclusion, that if I have a network that has, I don't know, a hundred kilowatt hour, if I send a hundred gigabit more, I will use ten kilowatt hour of-

Chris Adams: It increases entire system by that rather than my share of this, for example.

Romain Jacob: Exactly. Except that it's not true. It's not true because the total number in watt, in energy per bit accounts encapsulate all the infrastructure costs. And those infrastructure costs are constant, they are independent of the amount of traffic. And so this summary statistics is useful in order to track the evolution of how, how much is used, how much is your network used over time.

But it's not good to predict the effect of sending more or less traffic. And it's a subtle thing that if you overlook this, you can make the very wrong statement and make bad decisions. And so this is what kind of like these papers you refer to try to highlight and explain. And say that you need to have a finer view on the, the energy per unit that you're interested in.

It's a bit more subtle than that. People should read the paper. It's a great paper. It's very accessible. It's not too technical, I think. And it's great for people that are interested in this area to get a good primer on the challenge of computing the energy efficiency of a network. I think it's really a great piece.

Chris Adams: Okay, cool. So basically the, I think one of the implications of that is that let's say I'm designing a website, for example, if I make the website maybe half the size, it doesn't necessarily mean I have half the carbon footprint of it because some of the models we use and they're popular, they are an improvement on having nothing, but there's extra nuance that we might actually have.

I say this as someone who works in an organization where we have a library, we have one, we have a software library called CO2.js. We have a transfer based model for this because this is one of the ones that's most common that is just like one of the defaults. We also have like an issue open specifically about this paper because there is, when you're starting out, you will often reach for some of these things for this.

And while there's benefit and there's some value in actually having some of these models to help you work out, it's also worth understanding that there is extra nuance to this. And they can end up with slightly different incentives for this. This is something we'll talk about carbon aware as well because again, different ways you measure the carbon intensity of electricity can create different incentives as well. So, like, this is one thing we'll be, I guess, we'll be developing over time, but the thing that I just, if we may, I'm just going to touch on this other thing before we move on to kind of wrap up on this.

This can give the impression that there is no proportionality between using digital tools, and, like roll out of extra infrastructure and if you said there's no link that would be an oversimplification as well and I think we're gonna one of the previous guests Daniel Schien he came on he spoke about some of this and maybe you might paraphrase some of this because I think his this perspective is also very helpful and kind of illustrates why we need to be doing coming up with better models to represent this stuff.

Romain Jacob: Yes, exactly. Thanks for bringing that up. That is also extremely important. So I've said before, power is kind of constant. It doesn't depend so much on how much you send. Two, two things to keep in mind. First, there is some correlation. So if you do send more traffic, there will be an increase in power and so you will consume more energy. That is true. And the work that I'm doing and fuel make that even more so in the future. So what we are trying to do is essentially say. We tried to find ways of reducing the power draw when you're under low utilization. And if we were successful in doing that, by sleeping and by other methods, then it will create a stronger correlation between traffic and power.

Right? So, and this is actually good, right? For energy efficient theories, the closer you are to proportionality, the better. So if we are successful, then the correlation will increase. And then sending more bits or, or having smaller website will have a stronger impact, in energy consumption in carbon footprint.

So that's one aspect.

Chris Adams: Mhmm.

Romain Jacob: The second aspect was also extremely important. You mentioned already the, the work of Daniel Schien that is great about this is to think about the internet in, in, in a different timescale. If you look at one point in time right now, the network is the static element. There's so many nodes in the network.

There's so many networks and therefore today, if I send more traffic; be low impact. However, if you put a longer timescale and you look at a one year, six month, or ten year horizon, what happens is that when people spend more traffic, you see the utilization of the links going up, and that will have a future consequence of incentivizing people to deploy new links, to increase the capacity of the network.

Chris Adams: Hmm.

Romain Jacob: Which means that over a year, so over time, as you send more traffic, you create more demand. As you create more demand, you will create more offer. That means scaling up your network, and every time you scale up, almost away, you will increase the energy consumption of the network, right? So, you will further increase the infrastructure costs.

So, as you send more traffic, as you watch more Netflix today, It does not consume more energy, not so much, but it will incentivize the network to be scaled up, and that will consume more energy. So, there is a good reason to advocate for what is known as digital sobriety, to be, to try to use less of the network or to make a more sensible use of the network, because if we use it more, It will incentivize future increase of the digital network size and therefore future increase of the energy consumption.

Chris Adams: I see. Okay. So basically, if you like set these norms of all this extra use, even though you're not making these changes in the meantime, on a kind of large, on a multiple, multi year timescale that people make investments, like infrastructure investments on, they would then respond to make sure that they've got that kind of headroom available over time.

And I think that's actually some of the work that Daniel has been doing to model, because there is no way that we can do that. We're going to deploy new infrastructure where there being zero carbon footprint, even if everything is green. So there is a, there's an impact there that we need to be mindful of.

That's some of the work that he's referring to there. We'll share a link to that paper as well, because I'm not quite sure how Okay, I know that we can't model that in co2.js, for example, in our library, but he's, this is literally the cutting edge work that I think he's been doing. And he's, last time he was on, he was hiring for some researchers to find out, okay, how do you represent this stuff?

Because when we think about large organizations of the scale of Amazon or Microsoft who are spending literally tens of billions of dollars each year, then you do need to think about these kind of multi year, decade style infrastructure kind of investment scale. Okay. All right. So, we've gone really into the details, then we've spoken at the kind of macroeconomic level now.

I wonder if I can just bring this back to the kind of frame for developers who are like, oh, this sounds really cool. How do I use some of this? Or what would I do? Like, if you wanted to have an internet that was able to kind of sleep more and could grow old, are there any ideas you might use? Like how might it change how you build, for example?

Are there any kind of sensibilities you might take into account? Because as I understand it, some of the things with Hypnos were primarily designed to say, this is how you can do this without forcing people to make too many changes at the end user level. But there may be things that as a practitioner you might make things more conducive to or something like that, for example.

Romain Jacob: Yes, definitely. So I think it goes back to the question we had just before about Daniel's work, about looking at the longer time scale perspective. I think as an end user, as a software designer, I think thinking about sobriety is something that everybody should be doing. It's not just for sustainability.

I think one very recent Environment Variables podcast was about the alignment between the sustainable practices and the financial operations, and in many cases, those two things align. In a similar mindset, if you think about web design, this, if you look at the system, the WebW3C sustainability guideline, they align pretty much almost perfectly with the accessibility guideline and with the performance optimization guideline. Why? Because a smaller website will also load faster and, you know, get people faster to what they want. This is the content they really want to consume. So, there's general value into being as, modest in your demand from the system or from the network as, as possible and for the compute as well.

It's the same thing. Today, it does not yet translate into net benefits. At the network level, but it might in the future. And, you know, somebody has to stop. So you need to, I need the efforts of all sides in order to, you know, make that work. In the networking domain, I know there have been some people studying this from a theoretical point of view where you would say, the end user could be able to say, I want to, I want to place a phone call.

But I'm willing to wait for, I don't know, 20 seconds or 30 seconds before my call is being played. And if you have an ecosystem of users of that network, where the sufficiently large share of users are so called delay tolerant, then you can optimize your network in order to save in resources. So saving energy and ultimately in reducing your carbon footprint.

In the more traditional networking domain, one could envision something like that. There is no work in this area, as far as I know. One way you can think about this would be, the incentive would be pricing. That you, you, you could say, okay, I'm, I'm winning. So not so much to wait the most, most likely to cap the bandwidth I can get out of the network.

But you would, if you were to say, I'm winning to get at most, I don't know, 100 megabits per second in high utilization times, then you would get a discount on your internet deal. I think that's sizable, that's possible. One quick working thing that could happen. But that would be more like a global effect, like it's between the user and the internet service provider.

If you're a software developer, If you think about how your application could be built in such a way, I honestly don't know. I think it's extremely easy today, technically speaking, to have any sort of flagging where your application can say, I'm data intolerant, I can wait, I will not use more than an egg.

One can do that, that's easy. Your network can get that information. The tricky bit is, how would the network then use that information to route traffic in a way that would save energy? That is much trickier.

Chris Adams: Ah, okay, so that sounds like a possible route that people might choose to go. Because I think, I, I know, for example, there is some work in the world of streaming, where there was a notion of a, I think it was the gold button, that was put together by the Greening of Streaming group. They were basically saying, look, most of the time, I, if I'm looking at television from across the room, I can't really tell if it's 8K or 4K.

So, allow me to, you know, have a default which lets me kind of, reduce the resolution or the quality so that when there's lots of people trying to use something, we can see the amount of data reduce somewhat. And that reduces the amount of kind of extra peak capacity people might need, for example. These are some ways to kind of make use of the existing capacity that lives inside the entire network to kind of smooth off that peak as it were, for example.

That's some of the stuff that we might be looking at. So it's very, in some ways, it might be, kind of providing hints to when you send things over the network. And I think there's actually some work that we've seen from some existing tools. I know that Facebook's serverless platform does precisely this.

And there is also some work in Intel, we'll share some links to this, where when you have a computing workload, you can basically say, well, I'm not worried about when this gets delivered, for example, or I have a degree of, as long as it happens before this time, it's okay. And this does provide the information for people running these systems to essentially, like, move things around to avoid having to increase the total capacity, for example.

So you make better use of the existing capacity you have before you have to buy new capacity or deploy new wires or anything like that. That seems to be what you're kind of suggesting.

Romain Jacob: In the cloud computing world, that does exist for real. Yeah, for sure.

Chris Adams: Okay, so there's a possible path for future, future research. And maybe this is the thing I, this is what we can kind of wrap up on. So we've spoke and we've done a dive into sleeping and getting old. Right. But that's not the only tool available to us. Are there any kind of papers or projects or things that you would direct people's attention to that you think is really exciting but may not necessarily be in your field that you think is worth, that you're excited about?

For example, because you spend a lot more time thinking about networks than I do, and I'm pretty sure there's some things you might say that that, that other people listening here might, might enjoy following, for example.

Romain Jacob: Yeah. Yeah. So I think two things come to mind. The first is that today I talked about protocol adaptation, about, you know, we, we would put things to sleep and save energy and so on. But one, one key problem we have, we as practitioners in this field and researchers and operators is the lack of visibility in the power data.

It's actually really hard to get a good understanding about how much power is going to be drawn by a given router, depending on the amount of traffic and depending on how this thing is configured and so on. So, this became very clear at the beginning where we started working on this. And so, a big part of my research has been to try to, to develop tools for building datasets to aggregate such power information in a way that people can contribute to and then use in their own research, do their own analysis.

And try to do some predictions about, okay, now, if I were to buy this device, for example, it will cost me so much in embodied carbon footprint that this is how much I could hope to save. Because I know how this device typically operates and how much it consumes. And so, this led to a dataset on a platform project we call now the Network Power Zoo.

Which is actually a reference to another very well known networking data set that was the Network Topology Zoo. It's kind of like a historical reference to that. But it is really a zoo in a sense that it's very broad. From devices that look the same can consume from, I don't know, two, three times more magnitude power.

Whereas it seems there's the same number of ports, the same part of the number of connections, but it can change drastically. So, it's still a work in progress. I mean, the database is in building and we're starting pouring data in. And very soon we'll do an open call for anyone to contribute their own data sources into this database.

So that people can have access to richer power related data.

Chris Adams: For a data informed discussion. Yeah.

Romain Jacob: So that's one thing that is very active for us. It's still very much related to networking, but that's one thing that is not related to protocols so much. More generally, what I think has become clear and clear to me is that if we want to address the sustainability problems in networking, what we need are not really networking researchers, because a lot has to do with the hardware design, and the hardware architecture, and writing good software for optimizing the hell out of the hardware we get.

And those are just not the typical expertise that you find in networking people. So, networking people are a bit at a protocol, but, you know, they don't know as much how the hardware is built and designed. Maybe I should not make such generalities, but it's definitely true for myself. So, we've been poking more and more people from the computer architecture area, from the hardware design to collaborate with us and say, okay, look, we have those sorts of needs.

This is mainstream in embedded systems for 20 years. We still don't have it yet in routers, it's hightime, we need it now. If anyone is working in this area and is interested, please reach out to me, I'll be happy to chat.

Chris Adams: Brilliant, thank you for that. Okay, and I suspect there's some, I'm just looking through our notes. We actually have been in touch on the E Impact mailing list, which is one project, but I think it's by the IETF. That's one of the things that we can share a link to where there's often quite a lot of, if you want to go into the networking, that's probably one of the deeper ones I've found.

Okay, great. I think, is there anything else? I should, we're just coming to time, so I just want to check. This has been really fascinating, and I've learned a huge amount from this. So, if people are interested in the work that you're doing or they want to learn more about this, where should people look to find more about this?

Like, we'll share a link to the paper that you worked on, for example. But beyond that though, where do we find out what's going on with Romain Jacob and his team of research, researchers?

Romain Jacob: So, nowadays the best place to find me would be on LinkedIn. So, we'll add a link to my profile, but my name is not that common, usually I'm findable on LinkedIn at least. Yeah, that's the best place for you to reach out, I'm quite reactive there.

Chris Adams: Okay, cool. So I'll share the link there. I'll also share the link to HotCarbon, which had the paper that you had. If there's a, and if there's a link with ETH, the research institution you're part of, I'll add a link to that as well. Brilliant! Well, this has been really enlightening for me and hopefully other people who've been listening along with this.

Thanks once again for being so generous with your time, Romain, and, yeah, have a lovely holiday over the summer, okay? Take care of yourself.

Romain Jacob: Thanks, bye bye.

Chris Adams: bye! Hey everyone, thanks for listening! Just a reminder to follow Environment Variables on Apple Podcasts, Spotify, Google Podcasts, or wherever you get your podcasts.

And please, do leave a rating and review if you like what we're doing. It helps other people discover the show, and of course, we'd love to have more listeners. To find out more about the Green Software Foundation, please visit greensoftware.foundation. That's greensoftware.foundation in any browser.

Thanks again and see you in the next episode!