Technology:Unpacked

This week Chris and Martin talk to J Metz, R&D Engineer in Cisco Systems’ Advanced Storage division. This is a two-parter – the second episode will be published on Monday 30th September 2019.

The NVMe 1.4 specification has recently been published and is in the final 45-day ratification process. This discussion provides some background on NVMe then dives pretty deep into the NVMe standards and how design decisions are refactoring the NVMe protocol for version 2.0.

NVMe for consumer devices will be different to the feature set used within enterprise storage (although the protocol is the same). With NVMe-over-Fabrics, the underlying protocol provides independence in fabric choice, although the specific platform used will have an effect on performance. It’s all about trade-offs.

In part 2, will look a bit more into the specifics of NVMe 1.4 and what’s new.

For information on NVMe, check out our NVMe in the data centre report, available here.

Further information on NVM Express can be found here – https://nvmexpress.org/. You can find details on Fibre Channel at http://fibrechannel.org and more on J at http://jmetz.com

Finally, here are some useful links on NVMe:

* NVMe for Absolute Beginners* NVMe over Fabrics for Absolute Beginners* NVMe Program of Study

Elapsed Time: 00:24:55

Timeline

* 00:00:00 – Intros* 00:01:50 – What is NVMe? Think memory.* 00:03:47 – NVMe over Fabrics – what is it?* 00:04:55 – Memory-to-Memory via q-pairs* 00:07:15 – How are NVMe standards developed?* 00:09:30 – Does it matter what release my device supports?* 00:12:00 – NVMe standards will be refactored with NVMe 2.0* 00:14:30 – Will fabric switches look different from each vendor?* 00:16:30 – Will terminology change with NVMe?* 00:19:15 – Protocols will be a technical choice, not idealogical* 00:20:30 – The Storage Triumvirate of Evil….* 00:24:50 – Wrap Up (for now)

Related Podcasts & Blogs

* #115 – The Transition to NVMe-oF Storage Solutions with Praveen Asthana* #97 – Building Storage Using NVMe/TCP with Kam Eshghi from Lightbits Labs* #89 – Choices in NVMe Architectures* #61 – Introduction to NVM Express with Amber Huffman

J’s Bio

J is currently an R&D Engineer for Advanced Storage in the Compute and Server Group for Cisco Systems, but has a broad and eclectic background of both academic, corporate, and industry experience. He is an award-winning public speaker, author, and contributor to industry trade publications. He has been active in industry standards, with membership on the Board of Directors for the Fibre Channel Industry Association (FCIA), Storage Networking Industry Association (SNIA),

This week, Chris and Martin catch up with some of the storage hardware on show at Flash Memory Summit 2019. FMS is essentially a trade show and so there are lots of technologies on display that haven’t quite made it to the enterprise. Technology vendors are selling to other vendors and this year there were some interesting hardware solutions to see.

Flash has been on the agenda for the last decade and continues to push the boundaries of capacity and performance. Chris and Martin discuss 96+ layer products and the promise of 500 layers in the future. PLC or Penta-Level Cell NAND may be on the horizon but how would it be used? Could Optane (3D-XPoint) get cheaper and kill off new solutions like Z-NAND and XL-FLASH?

The discussion moves on to look at computational storage, a subject we recently discussed with NGD Systems. There still needs to be some work done to make these technologies easily adoptable, such as trusted code and device management. Storage Acceleration looks to improve performance of traditional databases, with Xilinx and Pliops showing solutions.

Finally, Martin and Chris discuss Burlywood and the optimising of SSDs. Storage hardware is definitely back!

Vendors covered in this podcast:

* Toshiba (XL-FLASH)* Samsung (Z-NAND)* NGD Systems (Computational Storage)* Xilinx & Pliops (FPGA acceleration)* Burlywood (SSD optimisation)* Liqid (Composable infrastructure)

Elapsed Time: 00:30:06

Timeline

* 00:00:00 – Intros – Friday 13th!* 00:00:44 – Flash Memory Summit 2019* 00:01:30 – Where is NAND Flash headed? 96 Layers!* 00:03:00 – The mythical 100TB drive* 00:03:56 – Is PLC (penta-level cell) a reality?* 00:06:00 – Intelligent flash drives could be on the way* 00:06:40 – Zee-NAND or Zed-NAND?* 00:07:54 – Could price declines in Optane kill of Z-NAND & XL-FLASH? * 00:09:15 – Compute and storage getting closer – computational storage * 00:10:40 – Compute on storage has been done before….* 00:12:14 – Who will write code for computational storage devices? * 00:15:10 – How will the codebase be managed across thousands of drives?* 00:16:20 – Storage Accelerator cards using FPGAs (Xilinx & Pliops) * 00:19:00 – Composable (e.g. Liqid) with accelerator cards* 00:20:50 – Are future storage startup exits likely to be via hyper-scalers?* 00:22:23 – Burlywood – SSD optimisation* 00:24:30 – Burlywood has so many acquisition choices….* 00:26:00 – Public cloud will be a rich seam of future storage podcasts* 00:28:04 – So are we seeing a renaissance in storage hardware?* 00:29:21 – Wrap Up

Related Podcasts & Blogs

* #117 – Introduction to Computational Storage with NGD Systems* #113 – The Expanding Storage Hierarchy with Erik Kaulberg* The Expanding Storage Hierarchy* What is Software Composable Infrastructure?

Copyright (c) 2016-2019 Storage Unpacked.

This week, Chris is in Austin, Texas at Pure Accelerate 2019. Prior to the event, Chris catches up with Patrick Smith, Field CTO for EMEA for Pure. Patrick moved to Pure after being a customer. The initial discussion covers why flash as a platform offers better reliability and consistency than traditional disk based storage.

At Accelerate 2019, Pure announced FlashArray//C, DirectMemory modules and the GA of Cloud Block Store. Patrick previews these changes and what benefits customers can expect to see from these new products.

Elapsed Time: 00:30:07

Timeline

* 00:00:00 – Intros* 00:03:00 – What are customer experiences in moving to Flash?* 00:06:30 – Flash is about consistency as much as performance* 00:08:50 – Dedupe & compression helped justify flash costs* 00:09:55 – Flash removed significant portions of design and deploy headaches* 00:12:00 – The swing from generalists to specialists and generalists again* 00:13:34 – Flash also improves availability* 00:15:00 – Businesses wanted to be like hyperscalers* 00:17:00 – AWS has offloaded storage functionality to hardware 00:18:30 – Announcements – FlashArray//C* 00:21:55 – DirectMemory Flash for read performance acceleration 00:24:00 – CloudSnap extends to Microsoft Azure* 00:24:50 – Cloud Block Storage is now GA* 00:29:30 – Wrap Up

Related Podcasts & Blogs

* #51 – Pure Accelerate Pregame* FlashBlade 2.0 with Rob Lee* Pure Storage Microsite

Copyright (c) 2016-2019 Storage Unpacked. No reproduction or re-use without permission. Podcast episode #8e6d.

This week’s episode is the final recording from Flash Memory Summit 2019. Chris is joined by the team from NGD Systems for a discussion on Computational Storage. On this podcast are Scott Shadley, VP of Marketing, Nader Salessi, CEO and Ashok Savdharia who has the lovely title of “technologist”.

Computational storage uses the processing power of modern SSD controllers to run application workloads directly on the SSD itself. Imagine each drive being a mini blade server, with compute, memory and storage. In the case of NGD Systems, this means a range of devices and form factors in a product family called Newport.

Aside from the fascinating ability to run applications directly on an SSD, what could be really attractive about in-situ processing with computational storage is the ability to use the local compute capability with little or no power overhead. The team explains how this is achieved and explain some of the use cases that make computational storage devices an interesting area to watch, especially for edge computing.

In the recording, Scott references the SNIA working group, which can be found online here – https://www.snia.org/computational.

Elapsed Time: 00:29:22

Timeline

* 00:00:00 – Intros* 00:01:30 – What is Computational Storage?* 00:02:40 – How real are the products & solutions today?* 00:03:55 – Form factors – what is available?* 00:07:30 – Running Containers & Linux for applications* 00:10:00 – Products should be as cheap as standard drives* 00:12:00 – ASIC than FPGA is a cheaper solution at scale* 00:13:30 – What applications run on a computational device?* 00:15:43 – Cluster file system – share the data with the host* 00:17:30 – Drives can run databases like MongoDB and TensorFlow * 00:20:00 – What manages workload processing across multiple devices?* 00:22:25 – Computational devices will have strong edge use cases* 00:26:00 – Customer creativity will generate new and interesting solutions* 00:28:00 – Where are the gaps?  What needs to be done? * 00:29:20 – Wrap Up

Related Podcasts & Blogs

* #57 – Storage on the Edge with Scott Shadley* #37 – State of the Storage Union with Chris Mellor* The Practicality of In-Situ Processing

Copyright (c) 2016-2019 Storage Unpacked. No reproduction or re-use without permission. Podcast episode #d260.

This week, Chris chats with Andy Watson, CTO at WekaIO in another episode recorded live at Flash Memory Summit 2019.

With the recent acquisition of E8 Storage and Elastifile by hyper-scale cloud companies, is this move an attempt to plug the gap in existing cloud storage offerings?

Andy provides his opinion on the two acquisitions and how they might fit into the existing ecosystem. Even 1GB/s may not be enough throughput as we see more focus on ML/AI applications and using cloud-based GPUs and TPUs. Some applications expect data in both NFS and SMB format. That could be a challenge for some solutions.

These purchases could be opportunistic and we may need to keep watching to see how cloud providers will move to the next level of storage performance.

You can find out more about WekaIO at and follow Andy on Twitter at https://twitter.com/the_andywatson.

Elapsed Time: 00:22:54

Timeline

* 00:00:00 – Intros* 00:01:00 – Quick overview of WekaIO* 00:05:00 – E8 Storage & Elastifile acquisitions by public cloud providers* 00:07:10 – Object storage isn’t good for performance, block is not scalable* 00:08:20 – AWS SMB support was implemented via a separate feature* 00:10:39 – Some applications are inherently multi-protocol* 00:12:20 – 1GB/s isn’t enough performance to keep a GPU busy* 00:15:30 – Could latest purchases be opportunistic?* 00:18:00 – The public cloud will have greater file demands in the future* 00:21:55 – Wrap Up

Related Podcasts & Blogs

* #52 – An Introduction to WekaIO Matrix with Liran Zvibel* Can the WekaIO Matrix file system be faster than DAS?* WekaIO Presents at Storage Field Day 18* #92 – Introduction to Elastifile with Jerome McFarland* Disaggregated Storage Part II with Zivan Ori from E8 Storage

Copyright (c) 2016-2019 Storage Unpacked. No reproduction or re-use without permission. Podcast episode #8818.

This week, Chris catches up with Praveen Asthana from Exten Technologies in an episode recorded at Flash Memory Summit 2019. With NVMe and NVMe over Fabrics set to become the dominant technologies for public cloud and the enterprise, exactly how will the transition from current solutions occur?

Praveen discusses the continuing need for centralised storage that can fully exploit the benefits of NVMe. This will need another transition in technologies in the same way flash created a whole new market of storage solutions some ten years ago.

An interesting aspect of NVMe adoption is the way hyperscalers will use the technology. The need for NVMe in public cloud was demonstrated by the acquisition of E8 Storage by AWS earlier this year. Hyperscalers will use NVMe not just for speed but to optimise the use of hardware and reduce costs.

Exten Technologies is developing solutions in software that allow the efficient use of hardware from a range of vendors. The architecture is designed to exploit micro-services with efficient processor utilisation.

You can learn more about how Exten Technologies is approaching the NVMe market at https://exten.io.

Elapsed time: 00:24:17

Timeline

* 00:00:00 – Intros* 00:01:55 – Is the adoption of NVMe reflective of SAN 1.0?* 00:03:30 – Whatever happened to application-level resiliency?* 00:05:30 – The move to flash – retro fit and new solutions* 00:07:00 – NVMe solutions aren’t all end to end – some use SAS* 00:09:30 – Will NVMe force a build from scratch for vendors?* 00:11:00 – Will all applications need the performance NVMe offers?* 00:12:30 – Use cases – HPC, analytics* 00:14:00 – How will hyperscalers use NVMe solutions?* 00:16:35 – What does enterprise adoption look like?* 00:18:00 – EXTEN is developing a solution via software* 00:20:35 – Enterprises like appliances* 00:20:55 – Will NVMe drive new (unexpected) solutions?* 00:23:30 – Wrap Up

Related Podcasts & Blogs

* What is Software Composable Infrastructure?* Performance Analysis of SAS/SATA and NVMe SSDs* #89 – Choices in NVMe Architectures

Copyright (c) 2016-2019 Storage Unpacked. No reproduction or re-use without permission. Podcast episode #C7A0.

This week, Chris catches up with Clay Ryder from the DCS (Data Centre Systems) group at Western Digital. The discussion focuses on the challenges for today’s CIO, with an emphasis on storage and data.

The division of responsibility between CTO and CIO isn’t always obvious. As we learn in this discussion, there’s a lot of overlap and a lot of interaction between the two. CIOs are focused on data and more specifically information, whereas the CTO has to deliver the right platform to meet application needs.

The conversation covers challenges around technical debt, determining what data to retain and what to discard, as well as how the public cloud should be used. Clay describes the idea of a “golden copy” of data, sitting on-premises, with cloud rented for compute.

Finally, the discussion closes with a view on whether the modern CIO has more or less to worry about than 20 years ago. Is it just more of the same?

You can find out more information about Western Digital online at https://www.westerndigital.com/ and in particular the blogs Clay references at https://blog.westerndigital.com/.

Elapsed Time: 00:37:13

Timeline

* 00:00:00 – Intros* 00:01:10 – How do CIO & CTO responsibilities differ?* 00:01:54 – Information or Data – what’s the difference?* 00:03:00 – What are CIO top concerns – security* 00:04:30 – Technical debt, worse in IT than other industries* 00:10:13 – Re-using data for future value, what are the challenges?* 00:14:00 – Automation will be needed to manage data sources* 00:15:30 – Is public cloud the solution for everything IT?* 00:20:00 – The disadvantage of opex-based charging* 00:24:00 – Creating a golden copy on-premises* 00:25:46 – Is the CIA listening?* 00:27:02 – Should CIOs care about technologies like NVMe?* 00:32:00 – Unknown unknowns, understanding what’s possible* 00:34:00 – Are modern CIO challenges the same as they ever were?* 00:36:03 – Wrap Up

Related Podcasts & Blogs

* #62 – The Future of Data Infrastructure with Scott Hamilton* Western Digital Redefines DRAM Caching* What is Software Composable Infrastructure?

Copyright (c) 2016-2019 Storage Unpacked. No reproduction or re-use without permission. Podcast episode #BB2D.

This week’s episode is the first in a series from Flash Memory Summit 2019. Chris catches up with Erik Kaulberg from INFINIDAT to discuss how the expanding hierarchy of solid-state storage media will drive new products and solutions in the future.

The topics of conversation cover how media has diversified, with new solutions like MRAM and 3D-XPoint. At the same time, we hear that 5-level NAND could be a reality. How are vendors combining these products with traditional storage?

Will the ability to learn from the field with the deployment of traditional storage designs, provide an advantage over SDS? Will vendors selling both hardware and software solutions have an advantage over those only playing in one market? These are just some of this week’s topics of conversation.

Elapsed Time: 00:25:55

Timeline

* 00:00:00 – Intros* 00:01:50 – What changes are we seeing in the media industry?* 00:04:00 – How does solid state media expansion parallel HDDs?* 00:07:00 – New flash solutions will not necessarily improve performance* 00:08:00 – It’s not tiering, it’s new techniques to rebalance active data* 00:10:00 – Storage needs to get simpler, not more complex over time* 00:13:00 – How will benchmarks work in the future, are they done?* 00:14:20 – How is INFINIDAT using a mix of media?* 00:16:00 – VAST Data – good example of new media in new systems* 00:17:00 – Will storage systems kill off SDS?* 00:21:30 – Where will the solid-state market go?  More products?* 00:23:30 – Will we see new a new generation of hardware platforms? * 00:25:00 – Wrap Up

Related Podcasts & Blogs

* #105 – Introduction to VAST Data (Part I)* #89 – Choices in NVMe Architectures* #82 – Storage Predictions for 2019* 2019 is the year of NVMe* Will TCO Drive Software Defined Storage?

Erik’s Bio

Erik Kaulberg is a Vice President at Infinidat (https://infinidat.com), leading cloud strategy including the Neutrix Cloud storage service and FLX private cloud business model, key alliance partnerships like VMware and Cisco, and analyst relations. He has broad expertise in enterprise storage and frequently engages key customers, partners, and analysts. Erik previously ran worldwide enterprise storage strategy and business development for IBM, after he sold all-flash array innovator Texas Memory Systems to the company.

Copyright (c) 2016-2019 Storage Unpacked. No reproduction or re-use without permission. Podcast episode #6C8C.

NetApp has developed a new backup service called NDAS or NetApp Data Availability Services. NDAS is based in the public cloud and provides the ability to run analytics against secondary data in AWS S3, without having re-hydrate through a backup platform. Chris met with Charlotte Brooks (Technical Marketing Engineer) and Joel Kaufman (Director of Technical Marketing) to discuss how NDAS works and what customers are doing with their cloud-based data.

NDAS provides some interesting features that make the product slightly different from existing backup software. The solution runs in AWS using the customer’s account. This ensures data is secure, but also provides the ability to access data directly and run other services (like analytics) against the backup data.

NDAS uses a feature called libC2C to directly access data in S3 as if it was a file system. This self-describing capability means data can be accessed long after the backup software has been shut down, or even decommissioned.

You can read more on NDAS in this blog post that also links to some great Tech Field Day presentations explaining NDAS in more detail.

Elapsed Time: 00:31:51

Timeline

* 00:00:00 – Intros

* 00:01:00 – What happened to Joel’s beard?

* 00:02:00 – NetApp Data Availability Services – NDAS

* 00:05:00 – NDAS is an enabler for getting data into the public cloud

* 00:06:30 – Full SaaS or run in the customer’s own AWS account?

* 00:08:05 – NDAS is deployed through the AWS Marketplace

* 00:10:28 – Today just AWS, possibly other clouds and on-prem in the future

* 00:12:20 – What do we mean by self-describing data?

* 00:16:00 – Scanning file systems isn’t efficient – how is metadata indexed?

* 00:17:44 – What else can customers do with this secondary data?

* 00:20:00 – IT relevance – building an internal catalogue of services

* 00:23:00 – How are customers using NDAS data?

* 00:29:39 – Any teasers for futures?

* 00:31:16 – Wrap Up

Related Blogs & Podcasts

* Exploiting Secondary Data with NDAS from NetApp* #91 – Storage Field Day 18 in Review

Copyright (c) 2016-2019 Storage Unpacked. No reproduction or re-use without permission. Podcast episode #4E81.

This week Chris is in Silicon Valley and catches up with Rawlinson Rivera, Field CTO at Cohesity. The company recently released a new feature called Marketplace that enables customers to run data-focused applications directly on the Cohesity platform.

The idea of running applications on data protection hardware has some benefits and potential disadvantages. Naturally, the focus is to provide a single point of truth for secondary data, reducing the risk of having many teams and departments storing their own data copy. But is DataPlatform capable of delivering the performance requirements of AI and ML?

Rawlinson takes us through some of the company strategy and sets the scene for what could be coming up in the future. We could speculate that Marketplace could run on public cloud and even be a profit centre for the business. We will have to wait and see.

You can find more details on Marketplace at the Cohesity website. Here’s a link to the Tech Field Day website where you can find a great introduction to Marketplace.

Elapsed Time: 00:31:11

Timeline

* 00:00:00 – Intros* 00:01:20 – Background, what is the Cohesity DataPlatform?* 00:04:00 – Perhaps “secondary” isn’t the right term.* 00:05:30 – What is the Cohesity Marketplace?* 00:09:01 – What applications does the platform run?* 00:11:24 – How is useful data extracted from virtual machines?* 00:14:00 – Are the marketplace applications free?* 00:17:09 – Marketplace enables application consolidation* 00:18:00 – Marketplace could become a profit centre* 00:19:40 – DataPlatform is totally API driven, apps via SDK* 00:23:00 – Is the process of managing apps a scalable one?* 00:25:29 – What examples are there of customer use cases?* 00:26:41 – How does Imanis Data fit into the platform?* 00:28:44 – Wrap Up

Related Podcasts & Blogs

* #104 – Creating a Data Management Strategy with Paul Stringfellow* #81 – Storage or Data Asset Management?* The $2 billon Gamble on Data Management* The Need for APIs in Storage and Data Protection

Copyright (c) 2016-2019 Storage Unpacked. No reproduction or re-use without permission. Podcast episode #E512.