Technology:Unpacked

This week’s podcast looks at storage networking and in particular the choice of using Ethernet vs Fibre Channel as the network protocol.  Traditionally, enterprise storage platforms have been Fibre Channel connected, with only a small amount of iSCSI usage.  However, the world isn’t just block storage and in fact, as guest Marty Lans (General Manager, Storage Connectivity Engineering & Global Interoperability Business Unit at HPE) tells us, 80% of storage is Ethernet connected.  This is because of the growth in unstructured data stored on NAS and object stores.

Ethernet storage now includes lossless Ethernet (DCB) and has done for some time.  Ethernet is faster than Fibre Channel in raw throughput, yet Fibre Channel continues to see modest growth, according to Lans.  The next wave of storage networking is coming along with the introduction of NVMe.  This will bring FC-NVMe and NVMe-oF protocols, both improving performance for shared storage.  So is it the end for Fibre Channel, should the default be to move to Ethernet?  The answer isn’t that clear cut, as Marty explains.

Elapsed Time: 00:27:57

Timeline

* 00:00:00 – Intros

* 00:01:30 – Summarising the storage networking landscape

* 00:02:35 – 80% of all storage is attached via Ethernet

* 00:04:00 – Lossless Ethernet and DCB.

* 00:06:00 – Why use Ethernet instead of Fibre Channel?

* 00:07:00 – HPE doesn’t see the Fibre Channel business shrinking.

* 00:08:50 – Whatever happened to Fibre Channel over Ethernet?

* 00:12:00 – Cloud service providers – all Ethernet, surely?

* 00:13:00 – NVMe, FC-NVMe and NVMe-oF

* 00:18:00 – How have form factors changed, TOR switches?

* 00:20:00 – Is there a storage equivalent of SDN?

* 00:23:50 – Should customers be actively moving away from FC?

* 00:25:00 – Where does storage networking go next?

Related Podcasts & Blogs

* #57 – Storage on the Edge with Scott Shadley

* #55 – Storage for Hyperscalers

* #47 – Enterprise Storage is Not Boring

* Garbage Collection #005 – Disaggregated Storage

* NVMe over Fabrics – Caveat Emptor?

* The Race towards End-to-End NVMe in the Data Centre

* One (Storage) Protocol to Rule Them All?

Marty’s Bio

Marty Lans (General Manager, Storage Connectivity, HPE Complete, Storage Qualification & Interoperability Engineering) joined HPE in 2013 and is responsible for innovating and developing next generation storage connectivity technology, implementing third party ecosystem partnerships with HPE Complete and storage qualification and interoperability engineering. Prior to joining HPE, Marty had more than 20 years of storage and networking industry experience with executive roles at leading technology companies including EMC, Cisco Systems, Brocade Communications and Juniper Networks.

Copyright (c) 2016-2018 Storage Unpacked.  No reproduction or re-use without permission. Podcast Episode I5FC.

This week, Chris Mellor is back and the team take a dive into the subject of storage vendor hero numbers.  What are hero numbers?  We’ve all seen them, they’re the huge performance figures quoted by all-flash storage vendors aimed at putting their products forward in the best light possible.

Are hero numbers believable or should we be looking at certified vendor benchmark testing as a guide to capability?  Do users even look at benchmark or hero numbers in the first place?  Could the whole exercise be a waste of time?  The conversation moves to talk about STAC (Securities Technologies Analysis Center) and whether more application specific benchmarks have greater value.  This includes the approach Pure Storage appears to have taken with FlashArray and FlashBlade, only quoting application-specific testing like that of AIRI.

As the discussion moves on, the team talk about more consistent deterministic latency (more on that here) and whether public cloud service providers will ever subject their service offerings to benchmark testing (TL;DR, probably not).  As a final wrap up, the team asks, what can customers actually do to ensure they are seeing the real truth behind benchmark claims.

Elapsed Time: 00:20:48

Timeline

* 00:00:00 – Intro

* 00:01:00 – What are vendor hero numbers – ten meeelion IOPS!!

* 00:02:00 – But certified benchmarks are expensive to perform.

* 00:04:00 – Storage arrays aren’t interesting, do hero numbers help?

* 00:05:00 – Do users even care about industry standard benchmarks?

* 00:06:00 – STAC – financial application benchmarks

* 00:07:48 – Should vendors be more realistic about hero numbers?

* 00:09:50 – Why has Pure Storage stopped producing FlashArray performance data?

* 00:11:40 – However Pure do produce numbers for CI like AIRI (see Soundbytes #008).

* 00:14:30 – Do people do proofs of concept any more?

* 00:16:00 – Maybe arrays are massively over specified anyway?

* 00:16:30 – Or actually, it’s about consistent deterministic performance.

* 00:17:44 – Do we need performance benchmarks for cloud vendors?

* 00:19:00 – So what should customers do?

* 00:20:08 – Wrap Up

Related Podcasts & Blogs

* Pure Storage AIRI – CI for AI

* Why Deterministic Storage Performance is Important

* Avoiding the Storage Performance Bathtub Curve

* Avoiding All-Flash Missing IOPS

* Are DataCore’s SPC Benchmarks Unfair?

* NFS Vendor Benchmarks – Caveat Emptor

* Storage Performance: Why I/O Profiling Matters

Vendors referenced in this post: Violin Systems, Dell EMC (PowerMax), Pure Storage (FlashBlade & FlashArray), NetApp.

Copyright (c) 2016-2018 Storage Unpacked.  No reproduction or re-use without permission. Podcast Episode UGE3.

This week, Chris Mellor is back and the team take a dive into the subject of storage vendor hero numbers.  What are hero numbers?  We’ve all seen them, they’re the huge performance figures quoted by all-flash storage vendors aimed at putting their products forward in the best light possible.

Are hero numbers believable or should we be looking at certified vendor benchmark testing as a guide to capability?  Do users even look at benchmark or hero numbers in the first place?  Could the whole exercise be a waste of time?  The conversation moves to talk about STAC (Securities Technologies Analysis Center) and whether more application specific benchmarks have greater value.  This includes the approach Pure Storage appears to have taken with FlashArray and FlashBlade, only quoting application-specific testing like that of AIRI.

As the discussion moves on, the team talk about more consistent deterministic latency (more on that here) and whether public cloud service providers will ever subject their service offerings to benchmark testing (TL;DR, probably not).  As a final wrap up, the team asks, what can customers actually do to ensure they are seeing the real truth behind benchmark claims.

Elapsed Time: 00:20:48

Timeline

* 00:00:00 – Intro

* 00:01:00 – What are vendor hero numbers – ten meeelion IOPS!!

* 00:02:00 – But certified benchmarks are expensive to perform.

* 00:04:00 – Storage arrays aren’t interesting, do hero numbers help?

* 00:05:00 – Do users even care about industry standard benchmarks?

* 00:06:00 – STAC – financial application benchmarks

* 00:07:48 – Should vendors be more realistic about hero numbers?

* 00:09:50 – Why has Pure Storage stopped producing FlashArray performance data?

* 00:11:40 – However Pure do produce numbers for CI like AIRI (see Soundbytes #008).

* 00:14:30 – Do people do proofs of concept any more?

* 00:16:00 – Maybe arrays are massively over specified anyway?

* 00:16:30 – Or actually, it’s about consistent deterministic performance.

* 00:17:44 – Do we need performance benchmarks for cloud vendors?

* 00:19:00 – So what should customers do?

* 00:20:08 – Wrap Up

Related Podcasts & Blogs

* Pure Storage AIRI – CI for AI

* Why Deterministic Storage Performance is Important

* Avoiding the Storage Performance Bathtub Curve

* Avoiding All-Flash Missing IOPS

* Are DataCore’s SPC Benchmarks Unfair?

* NFS Vendor Benchmarks – Caveat Emptor

* Storage Performance: Why I/O Profiling Matters

Vendors referenced in this post: Violin Systems, Dell EMC (PowerMax), Pure Storage (FlashBlade & FlashArray), NetApp.

Copyright (c) 2016-2018 Storage Unpacked.  No reproduction or re-use without permission. Podcast Episode UGE3.

This week, Chris Mellor is back and the team take a dive into the subject of storage vendor hero numbers.  What are hero numbers?  We’ve all seen them, they’re the huge performance figures quoted by all-flash storage vendors aimed at putting their products forward in the best light possible.

Are hero numbers believable or should we be looking at certified vendor benchmark testing as a guide to capability?  Do users even look at benchmark or hero numbers in the first place?  Could the whole exercise be a waste of time?  The conversation moves to talk about STAC (Securities Technologies Analysis Center) and whether more application specific benchmarks have greater value.  This includes the approach Pure Storage appears to have taken with FlashArray and FlashBlade, only quoting application-specific testing like that of AIRI.

As the discussion moves on, the team talk about more consistent deterministic latency (more on that here) and whether public cloud service providers will ever subject their service offerings to benchmark testing (TL;DR, probably not).  As a final wrap up, the team asks, what can customers actually do to ensure they are seeing the real truth behind benchmark claims.

Elapsed Time: 00:20:48

Timeline

* 00:00:00 – Intro

* 00:01:00 – What are vendor hero numbers – ten meeelion IOPS!!

* 00:02:00 – But certified benchmarks are expensive to perform.

* 00:04:00 – Storage arrays aren’t interesting, do hero numbers help?

* 00:05:00 – Do users even care about industry standard benchmarks?

* 00:06:00 – STAC – financial application benchmarks

* 00:07:48 – Should vendors be more realistic about hero numbers?

* 00:09:50 – Why has Pure Storage stopped producing FlashArray performance data?

* 00:11:40 – However Pure do produce numbers for CI like AIRI (see Soundbytes #008).

* 00:14:30 – Do people do proofs of concept any more?

* 00:16:00 – Maybe arrays are massively over specified anyway?

* 00:16:30 – Or actually, it’s about consistent deterministic performance.

* 00:17:44 – Do we need performance benchmarks for cloud vendors?

* 00:19:00 – So what should customers do?

* 00:20:08 – Wrap Up

Related Podcasts & Blogs

* Pure Storage AIRI – CI for AI

* Why Deterministic Storage Performance is Important

* Avoiding the Storage Performance Bathtub Curve

* Avoiding All-Flash Missing IOPS

* Are DataCore’s SPC Benchmarks Unfair?

* NFS Vendor Benchmarks – Caveat Emptor

* Storage Performance: Why I/O Profiling Matters

Vendors referenced in this post: Violin Systems, Dell EMC (PowerMax), Pure Storage (FlashBlade & FlashArray), NetApp.

Copyright (c) 2016-2018 Storage Unpacked.  No reproduction or re-use without permission. Podcast Episode UGE3.

What is computational storage and how do we manage storage on the edge of the network?  This week’s conversation is a discussion on how storage and data is managed within IoT devices and partially covers the new segment of computational storage.  Martin and Chris talk to Scott Shadley, VP of Marketing for NGD Systems.

Edge storage means more than simply managing the data on a Raspberry Pi or other small form-factor device.  During the podcast, Scott uses the example of aircraft analytics, which generate terabytes of content per flight.  How do these implementations work?  How will data be secured and safely accessed in an architecture where none of the devices are in the confines of the data centre?

At this point it seems like there aren’t any real standards for distributing and managing code, or tracking the concurrency of content consolidation.  However with new bodies like the Open Fog Consortium forming, we can expect this sector to develop and mature.

Elapsed Time: 00:28:57

Timeline

* 00:00:00 – Intros

* 00:01:00 – What is Edge computing and what are the storage needs?

* 00:02:00 – Didn’t we already have distributed computing?

* 00:04:30 – Why process on the edge at all?

* 00:05:30 – A real example – aeroplane sensor/analysis data.

* 00:09:00 – What are the implementation models?

* 00:10:40 – What is Fog Computing?

* 00:14:00 – How would distributed data be secured?

* 00:15:30 – EFTPOS – the SEN room?

* 00:17:30 – What type of code could actually be run on the edge?

* 00:20:40 – What standards exist for code distribution/management?

* 00:21:50 – Data management – how do we protect content?

* 00:23:50 – How would a computational storage device work?

* 00:26:45 – Where is the market for computational storage headed?

* 00:27:56 – Wrap up!

Related Podcasts & Blogs

* #36 – The Persistence of Memory with Rob Peglar

* #37 – State of the Storage Union with Chris Mellor

* Garbage Collection #006 – LTO-8 and the Future of Tape

* The Practicality of In-Situ Processing

Scott’s Bio

Scott Shadley, Storage Technologist and VP of Marketing at NGD Systems, has more than 20 years of experience with Storage and Semiconductor technology. Working at STEC he was part of the team that enabled and created the world’s first Enterprise SSDs. He spent 17 years at Micron, most recently leading the SATA SSD product line with record-breaking revenue and growth for the company. He is active on social media, a lover of all things High Tech, enjoys educating and sharing and a self-proclaimed geek around mobile technologies.

Copyright (c) 2016-2018 Storage Unpacked.  No reproduction or re-use without permission. Podcast Episode GI05.

What is computational storage and how do we manage storage on the edge of the network?  This week’s conversation is a discussion on how storage and data is managed within IoT devices and partially covers the new segment of computational storage.  Martin and Chris talk to Scott Shadley, VP of Marketing for NGD Systems.

Edge storage means more than simply managing the data on a Raspberry Pi or other small form-factor device.  During the podcast, Scott uses the example of aircraft analytics, which generate terabytes of content per flight.  How do these implementations work?  How will data be secured and safely accessed in an architecture where none of the devices are in the confines of the data centre?

At this point it seems like there aren’t any real standards for distributing and managing code, or tracking the concurrency of content consolidation.  However with new bodies like the Open Fog Consortium forming, we can expect this sector to develop and mature.

Elapsed Time: 00:28:57

Timeline

* 00:00:00 – Intros

* 00:01:00 – What is Edge computing and what are the storage needs?

* 00:02:00 – Didn’t we already have distributed computing?

* 00:04:30 – Why process on the edge at all?

* 00:05:30 – A real example – aeroplane sensor/analysis data.

* 00:09:00 – What are the implementation models?

* 00:10:40 – What is Fog Computing?

* 00:14:00 – How would distributed data be secured?

* 00:15:30 – EFTPOS – the SEN room?

* 00:17:30 – What type of code could actually be run on the edge?

* 00:20:40 – What standards exist for code distribution/management?

* 00:21:50 – Data management – how do we protect content?

* 00:23:50 – How would a computational storage device work?

* 00:26:45 – Where is the market for computational storage headed?

* 00:27:56 – Wrap up!

Related Podcasts & Blogs

* #36 – The Persistence of Memory with Rob Peglar

* #37 – State of the Storage Union with Chris Mellor

* Garbage Collection #006 – LTO-8 and the Future of Tape

* The Practicality of In-Situ Processing

Scott’s Bio

Scott Shadley, Storage Technologist and VP of Marketing at NGD Systems, has more than 20 years of experience with Storage and Semiconductor technology. Working at STEC he was part of the team that enabled and created the world’s first Enterprise SSDs. He spent 17 years at Micron, most recently leading the SATA SSD product line with record-breaking revenue and growth for the company. He is active on social media, a lover of all things High Tech, enjoys educating and sharing and a self-proclaimed geek around mobile technologies.

Copyright (c) 2016-2018 Storage Unpacked.  No reproduction or re-use without permission. Podcast Episode GI05.

What is computational storage and how do we manage storage on the edge of the network?  This week’s conversation is a discussion on how storage and data is managed within IoT devices and partially covers the new segment of computational storage.  Martin and Chris talk to Scott Shadley, VP of Marketing for NGD Systems.

Edge storage means more than simply managing the data on a Raspberry Pi or other small form-factor device.  During the podcast, Scott uses the example of aircraft analytics, which generate terabytes of content per flight.  How do these implementations work?  How will data be secured and safely accessed in an architecture where none of the devices are in the confines of the data centre?

At this point it seems like there aren’t any real standards for distributing and managing code, or tracking the concurrency of content consolidation.  However with new bodies like the Open Fog Consortium forming, we can expect this sector to develop and mature.

Elapsed Time: 00:28:57

Timeline

* 00:00:00 – Intros

* 00:01:00 – What is Edge computing and what are the storage needs?

* 00:02:00 – Didn’t we already have distributed computing?

* 00:04:30 – Why process on the edge at all?

* 00:05:30 – A real example – aeroplane sensor/analysis data.

* 00:09:00 – What are the implementation models?

* 00:10:40 – What is Fog Computing?

* 00:14:00 – How would distributed data be secured?

* 00:15:30 – EFTPOS – the SEN room?

* 00:17:30 – What type of code could actually be run on the edge?

* 00:20:40 – What standards exist for code distribution/management?

* 00:21:50 – Data management – how do we protect content?

* 00:23:50 – How would a computational storage device work?

* 00:26:45 – Where is the market for computational storage headed?

* 00:27:56 – Wrap up!

Related Podcasts & Blogs

* #36 – The Persistence of Memory with Rob Peglar

* #37 – State of the Storage Union with Chris Mellor

* Garbage Collection #006 – LTO-8 and the Future of Tape

* The Practicality of In-Situ Processing

Scott’s Bio

Scott Shadley, Storage Technologist and VP of Marketing at NGD Systems, has more than 20 years of experience with Storage and Semiconductor technology. Working at STEC he was part of the team that enabled and created the world’s first Enterprise SSDs. He spent 17 years at Micron, most recently leading the SATA SSD product line with record-breaking revenue and growth for the company. He is active on social media, a lover of all things High Tech, enjoys educating and sharing and a self-proclaimed geek around mobile technologies.

Copyright (c) 2016-2018 Storage Unpacked.  No reproduction or re-use without permission. Podcast Episode GI05.

This week’s podcast is a conversation between Martin and Chris, talking about how we define scale-up and scale-out storage.  A recent discussion on Twitter about Pure1 and the idea of federated scale-out generated some interesting feedback, so we thought it might be good to get some definitions in place.

The opening discussion talks about how scale-up and scale-out should be defined and what definitions of scale-out exist.  Volume managers used to be the old-school way of implementing federation, as was storage virtualisation.  So perhaps federation is a genuine use case.

Martin and Chris move to talk about NAS and object, then reflect on how the cloud computing providers implement their storage platforms.  Surely these are scale-out?  The truth is, that level of detail is generally kept secret.  Finally, the conversation talks about one of the biggest storage headaches – rebalancing.

Elapsed Time: 00:19:38

Timeline

* 00:00:00 – Intros

* 00:01:00 – Scale-out versus scale-up – how do we define them?

* 00:02:30 – tightly-coupled, loosely-coupled, federated scale out.

* 00:03:30 – So what solutions are scale-up?

* 00:05:00 – Is federated storage a fair version of scale-out?

* 00:07:00 – Does scale-up have a value with very large drive types?

* 00:07:30 – Volume Managers – old school federation

* 00:09:30 – What about NAS and object store solutions?

* 00:12:00 – How does public cloud storage implement scale-out?

* 00:15:00 – The biggest storage management issue – rebalancing.

* 00:18:20 – So should we care about scale-out or scale-up?

* 00:19:15 – Wrap Up

Vendors mentioned in this podcast: Nimble, SolidFire, NetApp, Pure Storage, Tintri, Dell EMC and IBM.

Copyright (c) 2016-2018 Storage Unpacked.  No reproduction or re-use without permission. Podcast Episode J879.

This week’s podcast is a conversation between Martin and Chris, talking about how we define scale-up and scale-out storage.  A recent discussion on Twitter about Pure1 and the idea of federated scale-out generated some interesting feedback, so we thought it might be good to get some definitions in place. The opening discussion talks about how […]

This week we talk to Mark Carlson, co-chair of the SNIA Technical Council, about the storage needs of hyperscalers.  Mark defines hyperscalers as those companies opening multiple data centres a year, most notably Amazon, Microsoft Azure, Google and Facebook in the US and Baidu, Alibaba and Tencent in China.  These vendors are deploying petabytes of storage a year, with specific requirements on storage media.  The hyperscaler applications have issues with HDD and SSD performance characteristics, such as tail latency and the effects of garbage collection.  As a result, drive manufacturers are building in new features to meet the needs of these companies.

What exactly are the issues?  The conversation covers some of the typical problems, like reusing existing partially failed media, the problems caused by tail latency, write amplification and why non-deterministic latency is such an issue.  Some vendor solutions include Depop, Fast Fail and Open Channel.  Not sure what these are?  Listen in to find out!

In the podcast, Mark mentions a white paper on hyperscaler storage.  You can find it here (PDF).  The SNIA YouTube channel can be found here.  Mark mentions his presentation at Tech Field Day, which can be found here.

Elapsed Time:  00:27:54

Timeline

* 00:00:00 – Intro

* 00:01:30 – Who do we mean by hyper-scalers?

* 00:02:30 – ODM (Not OEM) – Original Design Manufacturers.

* 00:04:00 – Aren’t HPE and Dell selling to hyper-scalers?

* 00:05:30 – Why do hyper-scalers need different storage media?

* 00:07:00 – Differences of approach to resiliency compared to Enterprise.

* 00:09:00 – Don’t we want to have partial device failures?

* 00:10:20 – Depop – depopulate and reset drive to factory settings.

* 00:11:30 – Back to tail latency – extended response times.

* 00:12:30 – NVM Sets to logically partition a drive for each application.

* 00:13:40 – Fast Fail for HDD reads.

* 00:15:00 – putting drives into a deterministic window – putting maintenance off.

* 00:16:30 – Open Channel approach – let the host do the maintenance work.

* 00:17:30 –  How are changes introduced?  ECNs and Technical Proposals.

* 00:22:00 – Will hyper-scaler features filter down to the enterprise?

* 00:26:00 – Wrap Up

Mark’s Bio

Mark A. Carlson, Principal Engineer, Industry Standards at Toshiba, has more than 35 years of experience with Networking and Storage development and more than 20 years experience with Java technology. Mark was one of the authors of the CDMI Cloud Storage standard. He has spoken at numerous industry forums and events. He is the co-chair of the SNIA Cloud Storage and Object Drive technical working groups, and serves as co-chair of the SNIA Technical Council.

Copyright (c) 2016-2018 Storage Unpacked.  No reproduction or re-use without permission. Podcast Episode HIJ4.