Technology:Unpacked

In this episode, Chris is joined by Phison CTO Sebastien Jean to discuss the transition from HDD to SSD. As pricing for the two media types comes close to reaching parity, this conversation looks at how the two media are evolving and how NAND flash SSDs are set to become the de facto choice for the enterprise data centre. The transition to an all-flash data centre has been predicted for some time, however end game could be on the horizon as soon as 2025 or 2026.

The Phison blog discussed by Sebastien can be found here - https://phisonblog.com/

Elapsed Time: 00:44:01

Timeline

00:00:00 - Intros

00:02:45 - Predictions of the death of the hard drive are overdone

00:04:15 - Enterprise SSD prices continue to drop - but don’t HDDs do that too?

00:05:30 - 2020 was the cutover year for units produced (SSD vs HDD)

00:06:45 - HDD capacity growth has slowed

00:07:50 - SSD growth is being achieved with 3D techniques

00:09:15 - Floors could be used to increase layer count

00:11:00 - Latency is increasing in newer flash designs

00:13:05 - What is the “acceptable” limit for SSD capacity?

00:14:15 - Dense SSDs will be the norm with NVMe connectivity

00:15:30 - New form factors will drive greater drive capacities

00:17:00 - Is $/GB too simple a measure?

00:19:40 - SSD rebuilds are quick, avoiding the need for RAID-6

00:20:45 - SSD failure is not predictable

00:25:00 - Are HDD shipping costs relevant?

00:26:30 - How are the hyper-scalers using HDDs and driving change?

00:29:15 - Where do we go next - PLC?

00:32:00 - Could an entire SSD be dynamic between SLC and QLC?

00:35:45 - What about new technology Optane 4.0 or MRAM?

00:40:30 - Look out for CXL-enabled SSDs

00:42:40 - 2Tb NAND dies will bring HDD parity

Copyright (c) 2016-2022 Unpacked Network. No reproduction or re-use without permission. Podcast episode #34ee.

The PCI-SIG recently announced the next generation of PCI Express - 7.0 - will be released in 2025. We've only just heard about PCIe 6.0 (see this recent podcast), so it's interesting to see that the working group is already committed to extending the performance and capabilities of PCIe further into the future. Chris caught up with Al Yanes (President and Chairperson of PCI-SIG) earlier this year at Flash Memory Summit to get an update. This podcast was recorded after the event and goes into the detail about the improvements. The timeline for PCI Express means we're seeing new PCIe 5.0 devices on the market today, with PCIe 6.0 devices expected in 2025 and PCIe 7.0 supported hardware in 2028. With a certain future for PCI Express, vendors can develop products and have confidence that forward and backward compatibility will be maintained until the end of the decade.

For more information on PCI-SIG, go to https://pcisig.com/.

Elapsed Time: 00:22:53

Timeline

00:00:00 - Intros00:01:25 - PCIe 6.0 was discussed last March, now we’re discussing PCIe 7.000:02:00 - PCIe 7.0 in 2025, a three year cadence.00:03:15 - Compliance is a big benefit of the PCIe standard00:04:00 - PCIe devices are backwards and forwards compatible00:07:00 - Throughput is important but low latency is critical too00:08:30 - PCIe 7.0 will evolve as the standard is developed00:09:15 - PCIe continues to improve in power efficiency00:11:20 - Using next gen PCIe could be to gain cabling or power efficiencies00:13:25 - PCISig is working on standardisation of cabling00:14:05 - Technology is becoming more integrated00:17:30 - Timelines - PCIe 5.0 devices now, announced 201900:18:40 - PCIe 6.0 - announced 2022, devices expected 202500:19:00 - PCIe 7.0 - ratification in 2025, devices expected 202800:20:30 - ESG will put a focus on the ability to use a range of PCIe speeds00:21:30 - Wrap Up

Related Podcasts & Blogs

#229 - Exploring the PCIe 6.0 Specification with Al Yanes#217 - Introduction to CXL with Jim Pappas

Copyright (c) 2016-2022 Unpacked Network. No reproduction or re-use without permission. Podcast episode #pci7.

Once again Chris is joined by Rick Kutcipal, board member with the SCSI Trade Association. It's been four years since episode #74 when Rick provided an update on SAS 24Gb. This time, Rick has details of new features in 24Gb+. What does the "plus" mean in this context? While the bus speed may not be increasing, new features (+) are continually being added to the SAS protocol, hence the additional designation.

In this episode, Rick discusses some interesting enhancements to drive management, mainly driven by the demands of hyper-scalers. These include Command Duration Limits, Format with Presets and Logical Depopulation. There are also additional features still in discussion that may yet be added in future releases.

To learn more about the work of the SCSI Trade Association, hop over to www.scsita.org, or of course, simply listen to our next podcast after Flash Memory Summit 2023.

Elapsed Time: 00:00:24:19

Timeline

00:00:00 - Intros

00:01:45 - How has the SCSI/SAS standard evolved?

00:03:00 - There’s no need (currently) to go faster than 24Gb

00:03:50 - What features keep SAS relevant in the market?

00:05:35 - Hyper-scalers are driving the use of HDDs and SAS

00:06:50 - What’s new in 24Gb+?

00:09:45 - Enterprise customers will use new features (driven by hyper-scalers)

00:10:15 - What is Command Duration Limits (CDL)?

00:11:25 - What is Format with Presets?

00:13:25 - What is Logical Depopulation?

00:17:30 - Drive failure is a part of “normal” operations

00:17:50 - What features are under consideration?

00:20:15 - How will 24Gb+ features be rolled out?

00:22:30 - Wrap Up

Related Podcasts & Blogs

#74 - All About Serial Attached SCSI with Rick Kutcipal

#161 - Seagate MACH.2 Dual Actuator Drive Deep Dive

#174 - Introduction to Zoned Storage with Phil Bullinger

Copyright (c) 2016-2022 Storage Unpacked. No reproduction or re-use without permission. Podcast episode #fveh.

In this episode, Chris chats to friend of the show Chris Mellor from Blocks and Files. Both Chris's attended Flash Memory Summit this year, held annually in San Jose and in person for the first time since 2019. Two big technology developments stand out from this year's event. Firstly CXL, which is current flavour of the month with all vendors. The second is the diversity of SSD products, from computational storage to Ethernet drives. Chris walks through some of his observations, including MemVerge, NeuroBlade, Intel, Micron, Neo Semiconductor and Solidigm.

There's a lot to unpack here, so buckle in and enjoy the ride! Note this podcast is recorded live (in a meeting room we "borrowed" from IBM), so the noise comes from the acoustics of the room. Apologies for the brief background interruption at 00:14:50, and of course Mr Mellor's desk thumping, which fortunately isn't too bad this time.

Elapsed Time: 00:26:50

Timeline

00:00:00 - Intros

00:01:03 - Three things - CXL, SSD diversity, event buzz

00:02:00 - No Micron stand at FMS

00:02:37 - Everybody loves CXL

00:04:10 - MemVerge on a rising tide?

00:04:55 - Is memory tiering a thing?

00:07:17 - NeuroBlade presented a “AI machine”

00:09:00 - Systems now need closely coupled designs

00:10:02 - 2024 could be a breakthrough for shared memory

00:11:55 - What is the “chipset” for ARM processors?

00:13:37 - Can Intel continue to dominate the server market with x86?

00:14:20 - The SSD market is exploding

00:14:50 - Disturbed!

00:16:09 - Neo Semiconductor has IP to optimise SSDs

00:17:40 - SSD designs could create internally tiered devices

00:19:13 - Devices could internally restack their data

00:19:40 - NGD Systems - are they gone?

00:22:03 - Solidigm introduced “synergy” software

00:22:56 - Standards protocols are focused on hyper-scaler requirements

00:25:01 - Meta is giving away technology (IP)

00:25:54 - Final conclusions - SSD is diversifying

00:26:00 - Wrap Up

Related Podcasts & Blogs

#236 - A FlashCore Module Primer with IBM Fellow & Storage CTO Andy Walls

#235 - FlashCore Futures with IBM CMO Scott Baker

#206 - An Update on NVMe 2.0 with Amber Huffman

#117 - Introduction to Computational Storage with NGD Systems

Copyright (c) 2016-2023 Unpacked Network. No reproduction or re-use without permission. Podcast episode #ky4r.

In this episode, recorded life at FMS 2022, Chris chats to IBM Fellow, Storage CTO and FlashSystem Architect, Andy Walls.

In this episode, Chris is joined by IBM Storage CMO Scott Baker. The discussion takes place at Flash Memory Summit 2022 (FMS), being held in person in Santa Clara, USA. FlashCore Modules are IBM's custom flash drives, developed from IP acquired with the acquisition of Texas Memory Systems. Scott is presenting at FMS, discussing how FlashCore could be used for anomaly detection and enhanced ransomware protection.

You can learn more about FlashCore modules in our deep dive into FlashSystem 5200 and in this podcast recorded in September 2020.

Elapsed Time: 00:13:36

Timeline

00:00:00 - Intros

00:02:00 - What is the vision for future FlashCore modules?

00:04:15 - FlashCore are 38.4TB in capacity

00:05:45 - FlashCore could provide hits and info into other anti-ransomware tools

00:07:30 - Malware software and processes are becoming more complex

00:09:45 - Is storage “done”?

00:11:00 - Joined up systems can focus on identifying threat activities

00:13:00 - Wrap Up

Related Podcasts & Blogs

#175 - IBM FlashSystem Deep Dive (Sponsored)

IBM FlashSystem 5200 Deep Dive

Copyright (c) 2016-2022 Storage Unpacked. No reproduction or re-use without permission. Podcast episode #d4rf.

In this episode, Chris talks to Rob Lee, CTO at Pure Storage about today's announcement covering FlashBlade//S and Evergreen//Flex. FlashBlade//S is a new hardware platform that evolves the solution into a disaggregated architecture. The design enables compute and storage resources to be scaled independently, while incorporating existing DFMs (DirectFlash Modules) that are currently used in the FlashArray family. To exploit the benefits of FlashBlade//S, Pure has also announced a realignment of the Evergreen programme, including the new Evergreen//Flex option. //Flex customers can retain hardware ownership while benefiting from "component" mobility.

The hardware changes in FlashBlade//S compared to the first hardware iteration of FlashBlade are considerable. We discuss them in this blog post. The benefits to the customer are greater flexibility in picking the right hardware configuration, with ongoing investment protection through Evergreen//Flex. From an environmental standpoint, separation of compute and storage helps reduce e-waste, while from the customer perspective, reduces the volume of data rebuilds when hardware is replaced. Of course, there's much more, but you have to listen to find out.

The updates to Evergreen fill the gap between hardware ownership and ongoing upgrades and the full "as-a-service" offering. Evergreen Gold is now rebranded Evergreen//Forever, while Pure as-a-Service becomes Evergreen//One.

We've been able to follow the progress of FlashBlade since inception (including an early podcast with Rob). Check out the details in our related posts below, or get more information from the Pure Storage website.

Elapsed Time: 00:39:17

Timeline

00:00:00 - Intros00:00:30 - Rob Lee last recorded to discuss FlashBlade 2.000:01:30 - Pure has two announcements - FlashBlade & Evergreen00:02:50 - FlashBlade is the scale-out object & file platform00:04:15 - FlashBlade//S transitions the platform into a family of solutions00:05:30 - What does //S signify? Speed, Simplicity, Scale & Sustainability00:07:00 - What’s new in FlashBlade//S - disaggregation00:09:00 - Customers love the bladed architecture and integrated networking00:10:30 - FlashBlade software always intended for the current product evolution00:12:15 - Each blade supports up to 4 removable DirectFlash Modules00:14:00 - FlashBlade//S and FlashArray provides component mobility00:16:10 - Pure1 provides the data to make informed rebalancing decisions00:18:00 - Sustainability is becoming an increasingly important factor in product choice00:20:05 - Standardisation will bring power efficiencies to FlashBlade//S00:23:00 - The new FlashBlade design allows optimisation for multiple use-cases00:26:30 - How is Evergreen changing to address the new consumption requirements?00:27:00 - Evergreen Gold becomes Evergreen//Forever00:27:30 - Pure as-a-Service becomes Evergreen//One00:27:40 - Evergreen//Flex is new and fits the gap between both options00:33:30 - //Flex gives customers the benefits of efficiency and optimisation00:35:15 - Do customers follow a journey of service consumption?00:37:40 - When will the new products & solutions be available?00:38:45 - Wrap Up

Related Podcasts & Blogs

Pure Storage Announces FlashBlade//SPure Storage Introduces a Revamped Evergreen ProgrammePure Storage accelerates growth ahead of Accelerate techfest22Pure //Accelerate 2016 - FlashBladeSoundbytes #008: FlashBlade 2.0 with Rob Lee at Pure Accelerate#232 - Building Power-Efficient Storage Systems with Justin Emerson from Pure Storage#209 - Discovering Unified Fast File and Object with Pure Storage

Copyright (c) 2016-2022 Storage Unpacked. No reproduction or re-use without permission. Podcast episode #r445.

In this episode, Chris is in conversation with VAST Data CMO, Jeff Denworth. The topic covers the recent announcement of Ceres, a new hardware platform combining ruler form-factor flash and BlueField DPUs. The new solution, which will eventually replace the current "Mavericks" D-boxes is a 1U system with flash, storage-class memory and network connectivity through DPU SmartNICs. The design can be vastly simplified compared to other storage solutions as the hardware component simply presents storage as drives over NVMe-oF. A single 1U system supports up to 22 ruler drives, with a capacity of 675TB.

Jeff also discusses the certification of VAST Universal Storage with NVIDIA SuperPods. The certification process confirms the suitability of storage systems to operate in the highest scaling SuperPod DGX systems, crucially offering customers a path to support. The certification is expected to be completed in Q3.

As Jeff highlights towards the end of the discussion, Universal Storage addresses many modern application requirements. Customers are only just starting their journey to learn what the VAST platform has to offer.

To find out more about Ceres, head over to https://vastdata.com/ - we've also added links to previous podcasts and content at the end of these show notes.

Elapsed Time: 00:31:04

Timeline

00:00:00 - Intros00:01:35 - What is Ceres? (In VAST terms)00:03:10 - Ceres uses “ruler” flash and DPUs00:03:45 - Ceres will replace “Mavericks”00:04:05 - What are the benefits of the “ruler” flash format?00:05:20 - 48/60TB ruler capacities could come soon00:07:30 - Base unit is 22 drives, 675TB capacity - in 1U00:08:55 - A PCIe switch connects drives and DPUs00:09:30 - The BlueField DPUs expose NVMe devices to the network00:10:45 - Ceres enables the DASE architecture00:12:20 - Scalability is achieved through additional Ceres shelves00:13:35 - Disaggregation is possible now, with faster networks00:14:15 - Other deployment modes will come along - including cloud00:14:40 - What is SuperPod?00:16:05 - How is VAST supported on SuperPod?00:16:55 - Universal Storage is the first NAS system for SuperPod00:18:25 - SuperPod certification is also about gaining support00:20:30 - Legacy design required cores to scale lanes00:22:40 - Is storage starting to centralise?00:24:55 - It’s good to be protocol independent00:27:20 - Customers are looking for simplicity of operation00:29:20 - Single systems are eclipsing 100PB - petabytes00:30:15 - Wrap Up

Related Podcasts & Blogs

#226 - VAST Data Universal Storage for Data Protection#216 - The Data Explosion with VAST Data CEO Renen Hallak#105 - Introduction to VAST Data (Part I) with Howard Marks#106 - Introduction to VAST Data (Part II) with Howard Marks#190 - NVIDIA BlueField SmartNICs and DPUsVAST Data has Vast AmbitionsVAST Data goes software-only - what does this mean for customers?VAST Data launches with new scale-out storage platform

Copyright (c) 2016-2022 Unpacked Network. No reproduction or re-use without permission. Podcast episode #w3e4.

In this episode, Chris talks to Justin Emerson from Pure Storage about modern design considerations for power-efficient storage. The cost of powering data centres has increased massively over the last 12 months, driven by the global economy and recovery from COVID-19. At a component level, servers are increasing their power usage, with faster processors and memory, plus the use of power-hungry GPUs. By contrast, storage power demands have reduced, as flash takes over from HDDs. Of course, this doesn't mean there aren't still considerations to make. Vendors such as Pure Storage have developed custom flash modules and introduced other design features to optimise on performance per watt, while deliver increasing densities. Justin explains how Pure has achieved this and what we should look out for in the future.

We discuss the Pure ESG report, which can be found here - https://www.purestorage.com/company/corporate-social-responsibility.html. Justin also mentions the Accelerate TechFest event on 8th June 2022, more details are available here - https://www.purestorage.com/accelerate.html.

Elapsed Time: 00:50:03

Timeline

00:00:00 - Intros

00:01:30 - Power costs are driving the need for efficiency

00:02:30 - Power is the limiting factor on data centre design

00:03:40 - Flash generations implicitly drive power efficiencies

00:04:30 - Higher clock speeds drive higher power usage

00:06:15 - What are the hyper-scalers doing with power efficiency?

00:07:10 - Cooling is also a big efficiency factor

00:08:05 - HDDs are still hugely relevant for hyper-scalers

00:10:10 - Can businesses hedge power costs for their data centres?

00:12:35 - If you’re in a co-lo, power costs could be unpredictable

00:14:30 - In-data centre power management is creating challenges

00:17:05 - Pure has released an ESG report including power efficiency

00:18:35 - Product longevity is important to sustainability

00:20:30 - Evergreen and product design allows Pure to reduce waste

00:23:15 - What is DirectFlash?

00:26:00 - Commodity drives have DRAM and make independent decisions

00:28:05 - Couldn’t I just use the Ruler form factor?

00:30:00 - FlashArray//C uses QLC - is this more power-efficient?

00:33:42 - What software enhancements are used to increase efficiency?

00:36:00 - Data reduction features need to balance write amplification with efficiency

00:38:10 - How do end-users make decisions about picking the right products?

00:40:15 - There are no recognised metrics for measuring storage power/performance efficiency

00:44:30 - What future improvements should we expect?

00:46:50 - Code and protocol efficiency will improve power usage

00:48:15 - Wrap Up

Related Podcasts & Blogs

#228 - Exploiting Data in the Third Wave of IT Agility with Murli Thirumale

#209 - Discovering Unified Fast File and Object with Pure Storage

#211 - Pure//Launch - Announcing Fusion and Portworx Data Services

#199 - Quantifying Data Storage Innovation

Pure Storage Introduces FlashArray//XL

Pure//Launch - Pure Storage Announces Fusion and...

In this episode, Chris talks to Eric Herzog, CMO at Infinidat about the announcement of SSA II, the second generation all-flash InfiniBox platform. The first generation SSA (solid state array) was developed to address a customer requirement that all long-tail latency I/Os must be delivered at all-flash levels of performance. The InfiniBox architecture is capable of delivering up to 95% of I/O from DRAM, but for some customers, that figure simply wasn't high enough.

With SSA II, Infinidat has moved to the Intel Cascade Lake architecture, adding a further 48 cores of processing. In addition, the InfiniBox software has been enhanced to include InfiniSafe and InfiniOps, a combination of AIOps and integrations with common application platforms. The enhancements enable SSA II to deliver I/O as low as 35µs, with real world 70/30 read/write mix of 50µs (read) and 70µs (write). A demonstration on customer performance I/O can be seen in this pre-recorded presentation (registration required).

The new InfiniSafe technology mirrors that introduced into the InfiniGuard platform a few months ago. We discussed this in depth with Eric in episode #227. Further information on SSA II can be found in the Infinidat press release (here).

Elapsed Time: 00:32:19

Timeline

00:00:00 - Intros

00:01:15 - SSA II has been announced - but what is SSA?

00:03:00 - InfiniBox uses neural cache to deliver low latency

00:06:00 - What's new in SSA II? Cores and Processors to start

00:07:15 - What are the metrics? 35 microseconds - cache hits

00:09:10 - High performance is based on new architectures

00:11:00 - Why do vendors hide away from real-world numbers?

00:13:20 - What software changes are in SSA II?

00:14:00 - InfiniSafe has been added to SSA II

00:19:00 - Cyber-resilience needs to be deployed on primary storage

00:19:30 - SSA II introduces InfiniOps to the InfiniBox family

00:22:00 - Containers are supported on SSA II with CSI, Red Hat OpenShift & VMware Tanzu

00:28:15 - SSA II is available from today and existing SSA gets a software upgrade

00:31:10 - Wrap Up

Related Podcasts & Blogs

#227 - Infinidat InfiniGuard Enhancements with Eric Herzog

#202 - Enterprise Storage Consolidation with Phil Bullinger from Infinidat

#166 - Infinidat Elastic Storage Pricing with Eran Brown

#222 - End of the Year Show 2021 - Part Two

#173 - Transparent Enterprise Storage Pricing

The Quiet Success of Infinidat

Predicting the Future of Technology Influencers

Copyright (c) 2016-2022 Unpacked Network. No reproduction or re-use without permission. Podcast episode #er8r.