Home Blog Page 102

Western Digital spinning out Flash business

By
Chris Mellor
-

Western Digital has concluded its strategic review and intends to separate its existing hard disk drive and flash/SSD businesses into two separate publicly listed companies by the second haf of calendar 2024.

David Goeckeler.

The strategic review was announced in June last year following input from Western Digital shareholder and activist investor Elliott Management, which calculated the flash and SSD part of the business was under-valued and overshadowed in investor sentiment terms by the HDD division. The message was loud and clear: a standalone flash and SSD business would have a higher value, and WD shareholders would benefit from this.

WD CEO David Goeckeler confirmed the decision: “Separating these franchises will unlock significant value for Western Digital shareholders, allowing them to participate in the upside of two industry leaders with distinct growth and investment profiles.”

He added that both the HDD and flash/SSD markets are emerging from a long downcycle: “Moving forward, as we progress through fiscal year 2024, we see an improving market environment in both businesses, and we will remain open to strategic opportunities that unlock further value in both our HDD and Flash investments and assets.” 

The HDD business will be called Western Digital, and the to-be-spun-out flash and SSD business has yet to be named. In WD’s first quarter for fiscal 2024, ended September 29, revenue generated by the flash/SSD division was $1.56 billion versus the HDD business unit’s $1.19 billion.

The jewel in the crown for the Flash division is the joint venture with Kioxia to fabricate NAND chips in Japanese foundries, with each company building their own SSDs using the chips. WD estimates Kioxia had an 18 percent share of the calendar 2022 NAND market, with WD having 13 percent, meaning a combined 31 percent share behind industry leader Samsung with 34 percent and ahead of SK hynix (19 percent) and Micron (12 percent).

WD got into the flash business by buying SanDisk for $19 billion in 2015. SanDisk had a NAND fab joint-venture with Toshiba, which WD obtained through this transaction. Toshiba spun off its NAND business into Kioxia in an $18 billion transaction with a Bain Capital-led consortium in September 2017.

An attempt to merge WD’s flash business with Kioxia failed a few days ago when Kioxia shareholder SK hynix objected. It would seem likely that WD execs were waiting to see what happened with Kioxia before revealing the separation plan for the Flash unit.

The final determination to separate will be subject to WD board approval, the execution of definitive documentation, and receipt of opinions or rulings as to the tax-free nature of the transaction for shareholders that are given shares in the spun-off flash business. It also needs satisfaction of customary conditions, including effectiveness of appropriate filings with the U.S. Securities and Exchange Commission, the completion of audited financials, and the availability of financing. 

Western Digital said its Board remains open to considering any alternatives that deliver superior value to the proposed separation should they become available. For the time being Ellliott Management can look forward to completing another activist investor engagement.

Hitachi Vantara gives birth to AI-led Digital Services unit

By
Chris Mellor
-
Hitachi Construction Excavator

Hitachi has spun out a Digital Services unit from Hitachi Vantara to provide generative AI-led services across the business.

Vantara is the enterprise storage data infrastructure business inside Hitachi, covering all markets outside Japan. Its product portfolio includes the VSP (Virtual Storage Platform) high-end and mid-range block arrays, HNAS file storage, VSS (software-defined storage), and HCP (Hitachi Content Platform) for object data. These standalone products are being combined into a single hybrid storage platform called Virtual Storage Platform One, managed through the Ops Center product.

As a Japanese conglomerate, Hitachi has wide interests, including Hitachi Energy, Hitachi Construction, and Hitachi Rail.

Keiji Kojima, president and CEO of Hitachi Ltd, said in a statement: “We see the emergence of generative AI as a breakthrough with enormous impact.” Hitachi’s company changes have “the aim of effectively utilizing generative AI to unleash the potential of the company. Through this reorganization, Hitachi will actively invest in the data infrastructure necessary for AI training and in OT and IT Integration, [in] which Hitachi has domain knowledge, to achieve company-wide digital transformation through the use of generative AI.”

OT refers to operational technology distinct from IT (information technology). Overall, the company is “reorganizing its global organizations to accelerate the creation of synergies with OT fields such as energy, transportation, and industry, while strengthening digital capabilities and value chains … Hitachi will promote company-wide digital transformation through the utilization of rapidly evolving generative AI and contribute to the realization of a sustainable society.”

A diagram shows the big idea:

Hitachi restructure diagram
GlobalLogic provides digital engineering services and the Silicon Valley company was acquired by Hitachi in 2021. DSS stands for Digital Systems and Services. GEM stands for Green Energy and Mobility. The CI sector refers to Connective Industries. OT and IT refer to Operational Technology and Information Technology.

Hitachi Digital Services (HDS) starts life on November 1 and will be spun out from Hitachi Vantara LLC’s digital solutions business. It will be led by CEO Roger Lvin, currently president of Hitachi Vantara’s Digital Solutions Business Unit.

HDS will offer services in cloud, data, IoT technologies, and OT and IT integration services. It will support enterprise transformation through automation of business processes and generative AI, and work with the separate Hitachi Digital, led by CEO Jun Taniguchi and president Gajen Kandiah.

Hitachi Vantara LLC will have a new CEO, Sheila Rohra,  who moves from being president of Hitachi Vantara’s Data Infrastructure business unit. The company, headquartered in Santa Clara, CA, will be responsible for development, construction, sales, operations, monitoring, and maintenance of data infrastructure products centered on storage for overseas markets, data management software, hybrid cloud infrastructure, and related services.

As of April 1, 2024, Hitachi’s IT Platform Products Management Division, which has been responsible for business development, R&D, and production of data infrastructure in Japan, will be succeeded by the newly established Hitachi Vantara Ltd – not LLC. Hitachi Vantara Ltd will be headquartered in Yokohama City, Japan, and perform the same roles as Hitachi Vantara LLC but for the domestic market.

So there are two Hitachi Vantara business units – one for Japan, and one for overseas. Sheila Rohra runs the LLC Hitachi Vantara and Akinobu Shimada is the president of Hitachi Vantara Ltd.

This mind-numbing corporate stuff is further complicated as Hitachi Vantara Ltd and Hitachi Vantara LLC will strengthen their mutual management as the realigned Hitachi Vantara. They will integrate global storage sales, and combine engineering, development, and manufacturing through an integrated Japan-US operation.

The two companies will provide hybrid cloud storage and a common platform for generative AI under a unified brand. The company will build a partner ecosystem and promote the development of a common platform for generative AI through strategic alliances, and strengthen its data infrastructure services including data management.

A Gajen Kandiah blog provides some background.

Pure Storage exec shuffle combines FlashBlade and FlashArray product managment

By
Chris Mellor
-

Pure Storage is ringing in some exec-level changes after combining the FlashBlade and FlashArray product management teams.

FlashBlade is an all-flash array for storing unstructured data. It includes a FlashBlade//S variant, with disaggregated and modular compute and storage design features, and a FlashBlade//E product optimized for capacity. FlashArray is a line of all-flash unified block+file storage arrays with three variants: high-performance //X models; tier 2 workload FlashArray//C products; and capacity-optimized //E models.

According to Pure, Amy Fowler is no longer the VP and general manager (GM) for FlashBlade, a position she was appointed to in November 2022. Instead she is now GM for the commercial line of business.

Shawn Hansen is now VP and GM of the Core Platform business unit and gets the FlashBlade product added to his FlashArray line of responsibility. He also leads the Fusion and Cloud Block Store cloud services product teams.

From left to right: Amy Fowler, Matt Burr, Ajay Singh, and Shawn Hansen.

When Fowler was appointed FlashBlade GM, she succeeded Matt Burr, who subsequently became interim Chief Marketing Officer. He has now become the Chief Operating Officer. This follows the retirement earlier this month of Mike Fitzgerald, VP Operations. Consequently, Pure no longer has a CMO.

Fowler and two other GMs are now running the lines of business, including Bill Cerreta, who is leading hyperscale, and Maciej Kranz, in charge of enterprise, appointed in September 2023.

Cerreta joined Pure Storage in 2013 and became VP and GM of the Platform business unit, responsible for the strategy and delivery of the hardware developed for FlashArray, FlashBlade, and ObjectEngine platforms. Fowler was hired as the FlashBlade GM in November 2022 after its product management was separated out. Now it has returned to the core platform group.

A spokesperson at Pure Storage denied the reorg was due to slowing sales of FlashBlade: “FlashBlade continues to be one of the most innovative products in the market and has achieved $2 billion in sales since inception. With the recent launch of FlashBlade//E, sales and pipeline have exceeded our expectations and it is experiencing the fastest growth of all prior new product releases.

“Today, FlashBlade is fully GPU-ready and battle tested at the highest levels of performance. Pure is one of, if not the first vendors to bring a product to market in the storage + AI space and we’ve been partnering with NVIDIA since 2018. Hundreds of our customers (including Meta) deploy our joint solutions for intensive and demanding workloads such as high-frequency trading, data analytics, and AI Training and Inference.

“What sets us apart in performance is FlashBlade’s metadata engine, engineered to handle AI workloads that ingest and manage billions of files, where many other competitive solutions struggle and even fall-over. Our metadata engine leverages a common pool of DirectFlash, rather than separate tiers of ‘exotic storage’ (i.e. Storage Class Memory) providing significant performance and scale in metadata processing versus the rest of the market. DirectFlash also provides FlashBlade with fast write performance in addition to reads, consistently across the entire dataset.”

Pure’s spokesperson added: “Pure’s comprehensive portfolio continues to be led by Ajay Singh, Chief Product Officer, and is an innovation engine. Shawn Hansen, leader of the Core Platform Business Unit, continues to report to Ajay Singh.”

As for exec-level moves in general: “As Pure continues to scale and innovate, we regularly evaluate our business to ensure that it aligns with our strategic priorities.”

IBM reports robust Q3 revenue growth, with GenAI a catalyst for expansion

By
Chris Mellor
-

IBM reported 4.6 percent revenue growth for Q3 2023, ended September 30, and banked $1.7 billion in profit – a radical turnaround from the $3.2 billion loss a year ago.

In prepared remarks, chairman and CEO Arvind Krishna said: “In the third quarter, we had solid growth across revenue, profit, and free cash flow while delivering innovations and positioning our business to capture future opportunities.”

From his vantage point: “We see both tailwinds and headwinds to overall spending. Nearly every business we talk to wants to leverage technology to offer better services, scale more quickly, and fuel growth without increasing their footprint. This has been driving demand for technologies that boost productivity and competitiveness, like hybrid cloud and artificial intelligence. Overall, we believe the tailwinds outweigh the headwinds, and technology spend will continue to outpace GDP.”

He spoke a lot about generative AI and IBM’s watsonx products. GenAI will be multi-modal and the “software stack starts with Red Hat OpenShift, which serves as the foundation that allows clients to operate in a hybrid environment,” Krishna said.

Yesterday Dell announced a validated design for Red Hat OpenShift AI on APEX Cloud Platform – meaning Dell PowerEdge server hardware and PowerFlex storage software.  

IBM generative AI tech stack

Krishna talked about a flood of watsonx AI products and data+GenAI consultants, more than 20,000 of them.

He told the earnings call that AI would enhance productivity without causing developer layoffs. But the same couldn’t be said for other areas. “When you go to customer service, there is a cost takeout aspect, but it’s not 100 percent. But there is probably a 20 percent to 30 percent. If you can have more calls, more chats answered by AI, that means you can have much more volume with a smaller number of people. So there’s both.”

He added: “AI does not get tired. It doesn’t get angry. It doesn’t get upset. So there is an NPS improvement to get along with it.”

Financial summary

  • Gross Profit Margin: 54.4 percent vs 52.7 percent a year ago
  • Operating cash flow: $3.1 billion, up $1.2 billion year-over-year
  • Free cash flow: $1.7 billion, up $0.9 billion year-over-year
  • Cash & marketable securities: $11 billion, up from $2.2 billion at end of 2022
  • Debt: $55.2 billion, up $4.3 billion since end of 2022.

Storage revenues are spread across the Software reporting line and the Infrastructure (hardware) reporting line. In the Infrastructure segment, with $3.27 billion in revenues, IBM noted that there are two sub-segments:

  • Infrastructure support: $1.3 billion, down 7 percent year-over-year
  • Hybrid Infrastructure: $1.9 billion, flat year-over-year

Within hybrid infrastructure, zSystems (mainframe) was up 9 percent year-over-year while distributed infrastructure was down 6 percent, reflecting Power10 server growth offset by storage declines.

Teasing out storage revenue from the Software segment is near-impossible. IBM software revenues of $6.3 billion were split between transaction processing ($1.8 billion; up 5 percent year-over-year), Hybrid Platforms and Solutions ($4.5 billion, up 7 percent), and four sub-segments with no identified revenue amounts:

  • Red Hat – up 8 percent year-over-year
  • Automation – up 13 percent
  • Data & AI – up 6 percent
  • Security – down 3 percent

In the Data & AI area, CFO and SVP Jim Kavanaugh said: “Growth areas include Data Fabric and Customer Care as enterprise clients are both preparing for and adopting generative AI solutions, leveraging watsonx.”

In a Wells Fargo-moderated Gartner report, IBM’s all-flash array revenues in the second 2023 quarter were calculated to be around $375 million, ahead of Pure, HPE, and Hitachi, and trailing Dell, Huawei, and NetApp. Add storage software to that and IBM could make a song and dance about its storage revenues if it wanted to, but prefers to say virtually nothing.

The outlook for the final quarter of IBM’s fiscal 2023 is revenue growth of 3-5 percent at constant currency.

VMware and Nutanix lead pack in Forrester HCI report

By
Chris Mellor
-
Under the Wave off Kanagawa
Public domain image: https://www.metmuseum.org/art/collection/search/45434

VMware and Nutanix are the only two leaders in Forrester’s Wave report on hyperconverged infrastructure (HCI) – Nutanix being slightly in front and everyone else – bar Microsoft – losing status or exiting the HCI landscape.

IT research consultancy Forrester provides so-called Wave reports on IT market segments with vendors categorized as Challengers, Contenders, Strong Performers, and Leaders. They are placed in a two-dimensional quarter circle space with a vertical weak-to-strong current offering axis, and a horizontal weak-to-strong strategy axis. Challengers, Contenders, Strong Performers, and Leaders have separate and concentric areas going from left (Challengers) to right (Leaders). The size of a vendor’s circle represents market presence.

Forrester Wave showing VMware and Nutanix in leadership positions

After the two leaders comes three strong performers: Microsoft, HPE (SimpliVity), and new entrant Sangfor Technologies. There are four Contenders: Scale Computing, IEIT Systems, Huawei, and new entrant SmartX. Cisco, which has just canned its own HyperFlex product in favor of a Nutanix-powered system, did not participate, and neither did IBM with its Storage Fusion product. There are no Challenger suppliers.

When we compare this chart with the one Forrester prepared three years ago, supplier consolidation becomes quite visible:

Forrester 2020 HCI Wave report

Cisco has given up on its DIY approach. NetApp has exited the market. HPE has fallen back, as has Huawei. DataCore has exited the market. Scale Computing has fallen back. Microsoft (Azure Stack HCI) has hung in there. Pivot 3 crashed and burned, selling its assets to Quantum. Red Hat, acquired by IBM, did not participate in Forrester’s report.

Let’s see what Forrester consultant and report author Brent Ellis says about individual suppliers:

  • HPE SimpliVity has new life as the HCI underpinnings of HPE GreenLake for Private Cloud Business Edition.
  • Sangfor stands out in identity management, but otherwise delivers expected functionality.
  • IEIT Systems’ dSAN and InCloud Rail constitute a good HCI platform, but needs further focus.
  • IBM’s bare metal container HCI platform is disruptive, but OCP coupling limits adoption.
  • Scale Computing scales out at the edge, but leaves scale-up workloads behind.
  • Huawei has mastered scaling up and down, though it provides a mediocre HCI product.
  • SmartX is gaining traction in APAC but must fill feature gaps as it grows.

Overall, Ellis says HCI suppliers should have good snapshot and replication facilities, disaggregated compute and storage resource scaling, and comprehensive virtualized network function support such that networks can stretch across cloud, datacenter, and edge environments.

Get a copy of the report from Nutanix, registration required.

Seagate revenue decline continues amid slow HDD sales

By
Chris Mellor
-
HDD storage
HDD storage

Revenue declines continue at Seagate – first quarter fiscal 2024 earnings were 29 percent lower than a year ago, when they were 35 percent down on the year before that.

Seagate’s customers haven’t been buying enough high-capacity disk drives for the company to make a profit for four quarters in a row. Turnover in the quarter ended September 29 was $1.54 billion, compared to $2.04 billion a year ago. It reported a loss of $184 million in contrast to $29 million profit a year ago.

We’re used to Seagate making profits in the hundreds of millions, but that was a year and a half ago.

CEO Dave Mosley was not upbeat in his statement: “Seagate’s first quarter results were in line with our revised expectations. During a longer-than-typical downcycle we have prioritized financial discipline, positive free cash flow generation and execution on our HAMR product roadmap, which will extend our industry leadership in areal density and improve the total cost of ownership for our customers.”

Seagate revenue and profit

He pointed to weak demand in China, worse-than-anticipated legacy drive demand, and continued inventory digestion in the hyperscalers. Mosley did say there were “indications that demand fundamentals in certain markets are starting to improve, including a modest uptick in US cloud.” Overall Seagate saw depressed demand across its three end markets: mission-critical (10K rpm, 2.5-inch), client, and consumer. Enterprises are generally cautious in their IT spending at the moment. We note that customer organizations may be prioritizing flash over disk storage in their generative AI-led IT spending.

Mosley’s earnings call statement said: “Spending priorities for CSPs have temporarily shifted toward AI-related infrastructure, which have further slowed the pace of demand recovery for mass capacity storage.” 

Seagate revenue by quarter by fiscal year
Mosley said Seagate is in a longer than usual downcyle and this chart confirms that, with more to come

Financial summary

  • Gross margin: 10.2 percent vs 23.7 percent a year ago
  • Operating cash flow: $127 million
  • Free cash flow: $57 million
  • Cash and cash equivalents: $795 million vs year-ago $761 million
  • Diluted EPS: -$0.88 vs $0.14 a year ago

Seagate said mass capacity HDD demand rose slightly quarter-on-quarter, led by the video surveillance market, but this was not enough to offset general market reluctance.

HDD revenues in the quarter were $1.3 billion, 84 percent of its overall revenues, with the nearline drives bringing in 79 percent of that, $1.02 billion, down 36 percent annually. Legacy drives (2.5-inch, PC, consumer) brought in $278 million, down 29 percent year-over-year.

Seagate HDD revenue
Two years of HDD revenues spinning down, and it’s set to continue next quarter

It shipped 89.6 EB of disk capacity, down 24 percent year-over-year, of which mass capacity represented 79.2 EB, down 24 percent year-over-year, and within that, nearline accounted for 56 EB, down 34 percent annually. The average capacity per drive was 75 TB, down 1 percent on a year ago, with nearline average capacity being 10.3 TB, down 13 percent annually.

Systems, meaning JBODs like Corvault and SSDs, generated just $159 million, down 40 percent year-over-year and 27 percent quarter-over-quarter.

Generally speaking, it’s a blood bath in the HDD market. Seagate is managing its costs and debt, $5.57 billion vs $6.25 billion a year ago, and waiting for hyperscalers, OEMs, and system builders to start buying in larger quantities again.

Looking ahead on the technology front, Seagate said its recently announced 24 TB PMR drive is set to enter a volume shipment phase in the first half of calendar 2024 (Q3 and Q4 FY 2024). A volume ramp would also begin in that period for its  3+ TB/platter class HAMR drives, meaning 32 TB drives initially. These have up to ten platters and will have enhanced capacity shingled magnetic recording (SMR) variants.

It said 4+ TB/platter HAMR tech would be introduced within the 2024-2026 calendar years, implying 40 TB+ drives with higher capacity SMR variants again. All this should enable HDDs to keep their cost/TB advantage over QLC (4 bits/cell) SSDs. Mosley said: “Simply put, we offer customers mass data storage at less than one-fifth the cost of comparable NAND solutions on a per-bit basis. We don’t foresee that value gap closing over the next decade relative to data center architectures.”

The company’s outlook for the second quarter of fiscal 2024 is revenues of $1.55 billion +/- $150 million, 18 percent lower than a year ago at the mid-point. In the next two quarters after that, Seagate will ramp up HAMR tech disk drive production and could have a significant capacity-per-disk lead over Toshiba and Western Digital to start capturing market share.

We expect Western Digital to report similarly depressed disk drive revenues on October 30.

OpenShift users now have a Dell APEX service just for them

By
Chris Mellor
-

Dell has launched an APEX Red Hat OpenShift service allowing users to subscribe to open source container wrangling software in an enterprise-style setting.

Red Hat OpenShift is the IBM-owned open-source company’s container orchestration service. It is based on Kubernetes and works with Ceph open-source storage. APEX is a set of services from Dell where it supplies its compute, storage and networking gear through a public cloud-like subscription model.

A Dell blog we saw pre-publication states: ”Dell APEX Cloud Platform is the first fully integrated application delivery platform purpose-built for Red Hat OpenShift.”

Dell quotes a Gartner prediction that more than 95 percent of global organizations will be running containerized applications in production by 2028. They’ll need a container app infrastructure platform on which to do it and Dell reckons OpenShift is just the job.

Dell slide from its presentation deck

This APEX Cloud Platform lets customers deploy, manage, and run containers alongside virtual machines, on-premises. It has been jointly developed with Red Hat. The PowerEdge MC-760/MC-660  servers involved have gen 4 Xeon processors, as you would expect, and PowerFlex software-defined block storage using SAS or NVMe SSDs. 

PowerFlex replaces the Ceph storage component. It can be used with Dell APEX Navigator to enable data replication between on-premises and APEX Block Storage for Public Cloud. 

The APEX Cloud Foundation software is integrated into the Red Hat OpenShift web console. Dell claims it’s the first integrated application delivery platform for Red Hat OpenShift on the market.

That means updating the components is theoretically simpler, with Dell claiming an up to 90 percent reduction in update time is possible compared to manual updating. Dell and Red Hat engineering collaboration is also said to give customers fast access to patched code. All-in-all its open source software, mostly, presented in an enterprise software fashion with none of the open-source gruntwork visible.

THere are also various Dell services to help customers get up to speed with this software.

AI angle

Dell is also announcing a validated design for Red Hat OpenShift AI on APEX Cloud Platform which offers customers a guide on how to deploy a digital assistant utilizing a Large Language Model (LLM) and the Retrieval Augmented Generation (RAG) framework. It’s a multi-cloud offering and shows how customers can use RAG with their own data – securely – on-prem.

Bootnote

IBM says retrieval-augmented generation (RAG) is an AI framework for improving the quality of LLM-generated responses by grounding the model on external sources of knowledge to supplement the LLM’s internal representation of information. Implementing RAG in an LLM-based question answering system has two main benefits: It ensures that the model has access to the most current, reliable facts, and that users have access to the model’s sources, ensuring that its claims can be checked for accuracy and ultimately trusted.

Backblaze B2 cloud / Stores HYCU backup data / In a low cost vault

By
Chris Mellor
-
Backblaze and HYCU against a Japanese haiku background

Cloud storage provider Backblaze has inked a deal with SaaS backup supplier HYCU.

Backblaze provides S3-compliant B2 cloud storage more affordably than Amazon S3 and with no egress charges. It costs $6/TB/month on a pay-as-you-go basis whereas Amazon S3 costs $26/TB/month with a $0.09/GB egress charge. HYCU is a fast-growing on-premises and in-cloud SaaS backup supplier with a strong focus on protecting customer data in SaaS apps. Now HYCU customers can store their HYCU DPaaS (Data Protection-as-a-Service) backups in Backblaze’s cloud repository.

HYCU SVP Product Subbiah Sundaram said: “Our collaboration with Backblaze now offers joint customers a cost-effective and scalable data protection solution combining the best in backup and recovery with Backblaze’s streamlined and secure cloud storage.”

HYCU and Backblaze reckon customers will get performance as B2 Cloud Storage has a 99.9 percent uptime service level agreement (SLA) and no cold delays or speed premiums. Its Object Lock feature provides data immutability, which protects backups from ransomware and satisfies evolving cyber insurance requirements.

Elton Carneiro, senior director of Partnerships at Backblaze, said: “Backblaze and HYCU is a game changer for higher education organizations and businesses who need flexible data management and storage. This partnership brings together HYCU’s speed and Backblaze’s affordability, and both platform’s simplicity.”

HYCU provides a number of other services beyond vanilla backup, such as data migration and disaster recovery. A Backblaze blog notes that HYCU “gives customers the ability to recover from disaster almost instantly, keeping them online and in business, with an average recovery time of 10 minutes.”

The two companies said that the integration is the first step in their partnership, with the intention to offer Backblaze B2 Cloud Storage as a part of HYCU’s data protection portfolio in the future.

NetApp tells analysts it’s focusing on full AI lifecycle

By
Chris Mellor
-

NetApp has been providing its storage and data management systems for AI work since 2018 or earlier and says it is well positioned as large language model workloads transition to general across-the-board enterprise use.

The company may be looking to correct perceptions it may be lagging in the AI field, because it can’t point to marquee AI engagements such as those promoted by, for example, DDN and VAST Data, nor product offer changes like those promoted by Dell Technologies and HPE.

But the company told financial analysts at its Insight 2023 conference that, with the surge in AI interest, it is focused on the full lifecycle of AI purchases, not just the speeds and feeds of pushing data to GPUs. It is focusing on where the data is being generated, how it’s being organized, how it’s being unified, how it’s being prepared, how it’s then being stuffed into GPUs for training; a very important part of it, but just one little bit of it.

On that point, it said it had introduced ONTAP Nvidia GPUDirect Storage (GDS) support capabilities in April.

NetApp is also looking at what happens to data after that, when the results, the validation of the AI model happens. It sees the entirety of the AI data lifecycle as what will be driving revenue for NetApp. 

CEO George Kurian said customers need “an intelligent data infrastructure that combines hybrid multi-cloud data storage with integrated data services and AI powered cloud operations, monitoring, optimization, and automation.” He claimed: “AI is powered by data and data runs on NetApp,” because many enterprises are NetApp customers. NetApp thinks that the customers are sitting on data that has value that they are unable to derive without AI and NetApp can help with that.

Director of Product Management Russ Fishman said: “AI isn’t really focused on any one particular deployment methodology, so it’s not really about on-prem, it’s not really about cloud, it’s really about all of it. And NetApp has this very unique position in the market, which is that we have a leading storage operating system, which is available on all the clouds and on-prem.”

NetApp said it’s working with partners, e.g. Nvidia, with whom it has delivered five or six unique solutions that it co-developed, co-innovated, and has brought to market and been adopted by hundreds of customers. 

Andy Sayare, director of Global Strategic Alliances for AI, said: “We paired our storage technology with video server technology, the DGX platform, and their Mellanox switching … So together that made what we call ONTAP AI, which is essentially our operating system managing AI workloads in this converged infrastructure stack.” Some customers don’t want this on-premises. “One alternative is what’s now called DGX Cloud that Nvidia offers. It’s currently available in OCI, but it will be moving to Azure and GCP and other clouds in coming months.”

He added: “We work with Nvidia on their SuperPOD, which is of course large-scale AI training, often focused on large language models and other very large training situations.”

Other partners include Domino Data Lab (MLOps), Run.AI, and computer system vendors Cisco, Lenovo, and Fujitsu.

The NetApp pitch is that virtually all of its products are AI ready and that it has hundreds of customers using NetApp in AI workloads already.

Wells Fargo analyst Aaron Rakers asked: “Does the DGX Cloud Solution utilize NetApp ONTAP AI as [its] primary storage backend versus other alternatives?”

Sayare replied: “So currently, no … DGX cloud is leveraging an alternative storage for its scratch space. So this is the area that the GPUs use as extra space from their internal storage. What NetApp is doing is working adjacent to that to help connect the data from multiple sources, multiple clouds and on-prem, to be able to bring that data together, to be able to train those models.”

He explained: “Scratch space in general is just that – it’s really the last mile, right? It’s ephemeral data, which means that it’s not stored long term, it’s not protected. The data management capabilities are generally not included there. What we’re finding from customers is that that is not us – that’s not a complete solution.”

Evercore analyst Irvin Liu asked: “Do you see a share gain potential or opportunity presented by AI or is or most organizations are going to stick with their incumbent vendors and avoid a major upgrade or a major transformation prior to jumping into the AI journey?”

Sayare said: “I think we’re extremely well positioned … We’ve been building for AI for five years, five and a half years. So I think we have a portfolio that makes us competitive in non-traditional NetApp and non-NetApp customers …  I think there’s always the opportunity.”

NetApp is not actually predicting it will make share gains through AI.

Comment

Both HPE and Dell are making big marketing pushes around AI. NetApp has been more muted in its approach, saying its Data Fabric spanning the on-premises world, co-lo services with Equinix, and first party ONTAP-based AWS, Azure, and GCP offerings enable its customers to access and use data anywhere in their enterprises’ reach. And do so better than any other supplier.

This is an incumbent’s pitch, and analysts recognized this.

Rakers told subscribers: “NetApp … emphasized that is increasingly pulled into discussions following GPU purchases as customers look to maintain high GPU compute utilization. AI is expected to drive more capacity-optimized flash solutions and consumption of its data management software.”

William Blair’s Jason Ader noted: “On the AI side, NetApp sees the ongoing hype around generative AI as highlighting the importance of enterprise data management (particularly related to unstructured data).” He told subscribers: “Management sees more of a revenue opportunity on the data management/inference side for generative AI models, since inferencing accounts for 85 percent of run time with generative AI models (though not necessarily 85 percent of the spending).”

He points out: “Within AI, there is competition ramping up from high-performance storage vendors like Pure and VAST Data.”

In effect, having made its product offers AI-ready, NetApp is talking to its existing customers rather than seeing AI as an opportunity to gain new customers.

SK hynix talks up AI demand as memory market recovers

By
Chris Mellor
-

SK hynix revenues for the third 2023 quarter ended September 30 were $6.9 billion, down 16.4 percent annually, with a net loss of $1.66 billion as the memory market recovers from recession. The results were helped by demand for its HBM3 memory used by Nvidia GPUs.

Update: added note about SK hynix opposing terms of Kioxia Western Digital merger deal; added note about merger talks ending. 26 October.

The drop represents an improvement on the dive in Q2, which was a whopping 47 percent, and a narrowing of losses, which stood at $2.14 billion in Q2.

The Korean company fabricates DRAM and NAND chips, and builds SSDs. This was its second successive growth quarter, with both DRAM and NAND contributing, as well as average selling price increases. The DRAM business has turned around after two quarters of losses. SK hynix said: “Revenues grew 24 percent, while operating losses narrowed 38 percent, compared with the previous quarter, thanks to strong demand for high-performance mobile flagship products and HBM3, a key product for AI applications, and high-capacity DDR5.”

A statement from CFO Kim Woohyun said: “Our strong position as the world’s key provider of the products such as HBM and DDR5 will help us discover new markets differentiated from the past. We will work toward strengthening our position as the world’s best provider of the high-performance, premium memory products.”

SK hynix revenues by quarter/fiscal year
Q4 2023 is likey to be a growth quarter

DRAM shipments increased 20 percent quarter-on-quarter, with a 10 percent ASP increase, and NAND shipments also rose with high-capacity mobile products and SSDs taking the lead.

The recovery in SK hynix’s DRAM business should quicken due to generative AI technology driving GPU demand with a knock-on effect for HBM3. It’s forecasting a 60-80 percent CAGR for the HBM DRAM market over the next five years. It shipped 1 billion HBM3 samples in August. The company sees a NAND market recovery in place as well.

SK hynix plans to increase investments in high-value flagship products such as HBM, DDR5, and LPDDR5. The company will increase the share of the products manufactured from 1-alpha and 1-beta process nodes, the fourth and the fifth generations of the 10nm process, while increasing investments in HBM and TSV (Through Silicon Via) technology.

Aaron Rakers, a Wells Fargo analyst, thinks the results are an overall positive for the memory market and other players such as Micron. He calculates SK hynix DRAM revenues as ~$4.627 billion in the quarter, down 11 percent year-on-year and up 34 percent quarterly.

He estimates NAND revenues were ~$1.865 billion, down 26 percent year-on-year but up 11 percent quarter-on-quarter. He reckons ~29 percent of the company’s NAND revenue is driven by SSDs (including discrete NAND for SSDs), while the mobile market accounted for 42 percent.

Kioxia and Western Digital NAND/SSD business merger

During the SK hynix earnings call, on being asked about SK hynix opposing a mooted Kioxia merger with a spun-off Western Digital NAND and SSD business, the CFO said: “SK Hynix is not agreeing on the deal given the overall impact on the value of the company’s investment in Kioxia.”

He added: “We cannot elaborate on further specific reasons and the progress of the merger due to confidentiality, but one thing is clear: we [SK hynix] will make a decision for all stakeholders, including shareholders and Kioxia.” Korea Joongang Daily reported this news.

That’s a pretty clear viewpoint. Subsequently the Nikkei reported that the merger talks had been abandoned. Following SK hynix’ public opposition and after Bain Capital could not provide any countervailing arrangements, Western Digital ended the talks. This is only the latest turn in what has become a convoluted saga. It may not be the actual end.

Storage news ticker – October 25

By
Chris Mellor
-

Data intelligence supplier Alation has signed up three new partners: Monte Carlo, Sigma Computing, and ThoughtSpot, all of which provide connectors to data sources.

…

Geo-distributed and decentralized storage provider Cubbit has a distribution agreement with Exclusive Networks, a global cybersecurity specialist for digital infrastructure, whose gross sales hit $4.9 billion in 2022. The deal achieves strategic positioning into geo-distributed secure cloud storage S3 services for Exclusive Networks. It includes joint initiatives to deliver the Cubbit DS3 offering on Exclusive Networks’ X-OD platform, an on-demand consumption channel for cybersecurity and infrastructure products and services. The platform enables channel partners to buy, manage, and distribute the technology, as well as create their own subscription-based service wrappers to suit bespoke customer demand.

Exclusive Networks will help Cubbit to dramatically scale its revenue growth objectives. Following this agreement, Cubbit will also directly open new offices in France, Germany, and the United Kingdom.

…

Lakehouse supplier Databricks is buying data pipeline supplier Arcion for $100 million. This follows Databricks’ $500 million funding round last month. Arcion’s software will get data from disparate sources into Databricks’ lakehouse for analysis. Databricks had participated in Arcion’s $13 million A-round in February 2022. Arcion’s software will be integrated into MosaicML’s GenAI software. Databricks bought MosaicML for $1.3 billion in June.

…

Multi-protocol software-defined storage supplier DataCore has bought Workflow Intelligence Nexus (WIN), which supplies workflow services and software. It will be folded into DataCore’s Perifery business unit. Abhi Dey, GM and COO of Perifery, said: “WIN has deep roots in the M&E sector; combined with our existing core technologies, we will deliver an even more powerful solution portfolio to transform the industries we play in.” WIN’s software integrates with media asset management systems that can automate over 400 manual workflows. Previously, Perifery announced its Swarm software integrated with Seagate’s CORVAULT for on-premises storage.

…

DataStax announced a new integration with LangChain, the most popular orchestration framework for developing applications with large language models (LLMs). The new integration lets developers leverage the power of the Astra DB vector database for their LLM, AI assistant, and real-time generative AI projects through the LangChain plugin architecture for vector stores.

…

Large (>20 GB) file transfer provider MASV now integrates with Storj Distributed Storage. Storj users can create a free MASV Portal and invite uploads from remote teams to their storage bucket, without an account. MASV says it has also increased file transfer speeds by 35x for users sending a single transfer containing thousands of individual files, like image sequencing. Case study here.

…

Micron is shipping production samples of its low-power double data rate LPDDR5X memory, built with a 1-beta process, for use with Qualcomm’s Snapdragon 8 Gen 3 mobile phone chip. It runs at 9 Gbps.

…

Wafer-scale AI compute supplier Cerebras is using VAST Data storage to enable optimal utilization of its Condor Galaxy, built in strategic partnership with G42, and its other AI supercomputers. Condor Galaxy is a network of nine interconnected AI supercomputers that will provide 36 exaFLOPS of AI compute, and was used to train Jais, the world’s highest quality Arabic large language model. More information here.

Veeam announced the Veeam Data Platform 23H2 update, including the new Veeam Backup & Replication v12.1 release. It includes more advanced data security, malware detection, automated recovery, and hybrid cloud capabilities, with new security and ransomware protection functionalities for Veeam Backup & Replication, Veeam ONE, and Veeam Recovery Orchestrator.

Veeam also flagged up a strategic partnership with cybersecurity supplier Sophos. The Veeam Data Platform will integrate with Sophos Managed Detection and Response (MDR), adding a layer of human-led threat detection and response. Together, Veeam and Sophos technologies detect cybersecurity threats, determine the warranted response, and rapidly recover affected data.

…  

Volumez cloud-based block storage composing software is available in the Azure marketplace. Users can select Volumez’s composable data infrastructure software-as-a-service (SaaS) to set up direct Linux-based data paths via a simple declarative interface for Azure Kubernetes environments (AKS) and Azure virtual machines (VMs). John Blumenthal, Volumez chief product and business officer, said: “Our composability service creates the highest performing data infrastructure in market, delivering unrivaled IOPS, latency, and throughput for databases and complex applications like AI/ML.” That’s up to 2 million IOPS and 200 microseconds latency.

…

Australian Bitcoin miner Iris Energy is partnering with Weka as part of its entry into the generative AI market. Weka will provide storage and data management for generative AI and performance-intensive workloads. Daniel Roberts, co-founder and co-CEO of Iris Energy, commented:  “We look forward to forging a long-term partnership with WEKA in servicing the growing generative AI and GPU compute markets.”

…

HPE’s Zerto unit has issued yet another report. This one says that more than a third of companies still do not have a well-rounded, holistic ransomware strategy in place. The survey also found that companies are reevaluating their data protection and cyber resilience strategies to align more closely with the threats they are facing. You can view the full report here.

NetApp intros QLC block array amid AI push

By
Chris Mellor
-

NetApp is attempting to win over customers with a pitch to reduce costs in all-flash arrays and cloud services as it strives for growth in an AI-infused market.

The headline item at the NetApp Insight customer shindig in Las Vegas was a C-series implementation of its ASA all-flash SAN array using QLC flash drives like the AFF C-Series announced eight months ago. QLC SSDs, with 4 bits per cell, cost less than the standard TLC (3 bits per cell) flash drives used in the mainstream AFF and ASA array products, providing a more affordable all-flash array at near-disk prices. NetApp is also increasing GenAI support with Google and improving Amazon Sagemaker integration.

Harv Bhela, chief product officer at NetApp, said: “As we unveil updates to our enterprise-grade storage portfolio, we’re setting the stage for a transformative era in data management … There are two major disruptions today for customers: the opportunity of AI and the threat of ransomware.” NetApp made announcements in these areas as well.

ASA C-Series

The C-Series AFF array has turned out to be the fastest ramping product introduction in NetApp’s history, showing customers have an appetite for lower-cost all-flash arrays. As there are more than 20,000 customers running block workloads on ONTAP systems, NetApp is now giving them access to C-Series affordability as well.

NetApp ASA C800 screengrab
Screen grab from ASA C800 installation video

We understand this is a straight repeat of the AFF C-Series introduction with three similar hardware products:

  • ASA C800 – 4RU x 48 slot controller chassis
  • ASA C400 – 4RU x 48 slot
  • ASA C250 – 2RU x 24 slot

In effect, these are QLC implementations of the TLC ASA A800, ASA A400, and ASA 250 arrays. They support an NS224 expansion shelf with a 2RU x 24-slot chassis.

NetApp array specs
Effective capacity is based on 5:1 storage efficiency ratio with all drives installed

The ASA C-Series has a Six Nines (99.9999 percent) data availability guarantee, and is eligible for the NetApp Ransomware Recovery Guarantee. They are supported by ONTAP v9.7 or later.

NetApp has a set of no-cost services to encourage customers to move to the ASA products: upfront solution design assessment that includes the evaluation of any existing architecture, free data migration, software license and services, or three months of complimentary storage support during the migration period, and free training workshops.

We’re told ASA C-Series has a cost of $0.50/GB. It is not yet clear if these are raw gigabytes or effective capacity gigabytes.

Comment

NetApp does not provide raw drive capacities nor does it say how many rack units are needed for the maximum configurations. That means we cannot work out a TB per RU number or compare its rack space footprint with Pure Storage FlashArray//C systems, which also use QLC flash drives. These are Pure’s own design with up to 75 TB capacity. The highest-capacity commercially available datacenter QLC flash drive is Solidigm’s DC P5336 with up to 61.44 TB capacity.

If NetApp is using 61 TB SSDs in the ASA C-Series then it is at a TB/RU disadvantage relative to Pure unless it has better data reduction ratios. The FlashArray//C was introduced in 2019.

More public cloud and anti-malware stuff

NetApp is announcing three cost-cutting arrangements for its first-party public cloud services:

  • Amazon FSx for NetApp ONTAP will lower overall TCO by taking advantage of VPC Peering to connect to VMware Cloud on AWS. 
  • Azure NetApp Files has enhanced capacity scaling, enhanced performance for Oracle databases and support for transparent automated tiering of data to the Azure Cool Blob tier for improved cost savings.
  • Google Cloud NetApp Volumes, introduced in August 2023, has a new Standard service layer, with an over 30 percent reduction in cost/GB.

NetApp and Google are announcing support for Google Cloud’s Vertex AI with Google Cloud NetApp Volumes to support hybrid Generative AI applications. Ronen Schwartz, SVP and GM, Cloud Storage at NetApp, said: “This offering allows NetApp data on-premises to connect directly to Google’s Vertex AI on Google Cloud NetApp Volumes, with no additional engineering.”

Vertex AI provides MLOps tools for data scientists and ML engineers to automate, standardize, and manage ML projects.

Amazon Sagemaker, which already supporting Amazon FSx for NetApp ONTAP, will add Amazon Sagemaker notebooks, which can access Amazon FSx for NetApp ONTAP via the native S3 object protocol.

NetApp announced the public preview of a new disaster recovery offering for VMware, integrated directly into the NetApp BlueXP unified control plane. With the new NetApp BlueXP DR for VMware software, a few steps enable failover and failback from an on-premises VMware infrastructure directly to the public cloud, starting with Amazon FSx for NetApp ONTAP, or to an alternative on-premises datacenter.

Guarantees

NetApp has extended its Ransomware Recovery Guarantee to make it available for all on-premises NetApp ONTAP-based storage systems, including AFF, ASA, and FAS. NetApp will warrant snapshot data recovery in the event of a ransomware attack. If data copies can’t be recovered through NetApp or one of its trusted partners, NetApp will offer compensation.

NetApp has its Ransomware Assurance Service, which helps ensure customer systems are protected against ransomware attacks. It validates a customer’s configuration for secured data snapshots and conducts a recovery test.

The Keystone Storage-as-a-Service offering has been given integrated performance and availability guarantees, but no details are available. NetApp is previewing a goal to offer a ransomware recovery guarantee for Keystone in the next 90 days.

We have no availability and pricing data for the ASA C-Series, beyond the $0.50/GB number.

1...101102103...437Page 102 of 437

ABOUT US

Blocks & Files is a storage news, information and analysis site covering storage media, devices from drives through arrays to server-based storage, cloud storage, networking and protocols, data management, suppliers and standards. We publish news, views, opinions and analysis of storage industry events and developments, aiming to keep storage users, developers, consultants, marketeers, salespeople, support staff, analysts and everyone else involved in IT storage up to date with what’s going on.

FOLLOW US

© Situation Publishing, 2018-2024