Home Blog Page 103

NetApp revenue surges as competitors struggle

By
Chris Mellor
-

NetApp revenue rose 5 percent for Q3 FY24 on the back of record all-flash array sales.

It reported $1.6 billion in revenues with profits of $313 million versus $65 million a year earlier. There was a record all-flash array (AFA) annual run rate of $3.4 billion, up 21 percent year-on-year. NetApp introduced its lower cost AFF C-Series product in October last year and it says sales exceeded expectations, as did the ASA series of all-flash SAN arrays.

CEO George Kurian said on the earnings call: “I’m pleased to report that we delivered exceptional performance across the board, despite an uncertain macro environment. Revenue was above the midpoint of our guidance, driven by the momentum of our expanded all-flash product portfolio. This strength coupled with continued operational discipline yielded company all-time highs for consolidated gross margin, operating margin, and EPS for the second consecutive quarter.”

Kurian is “confident in our ability to capitalize on this momentum, as we address new market opportunities, extend our leadership position in existing markets, and deliver increasing value for all our stakeholders.”

The revenue rise reverses four successive down quarters and contrasts with competitors. Product sales of $747 million rose 9.5 percent and services went up to $859 million from $844 million.

Financial summary

  • Consolidated gross margin: 72 percent
  • Operating cash flow: $484 million
  • Free cash flow: $448 million
  • Cash, cash equivalents, and investments: $2.92 billion

Kurian said NetApp expects “a sustainable step-up in our baseline product gross margin going forward with the continued revenue shift to all-flash.” If Pure’s assertion that AFAs will replace disk and hybrid arrays by 2028 comes true, NetApp, with base conversion prospects, could be a bigger beneficiary of that than Pure.

NetApp’s all-flash revenue of $850 million was 7 percent ahead of rival Pure’s latest quarterly revenue of $789.8 million, which was down 3 percent annually.

Why did NetApp do so well this quarter? Lower cost all-flash arrays is the short answer. In contrast to Dell and HPE, it has no PC/notebook, server or networking businesses. Like Pure, it is a dedicated storage player and not exposed to these other markets that are struggling. Kurian said the hybrid cloud segment strength was ”driven by momentum from our newly introduced all-flash products and the go-to-market changes we made at the start of the year.”

He added: ”Entering FY24, we laid out a plan to drive better performance in our Storage business and build a more focused approach to our Public Cloud business, while managing the elements within our control in an uncertain macroeconomy to further improve our profitability. These actions have delivered strong results to-date [and] support our raised outlook for the year.”

The results don’t yet extend to the public cloud business, which had a minor revenue lift of just 0.666 percent. Kurian said: “As I outlined last quarter, we are taking action to sharpen our approach to our public cloud business. As a part of this plan, we exited two small services in the quarter. We also began the work of refocusing Cloud Insights and InstaClustr to complement and extend our hybrid cloud storage offerings and integrating some standalone services into the core functionality of Cloud Volumes to widen our competitive moat.”

NetApp is focusing its public cloud efforts on first-party and hyperscaler marketplace storage services. These “are growing rapidly, with the ARR of these services up more than 35 percent year-over-year.”

Kurian says NetApp is increasing its AFA market share. “As customers modernize legacy 10k hard disk drives and hybrid flash environments, we are displacing competitors’ installed bases with our all-flash solutions, driving share gains.”

GenAI hype is helping NetApp, with Kurian saying: “We saw good momentum in AI, with dozens of customer wins in the quarter, including several large Nvidia SuperPOD and BasePOD deployments. We help organizations in use cases that range from unifying their data in modern data lakes to deploying large model training environments, and to operationalize those models into production environments.”

He said NetApp’s “Keystone, our Storage-as-a-Service offering, delivered another strong quarter, with revenue growing triple-digits from Q3 a year ago.”

Guidance for next quarter’s revenues  is $1.585 billion to $1.735 billion, 5 percent up on the year-ago Q4 at the midpoint. NetApp is raising its full-year revenue outlook to $6.185 billion to $6.335 billion, a 1.6 percent decrease on last year at the midpoint.

Comment

NetApp is a stable profit-making machine that has not grown its revenues for 11 years, as a chart of revenues by quarter by fiscal year illustrates:

A direct chart of annual revenues shows the same thing:

George Kurian became the CEO in 2015. Since then he’s weathered two revenue dips, in 2016-2017 and 2020, but he’s not been able to to drive NetApp’s revenues markedly higher than they were in 2013. It’s a company whose foray into the public CloudOps business has not yet proved worthwhile, despite making around $149 million of acquisitions between 2020 and 2022 . NetApp is pulling in just $151 million of public cloud revenues presently. Much of that is due to ONTAP running in the public cloud and not actually due to the acquisitions.

NetApp is generating cash and it’s not directly threatened by any competitor in the near term.

Dell eyes AI-fueled revival after fiscal 2024 downturn

By
Chris Mellor
-

Strong sequential growth in storage revenues at Dell was not enough to prevent an 11 percent revenue fall for the final fiscal 2024 quarter as the company waits in hope of an AI-driven recovery.

Dell reported $22.3 billion in revenues for the quarter ended February 2, with a $1.2 billion profit, up 91 percent year-on-year. Full year revenues were $88.4 billion, down 13.5 percent from the year before, with a profit of $3.2 billion, 32 percent higher year-on-year. Squeezing more profit from lower revenues is quite the achievement.

Jeff Clarke, Dell
Jeff Clarke

Vice chairman and COO Jeff Clarke said in prepared remarks: “In a year where revenue declined, we maintained our focus on operational excellence delivering solid earnings per share and outstanding cash flow. FY24 was one of those years that didn’t go as planned, but I really like how we navigated it. We showed our grit and determination by quickly adapting to a dynamic market, focusing on what we can control, and extending our model into the high growth AI opportunity.”

CFO Yvonne McGill said: “We generated $8.7 billion in cash flow from operations this fiscal year, returning $7 billion to shareholders since Q1 FY23. We’re optimistic about FY25 and are increasing our annual dividend by 20 percent – a testament to our confidence in the business and ability to generate strong cash flow.”

Dell revenues by quarter/fiscal year
Six successive down quarters with Dell’s Q4 FY24 revenues below the 2019 level

Quarterly financial summary

  • Gross margin: 23.8 percent vs 23 percent a year ago
  • Operating cash flow: $1.5 billion
  • Free cash flow: $1 billion vs $2.3 billion last year; 55 percent lower
  • Cash, cash equivalents, and restricted cash: $7.5 billion vs $8.9 billion last year
  • Diluted earnings per share: $1.59
Dell earnings

Dell has two main business units – Infrastructure Solutions Group (ISG) and Client Solutions Group (CSG). The larger CSG, with its PCs and laptops, had revenues of $11.7 billion, 12 percent down year-on-year, while ISG, with its servers, storage and networking, reported $9.3 billion, six percent lower annually. 

Dell annual revenues, profit
Dell annual revenue history shows FY24 having the lowest revenues since 2018

Servers and networking brought in $4.9 billion, the same as a year ago, while storage was responsible for $4.5 billion, down 10 percent year-on-year but up 16 percent quarter-on-quarter.

Dell server and storage revenues

Clarke said: “Our strong AI-optimized server momentum continues, with orders increasing nearly 40 percent sequentially and backlog nearly doubling, exiting our fiscal year at $2.9 billion,” implying that the next few quarter’s results should be better.

“We’ve just started to touch the AI opportunities ahead of us, and we believe Dell is uniquely positioned with our broad portfolio to help customers build GenAI solutions that meet performance, cost and security requirements.”

Why the quarterly storage revenue rise? Dell said there was demand strength across the portfolio, more than any expected seasonal improvement. Clarke said on the earnings call: “We had year-over-year demand growth in the unstructured space, ECS, as well as PowerScale. They grew quarter-over-quarter and year-over-year on a demand basis. Those are generally good indicators … around AI file and object, which are the data classes that generally feed the AI engines … Our progress in traditional storage was good too. We were ahead of our normal seasonality … It was down year-over-year, but better than we expected across mid-range, across our data protection products and our high-end storage products.”

The outlook is for growth. Clarke said: “We believe the long-term AI action is on-prem where customers can keep their data and intellectual property safe and secure.  PCs will become even more essential as most day-to-day work with AI will be done on the PC. We remain excited about the long-term opportunity in our CSG business.”

He added: “Our storage business will benefit from the exponential growth expected in unstructured data … We think AI moves to the data. More data will be created outside of the data center going forward than inside the data center today. That’s going to happen at the edge of the network. A smart factory, an oil derrick or platform, a deep mine, all variations of this. We believe AI will ultimately get deployed next to where the data is created driven by latency.” 

In his view, enterprises will “quickly find that they want to run AI on-prem because they want to control their data. They want to secure their data. It’s their IP and they want to run domain specific and process specific models to get the outcomes they’re looking for.”

Dell thinks an AI-fueled recovery in storage demand will lag stronger server demand by a couple of quarters.

Revenues for Q1 FY25 are expected to be between $21 billion and $22 billion, 3 percent higher annually at the midpoint. But then growth will accelerate. Clarke sees “modest growth in traditional [servers], stronger growth in AI-enabled servers, and an opportunity with storage as the year progresses.”

Full FY25 revenues should be between $91 billion and $95 billion, up 5 percent year-on-year at the midpoint. McGill said: “We expect ISG to grow in the mid-teens fueled by AI, with a return to growth in traditional servers and storage, and our CSG business to grow in the low single digits for the year.”

HPE storage results vanish in financial reporting revamp

By
Chris Mellor
-

HPE has changed its quarterly financial reporting structure with storage results disappearing into a hybrid cloud category.

Revenues in its first fiscal 2024 quarter ended January 31 were $6.8 billion, 14 percent lower than the year-ago quarter and also below its guided estimate for $6.9 billion to $7.3 billion. There was a $387 million profit, 23 percent less than last year. The move to GreenLake is causing overall revenue declines as subscription revenue is recognized over time whereas straightforward product sales are recognized when the product ships.

Antonio Neri, HPE president and CEO, said in prepared remarks: “HPE exceeded our profitability expectations and drove near-record year-over-year growth in our recurring revenue in the face of market headwinds, demonstrating the relevance of our strategy.” HPE did not exceed its revenue expectations, however. The annual revenue run rate was $1.4 billion, 42 percent more than a year ago, and primarily due to the GreenLake portfolio of services.

HPE revenues
HPE revenues have basically flatlined since 2019

HPE’s new organizational structure consists of the following segments: Server; Hybrid Cloud; Intelligent Edge; Financial Services; and Corporate Investments and Other. It has amalgamated previously separate Compute and HPC & AI reporting lines into a single Server category. The new Hybrid Cloud reporting line includes three elements:

  • The historical Storage segment
  • HPE GreenLake Flex Solutions (which provides flexible as-a-service IT infrastructure through the HPE GreenLake edge-to-cloud platform and was previously reported under the Compute and the High Performance Computing & Artificial Intelligence (“HPC & AI”) segments) 
  • Private Cloud, and Software (previously reported under the Corporate Investments and Other segment)

This means we no longer have a direct insight into the business health of its storage portfolio. HPE says it’s doing this to better reflect the way it operates and measures its business units’ performance. This new hybrid cloud segment is meant to accelerate customer adoption of HPE’s GreenLake hybrid cloud platform.

Financial summary

  • Gross margin: 36.4 percent, up 2.4 percent year-over-year
  • Operating cash flow: $64 million
  • Free cash flow: -$482 million
  • Cash, cash equivalents, and restricted cash: $3.9 billion vs $2.8 billion last year 
  • Diluted earnings per share: $0.29, 24 percent lower than last year
  • Capital returns to shareholders: $172 million in dividends and share buybacks

Server segment revenues were $3.4 billion, down 23 percent year-over-year. Hybrid Cloud pulled in $1.3 billion, down 10 percent, with Intelligent Edge reporting $1.2 billion, up 2 percent. Financial services did $873 million, down 2 percent, while Corporate Investments and Other was responsible for $238 million, up just 1 percent.

CFO Marie Myers said on the earnings call: “Demand in Intelligent Edge did soften due to customer digestion of strong product shipments in fiscal year ’23, which is lasting longer than we initially anticipated and is the primary reason Q1 revenue came in below our expectations.” Specifically, “campus switching and Wi-Fi products eased materially, particularly in Europe and Asia.”

Neri’s top-down view was that “overall, Q1 revenue performance did not meet our expectations.” It was “lower than expected in large part because networking demand softened industry-wide and because the timing of several large GPU acceptances shifted. Additionally, we did not have the GPU supply we wanted, curtailing our revenue upside.”

”This quarter is a moment in time and does not at all dampen our confidence in the future ahead of us.”

Myers commented: ”Demand for our traditional server and storage products has stabilized,” although “our traditional storage business was down year-over-year on difficult compares, given backlog consumption in Q1 ’23. Total Alletra subscription revenue grew over 100 percent year-over-year and is an illustration of our long-term transition to an as-a-service model across our businesses. We are starting to see AI server demand pull through interest in our file storage portfolio. We are also already seeing some cross-selling benefits of integrating the majority of our HPE GreenLake offering into a single business unit.”

Neri said HPE was “capturing the explosion in demand for AI systems” with GPU-enhanced servers, called Accelerator Processing (AP) units, orders rising. AP orders now represent nearly a quarter of HPE’s entire server orders since fiscal 2023’s Q1. Neri said: ”Our pipeline is large and growing across the entire AI life cycle from training to tuning to inferencing.”

Server revenues should grow due to AI demand and better GPU availability, and “hybrid cloud will benefit from continued HPE GreenLake storage demand and the rising productivity of our specialized sales force.”

HPE’s next quarter outlook is $6.8 billion in revenues up or down $0.2 billion, a 2.5 percent annual decrease at the midpoint. Myers said: ”For hybrid cloud, we expect sequential increases through the year as our traditional storage business improves and HPE GreenLake momentum continues. We expect meaningful progress through the year.”

Snowflake replaces CEO as losses widen

By
Chris Mellor
-

After reporting growth in both annual revenues and losses in its latest results, Snowflake chairman Frank Slootman decided to replace Snowflake CEO Frank Slootman, and appointed SVP Sridhar Ramaswamy in his place.

Revenues in cloud data warehouser Snowflake’s final fy2024 quarter, ended Jan 31, were $774.7 million, 31.5 percent more than a year ago, with a $170 million loss, better than the year-ago’s $207.5 million of red ink. The full year revenue number was $2.8 billion, 35.9 percent more than last year. There was a full year loss of $836.1 million, yes, $0.84 billion, contrasting with fy23’s loss of $797 billion.

Frank Slootman.

Chairman Frank Slootman’s prepared statement said: “We are successfully campaigning the largest enterprises globally, as more companies and institutions make Snowflake’s Data Cloud the platform of their AI and data strategy.”

About the resignation of himself as CEO he said: “As the leading cloud data platform, Snowflake is at the epicenter of the AI revolution. There is no better person than Sridhar to lead Snowflake into this next phase of growth and deliver on the opportunity ahead in AI and machine learning. He is a visionary technologist with a proven track record of running and scaling successful businesses. I have the utmost confidence in Sridhar and look forward to working with him as he assumes this new role.”

Presumably Slootman wants Ramaswamy to “Amp it up,” that being the title of Slootman’s recent book about being a CEO and leader. In that book he writes this about a board removing a CEO: “It will treat removing the CEO as a gut-wrenching, high-risk, last-resort move, never to be done casually or for frivolous reasons.” In his view “hiring a new one is a lengthy process.”

Ramaswamy joined Snowflake in May 2023 when it acquired Neeva, the world’s first private AI powered search engine. He was a Neeva co-founder and a Google SVP before that, and led the launch of Snowflake’s Cortex. This is a fully managed service that offers access to AI models, LLMs, and vector search functionality to enable organizations to analyze data and build AI applications. 

Sridhar Ramaswamy.

We note that Ramaswamy’s LinkedIn profile says of his first 8 months at Snowflake: “Learning Snowflake.” Our bet is that he was a CEO-elect from the start.

What Slootman said in the earnings call bears that out: “I was brought to Snowflake five years ago to help the company break out and scale. I wanted to grow the business fast, but not at all costs. It had to be efficient and establish a foundation for long-term growth.”

“I believe the company succeeded in that mission. The board has run a succession process that wasn’t based on arbitrary timeline, but instead, looked for an opportunity to advance the company’s mission, well into the future. The arrival of Sridhar Ramaswamy through the acquisition of Neeva last year represented that opportunity. Since joining us, Sridhar has been leading Snowflake’s AI strategies, bringing new products and features to market at an incredible pace.”

Snowflake hasn’t made a profit since going public, being run for growth. The company now has 9,437 customers, with 461 having trailing 12-month product revenue greater than $1 million, 39 percent more than a year ago, and 691 Forbes Global 2000 customers, representing 39% and 8% year-over-year growth, respectively. It obviously isn’t making a profit from them. 

If 9,437 customers brought in an average of $82,092 each this quarter and Snowflake made a $170 million loss, then it would have needed 2,071 more customers just to break even. Put another way; on these numbers Snowflake needs a 22 percent customer count increase to start making a profit.

Quarterly financial summary

  • Free cash flow: $324.5 million vs $205 million a year ago
  • Cash, cash equivalents and restricted cash: $1.78 billion
  • Gross product margin: 74 percent

AI is the key market driver, with Ramaswamy saying: “You heard the team say many times, “There’s no AI strategy without a data strategy.” And this has opened a massive opportunity for Snowflake to address. To deliver on the opportunity ahead, we must have clear focus and move even faster to bring innovation on the Snowflake platform to our customers and partners. This will be my focus.”

Apart from AI in the USA, Snowflake has a large revenue opportunity outside the USA, where 80 percent of its revenues are based. The EMEA region accounts for 15 percent of them and Asia Pacific a trivial 5 percent.

Snowflake is guiding next quarter’s product revenue to be $747.5 million +/- $2.5 million; a 26.5 percent increase at the midpoint year on year. Full fy2025 revenue guidance is $3.25 billion, with a lower 22 percent Y/Y growth rate. That’s slower growth than Snowflake has been used to enjoying. Perhaps competitors suchas Databricks and Dremio are finally having an impact on its growth.

Pure projects double-digit growth despite revenue dip

By
Chris Mellor
-

Pure Storage has ended its fiscal year as it started – with a drop in revenues. This time it’s due to subscriptions becoming more popular than anticipated, but the company says it expects to resume double-digit growth from the next financial quarter.

Revenues in Pure’s Q4 of fiscal 2024, ended February 4, declined three percent year-on-year to $789.8 million yet beat its own guidance. Net profit dropped 12 percent to $65.4 million. Full year revenues were $2.8 billion, up 3 percent to surpass guidance, and net earnings fell 16 percent annually to $61.3 million profit.

Charles Giancarlo, Chairman and CEO at Pure Storage, said on the earnings call: “We had a solid Q4 performance and ended the year with increasing sales momentum and balanced performance across our theaters and product portfolio. This momentum and growing customer interest in our platform strategy provides us with increased confidence for the coming year.”

Pure noted that subscription revenue of $329 million was up 24 percent while turnover from product sales was down 15 percent to $460.9 million. Giancarlo said: ”We certainly see our way clear to a majority of revenue over time being in the subscription category.”

Quarterly financial summary

  • Gross margin: 72 percent
  • Free cash flow: $200.9 million
  • Operating cash flow: $244.4 million
  • Total cash, cash equivalents and marketable securities: $1.5 billion

What was responsible for the revenue decline? Pure had better than expected sales of Evergreen//One storage-as-a-service (STaaS) and Evergreen//Flex pay-for-what-you-use storage deals at the expense of perpetual license sales. The latter’s revenues are recognized on shipment while the former’s are recognized over the life of the services contract.

During the conference call, Pure’s confirmed its file and object FlashBlade array has brought in more than $2 billion in total sales since launch. Pure has won an eight-figure Evergreen//One FlashBlade deal with one of the largest GPU cloud providers for arrays to be used in AI processing and training stages, it said. Pure, however, is not the only storage supplier to this unnamed customer.

Giancarlo said Pure’s Portworx container storage product “had a record year and accelerated growth based on customers increasingly graduating their container-based development projects to production scale.” The financial sector was highlighted as adopting Portworx in this way.

Answering an analyst’s question, Giancarlo said there were three aspects to AI for Pure: modeling (training) needing lots of GPUs, inference with fewer GPUs, and storage. Storage for AI needs to be in a single fast-access data environment, one that is not spread across silos. He reckons the training market is the smallest one of the three in terms of total market size.

Pure also thinks that the market researchers including Gartner underplay Pure’s market share because they are biased towards perpetual license (capex) sales. Giancarlo said: “They do not incorporate our Evergreen//Forever subscription, which, as you know, means that we don’t resell the same storage when an array becomes obsolete because in our case, arrays don’t become obsolete.”

He said that an analyst’s “basic premise that the market is under accounting, what we believe would be our growth if our Evergreen//One sales were actually in standard product sales, is absolutely correct.”

Pure did not see more sales into Meta’s Research Super Cluster (RSC) this quarter, which meshes with what Hammerspace is claiming, but Pure CTO Rob Lee said: “Our relationship with Meta is stronger than ever. We’re working with them on almost a continuous basis.”

Inevitably Pure was asked about flash replacing disk, and Giancarlo stepped up to that with no hesitation: “I started saying last year that I expected the last disk storage array to be sold in about five years’ time. We’re now four years in front of that. I’m going to stick with that timetable … I am very bullish on this, and I’ll stick with it, that I think the next three to four years, we’re going to see the decline of disk systems.”

Turning to the future, Giancarlo said: “We are beginning to see some encouraging signs of improvement in the macro environment.” CFO Kevan Krysler said in a prepared statement: “We expect double-digit revenue growth and strong growth of RPO, fueled by our highly differentiated data storage platform, and strength of our Evergreen and Portworx consumption and subscription offerings.”

With this in mind, next quarter’s revenues are expected to be $680 million, 15.4 percent up on the year-ago quarter. Pure is looking for 2025 revenues to be around $3.1 billion, meaning a 10.7 percent rise on 2024.

Cohesity pushes out RAG-enhanced Gaia GenAI backup search chatbot

By
Chris Mellor
-

Cohesity has announced a Gen AI chatbot called Gaia that can search through a customer’s backups to find answers to conversational questions.

This builds on its Turing AI/ML gateway initiative which used AWS Bedrock to generate business context-aware answers to questions. Cohesity claims it’s the industry’s first Gen AI-powered conversational search assistant to help businesses transform secondary data into knowledge. That’s because retrieval augmented generation (RAG) adds a company’s backup data in the Cohesity Data Cloud to the Gen AI’s Large Language Model (LLM) inputs.

Sanjay Poonen

Cohesity CEO and President Sanjay Poonen said in an canned statement: “Enterprises are excited to harness the power of generative AI but have faced several challenges gaining insights into secondary data, including backup, archived and vaulted data – because every approach requires re-hydrating the data, and painfully waiting weeks for the data to be available for analytics and insights. Cohesity Gaia dramatically simplifies this process with our patent-pending approach using Retrieval Augmented Generation.”

This means that tasks previously carried out by skilled and expensive data scientists querying data warehouses and the like, with specialized coded programs, can now be done by ordinary managers and employees querying backup data.

Poonen said: “Our approach delivers rapid, insightful results without the drawbacks of more manual and risky approaches. In short, it turns data into knowledge within seconds and minutes.”

Gaia is a Saas offering integrated with Cohesity’s Data Cloud and features:

  • RAG AI conversational search experiences across cloud and hybrid environments. 
  • A fully indexed database of all files, across all workloads, and at all points in time that supports the creation of AI-ready indexes for rapid conversational search and responses. Initially, Cohesity will support Microsoft 365 and OneDrive data and will expand to more workloads over time.
  • All indexed data is immediately available for reading without the need for backups to be reconstructed, allowing the Cohesity Data Cloud to function like a data lake with real-time access to data
  • Granular role-based access controls and zero-trust security principles to protect access to sensitive information and help maintain compliance with various regulatory requirements.

Cohesity says customers can use Gaia:

    • To assess an organization’s level of cyber resilience.
    • To perform financial and compliance audit checks.
    • To answer complex legal questions.
    • To serve as a knowledge base to train new employees.

    It has announced plans with the public cloud big three; Azure, AWS and Google, to bring their LLM services to Gaia.  

    As an example of Gaia use, Cohesity’s Greg Statton from its Office of the CTO, said: “If you notice a rise in costs in a region, typically, you would search for dozens of invoices, review and compare them, and see if you can discover the reason for the cost increases. It could take hours, days, or weeks to resolve. With Cohesity Gaia, you simply ask, ‘Why have costs increased in the region?’, and Cohesity Gaia will pull the relevant data from your stored data, analyze it, and return an answer to your question. It’s that simple.” 

    A spokesperson for Rubrik, which announced its Ruby Gen AI chatbot in November last year, said: “If you don’t have a GenAI product available today, you’re not in the game. We rolled out Rubrik Ruby last year and it’s already available.”

    Gaia will be made generally available on March 15. Get a PDF solution brief here.

    Nutanix makes its first ever profit

    By
    Chris Mellor
    -
    3D rendering of AI brain

    Nutanix has made its first ever profit as its revenues grow 16 percent year-on-year in its ninth successive growth quarter. VMware and CIsco represent clear opportunities while its Gen AI strategy is still emerging.

    Fifteen years after being founded and eight years after its IPO, hyperconverged infrastructure vendor Nutanix has reached business maturity. Revenues in its second fiscal 2024 quarter, ended Jan 31, were $565.2 million compared to the year-ago $486.5 million, and there was a profit of $32.8 million. Quite a turnaround from the $70.8 million loss last year. The company is still growing its customer base, with 440 new customers in the quarter, taking the total to 25,370.

    President and CEO Rajiv Ramaswami was, as usual, measured in his comment, saying: “We’ve delivered a solid second quarter, with results that came in ahead of our guidance. The macro backdrop in our second quarter remain uncertain, but stable relative to the prior quarter.”

    Nutanix is maintaining a quite consistently high growth rate after a slowdown in 2018-2021 period. Ramaswami became the CEO at the end of 2020. He rescued it from a revenue slump.

    He said: ”We achieved quarterly GAAP operating profitability for the first time in Q2, demonstrating the progress we continue to make on driving operating leverage in our subscription model.”

    Financial summary

    • Gross Margin: 85.6 percent vs 82.2 percent a year ago
    • Annual Recurring Revenue: $1.74 billion
    • Free cash flow:  $162.6 million vs year-ago $63 million

    William Blair analyst Jason Ader said it was another beat (the guidance) and raise quarter, with Nutanix “beating consensus across all key metrics, despite ongoing macro pressure.” 

    The future

    Nutanix sees steady demand ahead. The Broadcom-VMware acquisition is still creating uncertainty for VMware customers and where else should a worried VMware customer go but to the safe and steady haven represented by Nutanix? This is a multi-year opportunity for Nutanix, especially with 3-tier (separate server, storage and client GUI access) client-server VMware customers who could embrace HCI for the first time with Nutanix instead of inside the VMware environment. And also, of course, vSphere and VSAN customers who could migrate fairly simply to Nutanix.

    The Cisco alliance, replacing Cisco’s Hyperflex HCI with Nutanix’ offering is a second tailwind, set to become significant revenue-wise in fy2025. Gen AI is a third potential tailwind.

    Nutanix’ presence in the GenAI market is represented by its GPT-in-a-box system on which to run Large Language Models (LLMs). This was announced six months ago. Nutanix still does not support Nvidia’s GPUDirect protocol and thus GPT-in-a-Box looks like a relatively slow deliverer of data to external Nvidia GPU servers.

    Ramaswami said in the earnings call: “Moving on, adopting and benefiting from generating AI, is top of mind for many of our customers. As such, interest remains high in our GPT in-a-Box offering, which enables our customers to accelerate the use of generative AI across their enterprises, while keeping their data secure.” He talked about a federal agency customer for the system, along with other wins, and: “While it’s still early days, and the numbers remain small, I’m excited about the longer term potential for GPT in-a-Box.”

    He added: “We think Nutanix systems could run Gen AI Inferencing on their X86 servers but are not suited to Gen AI training as they have no GPU support. Ramaswami said the GPT-in-a-Box use cases include “document summarization, search, analytics, co-piloting, customer service, [and] enhanced fraud detection.”

    He characterized customer’s AI involvement as being hybrid multi-cloud in style: “It’s starting out with a lot of training happening in the public cloud, but it’s moving towards, okay, I’m going to run my AI, I’m going to fine tune the training, on my own data sets, which are proprietary that I want to keep carefully. And I’m going to have to potentially look at slightly different solutions for inferencing, which are going to be running closer to my edge locations.”

    The proprietary dataset fine tuning could run on-premises if there were GPU servers there. The GPU-in-a-Box supports GPU-enabled server nodes. These are x86 servers with added GPU accelerators, such as HPE’s ProLiant DL320, DL380a and DL395 which come with Intel or AMD CPUs and Nvidia L4/L40/L40S GPUs. It could run in GPU server farms, such as ones operated by CoreWeave and Lambda Labs. The data would need to be sent there and, again, Nutanix does not support GPUDirect

    Comment

    In the Ramaswami era Nutanix is a judicious company. It has not yet seen a need to add GPUDirect support to pump files faster out to GPU servers. This could mean that it is effectively excluded from large-scale Gen AI training and is an on-premises Gen AI inferencing supporter and small-scale, fine tuning trainer instead. 

    BitRipple moves liquid data using RaptorQ

    By
    Chris Mellor
    -

    Profile. Startup BitRipple has a data mover that can deliver large volumes of data in real-time across wireless networks by using RaptorQ erasure coding and a liquid data concept.

    It says its technology is used in a variety of applications, including defense, streaming video, automotive, space satellite communications, and more. BitRipple delivers massive volumes of data at high-speed with robust security, and ultra-low latency.  This positions BitRipple as a high-speed data mover, along with Vcinity and Zettar.

    BitRipple was co-founded  up in April 2020 by CEO Michael Luby,  VP Systems Pooja Aggarwal and VP Engineering Lorenz Minder. Luby spent 5½ years at the International Computer Science Institute in Berkeley, CA, finishing as a Senior Research Scientist in early 2019. Before that he worked at Qualcomm for almost 10 years, becoming VP Technology. Back in 2009 he founded Digital Fountain and was its CTO until early 1998. It developed broadcast and real-time data transport software and was a predecessor of BitRipple.

    RaptorQ is an example of a computer science fountain code or rateless erasure code. Their defining property is that a potentially limitless sequence of encoding symbols can be generated from a given set of source symbols. The original source symbols can ideally be recovered from any subset of the encoding symbols of size equal to or only slightly larger than the number of source symbols, which limits network overhead.

    The abstract of a 1998 ACM document, to which Luby contributed, said: “A digital fountain allows any number of heterogeneous clients to acquire bulk data with optimal efficiency at times of their choosing. Moreover, no feedback channels are needed to ensure reliable delivery, even in the face of high loss rates.We develop a protocol that closely approximates a digital fountain using a new class of erasure codes that for large block sizes are orders of magnitude faster than standard erasure codes.”

    Digital Fountain pioneered RaptorQ use, selling Linux appliances, and was bought by Qualcomm in 2009, which is how Luby ended up in Qualcomm.

    BitRipple describes its technology like this: “The essential idea is that, using RaptorQ, a sender generates and sends liquid packets for each data block to be delivered to a receiver. They are called liquid packets because, like drops of a liquid, they are interchangeable: using RaptorQ, the receiver recovers each block as soon as enough liquid packets arrive at the receiver, independent of which liquid packets arrive.”

    RaptorQ erasure code is specified in IETF RFC 6330. Qualcomm provides a RaptorQ Technical Overview, which says: “Raptor is a forward error correction (FEC) technology implemented in software that provides application-layer protection against network packet loss. … The RaptorQ encoder and decoder software libraries allow streaming and file delivery services to recover data lost in transit and completely reconstruct it, without using a backchannel.”  

    Specifically: “RaptorQ encodes and decodes a block of source data, called a source block, which is partitioned into equal-size pieces of data, called source symbols. The source block size is configured by the application that incorporates the RaptorQ software library based on the application requirements. The RaptorQ encoder generates repair symbols from the source symbols of a source block, where the repair symbols are the same size as the source symbols and the encoded symbols that can be sent consist of the combination of the source symbols and the repair symbols.”

    “Typically, each encoded symbol is sent in an individual packet together with a 32-bit header, called the FEC Payload ID consisting of an 8-bit source block number and a 24-bit encoded symbol identifier (ESI) that allows the receiver to identify the encoded symbol carried in the packet.” 

    YouTube BitRipple video.

    BitRipple says it uses an intelligent delivery protocol which “automatically adjusts on-the-fly the number of liquid packets to send for each block based on continual feedback from the receiver. The number of liquid packets proactively sent for a block is close to the minimal needed to recover the block when packet loss is low, saving on bandwidth. More liquid packets are proactively sent for a block when packet loss is higher, ensuring that enough liquid packets arrive at the receiver to recover the block without re-transmisson delays.”

    A YouTube video shows remote BitRipple-delivered 4K video almost as fast as local playback and much faster than HTTP Live Streaming (HLS). 

    There are three BitRipple products:

    • Mono – data package delivery over a 1-way connection with no back channel from receiver to sender.
    • Tunnel – accelerates data movement between pairs of endpoints over any network.
    • Multipath Tunnel -accelerates data movement between pairs of endpoints utilizing multiple network paths.

    BitRipple says it enable immersive experiences, such as cloud gaming, remote collaboration, augmented and virtual reality, delivered to geographically-distributed endpoints across wireless networks. It is a quiet, relatively low-profile company, making ripples in the data moving space, and not noisy splashes.

    Mechanical Orchard updates mainframe migration with AI

    By
    Chris Mellor
    -

    Two-year-old startup Mechanical Orchard migrates mainframe apps to the public cloud using generative AI technology.

    Mainframe systems are used by 44 out of the top 50 banks, ten out of ten top insurers, and 23 out of 25 top US retailers. Some 71 percent of the Fortune 500 are still reliant on mainframes, and spend in excess of $20 billion a year buying and maintaining them. The mainframe is a restricted and costly environment. Moving apps from it to the public cloud means they can interoperate easily with open source code, get away from historic COBOL programming, become SaaS-style entities, and be more agile and able to develop their previously fossilized code.

    Rob Mee, Mechanical Orchard
    Rob Mee

    Mechanical Orchard was founded in San Francisco in 2022 by three ex-Pivotal Labs guys, CEO Rob Mee, COO Matthew Work, and VP R&D Dan Podsedly. It took $7 million in a seed round that year and has just raised $24 million in an A-round, having demonstrated substantial progress. It has 50 staff and offices in the UK, Ireland, Italy, and Germany. Customers include Omni Logistics.

    ‍It says it uses iterative, AI-enhanced, and reverse engineering approaches to move large enterprise apps off mainframe systems and into the cloud. During a briefing in London, Mee told us that Mechanical Orchard’s method was to look at a mainframe or other legacy system application and migrate it to the public cloud incrementally, module by module. Mee said “there is no silver bullet, and never will be” that can be used to move legacy apps and their components en masse to the public cloud. This migration requires a meticulous, iterative, and incremental approach.

    Mainframes have distinctive hardware and system software architectures that do not translate easily to x86 compute, commodity-based storage, and networking-based infrastructure. There is no straight equivalence between mainframe Direct Access Storage Devices (DASD) and the main SSD and HDD instances available in AWS, Azure, and Google. Mainframe apps could also have code components that are decades old and may not have documented source code.

    A typical customer engagement starts with a pilot and looks like this. It selects a piece or component of the application that has defined inputs and outputs. If it doesn’t have the source code, Mechanical Orchard treats it as a black box with data flows, which it reverse engineers into a cloud-native application. This is loaded into an isolated sandbox and tested with various inputs, amended as needed when the output is different from what is expected, until it parallel runs with the same outputs as the source legacy module and is just as fast if not faster. This pilot can take six to eight weeks. 

    That is then subject to the customer’s judgement as to whether to take it further. If it’s a yes, component by component, the mainframe application is reverse engineered to use cloud-native code based on public cloud compute, networking, and storage instances. It then runs in parallel with the source mainframe component with equivalence monitoring looking for and detecting different outputs that need understanding and correcting, and checking performance.

    Generative AI is being used to help build a visual representation of the mainframe app component and what it does. Mee said AI can synthesize data, identify coding patterns, and speed up code development and documentation. It helps Mechanical Orchard staff be more productive.

    When full equivalence is attained, says the firm, the customer can switch over from the mainframe to the public cloud component. The switchover point varies with the criticality of the software. A financial trading module will be tested far more severely than a retail system for example. AWS, Azure, and Google’s public clouds are supported. In fact, the three partner with Mechanical Orchard because it represents a mainframe app and data on-ramp for them. Mechanical Orchard represents a different approach from using public cloud mainframe emulation, which leaves the source app still frozen.

    This is a service-based operation such as Arthur Andersen or Price Waterhouse might provide. For customers, they see that the Mechanical Orchard process costs money, but when it is complete they have a migrated mainframe app that gives them the ability to reduce mainframe costs and move into a far more scalable, agile, and fast-moving development environment than mainframe land. Mainframe digital transformation moves at glacial speed. Cloud-native development – DevOps with continuous integration and delivery – is much faster and there is a far larger base of skilled coders available. 

    That’s Mechanical Orchard’s would-be USP – dependable mainframe jailbreaking that is carried out proof point by proof point, component by component, until the frozen mainframe app is public cloud resident and ready to be developed as the business needs.

    Bootnote

    Pivotal Labs, an application development, containerization, and Kubernetes business, was founded in 1989. It was acquired by EMC in 2012 and spun-out in 2013. Rob Mee founded Pivotal and became the CEO that year. The company went public in April 2018. VMware completed a $2.7 billion acquisition of Pivotal in December 2019, at which point Mee left.

    Storage news ticker – February 28

    By
    Chris Mellor
    -

    German system house weSystems is working with partner Atempo and its Miria data mover to archive 40 PB of data in the Azure public cloud’s Blob Storage Archive Tier. The archive belongs to a German automotive manufacturer and is growing at 2 PB/month. A September 2023 weSystems case study said: “Until now, our customer, a well-known German automotive supplier, has stored the data volumes accumulating in daily production on an EMC Isilon storage. A single file can reach a size of 500 GB. At regular intervals, the data was transferred to tape systems for backup and long-term archiving.” Now a hybrid backup and archiving system transfers data to central Ceph storage in a secure data centre at high data throughput. Specifically, most of the data is first transferred to weSystems FlexStorage, based on Ceph, in a central German datacenter and stored there for approx. 3 months. Data can be retrieved from it using Miria. In a second step, the data is then transferred to the Azure Archive Tier, as the customer specified, with Miria retrieval again.

     …

    Cloudian has released new open source software that integrates machine learning library PyTorch with local data lakes running on Cloudian HyperStore S3-compatible object storage, eliminating the dependency on a dedicated parallel file system for ML workflows. Data scientists and AI developers can run ML on data resident in local Cloudian object storage, without the need to move and stage the data into another system. The ML tasks can also run on local compute resources such as AWS Outposts and Local Zones. Cloudian is a certified Service Ready partner for AWS Outposts and Local Zones, and is commercially available through the AWS Marketplace.

    Cloudian contributed enhancements to AWS Labs’ open source S3-Connector-for-PyTorch to enable PyTorch ML algorithms to access data in Cloudian’s HyperStore object storage system via the AWS S3 API. The enhanced S3 connector is available from the GitHub repositories of AWS Labs and Cloudian.

    …

    The 2024 AEC Data Insights Report from data collaborator Egnyte highlights the exponential growth in the AEC industry’s transition from local storage to cloud storage and collaboration solutions. Data storage requirements of the companies surveyed grew at an average 50.3 percent CAGR. This rapid growth trend began during the pandemic, but as companies started to see the benefits of the cloud, growth has maintained that pace. Transitioning to the cloud allows companies to leverage technology integrations, artificial intelligence, real-time collaboration, and meet strict regulatory requirements, such as the Cybersecurity Maturity Model Compliance (CMMC) in the United States and the Building Safety Act in the United Kingdom. To download a copy of the report, click here.

    …

    Can Hammerspace’s HyperNAS really add GPUDirect speed to non-GPUDirect-supporting filers? An industry source says GPUDirect does RDMA direct from the storage system into GPU memory. Let’s suppose you add HyperNAS to a Qumulo system, which doesn’t natively support GPUDirect. HyperNAS provides GPUDirect Hammerspace-held metadata client access but doesn’t provide an RDMA path out of the Qumulo system. Its data, the source suggests, wouldn’t flow at RDMA-based GPUDirect speed. A benchmark test would be necessary to verify this claim.

    …

    S3 tiering to tape with Red Hat-acquired NooBaa’s gateway is discussed in a four-part IBM Storage blog by Senior Technical Staff Member Nils Haustein and others. The blog talks about a new and modern S3 object storage services, using open source NooBaa, that can be installed on IBM’s Storage Scale. NooBaa provides S3 endpoints to users and applications, and allows full control over data placement with dynamic policies per bucket or account. The NooBaa-core standalone software package can be deployed on a Storage Scale cluster providing the S3 endpoints while the buckets and objects are stored in Storage Scale file systems. By leveraging the integration of Storage Scale with Storage Archive, S3 buckets and object stored in the file system can be tiered to tape.

    …

    SaaS app backup supplier Keepit has opened a UK headquarters in the City of London and expanded its UK team, led by industry veteran Jerry Mumford. Over the past three years, Keepit has seen 100 percent year-over-year growth in the UK and is on target to repeat this in 2024. Keepit counts The National Gallery and Oxford University Innovation Ltd among its UK customers.

    …

    Tech Radar has written about MaxDigitalData, a GoHardDrive brand, which sells renewed (used and refurbished) and end-of-life disk drives. A 12TB drive with a three-year warranty costs less than $90. A 14 TB SATA, 7,200 rpm drive costs $129.95. Reddit has a thread about the company and its products.

    …

    Nimbus Data CEO Tom Isakovich presented a new FlashRack Turbo all-flash system at an A3 Technology Live event in London. It’s a follow-on from the existing ExaFlash array, and composed of a scaled-out, independent sets of redundant 2RU nodes, which use off-the-shelf NVMe TLC SSDs. There is no high-speed cluster interconnect. A node’s base chassis has two stateless x86 controllers and 24 drives, with two 2RU x 24-drive expansion shelves supported. Max capacity is 4.5 PB raw (to 15 PB usable); raw 1.5 PB/chassis, meaning 62.5TB drives. Asked if these were Solidigm QLC or TLC drives, Isakovich said: ”I can’t talk about that.”

    The software supports block, file, and object access with one tier and logical space, with Ethernet (400g), FC, and InfiniBand connectivity supported. The array does about 50 GBps read and 40 GBps writes. Isakovich called it “the promised dream of the ideal storage platform” and claimed 10 PB of Nimbus costs 80 percent less than 10 PB of VAST. He said Nimbus is 20 years old, privately financed, has 30 staff, 200 customers (95 percent in the US), and 700 deployed systems. Nimbus is working on having its software delivered on an adapter card (DPU-esque) and on an intelligent, Ethernet-connected SSD.

    …

    Onna Technologies enables users to gather, analyze, and act on data to help minimize legal risk. It announced its data management platform has achieved Powered by Oracle Cloud Expertise and is now available in the Oracle Cloud Marketplace. Onna’s scalable platform supports multiple workflows across an organization by offering more than 20 pre-built connectors, multiple export options, and an intuitive interface.

    …

    Disk drive supplier Seagate has launched a new ecommerce website in the US. Customers who register will have first access to special pricing, free shipping offers, and other promotions. At launch, Seagate is offering an exclusive discount on its most popular gaming storage drives including the Xbox Storage Expansion Card for Xbox Series X|S 1TB at $129.99 (regularly $219.99), and Game Drive PS5 NVMe SSD for $89.99 (regularly $124.99). The company plans to expand the website to other regions later this year.

    …

    StorPool Storage has extended the capabilities of its Apache CloudStack integration in light of skyrocketing demand from VMware customers. CloudStack is an open source cloud computing software platform designed to deploy and manage large arrays of virtual machines as a highly available (HA), highly scalable IaaS system. CloudStack’s current KVM-hypervisor host HA feature depends on having NFS primary storage detect which nodes are online, even if other types of primary storage are used, such as faster block-level storage. StorPool extends the HA feature by adding a mechanism that overcomes the NFS primary storage heartbeat dependency to extend KVM host high availability for non-NFS storage in CloudStack environments. Future enhancements will support KVM host HA for hyperconverge infrastructure (HCI) deployments with CloudStack compute and StorPool storage running concurrently on the same servers.

    …

    TerraMaster has launched the F4-212 four-bay private cloud NAS for backup and home multimedia center use. The F4-212 is equipped with an ARM V8.2 Cortex-A55 64-bit 1.7GHz quad-core Realtek 1619B processor and independent NPU, built-in floating point unit (FPU) and NEON SIMD engine, video DSP hardware acceleration. Compared with the previous generation RTD1296, the performance is improved by 40 percent. The F4-212, running the latest TOS 5.1, meets diverse business needs, and enhances speed and security. It backs up data from Windows PC, Mac, mobile, HDD enclosure, and network disk for centralized management.

    Micron touts beefy HBM chips, and hot DRAM, Samsung’s releasing fresh stacked memory tech too

    By
    Chris Mellor
    -

    Micron has launched both a fingernail-sized flash storage device for smartphones and a large high bandwidth memory chip for Nvidia GPUs.

    Flash and DRAM foundry operator Micron builds and sells SSDs. The flash chip is an enhanced version of a UFS 4.0 format device, 9 x 13 mm in size, with up to 1 TB in capacity using Micron’s 232-layer 3D NAND technology. It delivers twice the performance of UFS 3.1 specification devices, pumping out up to 4,300 MBps sequential reads and 4,000 MBps sequential writes. Smartphone apps can be launched in less time and run faster, and the device can be used in automotive applications as well.

    Micron’s Mobile Business Unit GM and corporate VP, Mark Montierth, said in a statement: “Micron’s latest UFS 4.0 solution enables world-class storage performance and reduced power in the world’s smallest UFS package. Supercharged with breakthrough firmware advancements to keep smartphones running like new, Micron UFS 4.0 raises the bar for mobile storage with enhanced performance, flexibility and scalability to accelerate the rollout of generative AI-capable smartphones.” 

    The company claims large language models in generative AI apps can be loaded 40 percent faster than Micron’s prior 176-layer UFS 3.1 product, resulting in a smoother experience when initializing conversations with chatbots. This prior product was physically larger, measuring 11 x 13 mm, and needed more electricity. The new one is said to be 25 percent more power-efficient.

    The chip’s firmware has three new features:

    • High-Performance Mode (HPM): This optimizes performance by prioritizing critical tasks over background ones during intensive smartphone use, with an over 25 percent speed improvement when launching multiple apps over the UFS 3.1 product due to twice-as-fast storage access during heavy usage.
    • One Button Refresh (OBR): Automatically cleaning and optimizing data so smartphones can continue operating in a like-new state. 
    • Zoned UFS (ZUFS): The host can specify different zones where data can be stored, improving the usefulness of the device over time. This ZUFS approach reduces write amplification to extend the flash lifespan, keeping devices feeling like new for longer. 

    Micron is now shipping samples of its upgraded UFS 4.0 drive in capacities of 256 GB, 512 GB, and 1 TB. It announced that both HONOR and Samsung are using its LPDDR5X memory and UFS 4.0 mobile flash storage in their latest AI-driven smartphones, the HONOR Magic6 Pro and the Samsung Galaxy S24 series.

    Nvidia HBM3e

    Micron is building 24 GB HBM3e (High Bandwidth Memory gen 3 extended), with 8 x 3 GB stacked DRAM dies, for use by Nvidia in its H200 Tensor Core GPUs, due to ship next quarter. The H200 is a high-end GPU, a refresh of the existing Hopper architecture H100 intended for AI processing work. It should come with up to 141 GB of HBM3e memory and 4.8 TBps of bandwidth, meaning up to six Micron HBM3e chips per H200.

    We’re told that the pin speed of Micron’s HBM3e device is >9.2 Gbps and its overall speed is >1.2TBps. Micron claims it has around 30 percent lower power consumption than unidentified competitive offerings.

    EVP and Micron Chief Business Officer Sumit Sadana said: “Micron is delivering a trifecta with this HBM3e milestone: time-to-market leadership, best-in-class industry performance, and a differentiated power efficiency profile.

    “AI workloads are heavily reliant on memory bandwidth and capacity, and Micron is very well-positioned to support the significant AI growth ahead through our industry-leading HBM3e and HBM4 roadmap, as well as our full portfolio of DRAM and NAND solutions for AI applications.” Getting Nvidia as a customer is a good win for Micron, and steals a march on rivals Samsung and SK hynix. In its last earnings call, back in December, Micron CEO Sanjay Mehrotra said: “We are on track to begin our HBM3e volume production ramp in early calendar 2024 and to generate several hundred millions of dollars of HBM revenue in fiscal 2024. We expect continued HBM revenue growth in 2025.” The Nvidia deal must have been visible to Mehrotra then.

    Micron will be sampling a larger HBM3e product in March, with a 12-high stack of DRAM chips yielding 36 GB in capacity, and >1.2 TBps performance again.

    But Samsung has announced its own 12-high, 36 GB HBM3e device, with up to 1.28 TBps bandwidth. It builds the 12-layer device using a thermal compression non-conductive film (TC NCF) to squeeze it down to the same height as a 8-layer device. This technique mitigates chip die warping that can come with a thinner die. Sammy says it’s lowered the thickness of its NCF material and achieved the industry’s smallest gap between chips at seven micrometers (µm), and eliminating voids between layers. This results in enhanced vertical density by over 20 percent compared to its HBM3 8-layer product.

    Yongcheol Bae, EVP of Memory Product Planning at Samsung Electronics, said: “This new memory solution forms part of our drive toward developing core technologies for high-stack HBM and providing technological leadership for the high-capacity HBM market in the AI era.”

    Expect more Micron HBM3e details at the Nvidia GTC event on March 18.

    IBM adds real-time malware IO detection to flash drives

    By
    Chris Mellor
    -
    Malware detection
    Malware detection

    IBM has added real-time detection of ransomware and other attacks using machine learning to the firmware of its latest FlashCore Modules (FCMs).

    Update. IBM comments added at end. 4 March 2024.

    These FCMs are proprietary flash drives used in IBM FlashSystem 5000 and Storwize arrays. They come in a U.2 form factor with an NVMe interface. The existing gen 3 FCMs come in 4.8, 9.6, 19.2, and 38.4 TB usable capacity levels, with the 19.2 and 38.4 TB FCMs having a PCIe 4 connection. The other capacities use PCIe gen 3. Onboard compression increases effective capacity by up to 2.3x.

    Sam Werner, IBM Storage Product Management VP, blogs: “The FCM4 technology in new FlashSystem arrays is designed to capture and summarize detailed statistics about every I/O in real time. FlashSystem uses machine learning models to distinguish ransomware and malware from normal behavior, enabling organizations to take immediate action and keep operating in the event of an attack.”

    He says: “Existing IBM FlashSystem products scan all incoming data down to block level granularity without impact to performance as it’s being written, using inline data corruption detection software and cloud-based AI to help identify anomalies that might indicate the start of a cyber-attack, thereby enabling the system to detect, respond, and rapidly recover with immutable copies. The new technology enabled by FCM4 is designed to continuously monitor statistics gathered from every single I/O using machine learning models to detect anomalies like ransomware in less than a minute.”

    The gen 4 FCMs interoperate with IBM Storage Defender software, which incorporates Index Engines’ CyberSense code and Cohesity’s DataProtect offering. Storage Defender uses AI with event monitoring across multiple storage platforms to help detect ransomware, human error, and sabotage.

    Werner says: “FCM works with Storage Defender to provide end-to-end data resilience across primary and secondary workloads with AI-powered sensors that provide earlier notification of cyber threats so enterprises can recover faster.”

    Storage Defender has expanded its threat detection capabilities with the AI-powered FCM hardware “and software sensors that inform an industry leading index of the relative trustworthiness of copies, whether backup or primary snapshots. Additionally, IBM Storage Defender includes sensors developed by IBM Research that are engineered to detect potential threats, such as ransomware, in near real time and raise high fidelity alerts to security tools.”

    IBM has also added “workload and storage inventory management capabilities to IBM Storage Defender to help organizations assess their applications and data so they can properly incorporate all their assets in a business continuity plan to recover a minimum viable company after a cyberattack.”

    In Werner’s view: “Threat actors are now deploying AI-based cyber-attacks, and we must fight fire with fire. Our new FlashCore Module hardware and Storage Defender software both leverage IBM’s AI capabilities to help them better address this challenge.”

    Were standard SSD suppliers such as Micron, Samsung, SK hynix, and Kioxia/Western Digital to add similar AI-scanning of SSD IOs in real time, they would need to send attack alerts upstream to some system management resource to respond to the alerts. IBM’s Storage Defender is that resource for its customers. The commercial-off-the-shelf SSD suppliers don’t have similar functionality unless they partner with an upstream software vendor.

    Bootnote

    Barry Whyte, Principal Storage Technical Specialist and IBM Master Inventor, told us: “The FCM4 are PCIe Gen 4 across all capacities now. They are also now using “Charge-Trap” NAND technology which allows us to use the 176-layer NAND which gives faster programming times, best economics – but of course needs the intelligence in the computation storage that is FCM – no other vendor has anything close!  Also, the FCM are across all NVMe based FlashSystem (we dropped the Storwize name some time ago) so 5200,7300,9500 (the other smaller 5000, 5015, 5045 are still SAS based and can’t use FCMs)”

    1...102103104...459Page 103 of 459

    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