Accounting investigations at Quantum, assailed by financial reporting problems and Nasdaq delisting woes, discovered more revenues and extra profits for the affected 2022 and 2023 fiscal years. it was revealed yesterday when the company issued an update on accounting matters.
Under the financial management of a new CFO, Ken Gianella, Quantum discovered problems in reporting Q2 of its fiscal 2024’s financial results concerning the reconciliation of standalone selling prices for components it sold in product bundles. These affected the reporting of its Q3 and Q4 results as well. It started an accounting investigation and had to ask for a reporting extension from the Nasdaq stock exchange, where its shares are listed.
Separately Quantum’s stock price fell below an average $1.00 value required by Nasdaq and it faces a delisting threat because of that.
Now the accounting review has been completed, and company states it will “release financial results for its full year fiscal 2024 ended March 31, 2024 on Monday, June 17, 2024 after markets close.”
The review, supported by outside experts, has found that reported revenue and profit numbers for the first three quarters of fiscal years 2022 and 2023 will be increased, as will those for the first fiscal 2024 quarter.
Quantum states that the adjustment “does not impact the Company’s invoicing, cash, or contractual obligations to its customers.” Also, during the review “the Company found no evidence of fraud or intentional misconduct associated with its revenue recognition process.”
There is more good news, for investors at least. During the investigation Quantum identified a series of outstanding warrant agreements issued to its prior and current lenders in 2018, 2020, and 2023. It needs to assess the impact of revised liabilities for these warrants in fy 2022 and 2023 and Q1 fy2024. Quantum said: “The impact from the revised liability accounting treatment for outstanding warrants is estimated to increase net income in all restated periods.” More profits in other words.
Quantum also announced a revised agreement with its lenders to amend the company’s existing term loan and revolving credit agreements.
All in all this is a good result. The stock price rose 5.1 percent to $0.45 at the end of trading on May 29 as a result. We think chairman and CEO Jamie Lerner and CFO Gianella will address the Nasdaq delisting issue on June 17.
Hitachi Vantara has launched a VSP One Block scale-out storage appliance for mid-sized biz.
Update. VSP One Block data multi-access protocol details, spec sheet, data sheet, images and installation video references added, 1 June 2024.
Hitachi V rebranded its storage portfolio of products to Virtual Storage Platform One in October last year, taking an HPE Alletra-like tack to overall brand conformity. Its portfolio included the VSP (Virtual Storage Platform) high-end (VSP 5000) and mid-range (E Series) block arrays, HNAS file storage, VSS (Virtual Software-defined Storage), and the HCP (Hitachi Content Platform) for object data. Specific VSP One products were launched earlier this year, comprising three products: SDS (Software-Defined Storage) Block, SDS Cloud and File, the old HNAS. The all-NVMe flash SDS Block supports up to 1.6PBe (effective capacity) in its 2RU chassis. SDS Cloud comes as an appliance, a VM or public cloud offering and will be available in the AWS marketplace. Ops Center Clear Sight provides cloud-based monitoring and management for these products.
VSP One Block is different from SDS Block in that, starting as a single all-flash appliance, it scales out to a 65-node cluster. The nodes operate Hitachi Storage Virtualization Operating System (SVOS) software, which can manage virtualized 3rd party arrays. SVOS supports block, file (NFS, CIFS/SMB) and S3 access protocols. A spokesperson told us: “You can run Block natively, then add File and HCP for Object. … All Block models are running SVOS, you layer Block with File together in a 5U solution. File is doing pass through to block so we now offer 4:1 data reduction no questions asked (no T&Cs nothing to sign up for, no exception! You don’t get 4:1 we make you whole on block and file) and 100 percent data availability. We also have one user GUI for consumption of block and file and cloud observability with analytics and sustainability on Clear Sight.”
Octavian Tanase
Hitachi V’s Chief Product Officer, Octavian Tanase, newly recruited from NetApp, said in a prepared remark: “Our Virtual Storage Platform One Block appliance is powerful and dense, delivering the data processing and reliability that mid-sized businesses need while minimizing rack space and reducing power and cooling costs for a more sustainable datacenter environment.”
It “sets a new standard for storage performance.”
The VSP One Block is likely positioned to replace the existing VSP E Series. Hitachi V claims it has “breakthroughs in simplicity, security, and sustainability.” There are “three dedicated models providing businesses with a common data plane across structured and unstructured data in block storage, specifically designed to remove complexity, enhance data protection, and reduce carbon emissions.” The three variants are the VSP One Block 24, 26 and 28 – see the table below. These are claimed to optimize rack space while reducing power consumption and cooling costs, but without saying what these are being compared to.
Hitachi V positions VSP One Block as being suitable for AI needs, saying the rise of AI and connected technologies has led to an exponential surge in data volumes, as businesses expect the amount of data they use to double between 2023 and 2025. As a result, businesses, especially mid-sized organizations, are being forced to rethink how to build and scale their data architectures. Enter VSP One Block.
The products are designed to be self-installable and feature:
Per-appliance 32TB effective capacity to maximum of 1.8 PB of effective capacity using the internal 24 drive slot in 2RU chassis.
Hitachi Thin Image Advanced (TIA) snapshot software creates copies for decision support and software development, and “defends structured data against ransomware. Every volume can have up to 1,024 Safe Snaps taken, and the array supports up to 1 million Safe Snaps.
Always-available production data copies for data protection, with TIA saving up to 90 percent of disk space by only storing changed data blocks.
Pre-configured, including the creation of Dynamic Drive Protection (DDP) groups which replace traditional RAID groups, providing the resilience of RAID6 with distributed spare space and support for an arbitrary number of drives (from 9-32 per group). It supports adding drives one (or more) at a time. DDP dramatically lowers rebuild times.
Dynamic Carbon Reduction technology reduces energy consumption by switching CPUs into eco-mode during periods of low activity.
“Always on compression” allows the system to switch from inline data reduction to post-processing which reduces energy consumption and contributes to a lower CO2 footprint by as much as 30-40 percent.
New, patented compression accelerator modules (CAM) with a new compression algorithm.
4:1 No Questions Asked data reduction guarantee.
Management tools, including an embedded graphical user interface (GUI) and the intuitive SaaS-based Ops Center Clear Sight portal make it easy to manage and consume the storage.
100 percent data availability guarantee.
FIPS 140-3 lvl 1 data at rest encryption automatically enabled from the distribution center for most customers
Supports running cloud-native apps alongside traditional block workloads.
Hitachi V has released datasheets, and specsheets for the VSP One Block products, and there is an installation video here. Hint; it takes c30 minutes.
Hitachi, Hitachi Vantara and Google
Hitachi Vantara’s parent company, Hitachi, has just announced a multi-year partnership Google focused on generative AI. It will form the Hitachi Google Cloud business unit focused on offering business customers Gemini models, Vertex AI, and other cloud technologies, and it will also adopt Google Cloud’s AI to enhance its own products and services.
Hitachi’s new products will support customers running both on-premises and in the cloud, enabling them to modernize operations while retaining existing IT environments. They will also be compatible with VSP One, so that users can build GenAI applications using data stored on Hitachi Vantara’s hybrid cloud storage products.
Analysis: Customers are sometimes faced with costly and sometimes unexpected permanent withdrawal fees when ending digital tape cartridge vaulting contracts. What lessons should be learned from this and how should organizations deal with multi-decade information retention?
For companies wanting to store physical media for long periods, Iron Mountain, for example, has long provided physical storage of boxed paper records in its vaults, initially inside the deep excavated caverns of an old iron ore mine, and subsequently in storage vault buildings around the globe.
Organizations with overflowing filed paper archives can send them offsite to Iron Mountain facilities for long-term storage, paying a yearly fee for the privilege. Extending this to storage of digital tape cartridges was a natural step for Iron Mountain.
But what happens if you want to end such a contract and remove your tapes? Well, there is a withdrawal fee. One customer claimed: “They are the Hotel California of offsite storage. You can check out your tapes any time you want – but you can never leave.”
“If you permanently take a tape out of Iron Mountain, they charge you a fee that is roughly five years of whatever revenue that tape would have made … So people just leave tapes there that they know have no purpose, because they can’t afford to take them out,” the source alleged.
It happens that Iron Mountain charges a similar fee for permanent withdrawal if customers want to remove boxed paper files and store them in a cheaper facility or simply destroy them. This is stated in Iron Mountain’s customer contract’s Ts&Cs but can still be a surprise to some customers as it’s charged on top of a simple retrieval fee.
Permanent withdrawal section of Iron Mountain boxed paper vaulting contract with withdrawal fee surcharge paragraph highlighted
A court document states: ”On August 12, 2004, Berens and Tate notified Iron Mountain that it wanted to remove all of its records from Iron Mountain and transfer those records to another storage company. Iron Mountain informed Berens and Tate that pursuant to the operative ‘Schedule A,’ Berens and Tate would be obligated to pay the ‘Permanent Withdrawal’ fee of $3.70 per cubic foot and a retrieval fee of $2.10 per cubic foot. According to Berens and Tate, the total charge to permanently remove its records would have been approximately $10,000.”
Iron Mountain told the court that the permanent withdrawal fee was necessary in order to compensate for the additional labor and services that are provided when large amounts of records are permanently removed. Because Berens and Tate had freely entered into the storage contract with Iron Mountain, the court decided the permanent withdrawal fee “was the parties’ agreed-upon compensation for services to be performed – specifically, the permanent removal of records.” Berens and Tate had to pay the fee.
Based on what we have been told, the vendor may be attaching a digital tape cartridge permanent removal fee along the same lines as its boxed paper record permanent removal fee. We sent an email to Iron Mountain on May 17 to check this but have heard nothing back despite sending follow-up requests for comment. If we hear back from Iron Mountain, we’ll add its reply to this story.
Our customer contact’s final thought was: “I’m sure it’s legal because it’s in the contract. But it’s wrong.”
It seems clear that prospective tape cartridge vaulting customers need to pay careful attention to all T&Cs and, if a permanent withdrawal fee is charged on top of a retrieval fee when ending the contract, be aware of this and be willing to pay it. The devil is often in the detail.
There are two questions that occur to us about this situation. One is how to deal with such contracts and the second, much bigger one, is how to obtain cost-effective and multi-decade information storage. Should that be on-site or off-site, and in what format?
Neuralytix
Ben Woo
Neuralytix founding analyst Ben Woo said: “Of course any legal contract should be reviewed by a qualified attorney-at-law. But, to abandon long term tape storage providers on the basis of high retrieval costs would be a poor decision. Long term archive providers store backup tapes. Backup tapes by its very nature contain data that a customer hopes [it] will never see again, but is stored there for a small number of reasons – incremental forever backups, regulatory reasons, or (at least for the more recent tapes) disaster recovery.
“If a tape is required to be retrieved, the need must be extraordinary. VTL and technologies like these allow the customer to hopefully never rely on the tapes stored in storage.
“Whether the tapes need to be retrieved or whether a customer desires to combine older formats into newer ones require a lot of human effort, both of which carry costs.
“The only real alternative to tape is long term storage in the cloud – the problem with this is that customers will be paying for the storage of data irrespective of whether they use it or not for 10, 20, etc. years which quickly becomes cost-prohibitive and fiscally/economically irresponsible.”
Architecting IT
Chris Evans.
Architecting IT principal analyst Chris Evans said: “To summarise the problem, businesses use companies like Iron Mountain to create data landfill, rather than sending their content to a recycling centre. The Iron Mountain contract is constructed in such a way to make it more expensive to resolve the problem of the data landfill, than to simply add to it.”
“This is a massive topic. In my experience, tape has been used as a dumping ground for data that is typically backup media, where the assumption is the business ‘might’ need the data one day. In reality, 99 percent of the information held is probably inaccessible without specialist help.
“Setting aside the obvious concept of fully reading and understanding a contract before signing it, the long-term retention of data really comes down to one thing – future value. Most companies assume one or more of the following:
We may need the data again in the future. E.g. legal claims, historic data restores.
We might be able to derive value from the data.
We have no idea what’s on our old media, so we better keep it “just in case”. (Most likely and the root cause of most problems)
“Storage costs decline year on year, for disk and tape. Generally, on a pure cost basis, it’s cheaper just to buy more storage and keep the data than to process it and determine whether it has value. So, businesses, which are an extension of humans and human behaviour, generally push the management process aside for ‘another day’, kicking the can down the road for someone else to solve in the future.
“The obvious technical answer to the problem is consolidation. … LTO-8 and LTO-9 are only 1 generation backward compatible. So, to recycle old media you will need old tape drives of no worse than 2 generations behind.
“[A] second problem is data formats. If you used a traditional data protection solution, then the data is in that format. Netbackup, for example, used the tar file format, storing data in fragments (blocks) of typically 2GB. Theoretically, you can read a tape and understand the format, but adding in encryption and compression can make this process impossible.
Specialists that can help make their money by having a suite of old technology (LTO drives, IBM 3490/3480, Jaguar etc) and tools that can unpick the data from the tape. However, none of these solutions restack. The restacking process would require updating the original metadata database that stores information on all backups, which is probably long gone. Even if it still exists, the software and O/S to run it will be hard to maintain and only adds to further costs. So stacking of old legacy data is pretty much impossible.
Data Retention Policy
Evans added: “The best way to look at the data landfill issue is with a business perspective. This means creating a data retention policy and model that helps understand ongoing costs. Part of the process is also to create a long-term database that keeps track of data formats, backup formats etc.
Imagine, for example, you created a data application in 2010, which stored customer data. In 2015 you migrated the live records to a new system and archived the old one. It’s 2024. Who within your organization can remember the name of the old application? Who can remember the server names (which could be tricky if they were virtual machines)? Who can remember the data formats, the database structures? Was any of that data retained? If not, the archive/backup images are effectively useless as no one can interpret them.
“So, businesses need to have data librarians whose job is to log, track, audit and index data over the long term.
“My strategy would be as follows:
Fix Forward – get your house in order, create data librarians and data retention policies if they’re not already in place. Work with the technical folks to actively expire data when it reaches end of life. Work with the technical folks to build a pro-active refresh policy that retires old media and moves data to newer media. If you change data protection platform, create a plan to sunset legacy systems, running them down over time.
For example, if your oldest data is 10 years, then you’ll be keeping some type of system to access that data for at least 10 years. Look at storing data in more flexible platforms. For example, using cloud (S3) as the long-term archive for backup/archive data makes that content easier to manage than tape. It also makes costs more transparent – you pay for what you keep until you don’t keep it any longer.
Remediate Backward – Create a process to remediate legacy data/media. Agree a budget and timescale (for example, to eliminate all legacy content in 10 years). Triage the media. Identify data that can be discarded immediately, the data that must be retained and the data that can’t be easily identified. Create a plan to wind each of these down.
“…The exact strategy is determined by the details of the contract. For example, say the current contract is slot/item based and is $1/item per year. If you have 10,000 tapes, then reducing that by 1,000 items per year should eliminate the archive in 10 years. If the penalty is determined by the last year’s bill, then (theoretically) the final bill might be 5x times $1,000 = $5,000 rather than $1,000, but significantly better than the $10,000/year being paid in year 1. This is speculation of course, because it depends on the specifics of the contract. I’d have a specialist read it and provide some remediation strategy, based on minimizing ongoing costs.
“Most businesses don’t want to solve the legacy data sprawl problem because it represents spending money for no perceived value. This is why Fix Forward is so important, as it establishes data management as part of normal IT business costs.
“The opposite side of the coin to costs is risk. Are your tapes encrypted and if not, what would happen if they were stolen/lost? What happens if you’re the subject of a legal discovery process? You may be forced to prove that no data exists on old media. That could be very costly. Or, if you can’t produce data that was known to be on media, then the regulatory fine could be significant.
“So, the justification for solving the data landfill can be mitigated by looking at the risk profile,” Evans told B&F.
Lenovo’s latest quarterly revenues rose as AI opportunities drive growth and profitability.
Revenue of $13.83 billion was generated in the fourth fiscal 2024 quarter ending March 31, up 9.5 percent year-on-year, with profits up 117.5 percent to $248 million. The Hong Kong-based firm said full fy2024 revenues were $59.6 billion, down 8.1 percent and net profit came in at $1 billion, 37.8 percent less than the prior financial year.
Chairman and CEO Yuanqing Yang’s prepared quote said: “Lenovo’s fourth quarter results clearly demonstrate that we have not only resumed growth across all our businesses but that our business momentum is accelerating, driven by the unprecedented opportunities brought by Hybrid AI. … Supported by our strong execution, persistent innovation, operational excellence, and ecosystem partnerships, we are confident we can deliver sustainable growth and profitability improvement in the coming year.”
Lenovo said that, from the second half of the fiscal year, it achieved year-on-year revenue growth of 6 percent and net margin recovered from a first half year-on-year decline to be flat in the second half.
The company has three business units, Solutions & Services Group (SSG), Infrastructure Solutions Group (ISG – servers and storage) and Intelligent Devices Group (IDG – PCs and smartphones), with all three showing year-on-year revenue growth.
The IDG business unit has a long way to climb to regain its fy2022 Q3 glory.
IDG, the largest business unit (BU), reported $10.5 billion in revenues, up 7.2 percent on the year. ISG reported $2.5 billion in net income, 13.6 percent higher, while SSG brought in $1.8 billion, an increase of 9.1 percent. IDG is so large that the other BUs can’t do much to swing Lenovo’s profitability dial one way or the other.
Still Lenovo was complimentary about all its BUs. It said the results “strengthened SSG’s position as a growth engine and profit contributor” by delivering its year-on-year revenue growth “and high profitability with an operating margin exceeding 21 percent.”
ISG “resumed growth” with a record fourth quarter but it made an operating loss in all four quarters of the year. Lenovo’s “storage, software and services businesses all achieved hypergrowth, with the combined revenue increasing more than 50 percent year-on-year. High Performance Computing revenue hit a record high.” The growth percentage was actually 52 percent. Lenovo resells a lot of NetApp storage so that Sunnyvale business will be pleased.
Lwnovo’s ISG financial summary slide
IDG had a solid quarter, Lenovo said: “strengthening its global market leadership for PCs with a market share of 22.9 percent.” Lenovo’s PCs, tablets, and smartphones all resumed growth in the second half of its fy2024.
Yuanqing said: ”We’ve built a full stack of AI capabilities and are at the forefront of pioneering the revolutionary AI PC market.” Lenovo expects the AI PC to grow from its current premium position to mainstream over the next three years, driving a new refresh cycle for the industry, and bolstering its revenues substantially.
Lenovo hopes to accelerate growth and have sustainable profitability increases across its entire business in fy2025. The driver will be hybrid AI with every product sector getting an infusion of AI.
Commissioned: Generative AI adoption within organizations is probably much higher than many realize when you account for the tools employees are using in secret to boost productivity. Such shadow AI is a growing burden IT departments must shoulder, as employees embrace these digital content creators.
Seventy-eight percent of employees are “bringing their own AI technologies” (BYOAI) to work, according to a joint Microsoft and LinkedIn survey. While the study acknowledges that such BYOAI puts corporate data at risk it downplays the sweeping perils organizations face to their data security.
Whether you call it BYOAI or shadow AI this phenomenon is potentially far worse than the unsanctioned use of cloud and mobile applications that pre-dated it.
As an IT leader, you’ll recall the bring-your-own-device (BYOD) trend that marked the early days of the consumer smartphone 15 years ago.
You may have even watched in horror as employees ditched their beloved corporate Blackberries for iPhones and Android smartphones. The proliferation of unsanctioned applications downloaded from application stores exacerbated the risks.
The reality is that consumers often move faster than organizations. But consumers who insist on using their preferred devices and software ignore integrating with enterprise services and don’t concern themselves with risk or compliance needs.
As risky as shadow IT was, shadow AI has the potential to be far worse – a decentralized Wild West or free-for-all of tool consumption. And while you can hope that employees have the common sense not to drop strategy documents into public GPTs such as OpenAI, even something innocuous like meeting transcriptions can have serious consequences for the business.
Of course, as an IT leader you know you can’t sit on the sidelines while employees prompt any GenAI service they prefer. If ignored, Shadow AI courts potentially catastrophic consequences for organizations from IP leakage to tipping off competitors to critical strategy.
Despite the risks, most organizations aren’t moving fast enough to put guardrails in place that ensure safe use, as 69% companies surveyed by KPMG were in the initial stages of or had not begun evaluating GenAI risks and risk mitigation strategies.
Deploy AI safely and at scale
Fortunately, organizations have at their disposal a playbook to implement AI at scale in a way that helps bolster employees’ skills while respecting the necessary governance and guardrails to protect corporate IP. Here’s what IT leaders should do:
Institute governance policies: Establish guidelines addressing AI usage within the organization. Define what constitutes approved AI systems, vet those applications and clearly communicate the potential consequences of using unapproved AI in a questionable way.
Educate and train: Giving employees approved AI applications that can help them perform their jobs reduces the incentive for employees to use unauthorized tools. You must also educate them on the risks associated with inputting sensitive content, as well as what falls in that category. If you do decide to allow employees to try unauthorized tools, or BYOAI, provide the right guardrails to ensure safe use.
Provide use cases and personas: Education includes offering employees use cases that could help their roles, supported by user “personas” or role-based adoption paths to foster fair use.
Audit and monitor use: Regular audits and compliance monitoring mechanisms, including software that sniffs out anomalous network activity, can help you detect unauthorized AI systems or applications.
Encourage transparency and reporting: Create a culture where employees feel comfortable reporting the use of unauthorized AI tools or systems. This will help facilitate rapid response and remediation to minimize the fallout of use or escalation of incidents.
Communicate constantly: GenAI tools are evolving rapidly so you’ll need to regularly refresh your AI policies and guidelines and communicate changes to employees. The good news? Most employees are receptive to guidance and are eager to do the right thing.
Solutions to help steer you
GenAI models and services are evolving daily, but there are some constants that remain as true as ever.
To deploy AI at scale, you must account for everything from choosing the right infrastructure to picking the right GenAI models for your business to security and governance risks.
Your AI strategy will be pivotal to your business transformation so you should weigh whether to assume control of GenAI deployments or let employees choose their own adventures, knowing the consequences of the latter path.
And if you do allow for latitude with BYOAI, shadow AI or whatever you choose to call it, do you have the safeguards in place to protect the business?
Trusted partners can help steer you through the learning curves. Dell Technologies offers a portfolio of AI-ready solutions and professional services to guide you along every step of your GenAI journey.
Data intelligence supplier Alation has achieved Federal Risk and Authorization Management Program (FedRAMP) “In Process” status and gained a listing on FedRAMP Marketplace at a Moderate impact level. Alation’s partnership with Constellation GovCloud and Merlin Cyber allows government agencies to search for and discover secure, FedRAMP-compliant data. Constellation GovCloud de-risks the FedRAMP authorization process for Alation by handling most compliance tasks and reducing costs.
…
Cloud backup and general S3 storage vault provider Backblaze has won SaaS-based legal biz Centerbase as a customer. Back in Dec 2023 Veeam using Centerbase chose Object First’s Ootbi appliance as its on-prem backup store for its Veeam backups. It has also decided to use Backblaze’s B2 service as a cloud-based offsite disaster recovery facility. The scheme has Veeam Backup and Replication tiering infrastructure to Backblaze B2 as well as sending it to the Ootbi appliance. If Centerbase’s primary site is affected by ransomware or natural disaster, it can turn to its B2 backups to restore data and meet recovery time objective (RTO) requirements.
…
Supercomputer, HPC and enterprise/AI storage supplier DDN has won Qatar’s telecommunications operator and information and communications technology (ICT) provider Ooredoo as a customer. It will adopt DDN’s AI infrastructure across its networks.
…
HPE Storage bloke Dimitris Krekoukias, blogging as Recovery Monkey, has posted a blog about resiliency aspects of HPE’s Alletra MP scale-out block storage system’s DASE (disaggregated, shared everything) architecture – like VAST Data. Alletra MP has no concept of dual controllers. He writes: “all the write cache resiliency has been moved out of the controllers. It goes hand-in-hand with not having a concept of HA pairs of controllers. Ergo, losing a controller doesn’t reduce write cache integrity. Which is unlike most other storage systems.” Also: “all controllers see all disks and shelves, all the time.” Users can add new controller nodes (compute) without having to add capacity (storage). The system rebalances itself.
If a controller goes down, the system rebalances itself. Extra capacity can be added at will. If a component disk shelf is lost, the system recovers. Controller resilience is high – (N/2)-1 nodes can be lost at the same time for Alletra MP Block. So in a six-node cluster, any two nodes can be lost simultaneously. “Conceptually, the architecture allows arbitrary node counts – so if in the future we do, say, eight-node clusters, (8/2)-1=3 so any three nodes could be truly simultaneously lost without issues, and so on and so forth, as cluster size increases.”
He notes: “We now allow N/2 rolling failures in R4, and later may potentially allow more.”
…
Seagate’s HAMR drive qualification at customers has been hindered, Wedbush analyst Matt Bryson suggests, by mechanical components being stored in a non-climate-controlled environment. This resulted in a material reaction to the environment and, when the components were used in the HAMR process, media contamination followed. Bryson argues this is basically a logistics problem and relatively easily fixed – much more easily than fixing a physics or chemistry problem. The delayed HAMR take-up will benefit WD, with its 24/28 TB PMR/SMR HDDs. He thinks that, overall, the HAMR situation will improve for Seagate from now on, with the next HAMTR iteration due in H2 2025 with a 40TB-class product launch providing a Seagate revenue boost.
…
Storage Newsletter reports Trendfocus SSD shipment stats for calendar Q1, 2024, showed capacity shipped rose 6.5 percent quarter on quarter to 90.87EB with units dropping 5.1 percent to 83.79 million. But these low-level changes hid dramatic market sector differences.
Client units decreased 9.5 percent to 65.65 million and EB shipped went down 11.5 percent to 43.67EB;
SAS SSD units dropped 19.9 percent to 751,000 and EB shipped went down 5.6 percent to 3.03EB;
Enterprise SATA units rose 2.1 percent to 3.64 million, with 5.12EB shipped, up 3.7 percent;
Enterprise PCIe units shipped soared 50 percent to 8.08 million, and EB shipped went up 45.5 percent to 33.71EB.
Supplier capacity market shares:
Samsung – 36.8 percent
WD – 15.1 percent
Kioxia – 7.8 percent
Combined WDC + Kioxia – 22.9 percent
Solidigm – 14.8 percent
SK hynix – 6.9 percent
Combined SK hynix & Solidigm – 21.7 percent
Micron – 8.2 percent
Kingston – 4.5 percent
SSSTC – 0.6 percent
Others – 5.3 percent
Units shipped supplier market shares:
Samsung – 31.3 percent
WD – 18.5 percent
Kioxia – 9.3 percent
Combined WDC + Kioxia – 27.8 percent
Solidigm – 5.4 percent
SK hynix – 10.3 percent
Combined SK hynix & Solidigm – 25.7 percent
Micron – 9.8 percent
Kingston – 6.7 percent
SSSTC – 1.6 percent
Others – 7.2 percent
SK hynix is benefitting greatly from Solidigm’s high-capacity SSDs.
…
Vector database supplier Qdrant claims its dedicated vector database is faster than other databases with vector extensions added in, and also other dedicated Vector databases such as Pinecone. It claims Qdrant enhances both speed and accuracy by ingesting additional context, enabling LLMs to deliver quicker and more precise results. Qdrant delivers higher retrieval capabilities in Requests Per Second (RPS). Qdrant is designed for high-capacity workloads, making it ideal for large-scale deployments like in the global medical industry, where patient data is continuously updated. For smaller applications, such as a startup’s chatbot, the performance and accuracy benefits may be less noticeable between Qdrant and other products. For extensive datasets, Qdrant’s optimization and scalability offer significant advantages.
Qdrant benchmark chart. Rps = requests per second.
Qdrant benchmarked its performance against named vector databases – Elastic Search, Milvus, Redis, and Weaviate, but not Pinecone – using various configurations of them on different datasets. You can reference the parameters and results, summarized in tables and charts here.
…
ReRAM developer Weebit Nano has partnered with Efabless to allow quick and easy access to prototyping of new designs using Weebit ReRAM. Key points:
This will enable a broad range of designers (startups, government agencies, product companies, research centers and academia) to prototype their unique designs with limited quantities, before they decide to proceed to full production.
The manufacturing will be done at SkyWater, where Weebit is already proven and qualified up to 125 degrees C.
Efabless has thousands of users and customers and it has a unique value proposition: allowing anyone to design their own SoCs for a fraction of the price of a full-mask production, for limited quantities only.
…
Veeam announced new technical training and certification programs through Veeam University, which delivers Veeam technical training to IT professionals on-demand anytime, anywhere. The online offering is the result of a global partnership with Tsunati, a Veeam Accredited Service Partner, revolutionizing on-demand technical certification training for partners and customers worldwide. Veeam University offers maximum flexibility and an immersive, engaging learning experience in a self-paced format. This innovative approach includes clickable labs that can be accessed 24x7x365, video-based demos, and technical deep dives allowing students to effectively absorb concepts and prepare for real-world cyber security and disaster recovery scenarios. Completion of on-demand courses offered through Veeam University qualifies learners for Veeam certification exams – including Veeam Certified Engineer (VMCE).
Azure Native Qumulo now integrates with Microsoft Copilot and Graph connectors. The aim is to access and analyze unstructured data, delivering detailed insights based on natural language prompts.
Sean Gwaltney, who looks after Azure Development & Strategic Engagements at Qumulo, made a LinkedIn post last month saying: “Quietly, without making big promises the Azure Native Qumulo team put their heads down and built a deceptively simple solution to connect your unstructured data to MicrosoftCopilot in a functional manner.”
What scaleout file storage supplier Qumulo has done is to use Microsoft Graph and its customized connector facility to make files stored in its Azure instantiation, Azure Native Qumulo (ANQ), available to Copilot. It can then respond to users’ conversational inputs by referring to ANQ-stored information. This is retrieval-augmented generation (RAG) in action. This is accomplished by having a Qumulo-provided connector feeding the files to Graph and having its semantic index facility vectorize them.
A Microsoft web doc says: “A vector is a numerical representation of a word, image pixel, or other data point. The vector is arranged or mapped with close numbers placed in proximity to one another to represent similarity. Unlike a standard keyword index, vectors are stored in multi-dimensional spaces where semantically similar data points are clustered together in the vector space, enabling Microsoft 365 to handle a broader set of search queries beyond “exact match.”
A Qumulo video talks about this. It demonstrates a financial services-oriented environment with millions of invoices, using the Microsoft Copilot interface to query for details on specific invoices and customers. Customised connectors can enable access and insights for multiple file types stored in the Azure Native Qumulo environment.
The video commentary says: ”if you’re a financial services institution you probably have digital records everywhere on-prem, in the cloud and in archives, and in every type of format; databases, PDF files, office documents, even text files.
“We’re talking literally millions of files, maybe even billions, with different formats and different layouts in all kinds of different places.
“You want to be able to search and analyze all those files but most legacy tools only work with structured data, and AI engines like ChatGPT can pose both security and competitive leakage risks.”
Once the custom Qumulo “connector has been activated, millions of PDF files can be read via Copilot’s Ai and used to answer business queries from within Microsoft 365. … Each connector can be fully customized so you can import and use whatever file data you need, whether you’re consolidating records after an acquisition, searching old data for legal discovery, or itemizing and tracking archived records. … you can also use custom connectors to import data from different file types or from different locations.”
There is no leakage of proprietary information outside the customer’s Azure environment, we’re told, and you don’t need to be a SQL wizard or wannabee data scientist to get answers to quite complex questions.
Comment
This work by Qumulo supports Nasuni’s contention that Microsoft CoPilot has already won the race to provide GenAI chatbot facilities inside Microsoft’s Azure Windows and 365 environments. It seems inevitable that all file storage suppliers for Windows and Microsoft 365 will do the same, aiding their customer’s RAG efforts. Will on-premises Windows server systems do the same thing? We don’t see that happening as CoPilot is an Azure cloud-only feature. This could facilitate a renewal of interest in having unstructured data migrated to Azure.
As chatbots become available on premises (see Nutanix’ GPT-in-a-Box AI initiatives), then making a file – and object – storage supplier’s customer-stored data available to the chatbot for RAG will become necessary.
SSD and memory module supplier Netlist has won a memory module patent infringement legal case against Micron with the jury awarding it $445 million in damages. Micron is appealing the verdict.
Update. Micron response added 29 May 2024.
The case, 2:22-cv-00294, was heard in the district court for the Marshall Division in the Eastern district of Texas. Two patents were involved, the ‘417, where the jury awarded Netlist $20 million, and the ‘912 patent with a $425 million damages award. The jury further found that Micron wilfully infringed the ‘912 and ‘417 patents. Such wilful infringement means the court may increase the damages by up to three times, to $1.34 billion.
We wrote in August 2022: “the ‘912 patent … refers to the use of rank multiplication in an LRDIMM (Long Range DIMM) memory module. Such DIMMs can have four ranks or blocks of memory, and the patent describes IP to present the LRDIMM logically as only having two ranks, thus getting over system memory controller limits on the maximum rank count.” The ’417 patent is entitled “Memory Module With Data Buffering” in a DIMM and applies to LRDIMMs as well.
Netlist has steadfastly defended its patents by variously suing Diablo Technology, Google, Micron, Samsung and SK hynix for patent infringement and breaking licensing or development deal conditions.
It sued Google for infringing the ‘912 patent in 2009 and is also involved in a similar legal dispute with Samsung. The Google case was suspended (stayed) until the related Netlist-Samsung case involving five Netlist patents was decided. This was because Samsung supplied the alleged infringing modules to Google.
A jury reached a verdict against Samsung in the Federal District Court for the Central District of California, finding that Samsung materially breached a Joint Development and License Agreement with Netlist, which was signed by both companies in November 2015.
Reuters reported that another jury awarded Netlist $303 million in damages against Samsung in April last year, again in the Marshall court, in case 2:21-cv-00463.
Netlist sued SK hynix for similar memory chip patent infringements and won $40 million in damages and a cross-licensing and supply arrangement deal in 2021. That dispute commenced in 2016 and took 5 years to come to a conclusion.
In 2017 Netlist sued Diablo for infringing its memory module patents but Diablo went bust later that year.
It seems likely that Micron may now negotiate a patent licensing deal with Netlist, as might Samsung.
Micron replied to an inquiry about this with a statement: “Micron respectfully disagrees with the jury’s verdict. The vast majority of damages awarded by the jury were based on Netlist’s ’912 patent, which was already declared invalid by the U.S. Patent Trial and Appeal Board of the United States Patent and Trademark Office over a month before this trial (IPR2022-00615, Paper 96). The jury’s verdict does not affect the prior ruling that Netlist’s patent is invalid. Additionally, Micron’s evidence presented during the trial demonstrates that Micron does not infringe either of the patents asserted by Netlist.
“Netlist’s ’417 patent, which reflects less than 5 percent of the damages, is also under review by the patent office after a determination that there is a reasonable likelihood it is also invalid (IPR2023-01141, Paper 7). Micron will appeal the jury’s verdict and will defend the U.S. Patent Office ruling that the ’912 patent is invalid.”
Netlist investors will be happy but hold their celebrations until the outcome of Micron’s appeal.
Analysis: Dell has said it will add a parallel file system capability its scaleout clustered PowerScale storage arrays. How might this be done?
At its recent Dell Technologies World 2024 event in Las Vegas, Dell’s Varun Chhabra, SVP for ISG Marketing, told his audience: “We’re excited to announce Project Lightning which will deliver a parallel file system for unstructured data in PowerScale. Project Lightning will bring extreme performance and unparalleled efficiency with near line rate efficiency – 97 percent network utilisation and the ability to saturate 1,000s of data hungry GPUs.”
Update: VAST Data nconnect multi-path note added. 28 May 2024.
PowerScale hardware, the set of up to 256 clustered nodes, is operated by the OneFS sequential file system software. When files are read or written the activity is carried out on the nodes on which the files are stored. With a parallel file system, large files or sets of small files have to be striped across the nodes with each node answering a read or write request in parallel with the others to accelerate the IO.
Dell has several options for adding parallel file system functionality to PowerScale, none of which are easy.
Retrofit a parallel filesystem capability to OneFS,
Write a parallel filesystem from the beginning,
Use an open-source parallel filesystem – Lustre or BeeGFS say, or DAOS,
Buy a parallel filesystem product or supplier,
Resell or OEM another supplier’s filesystem product.
We have talked confidentially to industry sources and experts about this to learn about Dell’s options. Its first decision will be whether to add parallel file system capability to OneFS, and retain compatibility with its installed base of thousands of OneFS PowerScale/Isilon systems, or to start from scratch with a separate parallel file system-based OS. The latter would be a cleaner way to do it but has significant problems associated with it.
If it resells an existing supplier’s parallel filesystem, striking a deal say with VDURA (Panasas as was) for PanFS or IBM for Storage Scale then this gives its installed base a migration issue and risks ceding account control to the parallel filesystem supplier. OneFS customers would have to obtain new PowerScale boxes and transfer files to them; giving themselves a new silo to manage and operate. This is perhaps a little messy.
If Dell buys a parallel file system or supplier then it would have the same problems, in addition to spending a lot of money.
Were Dell to use open-source parallel filesystem software then it wouldn’t have to spend that cash but it would have to set up an internal support and development shop. It would also cede account control to its customers who could walk away from Dell to another supplier of that software or do it themselves, and it would be adding another silo to their data centers and losing compatibility with OneFS.
Account control and a clean software environment would be advantages for Dell writing its own parallel filesystem software. But this would be an enormous and lengthy undertaking. One source suggested it would take up to 5 years to produce a solid, reliable and high-performing system.
Even with is own parallel file system software, Dell would still be adding another silo to its customer’s storage environment and would lose OneFS compatibility, giving its customer base migration and management problems plus adding the risk of them not wanting to adopt a brand new and unproven piece of software.
All these points bring us to option number one in our list, adding a parallel filesystem capability to OneFS.
Coders could set about gutting the heart of OneFS’ file organization and access software and replacing it or, somehow, layer a parallel access capability onto the existing code. The former choice would be like doing a brain transplant on a human; taking so much effort that writing a clean operating system from scratch would be easier.
Fortunately OneFS supports NFS 4.1 and its session trunking, with support added in October 2021. OneFS v9.3 and later also support NFS v4.2.
The NFS v4.2 client has support for parallel NFS (pNFS) using a Flex Files feature. The pNFS functionality is dependent on session support, which became available in NFS 4.1. According to an SNIA slide deck, the Flex Files pNFS layout provides “flexible, per-file striping patterns and simple device information suitable for aggregating standalone NFS servers into a centrally managed pNFS cluster.”
Effectively, existing OneFS cluster nodes read and write NFS files as standalone servers. In the pNFS world the cluster nodes have an added metadata server. Access clients talk to this server using a specific pNFS protocol. It talks to the cluster nodes, called data nodes, using a control protocol, and they and the clients send and receive data using a storage access protocol as before.
Screenshot
The metadata server provides a file layout or stripe map to the client and this enables it to have parallel read and write access to the data servers (cluster nodes). The next slide in the SNIA deck makes a key point about the pNFS metadata server:
Screenshot
It need not be a separate physical or virtual server and can in fact run as separate code in a data server.
This means that if OneFS was given a pNFS makeover then one of the existing cluster nodes could run the metadata server function and we have a road to an in-place upgrade of an existing OneFS cluster.
It is interesting that we can see a parallel [sorry] to what Hammerspace is doing with its data orchestration software. This is based in pNFS and uses Anvil metadata service nodes.
Also, as we wrote in November 2023: “The Hammerspace parallel file system client is an NFS4.2 client built into Linux, using Hammerspace’s FlexFiles (Flexible File Layout) software in the Linux distribution.”
This approach of activating the nascent pNFS capability already supported in principle by OneFS would enable Dell to carry its OneFS franchise forward into the high-performance parallel access world needed for AI training. It wouldn’t be a true parallel file system from the ground up but it would enable a profound increase in OneFS file transfer bandwidth. Existing installed PowerScale clusters would become parallel access clusters.
Such a performance increase would immediately differentiate PowerScale from competitors Pure Storage (FlashBlade) and Qumulo, match NetApp, give it anti-Hammerspace ammunition, provide a counter to VAST Data overtures to its customers and hold the fort against encroachment by HPE OEM’d VAST Data, and also IBM’s StorageScale. The advantages of this approach just keep mounting up.
Note that any NFS standards supporting supplier could do the same, meaning in particular NetApp, which already has, and Qumulo. The latter’s CTO, Kiran Bhageshpur, said: “Qumulo is focused exclusively on a standards based system, end-to-end. Commodity platforms; on-premises, at the edge and in public clouds; no special hardware, standards based interconnectivity (Ethernet and TCP/IP) and standards based access protocols (NFS, SMB, FTP, S3, etc.)”
“We know, we at Qumulo can outperform any real-work workload in a standards based environment without requiring specialized hardware, specialized interconnectivity (Infiniband) or specialized clients ( “parallel file systems”) and at a much better total cost of ownership for our customers.”
Qumulo’s Core software added NFS 4.1 support at the end of 2021 but has not yet added 4.2 support. NetApp’s ONTAP supports both NFS 4.1 and 4.2. Qumulo could add parallel access by adding NFS .2 support and using pNFS. NetApp has already done this.
NetApp principal engineer Srikanth Kaligotla blogged about pNFS and AI/ML workloads: “In NetApp ONTAP, technologies such as scale-out NAS, FlexGroup, NFS over RDMA, pNFS, and session trunking – among several others – can become the backbone in supporting AI/ML workloads.”
ONTAP users can enable pNFS operations now. We could say NetApp is showing a path into parallelism for OneFS.
All-in-all, our bet is that Dell is adding a pNFS capability to OneFS, so that it becomes as we might say, “pOneFS”.
Bootnote
An October 2021 blog by VAST Data’s Subramanian Kartik about NFS enhancements from VAST Data mentions nconnect multi-path which enables the ability to have multiple connections between the NFS client and the storage. It says: “The performance we are able to achieve for a single mount far exceeds any other approach. We have seen up to 162 GiB/s (174 GB/s) on systems with 8×200 Gb NICs with GPU Direct Storage, with a single client DGX-A100 System. Additionally, as all the C-nodes participate to deliver IOPS, an entry level 4 C-node system has been shown to deliver 240K 4K IOPS to a single client/single mount point/single client 100 Gb NIC system. We are designed to scale this performance linearly as more C-nodes participate.”
It is a highly informative blog about adding parallelism to a VAST Data NFS system and well worth a read.
Cribl, which supplies datalake and data engine software, has announced the launch of the Cribl Technology Alliance Partner (TAP) program, a global ecosystem of technology partners bringing new integrations and validated offerings to customers to transform their data management strategy. With hundreds of existing integrations being used by customers today, we’re told the Cribl TAP provides new integrations with the world’s most widely-used technology providers, expanded partner support, and increased choice for customers to select the data management tools that best fit their needs.
…
Databricks Ventures (DV) has launched its 2nd fund: the Databricks AI Fund. DV head Andrew Ferguson said: “Just as software ate the world, we believe AI is now eating software. As such, the pace of change in the AI ecosystem has dramatically accelerated. The era of AI in the enterprise has arrived, and our new AI Fund embodies Databricks Ventures’ commitment to supporting a new generation of founders and startups in this critically important ecosystem. To bolster the ecosystem around the Databricks Data Intelligence Platform, we will be aggressively seeking out investments in innovative, early-to growth-stage startups that are utilizing or enabling AI in innovative ways on top of or alongside our platform.”
“Just since last fall, we’ve announced investments in six AI-focused companies: Anomalo, Cleanlab, Glean, Mistral AI, Perplexity and Unstructured. These portfolio companies fall in very different sectors of the landscape — from open-source LLM development (Mistral AI) to AI-powered data quality monitoring (Anomalo) — but all leverage the power of AI to deliver superior customer experiences. And with these varied investments, we have forged deeper partner and integration relationships across the AI value chain that benefit our portfolio companies and our common customers — and help us build a strong, differentiated ecosystem around the Databricks platform.”
…
Astra DB NoSQL database supplier DataStax has launched Astra Vectorize, a feature that performs embedding generations on the server-side. It includes new integrations with OpenAI and Microsoft Azure OpenAI Service, designed to accelerate and simplify embedding generation for developers. These integrations allow organizations to compare different embedding models with just a couple of clicks, saving hours and days of development time. Astra Vectorize is an addition to the Astra Data API that will allow users to provide raw, unstructured data – like a piece of text or an image – as a part of an insert, update, or vector search operation, and the vector embedding for that data will be automatically generated by Astra DB.
…
Dell will offer an integrated turnkey hyperconverged appliance combining the Nutanix Cloud Platform and Dell servers. Nutanix Cloud Platform for Dell PowerFlex will combine the Nutanix Cloud Platform and its AHV hypervisor, for compute with Dell PowerFlex for storage. The companies will collaborate on engineering, go-to-market, support and services, and the joint systems will be sold by Dell sales teams and partners worldwide. The joint systems from Dell and Nutanix are currently under development and will be available to customers in early access later this year.
…
Stream data lake startup Hydrolix has closed a $35 million Series B round led by S3 Ventures. Prior investors such as Nava Ventures, Wing Ventures, AV8 Ventures and Oregon Venture Fund also participated. The company’s total funding to date to is $68 million. Hydrolix said it doubled revenues in Q3 and Q4 in 2023, developed new partnerships and grew another 75% in Q1, 2024. Hydrolix says it offers a streaming data lake built to power log-intensive applications. Its SW combines real-time stream processing, low-latency indexed search, decoupled storage and high-density compression to create a high-performance, lower-cost data management platform designed to handle the hyper-growth industry demand for long-term data retention. All Hydrolix data is “hot,” eliminating the need to manage multiple storage tiers. This approach allows Hydrolix to offer customers real-time query performance at terabyte scale for a radically lower cost compared to other cloud data platforms, we’re told.
…
IBM ran performance tests with StorageScale CES S3 (non-containerized S3), using COSBench and large objects (1GB) and small objects. The blog concludes: “With the current cluster setup, the bandwidth measured for CES S3 for reading large objects is 63 GB/s and 24 GB/s for writes. For reading small objects, the maximum number of operations per second was in the range of 56000, using object sizes of 1KB and 4 KB. Interesting bandwidth results were observed with 4MB object size in combination with 256 and 512 workers, getting peaks of 70GB/s.
“Also, it was observed that CPU utilization increased based on the number of COSBench workers. Starting with a very low utilization for 1 and 8 workers and having a max utilization for greater number of workers. Performance engineering work will continue with the execution of diverse tests. In future entries, we will describe performance evaluations using COSBench with different workload characteristics as well as other benchmarking tools.”
…
France-based Kalray announced Ngenea for AI, a new edition of its Data Acceleration Platform that it says is fine-tuned for AI data pipelines. AI demands new ways to ingest and access massive volumes of data. The pitch for Ngenea for AI is that it helps speed up performance and simplify data management. Ngenea for AI is the companion to two Ngenea editions for Media & Entertainment and HPC customers. The Ngenea for AI empowers AI innovators to speed up their ingest performance and access their unstructured data from a unified, global name space, the company says.
Ngenea for AI adds data indexing and search capabilities, which users can use to feed any data to Smart Vision, GenAI and RAG applications. It incorporates a high-performance storage tier for the most data-intensive AI workloads, powered by a high-performance parallel file system that can manage petabytes of data and billions of files – whether data lives on the edge, in the cloud, or on premises.
Optional hardware acceleration via Kalray’s DPUs including the TURBOCARD4 (TC4) allows parallel processing in an asynchronous way for ultra-demanding workflows, via a complementary architecture to GPUs.
…
A record 152.9 EB of total LTO tape capacity (compressed @2.5:1) shipped in 2023, with a growth of 3.14 percent over 2022, driven in part by rapid data generation and the increased infrastructure requirements of hyperscalers and enterprises.
…
N-able, which supplies data protection software to MSPs, is working with the MSPAlliance to help equip MSPs to meet compliance requirements using MSPAlliance’s Cyber Verify program. The Cyber Verify program is available globally to N-able MSP partners and is built to help MSPs identify and adhere to industry gold standards, stand up and grow a Compliance-as-a-Service practice, and comply with cyber-regulations. The program also offers N-able partners a customized experience, helping them understand how to potentially utilize the N-able offerings in their tech stack to strengthen their compliance initiatives.
…
IBM-owned Red Hat and Nutanix has announced an expanded collaboration to use Red Hat Enterprise Linux (RHEL) as an element of the Nutanix Cloud Platform. The platform foundation is AOS, which combines components of a traditional OS with additional services and packages. AOS will now build on RHEL for traditional operating system capabilities. Nutanix will also contribute to CentOS Stream, working with Red Hat and the broader open source community on hypervisor functionality, networking and storage performance for emerging artificial intelligence (AI) workloads on RHEL.
…
NVIDIA and Microsoft have expanded their collaboration:
The latest AI models developed by Microsoft, including the Phi-3 family of small language models, are being optimized to run on NVIDIA GPUs and made available as NVIDIA NIM inference microservices.
NVIDIA cuOpt, a GPU-accelerated AI microservice for route optimization, is now available in Azure Marketplace via NVIDIA AI Enterprise. Developers can now easily integrate the cuOpt microservice, backed by enterprise-grade management tools and security, into their cloud-based workflows to enable real-time logistics management for shipping services, railway systems, warehouses and factories.
NVIDIA and Microsoft are delivering a growing set of optimizations and integrations for developers creating high-performance AI apps for PCs powered by GeForce RTX and NVIDIA RTX GPUs.
…
Vector database supplier Pinecone this week launched Pinecone serverless into general availability. This vector database, which is designed to make generative artificial intelligence (AI) accurate, fast, and scalable is now ready for mission-critical workloads. Over the past few months, more than 20,000 organizations have been using it in public preview, including names like Notion, Gong and You.com, as well as many smaller companies and individual developers. Users have been able to reduce costs by up to 50x, we’re told, while building more accurate AI applications at scale. Pinecone research shows that the most effective method to improve the quality of generative AI results and reduce hallucinations – unintended, false, or misleading information presented as fact – is by using a vector database for Retrieval-augmented Generation (RAG).
…
Samsung is working on 3D DRAM according to a wccftech report, with vertically-mounted transistors via a VTC (Vertical Channel Transistors) technique in a 4F Square cell structure. Sammy has a 16-layer DRAM stacking target. This leads to greatly increased DRAM chip capacity. A Fred Chen tweet showed a Samsung slide about the technology. It suggests product won’t appear until the 2030s.
Samsung 3D DRAM slide from Fred Chen.
…
Snowflake’s revenue growth continues unabated.
Cloud datawarehouser Snowflake has announced its FY25 Q1 financial results and the acquisition of TruEra, an AI observability platform. Revenues were $828.7 million, up 32.9 percent hyear-on-year, with a loss of $317.8 million, deeper than the year-ago $226 million loss. The customer count rose to 9,822 from 8,167.
TruEra’s technology helps evaluate the quality of inputs, outputs, and intermediate results of LLM apps, as well as identifying risks such as hallucination, bias, or toxicity. This is investment will bring LLM and ML observability to Snowflake’s AI Data Cloud, and is aimed at enabling it to provide deeper functionality to help organisations drive AI quality and trustworthiness.
…
Dr. Julian Chesterfield.
Edge HCI supplier StorMagic has appointed Dr. Julian Chesterfield as its CTO. Based in Cambridge, U.K, Dr. Chesterfield has a Master of Science from University College London and a Ph.D. in computer science from Cambridge University. At Cambridge, he was one of the creators of the Xen OpenSource hypervisor that was ultimately acquired by Citrix Systems. Dr. Chesterfield then founded the Sunlight.io hyperconverged infrastructure (HCI) platform, and served as the company’s CTO to support its growth. During his career as CTO and technology architect at companies such as Xensource, Citrix, and OnApp, he has developed hypervisor, software-defined storage and application management technologies.
…
Enterprise data manager Syniti will showcase a proof of concept that demonstrates using autonomous AI agents, known as the Syniti Squad, to address data quality issues. This will be done at the SAP Sapphire & ASUG Annual Conference in Orlando, Florida from June 3-5 and at SAP Sapphire Barcelona from June 11-13. This Syniti Squad will ask customers about their business challenges or data problems, and the AI agents will automatically collaborate to analyze the user-defined focus area, identify insights, and generate custom business rules tailored to the customer’s specific goals. The agents have an inner dialogue to deliver more accurate results. Syniti told us: ”Our proof of concept includes custom tools that are assigned to the agents; the agents are assigned an LLM, custom tools and task, they then have inner dialog and external dialog with humans and other agents” to deliver more accurate results.
…
Jason Yeager.
Virtualized data center suppler VergeIO has hired Jason Yaeger as its new SVP of Engineering. Yaeger will oversee all engineering operations, focusing on advancing product strategy, optimizing technology governance, and fostering an environment of innovation. Yeager has been a self-employed strategic advisor at NYJL advisors for almost 5 years since being co-founder and CEO at TenacityAI for three and a half years.
…
Veritone, which designs human-centered AI offerings, has announced a strategic partnership with Creative Artists Agency (CAA), an entertainment and sports agency, to power the CAAvault, a synthetic media vault conceived by CAA to serve the entertainment community and its participants (talent). CAA is using Veritone’s Digital Media Hub (DMH) technology to store the intellectual property of the participating talent’s name, image, likeness and all associated metadata, like synthetic counterparts, including digital scans and voice recordings, in the vault.
Apple’s Sonoma MacOS screws around with non-Apple format external drives stopping content access, users have complained, saying that it has done so for months and claiming the US tech giant has ignored the problem.
Mac users sometimes need external disks or SSDs to move files between Macs and other systems, such as Windows PCs and servers. Drives obviously have to be in a format accessible by both the MacOS and Windows operating system. One such is exFAT, the Extensible File Allocation Table format, developed by Windows in 2006 as a licensable item, which then became open for use by everyone in 2019.
exFAT is now pretty much a standard for moving files on disks, SSDs and SD cards between Windows, Linux and smart devices. Users are complaining that after upgrading from MacOS Ventura to Apple’s current MacOS v14 (Sonoma) is giving users problems with exFAT drives.
An Apple customer told us: “MacOS Sonoma messed up an exFAT formatted SD card containing dashcam footage. Literally all files vanished and Spotlight Search made sure recovery was going to be fruitless by stamping its mark on the filesystem although TestDisk/PhotoRec got some files back but mostly truncated MP4.”
“In the process Sonoma corrupted two Kingston SSDs. These appear to work fine on my Windows NUC but don’t want to risk plugging them in to my MacBook Air M1. They would literally not appear in Finder and I couldn’t figure out what MacOS was doing. Windows came to the rescue here.”
Poster BungalowBill92 is one of many reporting the issue, and in October 2023 he said: “Since I updated to macOS Sonoma, I’ve been experiencing issues with my external drives (in ExFAT). I can’t access them, nor can I access Disk Utility (which continuously displays a “Loading disks” message). Only when I use the Disk Arbitrator app to prevent the drives from mounting, it allows me to run Disk Utility and “First Aid,” enabling access to the external drive. However, the folder icon images are missing, and in the Sharing & Permissions info, it reads “You have custom access.” Is anyone else experiencing similar problems, or does anyone know how to resolve this?”
Poster feketegy reported experiencing an identical issue: “Having the same issue with external USB drives, USB sticks and external keyboards that are connected through USB on macOS Sonoma 14.2 and MacBook M3 Pro.”
There are 20 pages on this thread in the forum with users still experiencing the problem seven months later. Poster earz said this month:
“In all frankness I have to wonder just how much longer this travesty is going to continue. So many folks cannot USE their Apple product due to an operating system that seems to not be of any concern to Apple, and it looks like the crippled businesses who try to use the current OS and are not happy are not a concern either.
“I truly do not know of ANY OTHER scenario where the product remains ‘messy’ and ‘somewhat crippling to its users’ for so long. I truly believe Apple could care less about remedying Sonoma. Of course, they just might be sitting on a solution they will release with a price tag on it …some day ..maybe…IF they can fix the OS at all…”
Several posters suggest reverting to MacOS v14 (Ventura) to get their exFAT drives working again.
Elsewhere on the internet there are many similar forum posting threads: Reddit – r/<MacOS, Apple Insider, MacRumors Forums, Mac Help Forums and iBoysoft, for example.
In the MacOS Sonoma 14 release notes, Apple mentions a “New Feature” under the topic of File System. It states: ”The implementations of the exfat and msdos file systems on macOS have changed; these file systems are now provided by services running in user-space instead of by kernel extensions. If the application has explicit checks or support for either the exfat or msdos file systems, validate the applications with those file systems and report any issues. (110421802).”
The “user-space” term refers to everything within MacOS that isn’t in the kernel of the OS. Anything running in user space is subject to user ID access rules. These can limit file use and function depending on how they are set up. Kernel file access services don’t have any such file access issues as they run with root privileges instead.
It is not known if this is a contributor to the exFAT drive access problem or not – since Apple has not said anything about it – but the timing of the change is suggestive.
We contacted Apple’s media access service on May 20 relaying the issues above, and asking what Apple is doing to fix the issue, and will the issue be resolved in the forthcoming MacOS 15 (Catalina)?
There has been no reply.
Comment
It appears that Apple released MacOS Sonoma with inadequate testing of external drive connectivity and access. Four Sonoma point releases have not fixed the problem and Apple has issued no public statement recognizing the issue or committing to fix it.
Infinidat has upgraded its InfiniBox arrays to fourth-generation hardware with a higher level of performance and added cyber-protection, Azure support, a controller upgrade program, and increased service offerings.
The company sells high-end and scale-up enterprise arrays with three controllers for reliability and Neural Cache-branded memory caching for very low latency storage request responses down to 35μs. InfuzeOS controls the arrays and also runs in the AWS cloud, providing an InfiniBox environment there. Infinidat has all-flash (SSA) and hybrid flash/disc versions of its array, both with memory caching. There is an InfiniGuard cyber-protection and backup system, and the InfiniVerse cloud-based monitoring system receiving telemetry from InfiniBox arrays.
Phil Bullinger
Infinidat CEO Phil Bullinger said: “We’re excited to announce the new InfiniBox G4 systems and the many new enhancements that expand our InfiniVerse platform and STaaS (Storage-as-a-Service) initiatives, cybersecurity capabilities, infrastructure lifecycle management, and hybrid multi-cloud support, culminating significant product development efforts and field engagement with our partners and customers.”
Hardware
The InfiniBox SSA F1400T arrays move away from Intel Xeons to controllers based on AMD EPYC 9554P single socket 64-core CPUs. This gives them 31 percent more CPU capability and a 20 percent power reduction on a per core basis compared to the existing F4304T and F4308T systems.
They are fitted with a PCIe gen 5 bus, replacing the prior gen 3 PCIe interconnect, and DDR5 DRAM, enabling them to deliver up to twice the performance of the current (G3) generation of InfiniBox and InfiniBox SSA II arrays.
There are four models in the SSA range – F1404T, F1408T, F1416T and F21432T – and they have the same active-active-actuve controller set up but different capacities:
The capacities are expressed as percentages of a petabyte. The F1432T will be available in a few months. The F1404T requires 14 RU of rack space at 155TB usable capacity, and the F1408t and F1416T also take up 14RU as will the F1432T. You will still be able to buy them from Infinidat in rack form though, as well as 14RU enclosures that fit in a standard rack. It will not be possible to upgrade from one F1400T model to the next.
Infinidat has also extended its hybrid InfiniBox 4400 range upward with a new F4420 model fitted with 20 TB disk drives, giving it 3.17 PB usable capacity, 55 percent more than the prior top end F4412:
The F4408T and F4416T will be available in a flexible storage architecture with 60, 80 and 100 percent capacity populated forms.
Infinidat is joining Pure with EverGreen, IBM with Storage Assure, and HPE’s Timeless initiatives with its own Mobius-branded, non-disruptive, controller upgrade program for the gen 4 arrays over their life cycle.
The array’s InfuzeOS is now available in the Azure cloud with the InfuzeOS Cloud Edition product, as well as AWS, giving Infinidat customers multi-public cloud capability for running Infinidat storage facilities with replication to and between AWS and Azure from on-premises Infinidat systems for test/dev, DR, backup and business continuity. The performance of the AWS and Azure InfuzeOS instances will depend upon the underlying infrastructure the CSP uses.
Cyber-protection and InfiniVerse
Infinidat has developed Automated Cyber Protection (ACP) where API calls from Syslog, Security Information and Event Management (SIEM) and Security Orchestration, Automation, and Response (SOAR) tools can trigger automatic snapshotting for all or selected parts of the array’s data volumes. That means a ransomware or other attack, when detected by the tools, can have its threat window drastically reduced because critical data can be immediately copied into an immutable InfiniSafe snapshot. The snaps can be inspected with InfiniSafe Cyber Detection scanning uses AI and ML to check the content for integrity.
A blog by Bill Bassinas, Infinidat’s Senior Director of Product Marketing, says: “InfiniSafe ACP is a simple concept – when you see something, do something! … Without thinking, it automatically triggers a protection scheme to create immutable snapshots of any data within your InfiniBox SSA and InfiniBox platforms. Why do it? Why not? It costs you nothing! … and can save you millions!”
Infinidat has extended its InfiniSafe Cyber Detection capabilities to VMware environments. Bassinas says: “Volumes or file systems that are used for VMware datastores can now be specifically scanned with the same accuracy as standard data volumes and file systems. VMs are reported on with the same accuracy and high levels of granularity as volumes, files, databases, etc.”
It will extend its coverage to its InfiniGuard purpose-built backup appliance in the second half of 2024. InfiniSafe core functionality and InfiniSafe ACP are included at no cost with all Infinidat arrays.
The InfiniVerse cloud monitoring facility has been upgraded to a so-called Platform-Native Architecture. It now includes Cyber Resilience Services, Consumption Services, Lifecycle Management Services, Data Services, and Manage and Monitor control plane services.
Infinidat systems can be purchased as Capex, FLX consumption-based, pay-as-you-grow STaaS or in a COD (Capacity On Demand) scheme. InfiniVerse Mobius applies to the Capex purchase scheme.
Comment
Customers should welcome these substantial improvements of Infinidat’s on-premises arrays and their software and services, with extended cyber-protection, controller upgrades, and hybrid cloud facilities. The F1404T, with its 155 TB capacity, provides a lower-cost entry point than before and should extend Infinidat’s appeal to new customers.
