Analysis: DDN launched its Inferno fast object appliance and xFusionAI hybrid file+object technology at Nvidia’s GTC 2025. We examine both technologies to see what they offer.
From the AI training and inference point of view, DDN has two storage technologies. Its originally HPC-focused EXAScaler line presents a Lustre-based parallel file system running on NVMe SSDs in scale-out storage nodes with client software executing in the GPU server nodes. The newer Infinia technology provides a base key-value store on which data access protocols are layered, with the S3 object protocol first. File, block, and others will be added over time.
Infinia v2.0, released in February, is designed to eliminate AI bottlenecks, accelerate workflow, and scale for complex model listing. It delivers real-time data services, multi-tenancy, intelligent automation, and has “a very powerful AI-native architecture,” DDN says.
But it does not support file access, and most AI development work to date has used fast file access with slower object as the protocol to access mass unstructured data stores. Loosely speaking, in AI today, file is the past, but participates in the present with growing object access, and object will be increasingly important in the future. It will be a hybrid file+object world for AI training and inference generally for the foreseeable future.
The Inferno product adds Nvidia’s Spectrum-X switch, with RoCE adaptive routing, to Infinia storage. DDN says testing showed Inferno outperforming AWS S3-based inference stacks by 12x with sub-millisecond latency and 99 percent GPU utilization. DDN states that Inferno “is a high density and low power 1RU appliance with 2 or 4 BlueFields.” These will be Nvidia’s Arm-powered smartNICS that link to BlueFields in Nvidia-powered GPU servers.
Inferno uses “high-performance NVMe drives for ultra-low latency … supports seamless expansion,” and “is optimized for AI model training and inference.” It is “fully optimized with DDN’s AI and HPC ecosystem, ensuring streamlined deployment.”
There is no publicly available Inferno configuration or availability information. If such an appliance used 122 TB QLC SSDs, then we could be looking at a ten-bay chassis with 1.2 PB capacity. A rack containing 30 of them would have 36 PB with an NVMe/GPUDirect for Objects-based network fabric comprised of BlueField-3s talking across Spectrum-X networking to BlueField-3s connected to GPU servers.
DDN describes xFusionAI technology as different, stating that it is “engineered as an integrated AI storage and data management solution that balances high-speed parallel file storage with cost-efficient object storage for AI workflows … There is a single pool of storage that is logically partitioned between EXAScaler and Infinia, rather than two completely separate systems. xFusionAI can be deployed on a unified hardware infrastructure, with both EXAScaler and Infinia software components running within the same system.”
It “features a single-user interface that provides visibility and control over both EXAScaler and Infinia environments, ensuring streamlined management.”
Infinia is not merely a backend object store, says the vendor, adding that it serves as an intelligent data management layer to complement EXAScaler’s high-speed file performance. Data can be moved between EXAScaler and Infinia using automated policies or manual tiering, allowing users to optimize storage costs and performance.
In effect, we have high-speed file storage (EXAScaler) being added to Infinia, possibly as a stopgap, until Infinia’s native file system support arrives. This means xFusionAI controllers will be more capable than Inferno’s (Infina object-only) controllers, as they must manage both file and object environments and “move” data between them. We put “move” in quotes because the data might not actually physically move; it could somehow be remapped so that it is transferred from an EXAScaler partition to the Infinia partition, and vice versa. Of course, if the Infinia partition used slower QLC drives with the EXAScaler partition using faster TLC drives, then data would physically move.
It will be interesting to understand the details of this hybrid system as they emerge. One insight is that xFusionAI gives DDN a combined file+object AI training and inference storage system to compete with VAST Data’s hybrid file+object storage, which also has block access, less important in the AI world up to now. DDN says the “product is coming soon. Pricing details are available upon request and depend on configuration, capacity, and deployment scale.”
SPONSORED FEATURE: AI is driving an explosion in infrastructure spending. But while GPU-enabled compute may grab the headlines, data management and storage are also central to determining whether enterprises ultimately realize value from their AI investments and drive broader transformation efforts.
The worldwide AI infrastructure market is expected to hit $100bn by 2027, according to IDC. Servers are expected to account for the lion’s share of this spending, but storage investment is increasing in line with overall growth as tech leaders cater for the massive datasets which AI requires along with the need for training, checkpoint and inference data repositories.
While AI is fueling this spending boom, many of the underlying challenges facing CIOs haven’t changed, explains HPE’s SVP and GM for storage, Jim O’Dorisio. These include driving innovation, streamlining operations, and reducing the total cost of operations, all within the maelstrom of a constantly evolving tech and business landscape.
Data, and therefore storage, all play into this. AI relies on data. But so do the myriad of other, more traditional, operations that companies regularly undertake. But it must be the right data, available to the right systems at the right time, and at the right speed, says O’Dorisio.
“If you go back 15 years ago, 10 years ago, storage was really just where data sat. Increasingly, it’s where we create value now, right,” he explains.
Dealing with the issues of data gravity and location is particularly challenging, a situation aggravated by the broader span and complexity in customer IT environments. The last two decades have seen a rush to the cloud, for example. But many enterprises are now wondering just how much they actually need to be off premises, particularly when it comes to managing all the data they need in order to realize value from AI.
And beyond creating value through AI or other advanced applications, enterprise data still needs to be protected and managed as well. The cyberthreat is more acute than ever – with threat actors themselves enthusiastically leveraging AI.
The cyber challenge is clearly right up there, says O’Dorisio, but this repatriation of data also creates additional hybrid complexity. There’s sustainability to consider as well, for example. Complex systems require energy to run, and data should be managed efficiently. But the underlying storage should also be as efficient as possible. That includes optimizing energy consumption but also considering the impact of overprovisioning and unduly short life cycles.
This is a legacy problem
The crucial question for an organization’s storage systems then is whether they can keep up with the speed of change. The answer, too often, is they can’t. For multiple reasons.
Traditional architectures that rigidly tie together compute and storage can pose problems when scaling up to meet increasingly complex or large workloads. Expanding storage capacity can mean spending on compute that isn’t really needed, and vice versa. This can lead to silos of systems built out for a particular business unit or workload, or a particular location, for example, core datacenters or edge deployments.
Likewise, legacy architectures are often targeted at specific types of storage: block; file; object. But AI doesn’t distinguish between data formats. It generally wants to chew through all the data it can, wherever it is.
This lack of flexibility can be aggravated by legacy systems that were designed for a particular type of organization or scale, e.g. “enterprise” or a medium sized business. Integrating a raft of standalone systems can present a clear architectural issue as well as management challenges.
Disparate hardware often means disparate management systems and consoles for example, meaning managers are left with a fragmented view of their overall estate. That situation can force team members to specialize in a subset of the organization’s infrastructure, which can often result in inefficiencies and increased operational costs.
These fragmented, siloed, and often hard to scale systems don’t lend themselves well to the hybrid operations that are increasingly becoming the norm. Any organization contemplating repatriating some or all of its data will likely balk at losing the ease of use of managing their data in the cloud.
This can all contribute to a massive bottleneck when it comes to maximizing the value of all the data available. “The architectures are typically complex, and they’re siloed, explains O’Dorisio. “And it makes extracting value from the data very difficult.”
Where is the value?
HPE has sought to address these challenges with its HPE Alletra storage MP platform. The architecture disaggregates storage and compute, meaning each can be scaled separately. So, as the demands of AI increase, infrastructure can be scaled incrementally, sidestepping the likelihood of siloes or wasteful overprovisioning, says HPE. This is bolstered by HPE’s Timeless program, which ensures a free, nondisruptive controller refresh, cutting TCO by 30 percent compared to standard forklift upgrades according to HPE estimates.
The MP stands for multiprotocol, with common underlying hardware optimized for particular applications. HPE Alletra Storage MP B10000 modernizes enterprise block storage with AI-driven cloud management, disaggregated scaling, and 100 percent data availability for all workloads, says HPE. Whereas, the HPE Alletra Storage MP X10000 is purpose built for intelligent high-performance object storage. The AMD EPYC embedded processors at their core are designed to offer a scalable X86 CPU portfolio delivering maximum performance with enterprise-class reliability in a power-optimized profile.
An upcoming release of the X10000 system will give the ability to tag data and add metadata as data is being stored. Users will be able to add vector embeddings and similar functions to support downstream Gen AI RAG pipelines. “Our whole notion is really to add the intelligence and create value as the data is being stored, which really significantly reduces time to value for our customers,” O’Dorisio says. Together with the unified global namespace in HPE Ezmeral Data Fabric, customers can aggregate data from across their enterprise to fuel AI initiatives.
But, even if tech leaders have good reason to situate some or even all their storage infrastructure outside the cloud, giving up the ease of management the cloud offers is a harder sell. Step forward the HPE GreenLake cloud, designed to deliver a single cloud operating model to manage the entire storage estate, across the core, edge and cloud.
Any form of disruption to IT operations, whether due to a disaster or a cyberattack, is now considered an inevitability rather than misfortune. However, by harnessing the Zerto ransomware detection and recovery software, organizations “can really recover in hours and days, versus maybe weeks and months when you’re trying to recover a bunch of data from a cloud,” says O’Dorisio.
Intelligent data savings
This intelligent approach to architecture and ownership also supports a reduction in associated emissions by half, O’Dorisio adds, by reducing overprovisioning and the need for forklift upgrades.
HPE’s own research shows that HPE Alletra Storage MP’s disaggregated architecture can reduce storage costs by up to 40 percent. Better still, intelligent self-service provisioning can deliver up to 99 percent operational time savings, calculates the company.
One major global telecom provider recently deployed HPE Alletra Storage MP B10000 to refresh its legacy storage arrays. In the process, the company dramatically reduced the costs associated with support, energy and cooling, as well as datacenter space, says HPE.
The move helped reduce operating expenses by more than 70 percent while allowing the telco to accommodate a higher volume of traditional databases as well as more modern applications. The increased storage capacity with a smaller footprint means the telco provider also now has space in their datacenter to accommodate future growth.
None of that is to suggest that storage in the AI age is anything less than complex. After all, as O’Dorisio says, “The data really spans, from private to the edge to the public cloud. Data sits across all those environments. Data is more heterogeneous.”
But deploying block, file or object storage services on a common cloud-managed architecture means both managing and extracting value from that data will be much easier and efficient.
Pure Storage has worked with Nvidia to enable existing and new FlashBlade customers to provide data storage for AI models running on Nvidia’s AI Data Platform.
Despite recently announcing its disaggregated architecture FlashBlade//EXA tech – which scales performance and capacity beyond existing FlashBlade products – Pure has ensured compatibility with Nvidia’s new AI Data Platform. Through reference design adoption and certifications, FlashBlade customers can integrate with Nvidia’s Blackwell GPUs – transferring data via BlueField-3 NICs, Spectrum-X networking, NIM and NeMo Retriever microservices, and the AIQ blueprint.
Rob Davis, Nvidia VP for Storage Networking Tech, stated: “By integrating Nvidia AI Data Platform capabilities into Pure Storage FlashBlade, enterprises can equip AI agents with near-real-time business data to achieve new levels of personalized customer service, operational efficiency, and unprecedented productivity.”
Pure supports the AI Data Platform reference design with its FlashBlade products and is now certified as high-performance storage (HPS) platform for Nvidia’s Partner Network Cloud Partner Reference Architectures, including HGX systems with B200 or H200 GPUs. It has also earned Nvidia-Certified Storage Partner approval at both the Foundation and Enterprise levels, affirming that Pure FlashBlade can serve as a storage component in Nvidia-style AI factories.
Rob Lee
The Foundation level is an entry point for Nvidia storage partners, validating baseline performance and compatibility with its AI infrastructure for training smaller large language models (LLMs), inference tasks, and initial retrieval-augmented generation (RAG) workflows. The Enterprise level targets large-scale AI deployments powering AI factories and handling massive datasets for agentic AI and other GenAI applications.
Pure provides storage for the converged FlashStack system, built with Cisco servers and networking. FlashStack customers gain a clear pathway to integrating with Nvidia’s AI Data Platform.
Pure CTO Rob Lee asserts: “The incorporation of the Nvidia AI Data Platform into FlashBlade provides the AI-ready storage” that customers need, adding: “Our recent Nvidia certifications affirm that Pure Storage is supporting the pace and scale that AI models need.”
SK Hynix’s SSD business Solidigm has unwrapped an SSD liquid cooling technology it says will result in smaller, fan-less GPU servers while increasing storage density.
GPU servers, and the datacenters that house them, are becoming hellishly hot. If powering the mega chips and associated storage and networking infrastructure wasn’t enough, the need for liquid cooling is increasing the power draw, reducing the real estate available for servers, and making datacenter design increasingly complex.
At the same time, Solidigm senior director AI and leadership marketing Roger Corell said a typical GPU-based AI server typically carried about 30TB of storage across eight slots, “And we don’t see any reason why that capacity per server growth will not continue at a high level”
That storage element is usually air cooled, added Avi Shetty, senior director AI market enablement and partnerships. But as density increases, this becomes a problem. The SSDs themselves get hot, raising the risk of shutdowns, while traditional cooling technologies, ie heat sinks and fans put a brake on density or reducing server size
Solidigm’s answer is the D7 PS1010 E1.S, which is a 9.5mm form factor SSD and a “cold plate technology kit”, aimed at the direct attached storage element in an AI setup.
Shrinking the drive to 9.5 mm provides space for the cold plate which is connected to and chilled the liquid cooling system already supplying the server. That means the drives themselves are still hot-swappable. And it allows for more density in a 1U rack.
The cold plate only touches one side of the drive. However, as Shetty explained, there are components on both sides of the drive, and chilling one side only isn’t really an option.
“What we have is an innovation which allows for heat to dissipate from the front side as well as well as on the back side,” he said.
This results in “an overall thermal mechanism, which allows us to go at much higher watts at the platform level, up to 30 watts, while maintaining full bandwidth of performance.”
There are some liquid cooled consumer devices available and other storage companies have demonstrated professional devices with integrated cooling. However, Solidigm claims its tech will be the first enterprise device to feature “complete liquid cooling”.
The technology, developed in collaboration with an as yet un-named partner is aimed at “future AI servers” – so no retrofitting. It has not set a precise launch date, beyond the second half of this year.
Solidigm is still working through the exact impact of the cooling technology on overall power consumption.
But Corell said, said there was potential to save power in a number of ways. “One, you don’t need to power the fans. And two, you don’t need potentially as low an ambient air temperature inside the aisles of racks to, you know, pull that cooler air over storage, so lowering both HVAC and cooling power requirements.”
Shetty told us, “The liquid cooling loop for storage can be in parallel to CPU/GPU without affecting the cooling efficiency. With fans removed, we expect overall cooling efficiency to beimproved at the server level along with other TCO improvements.”
Twelve storage suppliers lined up in a carefully choreographed parade at Nvidia GTC 2025 to help Jensen Huang sell more GPUs, NICs, switches, and software to businesses buying AI stack systems, anticipating an agentic AI boom.
We have Cloudian, Cohesity, DDN, Dell, HPE, Hitachi Vantara, IBM, NetApp, Nutanix, Pure Storage, VAST Data, and WEKA. The twelve suppliers are tying their announcements to Nvidia’s news about its Blackwell Ultra GPUs and AI Data Platform, a reference design integrating Nvidia’s GPUs, BlueField NICs, Spectrum-X switches, and AI software with storage systems holding the block, file, and object data to be sent via the NICs and switches to the GPUs, there to be processed with Nvidia’s AI software.
Object storage vendor Cloudian said its HyperStore object storage platform supporting GPUDirect for objects can supply both data lake capacity and HPC-class high-performance data access. It can compete directly with HPC file products for even the most strenuous AI training and inferencing use cases, at a third of the cost of systems that do not support GPUDirect. Cloudian announced Nvidia-based reference architectures (RA) on Lenovo and Supermicro server platforms with all-flash HyperStore supporting GPUDirect and RDMA networking. The Lenovo/Nvidia RA delivers throughput of 20 GBps per node with linear scaling capabilities and supporting large language model (LLM) training, inference operations, and checkpointing functions.
Cloudian diagram showing AI environment complexity
Access the Lenovo RA document, with its ThinkSystem SR635 and SR655 V3 server options here, and the Supermicro RA with its Hyper A+ Storage Server here.
Cohesity has updated its Gaia GenAI search assistant “to deliver one of the industry’s first AI search capabilities for backup data stored on-premises.” Sanjay Poonen, Cohesity president and CEO, stated that “by deploying Cohesity Gaia on-premises, customers can harness powerful data intelligence directly within their environment and not worry about any of that data leaving their infrastructure.”
The updated Gaia relies on Nvidia GPUs, NIM microservices, and NeMo Retriever, and can search petabyte-scale datasets. It supports multi-lingual indexing and querying, and comes with reference architectures and pre-packaged on-premises LLMs. Cohesity and HPE will validate and deploy Gaia on HPE Private Cloud AI, a turnkey, cloud-based offering co-developed with Nvidia. Cisco and Nutanix will also offer Gaia with their full stack systems.
DDN said it’s integrating the Nvidia AI Data Platform reference design with its EXAScaler and Infinia 2.0 storage products, both part of its own AI Data Intelligence Platform, and announcing formal support for Nvidia Blackwell-based systems, including its DGX and HGX systems. The AI Data Intelligence Platform, hooked up to Nvidia’s GPU server HW and SW provides BlueField-3 DPU and Spectrum-X network switch integration, access to NIM and NeMo Retriever microservices, and reference architectures.
DDN’s AI400X2 and AI400X2 QLC storage arrays have achieved Nvidia-certified Storage status. They are fully validated with latest DGX SuperPOD with DGX GB200 systems and the GB200 NVL72 and optimized for Nvidia’s Spectrum-X networking. In general, they deliver sub-millisecond latency and 1 TBps bandwidth.
In testing, a single DDN AI400X2-Turbo achieved 10x the usual minimum requirement of 1 GBps/GPU for read and write operations, paired with a Nvidia DGX B200. Multiple DDN AI400X2-Turbo appliances deliver up to 96 percent network utilization per DGX B200, saturating nearly 100 GBps (800 Gbps) of bandwidth in both read and write operations. More benchmark testing details can be inspected here.
The vendor launched DDN Inferno, claiming it’s “a game-changing inference acceleration appliance” integrating DDN’s Infinia storage with Nvidia’s Spectrum-X AI-optimized networking. Early testing showed Inferno outperforms AWS S3-based inference stacks by 12x and can “provide 99 percent GPU utilization,” but there are no workload, configuration or storage capacity details available.
Omar Orqueda, SVP, Infinia Engineering at DDN, stated: “Inferno delivers the industry’s most advanced inference acceleration, making instant AI a reality while slashing costs at enterprise scale.”
The company is also combining its ExaScaler Lustre parallel file system storage with its Infinia object storage in a hybrid xFusionAI offering. Infinia has S3 support now, and coming block, file, and SQL access protocols, so ExaScaler provides a parallel file system complementing Infinia’s current S3-only storage. There are no details available of how AI processing workloads are spread across the component ExaScaler and Infinia systems.
DDN says Supermicro reports 15x faster AI data center workflows, “unlocking unprecedented efficiency in model training and deployment.” Also, enterprise customers achieve radical improvements in multimodal AI, from high-speed RAG pipelines to autonomous decision-making systems. There is “seamless AI scaling across environments, including on-premises, cloud, and air-gapped systems.”
DDN CTO Sven Oehme stated: “xFusionAI is the convergence of AI’s past, present, and future. It brings together the raw performance of ExaScaler with the intelligent scalability of Infinia, delivering a true ‘best of both worlds’ platform that revolutionizes AI infrastructure.”
The company will provide fully validated Nvidia-Certified Storage reference architectures in the near future.
Dell is announcing new infrastructure, software, and services as part of its Dell AI Factory with Nvidia products, which combine Dell storage with Nvidia GPUs, networking, and AI software, integrated with Nvidia’s AI Data Platform reference design.
CEO Michael Dell stated: “We are celebrating the one-year anniversary of the Dell AI Factory with Nvidia by doubling down on our mission to simplify AI for the enterprise … We are breaking down barriers to AI adoption, speeding up deployments, and helping enterprises integrate AI into their operations.”
The storage product is the PowerScale scale-out file system now validated for Nvidia’s Cloud Partner Program and with Nvidia’s Certified Storage designation for enterprise AI factory deployment. Recent PowerScale updates deliver 220 percent faster data ingestion – streaming writes – and 99 percent quicker data retrieval than previous generation systems.
Dell is adding an open source RAG Connector for LangChain and Nvidia NIM microservices to PowerScale. It will integrate the Nvidia RAPIDS Accelerator for Apache Spark with Dell Data Lakehouse software to speed data prep. Dell also supports Nvidia Dynamo, which frees up GPU memory by offloading key-value cache data from GPU server nodes to PowerScale or other external storage.
There are new Dell professional services for data management strategy and data cleansing to optimize Dell’s AI Data Platform with the Nvidia features, with a systematic approach to data discovery, integration, automation, and quality. A blog discusses how customers can improve RAG data ingestion with PowerScale’s RAG connector.
HPE storage staked its claim to a little bit of Nvidia’s GTC limelight this week, touting a new “unified data layer” it claims will help drive the “agentic AI era.” The unified data layer “brings together both structured and unstructured data” to speed up the AI data lifecycle and spans its high-performance data fabric and “enterprise storage infrastructure with sophisticated data intelligence.”
“As we integrate our Alletra storage MP [platform], our private cloud AI assets and our cross multi cloud environments, we create a truly unified data layer that moves AI closer to the data, and that’s absolutely critical,” SVP GM HPE storage Jim O’Dorisio said ahead of the conference.
“This ability to provide a single name space from edge to cloud across a heterogeneous set of data sources, to deliver universal access, multi-protocol support, automated tiering, and security is absolutely foundational to enabling efficient data-ready, AI use cases.”
Asked to explain where the fabric ended and the layer begins, O’Dorisio said: “The global name space is really that unified access layer within the data fabric. So today it’s kind of ubiquitous. And from that perspective, we’re going to be continuing to invest here and evolve things over time.”
The company announced new software features across its Alletra range. The MP B10000 gets unified file access, as well as enhanced ransomware protection with HPE’s Zerto service. HPE also announced Alletra Block Storage for Azure, in addition to pre-existing support for AWS. The vendor said this would simplify data management and workload placement across hybrid cloud setups.
The object storage focused MP X10000 platform gets automated inline metadata tagging, which HPE said meant enterprises can “infuse their object data – as it is stored – with intelligence that accelerates ingestion by downstream AI applications.”
O’Dorisio said this “allows customers to literally chat with their data almost immediately upon ingestion.”
HPE “expects to further accelerate” the platform’s performance through GPUDirect for object support, in collaboration with Nvidia, to “enable a direct data path for remote direct memory access transfers between GPU memory, system memory and the X10000. This will be rolled out in the coming release,” O’Dorisio said.
The storage enhancements were just one element in HPE’s GTC tie-in. The firm also highlighted an “instant AI development environment” to its Private Cloud AI portfolio – Nvidia-fueled, naturally. It includes an integrated control node, end-to-end AI software and 32 TB of integrated storage. Private Cloud AI now supports rapid deployment of pre-validated Nvidia blueprints, including the Multimodal PDF Data Extraction Blueprint and Digital Twins Blueprint, “enabling instant productivity from Nvidia’s extensive library of agentic and physical AI applications.”
There are new HPE professional services for agentic AI. Deloitte’s Zora AI for Finance on Private Cloud AI is a new joint offering that will be available to customers worldwide, plus CrewAI combined with Private Cloud AI can deploy, and scale agent-driven automation tailored for specific business needs.
And HPE unveiled the AI Mod POD, a performance optimized datacenter in a physical container. This supports up to 1.5 MW per module and can be “delivered with speed.” HPE’s Trish Damkroger, SVP GM HPC Computing and AI, said this meant companies that don’t have datacenter capacity or can’t install liquid cooling in existing datacenter space could be up and running in months rather than years.
“We have examples of our customers siting these in parking lots where they used to have employees,” she said. “But with the work from home from COVID, they have the space.”
HPE Data Fabric with support for HPE Private Cloud AI and HPE Alletra Storage MP X10000 will be available in summer 2025. The updates to the Alletra Storage MP B10000 and the X10000 will be available in May 2025.
Hitachi Vantara has added an M Series to its iQ AI infrastructure product portfolio, combining Nvidia GPUs, Virtual Storage Platform One (VSP One) storage, integrated file system choices, and optional Nvidia AI Enterprise software, meaning NIM microservices, Riva, NeMo retriever, RAPIDS, and other software libraries and tools. The result, Hitachi Vantara says, is “a scalable, adaptable, cost-effective AI infrastructure” offering with separate compute and storage scaling.
The file system choices are either a high-performance file system or a global namespace file system; Hitachi Vantara is now reselling Hammerspace’s Global Data Environment data orchestration software, integrated with the VSP One storage product, and building on a November 2024 partnership agreement. This “ensures distributed data is easily and transparently accessible from anywhere for GenAI workloads.”
There is also an optional object storage repository in the Hitachi iQ portfolio. Lastly, Hitachi iQ is integrating the Nvidia AI Data Platform reference design with its storage with Nvidia’s GPU, networking, and AI software to “enable AI agents with near real-time business insights.” A blog discusses the Nvidia Hitachi Vantara offerings.
NetApp ONTAP storage now has Nvidia validation for SuperPOD, Cloud Partners, and Nvidia-certified systems. Specifically, the AFF A90 product gets DGX SuperPOD validation and is certified as High-Performance Storage for Nvidia Cloud Partners with HGX B200 and H200 Systems. NetApp’s AIPod has achieved the new Nvidia-Certified Storage designation to support Nvidia Enterprise Reference Architectures with high-performance storage. NetApp is releasing a new version of its AIPod with Lenovo, including the Nvidia AI Enterprise software platform, which it says provides more flexibility for customers deploying AI infrastructure for inferencing and fine-tuning
NetApp says these certifications “ready NetApp to tap the Nvidia AI Data Platform reference design” and use ONTAP storage with AI agents for reasoning model inference workloads. This reference design includes Blackwell GPUs, Nvidia networking, and the Nvidia Dynamo open source inference library.
It means NetApp customers will be able to connect their data with agents using Nvidia AI Enterprise software, including the AI-Q Blueprints, NIM microservices for Nvidia Llama Nemotron Reason, and other models.
NetApp says it is developing features to accelerate end-to-end AI processing:
Global Metadata Namespace in which customers can discover, manage, and analyze all their data across the hybrid multicloud to enable feature extraction and data classification for AI.
Integrated AI Data Pipeline so customers can more automatically prepare their unstructured data to use in AI applications by tracking incremental changes, leveraging “incredibly efficient replication with NetApp SnapMirror,” classifying data, and creating highly compressed vector embeddings to enable semantic searches on data for retrieval augmented generation (RAG) inferencing.
Disaggregated Storage Architecture to enable customers to optimize their network and flash speeds and infrastructure costs to achieve high performance with minimal space and power requirements for compute-intensive AI workloads.
VAST Data, now saying it’s a $9.1 billion AI Platform company, is calling its Nvidia GTC 2025 news its “biggest launch of the year.” This involves an enterprise-ready AI Stack combining VAST’s InsightEngine with Nvidia DGX products, BlueField-3 DPUs, and Spectrum-X networking. It converges “instant automated data ingestion, exabyte-scale vector search, event-driven orchestration, and GPU-optimized inferencing into a single system with unified global enterprise-grade security.”
The InsightEngine incorporates Nvidia’s AI Data Platform reference design, including AI agents that use Nvidia AI-Q Blueprint, video search and summarization (VSS) blueprint, and Llama Nemotron Reason model NIM microservices. It will be available with Nvidia-certified systems from other server providers in the future.
The key features are:
Vector Search & Retrieval: AI-powered similarity search in VAST DataBase for real-time analytics and discovery.
Serverless Triggers & Functions:Event-driven automation for AI workflows and real-time data enrichment in VAST DataEngine.
Fine-Grained Access Control & AI-Ready Security:Advanced row and column-level permissions, ensuring compliance and governance for analytics and AI workloads, while unifying permissions for raw data and vector representations. Plus, there is built-in encryption and real-time monitoring that spans unstructured, structured, vector and stream data.
VAST InsightEngine for Nvidia DGX is available now. Read more in a VAST blog.
WEKA, the high-speed parallel file system data supplier, supporting up to 32,000 Nvidia GPUs in a single namespace, has achieved data store certification for Nvidia GB200 deployments, supporting Nvidia Cloud Partners (NCP). Specifically, WEKApod Nitro Data Platform Appliances have been certified for Nvidia Cloud Partner (NCP) deployments with HGX H200, B200, and GB200 NVL72 products.
WEKApod appliance details.
The company says its WEKApod appliances deliver “incredible performance density and power efficiency to Nvidia Cloud Partner deployments.” Each WEKApod node achieves 70 GBps read (560 GBpsec per minimum configuration) and 40 GBps write throughput (320 GBps per minimum configuration).
WEKA says its zero-tuning architecture optimizes dynamically for any workload, delivering sub-millisecond latency and millions of IOPS. A single 8U entry configuration meets the I/O demands of a GB200 Blackwell scalable unit (1,152 GPUs). WEKA’s S3 interface delivers ultra-low latency and high throughput, speeding small object access for AI, ML, and analytics workloads.
A new Augmented Memory Grid feature enables AI models to extend memory for large model inferencing to the WEKA Data Platform. It’s a software-defined extension, which provides exascale cache at microsecond latencies with multi-terabyte-per-second bandwidth, delivering near-memory speed performance. This provides additional petabytes of capacity, 1,000x more “than today’s fixed DRAM increments of single terabytes.”
It integrates with the Nvidia Triton Inference Server and caches or offloads prefixes or key value (KV) pairs from a GPU server’s high-bandwidth memory (HBM). The company says that, when processing 105,000 tokens, the Augmented Memory Grid reduced time to first token by 41x compared to recalculating the prefill context and “dramatically changing the response time to an end user’s query from 23.97 seconds to just 0.58 seconds.” It avoids the need to add more GPUs so as to get more memory.
The Augmented Memory Grid and WEKA’s Data Platform software is integrated with Nvidia GPUs, networking, and enterprise software to accelerate AI inference, “maximize the number of tokens processed per second, and dramatically increase token efficiency.”
WEKA’s NCP reference architecture for Nvidia Blackwell systems will be available later this month. The WEKA Augmented Memory Grid capability will be generally available for WEKA Data Platform customers in spring 2025.
Analysis: The storage array industry is undergoing a massive pivot to extreme scale and parallel, multi-protocol data delivery for AI training and inferencing, with dual-controller arrays and scale-out filer clusters being left behind as legacy technology.
The storage needs of AI training butted up against HPC’s parallel file system technology, led by DDN’s ExaScaler and IBM’s Storage Scale, and found it wanting because AI training shops did not want to learn parallel file system intricacies, needing access to basic file data and then object data. They wanted low access latency, meaning all-flash systems, not disk-based, fast, large, and small file performance and very high capacity, beyond the hundreds of petabytes and extending to exabyte levels.
VAST Data was one of the leaders of the charge, announcing its technology six years ago. Since then it has built up a lead in supplying storage and an AI data stack (Data Space, Data Base, Data Engine, Insight Engine) to AI training companies such as xAI and GPU cloud operations such as CoreWeave and Lambda.
Another leader was WEKA with its parallel-capable WekaFS, using standard NFS and SMB, delivering file and then S3 object data faster than scale-out filer suppliers such as Dell’s PowerScale and Qumulo, until Qumulo eventually caught up in the cloud. Dell was early with GPUDirect support, adding it to PowerScale in 2021. NetApp followed in April 2023. Hitachi Vantara announced GPUDirect support in March last year. GPUDirect support has become table stakes but it is not enough on its own to provide full AI storage capabilities.
Hammerspace added to the incumbent’s distress with its data orchestration technology. This, combined with its GPUDirect support, parallel NFS support, and embrace of GPU server’s local, tier zero, SSD storage, meant it could ship data very fast to GPU servers from relatively slow dual-controller file arrays and any other NAS and object storage as well, treating it as a universal data space.
The success achieved by VAST, WEKA, and Hammerspace presented a problem to the incumbent file and object array and parallel file system suppliers. In response, NetApp announced an ONTAP Data Platform for AI project. Dell said it would parallelize PowerScale. HPE OEM’d VAST Data file technology and developed its own Alletra Storage MP disaggregated compute and storage hardware.
DDN announced its Infinia software providing fast access to block, file, and object in late 2023 with a v2.0 update in February this year claiming an up to 100x improvement in AI data acceleration and 10x gain in datacenter and cloud cost efficiency. This was, in effect, a recognition that its Lustre parallel file system-based ExaScaler technology faced limitations and a new approach was needed.
Our understanding of DDN’s Infinia architecture
Huawei launched its A800 AI storage system in May 2024, saying it had a scale-out architecture with a separated data and control plane and an OceanFS high-performance parallel file system supporting NFS, SMB, HDFS, S3, POSIX, and MP-IO. The A800 can provide 100 million IOPS and petabytes-per-second bandwidth. This will not affect North American organizations but will feature in the rest of the world.
Pure Storage announced its FlashBlade//EXA last week and its announcement material identified three technology phases for fast file and object access, starting with Lustre-type parallel file systems:
This separates file metadata from the underlying object data to provide a two-layer system: object data nodes and separate metadata nodes. Access client systems would find out from the metadata nodes where the data they wanted was stored, with files being striped across many data nodes, and then multiple data nodes would pump their parts of the file in parallel to speed delivery. Pure says this runs into problems when there are many small files because the metadata nodes become bottlenecks. Also, the client system software is complicated.
The next phase was to have both the metadata and the data stored in data or storage nodes with separate and scale-out compute nodes doing the data access calculations – the VAST-style approach:
In its initial marketing material, VAST said there could be up to 10,000 stateless compute nodes and 1,000 data nodes, emphasizing the enlarged capacity it was offering. Pure identified problems with this as well, saying that there can be write bottlenecks on the data nodes providing variable performance and that network complexity could be an issue as well.
Step back a moment and reflect that Pure Storage is now an incumbent with many FlashBlade customers and needs to introduce a VAST-type disaggregated compute and storage technology without leaving its customer base behind. In a stroke of genius, co-founder John Colgrove decided to have separate metadata and data storage nodes, roughly similar to Lustre, but make FlashBlade arrays the metadata nodes:
Pure’s Fusion, with its fleet-level global storage pools, could move existing FlashBlade data to the EXA’s data nodes. These are simple JBOFs, using 24 of Pure’s proprietary Direct Flash Modules and their 75 TB and 150 TB capacity, with 300 TB and then greater capacities coming later. They provide relatively low-cost, high-density storage.
Pure says that accessing client systems, such as GPU servers, have simpler agent software and get consistent write performance at scale. The EXA system metadata nodes communicate with compute cluster clients using pNFS (NFSv4.1 over TCP) with data transmitted using NFSv3 over RDMA.
The EXA system scales to exabytes and delivers more than 10 TBps of bandwidth, with 3.4 TBps from a single rack. Its general availability will be in the summer, with S3 over RDMA, Nvidia certification and Fusion integration coming after that.
Now Pure has an AI training-capable storage system and an answer to DDN’s Infinia, Hammerspace, HPE with its Alletra Storage MP, Huawei’s A800, VAST Data, and WEKA.
VDURA will deliver RDMA and GPUDirect optimizations later this year. Object storage supplier MinIO has announced support for S3 over RDMA, while Cloudian and Scality have also announced fast object delivery to Nvidia GPU servers.
That leaves four storage suppliers waiting in the wings: Dell with its future PowerScale parallelism and NetApp with its ONTAP for AI project both yet to deliver the goods. Qumulo has not committed to deliver GPUDirect support, although it has said it can do so quite quickly. Nor has Infinidat. Once Infinidat has been absorbed by Lenovo, it might support GPUDirect alongside its existing RAG workflow deployment architecture for generative AI inferencing workloads.
We should note that Dell has been energetic in supporting AI workloads with its servers and its AI Factory initiatives.
Apart from these four, the rest of the mainstream incumbent file and object storage suppliers have all substantially reshaped their technologies to support generative AI’s need for extreme, exabyte-level storage capacity, RDMA levels of latency, and parallel-style read and write data access.
Architecting IT analyst Chris Evans has published his latest Primary Storage report, a 30-page eBook, looking at block storage offerings from Dell, HPE, Hitachi Vantara, IBM, Infinidat, NetApp and Pure Storage. The results of his analysis show NetApp and Pure Storage as the strongest performers, with Dell, HPE, and Hitachi achieving the lowest results, while IBM and Infinidat fall between the best and the worst achievers. You can get a copy via a monthly subscription of $99.
…
DDN CTO Sven Oehme has been appointed to its board – an unusual move for a CTO – along with Blackstone Head of Americas Jas Khaira. Blackstone recently funded DDN with $300 million.
…
MRAM developer Everspin announced additions to its PERSYST EMxxLX family, the EM064LX HR and EM128LX HR. They will operate at temperatures from -40°C to +125°C, meeting the AEC-Q100 Grade 1 standard for automotive applications. This expanded temperature range addresses the growing demand for persistent, high-speed memory in aerospace, defense, and extreme industrial environments.
From left, Gregg Machon and Jeff Echols
…
Data orchestrator Hammerspace has hired Gregg Machon to be its Global Vice President of Channel Sales and Jeff Echols as Vice President of Strategic Partnerships, Jim Choumas, who was VP Global Channel Sales for Hammerspace and built a great channel program as its foundation, is departing. Machon comes from his role as head of global channel sales at VAST Data, and his CV includes stints at Qumulo, HPE, Nimble Storage, SolidFire, NetApp, Isilon, and EMC. Echols, formerly VP of Global Partner Sales at WEKA, has worked at Nutanix, CommVault, and Dell.
…
HPE commissioned DCIG to write a competitive analysis report comparing Pure Storage’s FlashBlade with its Alletra Storage MP B10000. This contrasts the disaggregated architecture of the MP B10000, with separated storage and compute elements, against FlashBlade’s active-passive dual controller design. But events have overtaken HPE and DCIG as Pure has just introduced its FlashBlade//EXA with a disaggregated design rendering the DCIG analysis largely moot.
…
There is an IBM Storage Scale System 6000 with Nvidia DGX SuperPOD Deployment Guide here:
Chapter 1. Introduction and technical overview
Chapter 2. Architecture
Chapter 3. Deployment
Chapter 4. Server tuning
Appendix A. IBM Storage Scale System 6000 hosts file for Nvidia DGX SuperPOD
Appendix B. IBM Storage Scale System Nvidia DGX SuperPOD Solution Network Installation Worksheet
…
Kioxia announced the development of its LC9 Series 122.88 TB NVMe SSD in a 2.5-inch form factor. It is a QLC SSD using 218-layer BiCS 8 3D NAND, with a PCIe gen 5 x 4 interface, NVMe 2.0 support, 0.3 DWPD endurance rating, and a 2 Tbit die. It’s planning to have customer samples available at the end of 2025 and start mass production in 2026.
…
A reminder about an interesting Metablog on the use of QLC SSDs in its datacenters. It says QLC SSDs form a middle tier between HDDs and TLC SSDs, providing higher density, improved power efficiency, and better cost than existing TLC SSDs, and faster performance than HDDs. Meta expects QLC SSD density will scale much higher than TLC SSD density in the near-term and long-term. “Meta’s storage teams have started working closely with partners like Pure Storage, utilizing their DirectFlash Module (DFM) and DirectFlash software solution to bring reliable QLC storage to Meta. We are also working with other NAND vendors to integrate standard NVMe QLC SSDs into our datacenters.”
…
Ex-GigaOm analysts Arjan Timmerman and Max Mortillaro have launched the Osmium Data Group as an EMEA-based analyst company that focuses on market research, analyst insights, and media/content creation. They’re also planning a podcast for later in 2025. They’re currently working on a prototype market research document focused on data protection. The two recently released a white paper in collaboration with Infinidat on the topic of green IT/sustainability.
…
Vector database supplier Pinecone is rolling out its new architecture, designed to meet the growing and varied demands of agentic AI. It’s made significant improvements to support search and recommendation workloads, with impressive benchmark numbers. The new Pinecone serverless architecture can serve 4x as many queries with roughly 1/8th the latency, with half the compute footprint of OpenSearch. Achieving similar scale and performance as Pinecone serverless would require OpenSearch clusters scaled 10x to 100x. Pinecone handles this natively at scales from thousands to billions of vectors.
Practically speaking, apps can efficiently support millions of users or sessions with personalized data contexts. So, for example, a company building an AI agent can now create separate namespaces for each user without worrying about performance degradation. When a user returns after being inactive for days or weeks, Pinecone’s architecture can retrieve their context with response times of approximately 10ms, ensuring quality of user experience.
…
Nicole Perlroth
Rubrik, along with cybersecurity author and former New York Times lead cybersecurity reporter Nicole Perlroth, launched a new podcast series, To Catch a Thief: China’s Rise to Cyber Supremacy. The podcast, hosted by Perlroth as she is appointed Chief Cyber Raconteur at Rubrik, tracks the Chinese hacking threat as it evolved from trade secret theft, to mass surveillance, to a far more alarming phase: embedding in US government agencies, power grids, transportation hubs, and water systems. To Catch a Thief is a nine-part series that unpacks the high-stakes world of digital espionage and sabotage, and sheds light on many stories of Chinese cyberespionage that have remained untold. It’s available on all major podcast platforms.
…
A recent Wall Street Journal article said two large cloud-computing customers had each ordered one exabyte’s worth of Seagate HAMR disk drives, meaning tens of thousands of disk drives.
…
Decentralized storage provider Storj is partnering NodeShift and providing storage for its decentralized AI cloud platform. NodeShift customers can deploy Storj’s decentralized storage alongside their compute VMs and GPU instances, “offering a cost-effective, secure, high-performance, scalable alternative to traditional cloud solutions.”
…
HDD manufacturer Toshiba told Wedbush analysts it has achieved 31 TB capacity using energy assist/SMR technology in a lab environment on an 11 platter HDD. It has demonstrated 32 TB on a ten-disk platform with HAMR technology with initial drive sampling at customers expected to be this year. Tosh believes the HAMR transition will be elongated as customers take their time to ensure reliability and ascertain appropriate use cases. It also believes that it would cost approximately $240 billion in incremental investments for the NAND industry to address the 1.2 ZB of storage currently produced by the HDD industry.
…
VAST Data added more AI features to its Data Platform software, saying its shared-everything approach allows for all servers to search the entire vector space in milliseconds:
Vector Search & Retrieval: The VAST DataBase is the first and onlyvector database that supports trillion-vector scale with the ability to search large vector spaces in constant time. It can automatically embed vectors for search and retrieval.
Serverless Triggers & Functions: The VAST DataEngine is the first and only system to create real-time workflows without background ETL tools and scanning to provide GenAI access from source data. With event-driven automation for AI workflows and real-time data enrichment, this system can embed and serve context to agentic applications instantaneously, providing high-speed queries, serverless processing, and automated pipelines to ingest, process, and retrieve all enterprise data (files, objects, tables, and streams) in real-time.
Fine-Grained Access Control & AI-Ready Security:VAST now offers row- and column-level permissions, unifying permissions for raw data and vector representations, ensuring compliance and governance for analytics and AI workloads.
VAST Data is partnering with Nebius, with its AI-native cloud, to deliver AI cloud computing for enterprises across the globe. VAST says: “Nebius AI Cloud leverages the latest compute, paired with the VAST Data Platform’s Disaggregated Shared-Everything (DASE) architecture and exabyte-scale DataStore to provide the speed, performance, multi-tenant security and reliability needed to ensure optimal operational efficiency … After a thorough proof-of-concept (PoC) evaluation against competing alternatives, the VAST Data Platform demonstrated superior performance, seamless integration with Nebius’ cloud environment, and a fast-track deployment model that reduced operational risks.”
…
Datacenter virtualizer and VMware alternative VergeIO updated its VergeFabric SDN offering, integrating SDN directly into VergeOS at no additional cost with no dedicated SDN VM required, no external firewalls nor VLAN complexities. VergeFabric is available immediately as part of VergeOS. For more details, visit VergeIO’s website.
…
Weebit Nano has completed AEC-Q100 150°C operation qualification of its Resistive Random-Access Memory (ReRAM) module in semiconductor manufacturer SkyWater Technology’s 130nm CMOS process. AEC-Q100 is the standard automotive stress test qualification for integrated circuits (ICs). The result validates Weebit’s embedded ReRAM non-volatile memory (NVM) technology for high-temperature automotive applications.
…
Software RAID supplier Xinnor announced a reseller partnership with Supermicro “that demonstrates extraordinary performance” in mission-critical storage environments. The agreement enables Supermicro to offer xiRAID Classic with their NVMe server portfolio, including:
Up to 226 GBps sequential read and 64.4 GB/s sequential write speeds
12 million IOPS random read and 1.53 million IOPS random write
95-100 percent RAID efficiency with zero downtime capability
Complete resilience against multiple drive failures and entire server node outages
The MinIO v2.0 AIStor update adopts Nvidia GPUDirect, BlueField superNICS and NIM microservices to bring object data faster to AI inferencing.
Back in February 2022, MinIO co-founder and CEO AB Periasamy claimed Nvidia’s GPUDirect had a “poorly thought-out design” and an “NVMe raw block storage interface with a control channel for metadata is terribly complicated for the AI/ML community.” He claimed he saw “no point in implementing GPUDirect because, in almost all of MinIO’s AI/ML high-performance deployments, the real bottleneck is either the 100 GbitE network or the NVMe drives and definitely not bounce buffers,” in which data from a storage drive is temporarily held in the host server’s DRAM.
How times change. Three years later, MinIO is adding support for Nvidia’s GPUDirect Storage for object data to its AIStor offering, saying it “drastically improves overall GPU server efficiency” and “delivers a significant increase in CPU efficiency on the Nvidia GPU server by avoiding the traditional data path through the CPU, freeing up compute for additional AI data processing while reducing infrastructure costs via support for Ethernet networking fabrics.”
It’s also embracing BlueField-3 superNICs to hook up its object storage to Nvidia’s GPU servers with MinIO’s object storage software running natively on the BlueField-3’s ARM compute platform. It says this is the “first and only object storage software,” with its 100 MB footprint, to run natively on BlueField-3, using Arm’s Scalable Vector Extension (SVE) instruction set. In effect, MinIO object storage can now operate on the BlueField-3 NIC, hooked up to a box of flash drives.
MinIO says AIStor is Spectrum-X ready, “ensuring seamless integration with Nvidia’s next-generation networking stack for AI and high-performance workloads.”
It’s adding AIStor’s promptObject API to Nvidia’s NIM microservices infrastructure, “which allows users to ‘talk’ to unstructured objects in the same way one would engage an LLM, to deliver faster inference via model optimizations for Nvidia hardware.” NIM provides pre-built Docker containers, Helm charts, and a GPU Operator to automate the deployment and management of drivers and the rest of the inference stack on the Nvidia GPU server.
Periasamy now says: “MinIO’s strong alignment with Nvidia allows us to rapidly innovate AI storage at multi-exabyte scale, leveraging their latest infrastructure. This approach delivers high-performance object storage on commodity hardware, enabling enterprises to future-proof their AI, maximize GPU utilization, and lower costs.”
The new AIStor features are open to beta customers under private preview. AIStor support for Nvidia GDS and native integration with Nvidia BlueField-3 networking platform will be released in alignment with Nvidia’s GA calendar.
AWS is promising more disciplined and cost effective AI data pipelines with new services through its SageMaker and S3 Tables tools that go live today.
As part of its contribution to Pi Day, the cloud giant has announced general availability of SageMaker Unified Studio, which Sirish Chandrasekaran, AWS vp of analytics, told Blocks and Files was a single development environment with an integrated set of services across AWS’s data analytics and AI/ML services.
It pulls together the vendor’s Lakehouse platform, its SageMaker Catalog, “which is the governance layer”, while with “The studio…you can do everything from SQL, analytics, data prep, data integration, model building, generative AI app development, all in one place.”
It has added new models under SageMaker AI, he said, such as Claude 3.7 Sonnet and Deepseek R1.
“We’ve added capabilities like latency sensitive inferencing for specific models from Anthropic, Meta and Amazon. And we’ve also made it simpler in terms of how you can use Bedrock to both prototype applications but also share them across team members.”
AWS has also announced the ability to access S3 Tables from within SageMaker Lakehouse. “You can now run …. SQL, Spark jobs, model building, Gen AI apps. You can combine your S3 Table data with other data in your Lakehouse, whether it’s in Redshift with what we call native party on S3, on premises and federated sources, all of it you can bring together.”
This would all help companies – or at least those using AWS services – build a better data foundation for their AI projects, he said.
“Our perspective is the way you differentiate is through your data, because every modern business is a data business, and what’s unique to your company is your data.”
“What we’re seeing increasingly with our customers is that … the silos are slowing them down,” he said, because of challenges bringing data into the same place or collaborating between different teams. At the same time, in other organizations, the silos were blurring, he said.
Clearly some companies are rushing to pull data together as they dive into AI. This had led to fears that traditional data management disciplines and skills were being left by the wayside. Chandrasekaran said he was seeing the opposite. “What I’m seeing a lot from companies is that they have this realization now that the way they move faster is by getting back to basics.”
“How we’re reimagining the SageMaker Lakehouse, a lot of it is being able to query data where it is. You do not need to now transfer data from Redshift to S3 or from S3 to Redshift. You can query Lake data from Redshift.”
This reduced duplication, he said. “And that obviously saves costs, same as federated sources.”
At the same time he said, companies were acutely aware of the need for governance, “But I think what’s different about this new world is that governance is no longer just about compliance. It’s about confidence.” That include confidence that AI projects are using and been trained on trusted data, “but also confidence that your AI is adhering to responsible AI use policies.”
IDrive has jacked up its e2 offering, adding object replication and a brace of other enterprise level services.
Customers of the IDrive e2 S3 Compatible Object Storage platform will be able to enable automatic replication of their data across geographies. This will mean critical data will always be available in multiple regions, even when they suffer critical service failures, boosting redundancy and compliance. It also means companies can ensure up to date data is closer to the people who need it, it rather than having to ping its way across regions.
The cloud data firm operates its own datacenters to underpin its backup and data services, and initially targeted home users and small businesses. However, its IDrive e2 service, launched in 2022, is squarely aimed at developers, large organizations and enterprises.
“Replication was a natural step for us,” a spokesman said, “and stacks right up against any of the current incumbents in the object storage market today. Our customers are looking for replication, so we always want to provide e2 customers with what they are needing.”
E2 users also now get Bucket Event Notifications and Bucket Logging.
The former can be used to underpin real-time monitoring, for example, by enabling notifications when new objects are added or removed from a bucket. Likewise, they can be used to build data workflows, such as updating website content when new objects are added.
Bucket Logging will provide deeper insights into how data is access and used. So the firm reckons customers will use it for access control and security auditing, as well as for usage monitoring and, in turn, cost management.
IDrive says its service starts at under 50 bucks per annum for 1TB of capacity. Two years ago it launched an all-flash version of the e2 service.
Toshiba has doubled down on the future of spinning rust by opening an HDD Innovation Lab in Germany.
The Dusseldorf site will expand Toshiba Electronics Europe’s “evaluation” services for customers across Europe and the Middle East for chunkier storage installations where it makes sense to use traditional platter-based drives rather than their trendier, flash-based cousins.
The biggest tech infrastructure projects – the AI factories and datacentres attracting economy destabilizing levels of capital investment – are generally thought of as flash only zones.
But Rainer Kaese, senior manager for HDD business development at Toshiba, said the growing amount of data being stored, including for AI, was too great for flash alone to support. This was partly because of cost – HDD is one seventh of the cost of flash – and partly because: “The flash industry is not be able to manufacture enough capacity to satisfy the growing demand, and still will not be for a significant while.”
He said SSDs had a speed advantage, “Which makes them the best choice for local/server attached working storage”. Capacity requirements here are moderate, he continued, so the high cost of SSD capacity can be “offset” by the performance gain over HDD local storage.
Yet for “big data” capacity requirements in the petabyte range, SSD would be too expensive while the performance premium is not necesary. Not least as many HDDs can be run in parallel.
“We have demonstrated that 60 HDDs in ZFS software defined storage can fill the entire speed of a 100GbE network,” he said.
Meanwhile, HDD costs have remained stable, while capacity has shot up. “There are 100+TB SSDs, but they cost several 10k+ dollars.”
The lab will focus on configuration evaluations for RAID, scale up and scale out storage systems, for enterprise, datacenter and cloud applications. It will also carry out evaluations on smaller “vertical” applications such as Soho NAS and video surveillance, ie. digital video recorder, network video recorder.
Kaese said the lab has single node systems running up to 78 disks, providing up to 2 Petabyte of storage with today’s high-capacity HDDs. “For scale out we are operating a basic CEPH cluster with three nodes and 36 HDDs,” he said, which will be expanded in the future.
The company already operates a smaller lab in Dubai. Its lab programme has hitherto thrown up innovations, he said, such as optimizing the number of HDDs for the highest possible performance.
“[We] found that a typical configuration of four HDDs (ie. in small Soho NAS) can fill the 10GbE networks. 12 HDDs match the 25GbE of Enterprise networks, and 60 HDDs would require high end 100GbE network speed to unleash the full performance of the many combined HDDs.”
Western Digital also pushed more rust into the market this week, aimed squarely at the creative sector.
These include the $8,199.99 208TB G-RAID Shuttle 8 designed for “massive storage and seamless backup consolidation – whether on location or in the studio”. It offers 1700MB/s read and 1500MB/s write in RAID 5. A 104TB device is also available.
It also unwrapped G-Drive Project devices sporting 52TB or 26TB capacities. And it has introduced the 26TB WD Red Pro HDD for NAS environments, priced at $569.99.
The move comes after IBM storage bloggers Barry Whyte and Andrew Martin were widely quoted last month saying “the writing is on the wall for spinning rust.” But then again, it always has been.
Nasuni has hitched its wagon to CrowdStrike, meaning information from the data platform will be picked up and passed through the latter’s Falcon Next-Gen SIEM platform.
According to Nasuni, the integration means users “gain a unified platform to ingest, process and manage syslog messages, enabling seamless search, reporting, dashboards, and alert action.”
A Nasuni spokesperson said this meant “security teams can take appropriate steps to protect the rest of their infrastructure.”
Further interactions between the two firms are on the cards, they added. “We’re moving onto the SOAR [security orchestration, automation, and response] side next to provide automated coordination between the two systems based on this data provided.”
The tie-up will make building workflows spanning the two platforms easier, the spokesperson claimed. “Customers won’t have to teach CrowdStrike how to understand Nasuni events; we have done that work for them.” The idea is that they’ll be able to take action as soon as the event info flows into CrowdStrike.
CrowdStrike’s Falcon is probably best known for an update blunder that caused a worldwide crash of Windows computers last year.
The spokesperson was at pains to point out that last year’s outage was due to a flaky software update rather than a cyberattack. “CrowdStrike has implemented many measures to restore trust within their platform, customer base, and the general market.”
CrowdStrike’s SIEM is used by many Nasuni customers, they said. “This integration is designed to help these customers add another layer of protection for their files and make the lives easier for the security teams that are using both CrowdStrike and Nasuni.”
Nasuni has a similar integration with Microsoft’s Sentinel platform and is casting its net further, the spokesperson said. “We are continuing to pursue additional integrations with security tools that our customers use to expand the impact of Nasuni’s security capabilities.”
For its part, Crowdstrike struck an integration deal with Commvault in January.
Last month Nasuni claimed to have over 500 PB of total capacity under management. Its most recent investment round, led by Vista Equity and joined by KKR and TCV, gave it a $1.2 billion valuation.
