VDURA & WD target AI with hybrid storage

VDURA has announced a collaboration with WD to certify WD Ultrastar Data60 and Ultrastar Data102 hybrid storage systems for use with the VDURA Data Platform.

The effort targets organisations building AI and high-performance computing environments that need large storage pools for training and inference. It also reflects a broader shift toward mixed storage estates that combine flash for speed with hard drives for capacity.

As AI datasets reach petabyte scale, storage cost and data movement have become major constraints. Many teams have turned to flash to increase throughput, but all-flash architectures can become expensive as capacity grows. VDURA and WD are positioning a configuration that combines flash and hard drive tiers under a single file system.

Hybrid platforms

WD's Ultrastar Data60 and Ultrastar Data102 are high-density enclosures designed for mixed-media deployments. The Data60 offers 60 drive bays in a 4U chassis and supports up to 1.92PB of raw capacity when populated with 32TB hard drives, according to WD.

The Data102 scales to 102 drive bays in the same 4U footprint and can reach 3.26PB of raw capacity with 32TB hard drives, WD says.

Both systems use 12Gb/s SAS-3 host connectivity and include redundant, hot-swappable power supplies, fans, and I/O modules. WD also cites IsoVibe vibration isolation and ArcticFlow thermal zone cooling as design features.

Each enclosure supports up to 24 hybrid SSD slots, allowing hard drives and SSDs in the same chassis. The platforms can also be daisy-chained to add capacity.

WD has also signalled a hard drive roadmap of 40TB and above. VDURA and WD say future platform generations will align with that roadmap, increasing raw capacity per chassis as larger drives become available.

VDURA software

The VDURA Data Platform is a software-defined parallel file system built on a microservices architecture. It separates the control plane from the data plane, an approach VDURA says helps scale storage while keeping management functions distinct from data operations.

VDURA's DirectFlow client provides POSIX-compliant access and cache-coherent parallel I/O between compute servers and storage nodes. VDURA contrasts this with traditional NFS-based designs, which it says can become a bottleneck as GPU clusters expand.

The software includes intelligent tiering, moving data between flash and hard drive tiers based on access patterns. In the configuration described, VDURA's V5000 all-NVMe flash tier handles higher-throughput activity, while the WD Ultrastar hybrid systems provide the capacity layer.

VDURA describes the result as a single namespace across both tiers, intended to simplify data management for AI pipelines where datasets may shift between performance and capacity storage over time.

Chris Girard, Vice President of Product Management at VDURA, framed the collaboration as a response to cost and performance constraints in large AI systems.

"The AI infrastructure market demands storage solutions that balance performance with economics at scale," said Girard. "Our collaboration with WD enables customers to deploy petabyte-scale mixed fleet architectures where VDURA's parallel file system intelligently tiers data across flash and HDD, keeping GPUs fed while dramatically reducing the cost per terabyte of their AI data pipelines."

WD emphasised the balance between hard drive density and data management at scale. Scott Hamilton, Senior Director of Product Management, Marketing and Customer Experience at WD, linked the qualification work to building AI data pipelines across different media types.

"AI infrastructure requires storage architectures that balance performance with economics at massive scale," said Hamilton. "By qualifying our Ultrastar Data60 and Data102 Hybrid Storage Platforms within the VDURA architecture, we're enabling customers to build AI data pipelines that combine the density and cost-efficiency of HDD with the intelligent data management needed to keep GPUs fully utilized."

Economics tool

VDURA also highlighted a tool it calls the Storage Economics & Flash Volatility Index. It describes the tool as a free interactive calculator that tracks quarterly flash pricing and models cost scenarios across different storage architectures.

The calculator compares all-flash designs with mixed configurations using varying proportions of SSD capacity. VDURA says it draws on market data to illustrate trade-offs between performance, capacity, and cost for teams planning AI storage deployments.

Neither company disclosed commercial terms or deployment timelines. Certifying the Ultrastar Data60 and Data102 systems within the VDURA architecture defines a supported configuration that customers and integrators can use as they expand AI and HPC storage environments.