VDURA Data Platform

The VDURA Data Platform is a software-defined storage infrastructure designed for AI and high-performance computing (HPC) workloads, combining a true parallel file system with object-storage resilience. It targets organizations managing data-intensive pipelines—from research labs and government institutions to Fortune 500 companies—that require high throughput, low-latency metadata operations, and hyperscale durability across ingest, training, checkpointing, and inference stages. Born from the Panasas PanFS architecture, which defined the pNFS standard over 20 years ago, VDURA evolved into a microservices-based platform that decouples control and data planes. The platform is offered as a software subscription, with the V5000 system providing integrated hardware options including all-flash and hybrid storage nodes.

The architecture separates director nodes (for metadata and control) from storage nodes (for performance), enabling linear scalability to thousands of nodes. VPOD virtualization groups storage nodes for resilience and efficiency, while dynamic data acceleration uses a smart storage fabric to optimize data placement. AI-aware placement automates tiering based on workload patterns, and V11 data reduction compresses data without performance penalties. V-ScaleFlow, introduced in V11.2, orchestrates data movement between 128 TB QLC NVMe SSDs and 30+ TB hyperscale HDDs, using V-Burst to absorb write spikes from AI checkpoints and then writing sequentially. This reduces flash requirements by over 50% and halves power consumption per petabyte compared to all-flash arrays. The platform also includes per-file RAID, erasure coding, and end-to-end encryption for data in transit and at rest.

VDURA competes directly with NetApp’s AFX, which CEO Ken Claffey criticized at INSIGHT 2025 as still built on a general-purpose architecture. Unlike NetApp’s approach, VDURA’s true parallel file system avoids the metadata bottlenecks common in NAS-based solutions. Other competitors include Dell PowerScale and IBM Storage Scale, but VDURA differentiates through its unified parallel-object design and AI-aware automation. The platform’s legacy in HPC—deployed across thousands of sites—gives it credibility in demanding environments, though it lacks the broad ecosystem integrations of cloud-native alternatives like MinIO or Pure Storage’s FlashBlade.

Key trade-offs include a higher initial learning curve for teams unfamiliar with parallel file systems, as configuration requires understanding director and storage node roles. The platform’s software subscription model may have higher long-term costs for static workloads compared to traditional hardware purchases. While V-ScaleFlow reduces TCO by 60% versus flash-only competitors, the hybrid tiering adds latency for hot data not cached on NVMe. Operational overhead is low—0.5 FTE for 1–100 PB—but scaling beyond 100 PB may require additional planning for network topology and power. Finally, the platform’s reliance on commodity hardware means performance depends heavily on the underlying network fabric, typically requiring InfiniBand or high-speed Ethernet.

Software subscription model; no public per-tier pricing. V5000 hardware bundles available upon request. V11.2 includes V-ScaleFlow at no extra cost.