Vdura Controls Costs While Providing Available And Durable AI Storage

Introduction to Vdura: Revolutionizing AI and High-Performance Computing Storage

Vdura, formerly known as Panasas, has undergone a significant transformation in its business model, shifting from proprietary hardware to software-defined storage solutions. This strategic move aims to cater to the growing demands of artificial intelligence (AI) and high-performance computing (HPC). Founded in 2000 by Garth Gibson, the co-inventor of RAID and creator of the first Linux-based parallel file system, PanFS, the company has a rich history in developing high-performance storage systems.

Vdura’s Software-Defined Data Infrastructure Platform

Vdura’s software-defined data infrastructure platform is designed to provide a unified global namespace with a single control plane and data plane. This platform combines the scalable, linear performance of a parallel file system with the resilience and cost efficiency of object storage. By doing so, Vdura enables sustained throughput at scale, which is essential for avoiding GPU starvation and wasting investments in AI workloads. The platform also expands data metadata capacity and performance, allowing for quick access to data.

Accelerating the AI Pipeline

Vdura’s platform accelerates all parts of the AI pipeline, including data ingesting, model loading, training, checkpointing, fine-tuning, and inference. This comprehensive approach sets Vdura apart from its competitors, which often focus on specific parts of the pipeline. By supporting the entire AI workflow, Vdura helps organizations optimize their AI investments and improve overall efficiency.

Vdura Data Platform (VDP) and Its Benefits

The Vdura Data Platform (VDP) can turn thousands of storage servers into a high-performance, resilient, and durable data platform. VDP is resistant to device and node failures, making it an attractive solution for organizations that require high availability and durability. The platform is also easy to deploy and manage, providing seamless integration of NVMe for fast access layers and optional hard disk drives (HDDs) for cost-efficient storage capacity.

Cost-Effective Storage Solutions

Vdura’s use of SATA HDDs instead of serial attached SCSI (SAS) HDDs, which are commonly used by hyperscalers for object storage, helps reduce costs. The company’s NVMe flash + HDD architecture also lowers long-term costs compared to all-flash competitors, especially when using commodity SSDs and HDDs. Additionally, Vdura’s Vdura-bility guarantee and Vdura Care Premier provide guaranteed service-level agreements (SLAs) for availability and durability, along with a 10-year cover and device replacement.

Enabling Fast and Reliable Performance

By combining SSDs and HDDs, Vdura offers a flexible and adaptable ratio of HDDs and flash in a system. This approach enables fast and reliable performance, making it an ideal solution for AI and HPC applications. Vdura recommends an 80%/20% HDD to SSD mix for most customers, which provides an optimal balance between cost and performance.

High Availability and Durability

To get the most out of GPUs, the storage system supporting them must have high availability, often referred to as the percent uptime. Vdura’s storage solutions provide high availability, which is critical for minimizing downtime and reducing the incidence and impact of data loss or corruption. Although storage accounts for only about 20% of AI spending, it enables the use of the 50% server/GPU investment, making it a crucial component of AI infrastructure.

Partnership with AMD and Future Developments

In September, Vdura announced a validated reference architecture with AMD, designed to eliminate data bottlenecks and simplify large-scale deployments for AI and HPC. This collaboration combines Vdura’s V5000 storage with AMD’s Instinct MI300 series accelerators. Vdura continues to innovate and improve its solutions, offering lower capital and operating costs, high storage availability, and durability for demanding at-scale AI applications.