Deploy all-flash infrastructure for all your active data sets. VAST's new advancements in commodity flash management, data protection and reduction deliver a lower TCO than hybrid storage alternatives.
Overview
Universal Storage for Deep Learning
The Data Platform for the AI Era
Universal Storage is the data foundation trusted by world-leading AI research teams to deliver the scale, speed and simplicity needed to train neural networks and infer in real-time. Universal Storage enables greater statistical accuracy by removing the barriers to training on all of an organization’s data at any scale.
Get the power of an HPC file system with the simplicity of a NAS. Ready all your data for the age of AI. Universal Storage is a next-generation distributed data store, unifying file & object storage into one scalable, affordable all-flash system.
Overview
The new age of computing requires a new approach to data.
Each and every industry is at the dawn of a new AI-powered era thanks to the exponential advancements in deep learning. As AI becomes woven into the enterprise computing agenda, IT organizations are now faced with new and unprecedented data and storage challenges. Modern AI processors demand unprecedented bandwidth and the IOPS needed to randomly walk through massive training data sets. These pathological AI workloads quickly overwhelm conventional NAS technologies, resulting in application bottlenecks and stranded processor performance. Neither traditional NAS nor HPC parallel file systems were designed to make flash affordable enough to be used as a single, global storage tier, so customers find themselves tiering data to HDDs, which is a complicated proposition because AI pipelines cannot prefetch random read operations.
With a unique disaggregated and shared everything (DASE) architecture, VAST has rethought every aspect of storage system design to break all the tradeoffs of DAS and shared-nothing architectures, delivering a system that provides unprecedented price-performance and exabyte scale. The result is a highly scalable and affordable all-flash, file and object platform that allows you to run petabyte and exabyte scale analysis at less than half of the cost of traditional all-flash solutions, while being many times faster.
VAST delivered an all-flash solution at a cost that not only allowed us to upgrade to all-flash and eliminate our storage tiers, but also saved us enough to pay for more GPUs to accelerate our research. This combination has enabled us to explore new deep-learning techniques that have unlocked invaluable insights in image reconstruction, image analysis, and image parcellation both today and for years to come.
Bruce Rosen
Executive Director, Martinos Center for Biomedical Imaging
Key Benefits
Consolidate tiers to unleash AI model accuracy.
For the first time in 20 years it became possible to develop a new type of distributed system architecture. VAST's Disaggregated and Shared Everything (DASE) storage system was built from the ground up to eliminate the trade-offs imposed by legacy scale-out architectures. The Universal Storage platform was engineered to be fast-enough for your most demanding ML/DL pipelines, scalable-enough to manage all your data, and affordable-enough such you never need tier your data stores again. Universal Storage is a simple and scalable enterprise NAS where every I/O is a real-time I/O, making it possible to optimize AI training and inference without IT operations compromise.
All Flash, at Archive Economics
Scale Without Feature Compromise
Get near-limitless snapshots, enterprise replication, user and admin auditing, remote monitoring, online upgrades and expansions and so much more. Break the scalability & data management simplicity compromise.
A Unified, Multi-Protocol Platform
Process data to any application: data can be delivered across NFS, SMB and S3. The VAST Element Store is a universal permissions management model that arbitrates access across supported protocols.
Scale to Exabytes
Scale performance linearly as your data volumes grow. Access is optimized to deliver the random access needed to keep deep learning & machine learning pipelines running efficiently.
AI-Optimized Client Access
With support for RDMA and GPUDirect Storage access, VAST’s NAS experience delivers the performance of a parallel file system without any of the parallel file system complexity.
Multi-Tenant Infrastructure
Dedicate front-end servers and the performance they provide to the most critical projects. VAST's server pooling capability provides dedicated Quality of Service for competing projects.
Reference Architecture
HPC performance meets enterprise storage simplicity.
Resources
Innovation begins with understanding
Solution Brief
Universal Storage for Deep Learning
See how the linear performance and capacity scaling of VAST Universal Storage simplifies and accelerates ML/DL pipelines
2 pages
White Paper
GPU Reference Architecture
Deep Learning demands modern IT infrastructure, unsuitable for traditional compute or legacy storage. VAST's Universal Storage reference architecture meets AI/ML I/O demands and integrates with NVIDIA DGX A100 systems, Mellanox Quantum InfiniBand and Spectrum Ethernet switches.
6 pages