VAST Data Launches LightSpeed Storage Architecture

September 15, 2020

By Bio-IT World Staff 

September 15, 2020 | VAST Data has launched its next generation storage architecture, LightSpeed, which combines the light-touch of VAST’s simple NAS appliance experience with leadership-class performance for GPU-based and AI Processor-based computing, taking the guesswork and configuration out of scaling up AI infrastructure.

VAST calls LightSpeed both a product and a concept, designed to address all facets of storage deployment that system architects face when designing next-generation AI infrastructure, that includes: 

  • Flash at HDD Economics to Eliminate Tiering: VAST’s Disaggregated, Shared Everything (DASE) storage architecture was invented to break down the barriers to performance, capacity and scale - where revolutionary flash economics make it finally possible to afford flash for all of a customer’s training data. By eliminating storage tiering, the exabyte-scale DASE cluster architecture ensures that every I/O request is serviced in real-time, thereby eliminating the storage bottleneck for training processors.
  • 2X Faster AI Storage HW Platform: VAST’s new LightSpeed NVMe storage enclosure delivers 2X the AI throughput compared to previous generations of VAST hardware. This makes the system ideal for read-heavy, random I/O workloads like those found within modern deep learning frameworks, such as computer vision (PyTorch, Tensorflow, etc.), Natural Language Processing (BERT), Big Data (Apache Spark), Genomics (Parabricks)  and more. LightSpeed enclosures help customers increase their computing power with less hardware.
  • VAST is Pioneering NFS Support for Nvidia GPUDirect Storage: VAST supports best-in-class performance with NVIDIA GPUDirect Storage technology, currently in beta. GPUDirect enables customers running NVIDIA GPUs to accelerate access to data and avoid extra data copies between storage and the GPU by avoiding the CPU and CPU memory altogether. In initial testing, VAST demonstrated over 90GB/s of peak read throughput via a single NVIDIA DGX-2 client, nearly 3X the performance of VAST's NFS-over-RDMA and nearly 50X the performance of standard TCP-based NFS. By eliminating all bottlenecks, VAST and NVIDIA have paved the way for GPU-efficient computing using the industry standard NFS file system. 

In addition, the company also announced LightSpeed cluster configurations that make it simple for customers and partners to balance their GPU cluster scale with capacity & performance-rich storage:

  • The Line: 2 x LightSpeed Nodes pair 80GB/s of storage bandwidth for up to 32 GPUs
  • The Pentagon: 5 x LightSpeed Nodes pair 200GB/s of storage bandwidth for up to 80 GPUs
  • The Decagon: 10 x LightSpeed Nodes pair 400GB/s of storage bandwidth for up to 160 GPUs

“Today, VAST is throwing down the gauntlet to declare that customers no longer need to choose between simplicity, speed and cost when shaping their AI infrastructure strategy,” said Jeff Denworth, Co-founder and VP Product at VAST Data, in a press release distributed this morning. “Let data scientists dance with their data by always feeding their AI systems with training and inference data in real-time. The LightSpeed concept is simple: no HPC storage muss, no budgetary fuss, and performance that is only limited by your ambition."