NFS Acceleration

Supercharged Without Losing the Simplicity

images
Overview

Why NFS Acceleration Matters

NFS, the Network File System, has been the standard file-sharing protocol for open systems since 1984. NFS simplicity led to widespread adoption for many traditional data workloads but it was considered too slow for performance intensive applications in HPC and AI where complex Parallel File Systems were thought to be the only option.

With AI becoming mainstream more customers are considering parallel file systems and finding that it’s a poor fit for enterprise support models with complex tuning and requiring disruptive upgrades (even when consumed as a service from major cloud vendors)​.

VAST accelerated NFS removes the limitations of legacy NFS without the need for proprietary clients so that every customer can support next-gen AI-driven workloads.

images
How It Works

Breaking the limits

Breaking the TCP Session Limit with nconnect

The first step to boosting NFS performance is to simply spread the traffic between an NFS client and server across multiple TCP sessions.

NFS Over RDMA for Throughput

RDMA (Remote Direct Memory Access) transfers data across the network directly from the memory of the local computer into the memory of the remote computer, eliminating the latency of making copies of the data in TCP/IP.​

GPU Direct Storage for Analytics and AI Performance

NVIDIA’s GPU Direct Storage reduces the overhead of accessing storage. Switching to GPU Direct Storage not only raised throughput but also reduces the amount of CPU required.​

images
Comparison

VAST Removes the Performance Constraints of NFS Making an Ideal Choice for Al Enabled Workloads

images
images