Скачать или смотреть New Linux Standards for High Performance & Distributed Data in AI Architectures

"Gary Grider (Leader Of Hpc Division) - Los Alamos National Laboratory
David Flynn (Founder & Ceo) - Hammerspace

This session will delve into webscale and large HPC center adoption of the latest advancements in NFS and Linux servers for AI data storage and orchestration use cases.\nWe will discuss the performance requirements needed to power GPU-based computing use cases for AI/DL and other high-performance workflows. Historically- such use cases have needed to deploy parallel file systems such as Lustre or others- which require networking and skillsets not typically available in standard Enterprise data centers.\nThe Linux community is actively developing advancements to the embedded parallel file system client that is part of the NFS protocol. With parallel NFS v4.2- standard Linux clients now can read and write directly to the storage- scale out performance linearly for both IOPS and throughput to saturate the limits of both storage and network infrastructures- and orchestrate data across systems and locations with non-disruptive client access to the data in flight or once moved."