Paper Title: NVMe-oF File System Accelerating Machine Learning Workloads

Paper Abstract: Any IT infrastructure that supports a machine learning workflow needs to be able to handle a large number of files, and large amounts of data storage with high throughput access to all the data. Legacy file systems can't supply high throughput and high file IOPS, as they were designed for HDD and are not suitable for low latency, small file IO and metadata heavy workloads that are common in machine learning. This results in I/O starvation to the GPUs and CPUs, a major problem for a machine learning system. We'll introduce a next-gen massively parallel shared file system that is NAND FLASH and NVMe-optimized, which is solving the I/O starvation problem. What you will learn: - Case study: how a large autonomous driving car manufacturer maximizes their investment in GPUs by ensuring they are saturated with data - How to make I/O compute bound again - How next-gen file systems are utilizing flash to turn the compute layer into a high performance storage system: keeping hot data closer to the applications.

Paper Author: Liran Zvibel, CTO And Co-Founder, WekaIO

Author Bio: As Co-Founder and CEO, Mr. Liran Zvibel guides longer range technical strategies at WekaIO. Prior to creating the opportunity at WekaIO, he ran engineering at social startup and Fortune 100 organizations including Fusic, where he managed product definition, design and development for a portfolio of rich social media applications. Liran also held principal architectural responsibilities for the hardware platform, clustering infrastructure and overall systems integration for XIV Storage System, acquired by IBM in 2007. Mr. Zvibel holds a BSc.in Mathematics and Computer Science from Tel Aviv University.