Paper Title: Requirements and Challenges Associated with the World's Fastest Storage Platform

Paper Abstract: Distributed Asynchronous Object Storage (DAOS) is a software-defined storage platform designed from the ground up to leverage storage class memory (SCM) and NVMe-capable flash devices. It implements a low-latency, high bandwidth, distributed key-value store (KVS) upon which several middleware interfaces can be built, including POSIX, MPIIO, Apache Spark, and HDF5, thus enabling existing applications to leverage the benefits of DAOS. Each object shared in DAOS is managed by the Versioned Object Store (VOS), which supports requirements such as distributed transactions, erasure coding and replication, end-to-end checksums, persistent snapshots, and garbage collection - all while maintaining low-latency and high-bandwidth requirements. This presentation will discuss VOS design and implementation, and will present the challenges in meeting these various requirements.

Paper Author: Jeff Olivier, Software Engineer, Intel

Author Bio: Jeff is a senior engineer and architect in Intel's Super Compute Group working on DAOS, a next generation, all flash storage platform tackling requirements of modern, data intensive workloads. He has been at Intel since 2002 and has worked on several software products including DAOS, Parallel Studio's Inspector and Advisor, and the OpenMP runtime library. He holds a Master's degree from the University of Illinois at Urbana-Champaign and a Bachelor's degree from Brigham Young University, both in Computer Science.