Birds of a Feather

IO500: The High-Performance Storage Community

Authors: Andreas Dilger (Whamcloud/DDN), Dean Hildebrand (Google Cloud), Julian Kunkel (GWDG), Jay Lofstead (Sandia National Laboratories), George Markomanolis (Advanced Micro Devices (AMD) Inc), Shuichi Ihara (DataDirect Networks (DDN), Whamcloud Inc), Hendrik Nolte (GWDG, Germany)

Abstract: As efficient IO becomes increasingly critical to reach peak computing performance, IO500 has become the de-facto standard for measuring HPC storage performance. Developed in 2017, the IO500 has released bi-annual lists at SC and ISC since then. This BoF’s highlight is the presentation of the new IO500 list.

This BoF’s goal is to foster the IO500 community to progress common goals of creating, sharing, and benefiting from a large corpus of shared storage performance data. We are also building a detailed repository of high-performance production storage systems as they evolve, providing a knowledge base for HPC researchers and system designers.

Long Description: Goals of the BoF are to 1) reveal the new IO500 list and provide highlights and insight on the state of HPC storage; 2) interact with the HPC storage community, discussing and steering the direction of community efforts and latest developments; 3) discuss the benefit and direction of community efforts in enhancing the benchmark and gain a better understanding of the results.

The IO500 benchmark consists of both data and metadata benchmarks to identify performance boundaries for optimized and suboptimal HPC applications. Evaluating fundamental storage performance characteristics across the full range of HPC I/O and Machine Learning patterns provides concrete guidance to users on new and emerging HPC storage systems. Together with comprehensive information about the supercomputers and storage, the community can analyze the in-depth system characteristics tracked by the list. In contrast to other lists, the IO500 collects the configuration and execution scripts for providing reproducibility information and a means of result verification, while sharing best practices for HPC sites. Through our communication channels, we exchange tuning and configuration ideas with community members to maximum storage performance.

The IO500 has released one official list at ISC HPC and one at Supercomputing at IO500 BOFs since 2017, with attendance regularly in the 50-80 person range. Moreover, the IO500 benchmark is used by numerous HPC sites to evaluate storage performance for RFCs and benchmark against actual application performance.

Current IO500 community efforts include: * creation of Production and Research lists * improved reproducibility documentation to score and categorize submissions * addition of new benchmarks to cover evolving HPC I/O characteristics * improving the submission mechanism to make it easier for contributors * easing access and analysis of IO500 datasets

Expected HPC audience are 1) I/O experts and developers from industry, academia, and government laboratories, 2) researchers/engineers working on high-performance I/O, 3) domain scientists and computer scientists interested in discussing I/O issues.

The outcome of this BoF will steer the direction of the community efforts.

Website: https://io500.org/pages/bof-sc23