State of I/O in HPC 2020

Authors: Keith A. Bateman (Illinois Institute of Technology, Lawrence Livermore National Laboratory); Stephen Herbein (Lawrence Livermore National Laboratory); and Anthony Kougkas and Xian-He Sun (Illinois Institute of Technology)

Abstract: Modern supercomputers are designed to allow users access to computing and I/O resources that exhibit better speed, higher scales and increased reliability compared to that available in commodity hardware (i.e., cloud infrastructures). The current era is one of transition from petascale to exascale, and it's important for researchers to study this era in particular to gain an understanding of modern I/O behavior. An understanding of modern behavior will motivate current I/O research and inform future design of storage systems at exascale. In this project we perform a comprehensive statistical analysis of historical data to drive insights about I/O behavior of modern applications. Results contribute to the HPC community in several ways, including informing software system design, guiding hardware purchases and architectural design, diagnosing common problematic application behaviors and motivating and supporting an improvement of job scheduling via the classification of I/O behavior.

Best Poster Finalist (BP): no

Poster: PDF
Poster summary: PDF


Back to Poster Archive Listing