A case study of scientific application I/O behavior

Abstract

Characterizing the I/O demands of scientific applications is an integral part of the solution to the I/O bottleneck problem in high performance systems. Knowledge of the type, volume, and frequency of events that occur in common patterns of I/O activity can provide insight into the policies and mechanisms needed at all levels of the memory hierarchy. This paper is a detailed case study of one scientific application's dynamic I/O behavior. Results from this study show that with appropriate measurement tools, dynamic I/O behavior can be characterized and regular, recurring patterns of I/O activity can be isolated.<>