A Performance Model of I/O-Intensive Parallel Applications

High performance computing (HPC) is widely used in science and engineering to solve large computation problems. The peak performances of computers increase in a continuous and rapid way. But the sustained performances achieved by real applications do not increase in the same scale as the peak performances do and the gap between them is widening. Performance model of parallel systems, which is one of effective ways to solve this problem, draws the attentions of the research community as well as the industry community. In this paper, an open performance model infrastructure PMPS(n) and a realization of this infrastructure—PMPS(3), aperformance model of I/O-intensive parallel application, are given and used to perform NPB benchmarking on PⅣ cluster systems. The experiment results indicate that PMPS(3) can forecast better than PERC for I/O intensive applications, and can do as well as PERC for storage-intensive applications. Through further analysis, it is indicated that the results of the performance model can be influenced by the data correlations, control correlations and operation overlaps. Then such factors must be considered in the performance models to improve the forecast precision. The experiment results also show that PMPS(n) has very good scalability.