A Statistical Analysis of HPC Network Tuning.

Distributed scientific applications run on a complex stack of software and network technologies. Each layer has configuration options for tuning performance. These can range from protocol thresholds to algorithmic changes for collectives. Micro-benchmarks are a common methodology to evaluate the communication stack and are relatively easy to tune. However they aren’t representative of application behavior. Proxy applications, however, offer a simplified, but realistic, representation of the main computational and communicative methods in scientific programs. Since these proxy applications contain realistic message passing patterns, the correlation between micro-benchmarks and proxy application performance is not obvious. In this paper we present a study of statistically analysing the impacts of tuning. Our results show how tuned micro-benchmark performance correlates with tuned proxy application performance.