Benchmarking performance: in uence of task location on cluster throughput

Abstract

A variety of properties characterizes the execution of scientific applications on HPC environments (CPU, I/O or memory-bound, execution time, degree of parallelism, dedicated computational resources, strong- and weak-scaling behaviour, to cite some). This situation causes scheduling decisions to have a great influence on the performance of the applications, making difficult to achieve an optimal exploitation with cost-effective strategies of the HPC resources. In this work the NAS Parallel Benchmarks have been executed in a systematic way in a modern state-of-the-art and an older cluster, to identify dependencies between MPI tasks mapping and the speedup or resource occupation. A full characterization with micro-benchmarks has been performed. Then, an examination on how different task grouping strategies and cluster setups affect the execution time of jobs and infrastructure throughput. As a result, criteria for cluster setup arise linked to maximize performance of individual jobs, total cluster throughput or achieving better scheduling. It is expected that this work will be of interest on the design of scheduling policies and useful to HPC administrators