Improving Core Selection on a Multicore Cluster to Increase the Scalability of an Atmospheric Model Simulation

Abstract

This work shows that the we can improve the application speed-up curve up to a maximum number of cores allocation according to two Ocean-Land-Atmosphere Model (OLAM) workload parameters on a multicore cluster environment. Previous experiments have shown that the scalability of the system is limited by output operations performance. We show that one of the reasons for such output operations overhead is the low aggregated I/O throughput capacity of the multicore system. A higher capacity would be desirable to achieve better concurrent access rates. We show that, for a given atmospheric model application configuration, there is an maximum number os cores allocation that improves application speed-up curve.