How SPMD applications could be efficiently executed on multicore environments?

Abstract

A challenge for programmers of parallel programming environments is to execute applications efficiently. For this reason, applications with high levels of synchronism and communications such as SPMD (Single Program Multiple Data) create a challenge regarding how to distribute tasks between PE (Processing Element) in a multicore cluster; this kind of environment presents high heterogeneity in communication parameters due to different communication paths present. For this reason, this work is centered around developing a methodology to distribute SPMD tasks between PEs in a multi-core Cluster. The task assignment process is realized through mapping and scheduling strategies based on controlling the communications heterogeneities. Finally, the objective is to obtain a good execution time while maintaining the efficiency level over a threshold. The results obtained show an improvement around 40% of efficiency in a heat transfer application, when our methodology is applied.