Methodology for Efficient Execution of SPMD Applications on Multicore Environments

Abstract

The need to efficiently execute applications in heterogeneous environments is a current challenge for parallel computing programmers. The communication heterogeneities found in multicore clusters need to be addressed to improve efficiency and speedup. This work presents a methodology developed for SPMD applications, which is centered on managing communication heterogeneities and improving system efficiency on multicore clusters. The methodology is composed of three phases: characterization, mapping strategy, and scheduling policy. We focus on SPMD applications which are designed through a message-passing library for communication, and selected according to their synchronicity and communications volume. The novel contribution of this methodology is it determines the approximate number of cores necessary to achieve a suitable solution with a good execution time, while the efficiency level is maintained over a threshold defined by users. Applying this methodology gave results showing a maximum improvement in efficiency of around 43% in the SPMD applications tested.