Performance analysis for MPI applications with SCALEA

The performance of message passing programs can be challenging to comprehend. In previous work we have introduced SCALEA, which is a performance instrumentation, measurement, analysis, and visualization tool for parallel and distributed programs. In this paper we report on experiences with SCALEA for performance analysis of two realistic MPI codes taken from laser physics and material science. SCALEA has been used to automatically instrument - based on user provided directives - the source codes, to compute performance overheads, to relate them to the source code, and to provide a range of performance diagrams in order to explain performance problems as part of a graphical user interface. Multiple-experiment performance analysis allows to compare and to evaluate the performance outcome of several experiments which have been conducted on a SMP cluster architecture.