The methods of precise measurement of the loop nests' execution time during JavaMPI-programs analysis in ParJava environment

The methods for estimating execution time of the model of a parallel program using instrumental computer are discussed. The methods are based on accurate prediction of the execution time of fragments of the parallel program using the target computational platform. The model was developed for SPMD programs using explicit data exchange by Java MPI library and is the part of ParJava IDE. Certain kinds of loops (homogeneous, reducible) are marked out in the model and then estimated on the node of a target computational platform (high performance cluster). The technique allows to reduce prediction error and to accelerate the model simulation on the instrumental computer.