Predicting the Performance of MPI Applications over Different Grid Architectures

Abstract

Nowadays, the high speed and accurate optimization algorithms are required. In most of the cases, researchers need a method to predict some criteria with acceptable accuracy to use it after in their algorithms. However, in the field of parallel computing the execution time can be considered the most important criteria. Consequently, this paper presents new execution time prediction model for message passing interface applications execute over numerous grid scenarios. The model has ability to predict the execution time of the message passing applications running over any grid configuration in term of different number of nodes and their computing powers. The experiments are evaluated over SimGrid simulator to simulate the grid configuration scenarios. The results of comparing the real and the predicted execution time show a good accuracy. The average error ratio between the real and the predicted execution time for three benchmarks are 4.36%, 5.79% and 6.81%.