Abstract
Performance analysis is an important issue when developing a new program, because information obtained from the analysis can help to develop efficient programs. An essential component of performance analysis is the computation model, which can be used to model the parallel programs of interest. We can predict the performance if the program fits this model.In this thesis, we study the use of a computation model, Bulk-Synchronous Parallel(BSP) model, to correctly predict the performance of the programs using MPI collective communication calls on PC clusters. The BSP model is portable and efficient computation model, which fits our purpose due to its synchronous characteristic. Our study is based on this model and revise it to fit the MPI collective communication calls. We also use two applications to prove the feasibility of the revised model.