Performance prediction model for SPMD massively parallel computing systems

The development of a prediction process for estimating the performance of parallel algorithms designed for distributed memory multiprocessor systems is discussed in this paper. This effort is targeted in reducing the waste of computing as well as man power in developing efficient parallel algorithms by providing rough analysis on the execution time of the developed algorithms without actually running the user programs on the actual parallel machines. A preliminary version of this tool has been in use leading to relatively reliable prediction results.<
>