Performance evaluation of sorting parallel algorithms using portable message passing environments

This paper presents a performance evaluation of the portability platforms PVM and MPI when running in both a distributed system and a parallel architecture - the IBM SP2. The evaluation is performed investigating number of sorting parallel algorithms. Four implementations of MPI and PVM are also considered: IBM MPI and IBM PVMe (running in the SP2), MPICH and PVM (running in a distributed system). Several parallel algorithms were developed for the evaluation and the results are used to present a comparison study between the different environments considered and the several sorting algorithms implemented. Sequential algorithms were also analysed in order to evaluate the speedup reached for each environment. The results obtained show that the PVM presents better performance in distributed systems (PVM was actually designed to work in a set of loosely coupled computers) while the MPI is more adequate for parallel architectures.