An empirical study of MPI over PC clusters

Message Passing Interface (MPI) is an important mechanism in P2P. Herein we assess how different types of MPI collective communication functions perform on a Gigabit Ethernet Homogeneous Beowulf PC cluster. In this way we provide an insight on the factors that affect P2P applications over an enterprise context such as the emerging Cloud-based services. By contrast to the literature, which includes mostly theoretical studies, we carry out an empirical study. We show that the behaviour of scatter and gather are most unpredictable in comparison with other collective functions