Comparison of different approaches for matrix vector multiplication on transputer network-a performance analysis

Various parallel algorithms of matrix vector multiplication are available. The algorithms have been implemented on different machines of varying specifications and corresponding performance are analyzed. However, the analyses given in the literature are not performed on a common benchmark which renders comparison of efficiency a difficult task. Most often, when analytical model is used, the model does not consider the case of running multiple processes in one processing element which is a common practice in a multicomputer environment. This paper therefore attempts to present a simple analytical model which can be used to compare the efficiency of various parallel versions of matrix vector multiplication on a transputer network.<<ETX>>