Analyzing the Performance of MPI in a Cluster of Workstations Based on Fast Ethernet

Recent improvements in LANs make network of workstations a good alternative to traditional parallel computers in some applications. MPI (Message Passing Interface) standard library implementation for network of workstation holds great promise in providing both portability and performance. However, some authors point out if this hardware improvements are not supported with a reduction of software overhead, this overhead will dominate communication cost, and many applications may no bene t from the advancements in network hardware. In this paper we present an exhaustive study of MPICH implementation of MPI standard for this environment, analyzing added overhead of each layer. Finally, we propose modi cations to network interface to reduce the xed overhead of transmission.