Protocol and performance analysis of the MPC parallel computer

This paper presents the MPC parallel computer and its MPI implementation performed at the Laboratoire LIP6 of Univ. Pierre and Marie Curie, Paris. MPC is a low cost and high performance parallel computer using standard PC main-boards as processing nodes connected through the specific FastHSL board to a high speed communication network using HSL 1 Gbits/s serial links, IEEE 1355 compliant. Two Asics are presented: RCUBE which is the HSL network router and PCI-DDC the network controller implementing the Direct Deposit State Less receiver protocol. The software part of the MPC parallel computer consists of 2 zero-copy layers leading to a latency of 5 to 40 μs and a throughput of 490 Mbits/s. An efficient MPI implementation based on MPICH is presented and evaluated on an MPC parallel computer. Measures show a latency of 26 μs and an useful throughput of 450 Mbits/s This paper presents also a performances study of the MPI implementation on the MPC computer. This reveals Several possible optimizations to reduce the overall MPI transfer latency on the MPC Computer