Analysis and Optimization of Pipelined Broadcast Algorithms on Gigabit Ethernet and InfiniBand Networks

Theoretical and experimental analysis of MPI Bcast algortihms is presented. The optimal tree degrees and segment sizes for pipelined versions of algorithms are obtained. Algorithms were investigated according to their implementation in the Open MPI library. Theoretical results are consistent with experiments on a computer cluster with Gigabit Ethernet and InfiniBand communication networks.