Communication Optimization Beyond MPI

Abstract

The Message Passing Interface (MPI) is the de-facto standard for parallel processing on high-performance computing systems. As a result, significant research effort has been spent on optimizing the performance of MPI implementations. However, MPI's specific semantics can limit performance when using modern networks with Remote DMA capabilities. We propose a compiler-assisted approach to optimization of MPI-based applications by transforming MPI calls to a one-sided (as opposed to message passing) communication model. In this paper we present a research plan for developing new optimizations using this approach, then show preliminary results with up to a 40% increase in bandwidth over MPI by using a simpler one-sided communication model.