Mad-MPI: An Efficient Implementation of MPI for Fast Networks

Runtime

LaBRI, INRIA Bordeaux - Sud-Ouest

High Performance Runtime Systems for Parallel Architectures

Download

Powered By GForge Collaborative Development Environment

» Mad-MPI files are hosted on the PM2 project at gforge.

» Documentation on how to install, and use Mad-MPI is available in HTML format or PDF format.

Mad-MPI Overview

Mad-MPI is a new light implementation of the MPI standard. This simple, straightforward proof-of-concept implementation is a subset of the MPI API, that allows MPI applications to benefit from the NewMadeleine communication engine. Mad-MPI is based on the point-to-point nonblocking posting (isend, irecv) and completion (wait, test) operations of MPI, these four operations being directly mapped to the equivalent operations of NewMadeleine.

Mad-MPI also implements some optimizations mechanisms for derived datatypes (see here). MPI derived datatypes deal with noncontiguous memory locations. The advanced optimizations of NewMadeleine allowing to reorder packets lead to a significant gain when sending and receiving data based on derived datatypes.

Contact

For any questions regarding Mad-MPI, please contact Alexandre Denis.
Alexandre.Denis@inria.fr