The project will improve MPICH-G2 collective operation performance to the state of the art and also improve MPICH-G2's point-to-point TCP performance. Since MPICH-G2 is a popular package and has been used in a variety of Grid projects and applications, improvement will increase its usage and effectiveness. MPICH-G2 has been used to set and utilize QoS parameters having real-time network latency and bandwidth requirements, and has also been used in client/server scientific visualization applications. The project will implement a solution to the new symbol collision problem, which will also broaden the base to include a new class of computing platforms, the Linux Cluster. The proposed enhancements coupled with the extension of MPICH-G2's platform base to include Linux clusters, are significant improvements. The broader impact of this work will be realized by MPI application developers as the proposed activities will significantly increase the utility of MPICH-G2 as an effective middleware tool.
This tool as been used in a number of traditional scientific applications, most notably receiving the Gordon Bell Prize, Special Category, for its role in an astrophysics application used for solving problems. The work will significantly improve teh bandwidth for large messages over non-vendor MPI communications and will further promote MPI as an effective Grid API for non-traditional scientific applications.
