Grid enablement and performance evaluation for simulators of nanoscale measurements of material properties

We have enabled an MPI-based parallel program which is developed by our group to simulate nanoscale measurements of materials properties for a grid computing environment. The performance of this program has been evaluated based on three kinds of MPI middleware, MPICH-G2, GridMPI, and PACX-MPI. The speedup of this program on the grid is comparable with that obtained on supercomputers when CPU number is less than 200, and PACX-MPI gives the best performance and reliability among the three kinds of middleware in our study.