A Comparison of Application Performance Using Open MPI and Cray MPI

Open MPI is the result of an active international Open-Source collaboration of Industry, National Laboratories, and Academia. This implementation is becoming the production MPI implementation at many sites, including some of DOE’s largest Linux production systems. This paper presents the results of a study comparing the application performance of VH-1, GTC, the Parallel Ocean Program, and S3D on the Cray XT4 at Oak Ridge National Laboratory, with data collected on up to 1024 process runs. The results show that the application performance using Open MPI is comparable to slightly better than that obtained using Cray-MPI, even though platform specific optimizations, beyond a basic port, have yet to be done in Open MPI.