The Applicability of the Universal HP-MPI to MPP LS-DYNA on Linux Platforms

In addition to its parallel algorithm, MPP LS-DYNA for cluster computing derives its parallelism from the MPI library and the interconnect. MPI has always had a standard since its inception. On the other hand, there are diverse interconnects and switches in the market, each with unique low-level interface and performance considerations. This variety of interconnects has resulted in a difficult support burden for LSTC, the software developer, and also resulted in inconvenience for MPP LS-DYNA users. To address this difficulty in portability HP-MPI was invented. HP-MPI has enabled a single MPP LS-DYNA executable to work on most prevailing interconnects on a given Linux hardware platform, HP or non-HP. In this paper, the universal applicability of HP-MPI to diverse interconnects on Linux platforms will be verified by the MPP LS-DYNA simulation on the well- known 3-Vehicle-Collision model. Furthermore, it will be shown that the performance of HP-MPI is on a par with, and often better than, other MPI libraries. HP-MPI is ensured with the highest quality because every one of its release is qualified with a suite of 2000 tests. Thus we can safely claim that HP- MPI, being a supported product and requiring no end-users licensing, is the only MPI that MPP LS-DYNA users need on most Linux platforms.