Characterizing LS-DYNA performance on SGI® Systems using SGI MPInside MPI profiling tool

SGI delivers a unified compute, storage and remote visualization solution to manufacturing customers reducing overall system management requirements and costs. LS-DYNA integrates several solvers into a single code base. In this paper, the explicit solver is hereby studied for better matching with the multiple computer architectures available from SGI, namely, multi-node Distributed Memory Processor clusters and Shared Memory Processor servers, both of which are capable of running in Shared Memory Parallelism (SMP), Distributed Memory Parallelism (DMP) and their combination (Hybrid) mode. The MPI analysis tool used is SGI MPInside. SGI MPInside features customary communication profiling. It also features "on the fly" modeling to predict potential performance benefits of the different upgrades available from the latest Intel® Xeon® CPU, interconnect and its middleware, MPI library, and the underlying LS-DYNA source code. We will also describe how the profile-guided mpiplace component is used to minimize inter rank transfer times. Also outlined in the paper will be the SGI hardware and software components for running LS-pre/post via SGI® VizServer® with NICE Software and how, CAE engineers can now allow multiple remote users to create, collaborate, test, optimize, and verify new complex LS-DYNA simulations without moving their data.