Scaling Study of LS-DYNA MPP on High Performance Servers

With LS-DYNA MPP, scalable SolarisTM operating system and the MPI library, Sun Microsystems’ StarfireTM server proved to be capable of producing a scalable solution for large-scale automotive crash simulation problems. It was found that a proper decomposition plays a significant role in achieving optimal scaling results for large-model, computation- intensive runs. Also, a large amount of external cache memory on the Starfire SMP server was found to be crucial for optimal runtime performance. With proper decomposition, a Starfire server was able to achieve 30X speedup and 93% efficiency with 32 processors in the simulation run of the NCAC Neon model consisting of 270,000 elements. Version 940.2a of LS-DYNA MPP showed good repeatability over largely different numbers of processes. It also displayed an exact repeatability on different runs when the command setting and number of processes are kept constant.