Benchmarking of three parallelized implementations of LS-Dyna on a HPC server cluster

摘要

This paper discusses the benchmarking of three parallelized implementations of the popular LS-Dyna® finite element code (Livermore Software Technology Corp.) on the STePS2 high performance computing (HPC) server cluster. SMP, MPP and SMP-MPP hybrid implementations of LS-Dyna® are benchmarked over various numbers of nodes, cpus and cpu cores.ududThe STePS2 HPC cluster consists of a dual-cpu head node and 16 homogeneous dual-cpu compute nodes. All nodes are interconnected via an Infiniband® network fabric. All cpus are quad-core Intel Xeon 64-bit and the operating system(s) is the ROCKS cluster management software implemented through Redhat Enterprise Linux 5 (x86_64). SMP, MPP and MPP-SMP hybrid implementations of LS-Dyna were compiled for use with the HPMPI message passing interface.