Considerations for LS-DYNA Efficiency in SGI IRIX and Linux Environments with a NUMA System Architecture

Stan Posey, Nick Meng SGI, Mountain View, CA Manufacturing industry and research organizations continue to increase their investments in structural analysis and impact simulations such that the growing number of LS-DYNA users continue to demand more from computer system environments. These demands typically include rapid single job turnaround and multi-job throughput capability in a multi-user high-performance computing (HPC) environment. What is more, many LS-DYNA simulation environments coexist with other mechanical computer aided engineering (MCAE) software for structural analysis and computational fluid dynamics that all compete for system resources such as CPU cycles, system bandwidth, memory, disk storage and I/O. This paper examines the computational efficiency of structural analyses and simulations for relevant applications in LS-DYNA. Parameters such as model size, element types, and simulation conditions can produce a wide range of computational behavior such that consideration should be given to how system resources are allocated and configured. The computational characteristics of the SGI® Origin® 3000 servers, based on IRIX® and MIPS®, and the SGI® Altix™ 3000 servers, based on Linux® and Itanium® 2 from Intel®, are examined for both turnaround and throughput requirements that include industrial-size examples. In addition, simple guidelines on proper system configuration and innovative use of available SGI system resource management tools are provided that are designed to maximize LS-DYNA productivity. Introduction Mechanical design and manufacturing organizations increasingly rely on highperformance computing (HPC) technology and mechanical computer-aided engineering (MCAE) applications to drive innovation in product development.