Migrating modeling and simulation applications on to high performance computers
Mark D. Barnell, Brian J. Rahn · Proceedings of SPIE, the International Society for Optical Engineering/Proceedings of SPIE · 2006
The modeling and simulation efforts within Air Force Research Lab (AFRL) Sensors Directorate (SN) and Information Directorate (IF) have become increasingly complex. In order to develop, test, and analyze surveillance assets complex simulations and Matlab tools are developed to provide a better understanding of the environment. The increasing amount of primary and secondary data used to emulate the real world further increases the demands on systems used to simulate the environment. One option to overcome some of the computer resources needed to solve the problem is to leverage High Performance Computers (HPC). Within AFRL several approaches have been taken, we will focus on two; Matlab MPI and STAR-P. Both applications leverage Matlab and allow the user to have a familiar interface while expanding the computational limits of the users desktop and creating an environment that will run on a HPC. In the world of parallel computing the Message Passing Interface (MPI) is the de facto standard for implementing programs on multiple processors. MPI defines C, C++ and Fortran language functions for doing point-to-point communication in a parallel program. MatlabMPI is set of Matlab scripts that implement a subset of MPI and allow any Matlab program to be run on a parallel computer. STAR-P currently has a Matlab interface that combines all four parallel approaches in one environment: embarrassingly parallel, message passing, backend support and compilation. Where possible, STAR-P leverages established parallel numerical libraries to perform numerical computation. STAR-P integrates these wide ranges of linear algebra and other routines seamlessly with Matlab for the user. The end result gives the user a familiar interface, Matlab, and ability to use the HPC resources (CPU's and Memory).