Accelerating the Computation and Verification of Molecular Collision Models: A Case Study in Legacy Code Parallelization
Greg Wolffe, Kurt A. O’Hearn, Christian I. Trefftz, George C. McBane · ScholarWorks - GVSU (Grand Valley State University) · 2011
This research project constituted a case study in computational science: applying modern parallel computing techniques to a mathematical model used to solve a scientific problem. The problem involved a physical chemistry model that evaluates simulations of molecular collision experiments, implemented via a 15,000-line FORTRAN 77 code. This problem was chosen for parallelization because of its scientific importance, its computational complexity, and its overall structure that was amenable to parallelization. Since the original program was written, experimental designs have changed in a way that require significant increases in execution time for the simulation. To address this issue, the simulation code was profiled, analyzed, and parallelized using the OpenMP/multithreaded paradigm. Nearly linear speedup was measured for the OpenMP version executing on a 16-core multiprocessor. Furthermore, experimental results suggest these speedups should scale well with an increasing number of processors.