Implementation of the direct simulation monte carlo method for an exhaust plume flowfield in a parallel computing environment

In this paper, results are presented which demonstrate that the execution time for the DSMC method can be improved significantly by the use of parallel processing. The initial algorithm chosen for study was one in which the collision processes within a cell and the molecular motions are done in parallel. Each of these processes is done for the molecules within a number of cells simultaneously, with each processor being assigned a new cell after completing the required computations from the previously assigned one. Even though the initial algorithm we chose was limited in that speed-up curves obtained using it leveled off rather quickly, it still resulted in a sevenfold decrease in execution time for 16 nodes. We feel quite certain that an optimized version of the parallel algorithm described above could approach a tenfold decrease, We also present preliminary results, which show that the addition of a process which scales as the number of molecules squared causes only minor increases in execution time for the sequential version of the code and no increase for the parallel version.