Parallelization of a direct simulation monte-carlo code for workstation cluster

Abstract

This chapter presents the optimization of a gas kinetical Monte-Carlo program by parallelization. The code considered in the chapter calculates 3-dimensional dilute flows with the direct simulation Monte-Carlo (DSMC) method. The work connected with the parallelization is cooperation between the Institute of Fluid Dynamics of the University of Technology, Braunschweig and the Scientific Center of IBM at Heidelberg and is developed for workstation cluster, i.e., distributed-memory multiprocessors. An application of the code is the reentry of a space vehicle during the flight from the orbit back to the earth. Starting from nearly free molecular flow at an altitude of 150 km, the reentry capsule passes the transitional regime and enters at an altitude of about 92 km, the regime of continuum flow. The studies have shown that the DSMC method is very easily parallelizable in an efficient way. For a small number of processors, the speedup is nearly linear. The communication has proven to be no bottleneck. The slightly smaller efficiency for a smaller number of processors can be explained by the non-optimal domain decomposition.