A parallel implementation of the direct simulation Monte Carlo method

The underlying assumptions used to formulate the Navier-Stokes equations preclude their use for the analysis of rarefied gas dynamic environments. Alternatively, the Direct Simulation Monte Carlo technique of Bird can be used to numerically investigate such flows. In this methodology, molecules are tracked through representative collisions and boundary interactions. This microscopic view of the flow allows the inclusion of internal energy modes and chemical reactions in a direct and physical manner. This well-established methodology is combined with innovative strategies to improve the performance and overall capabilities of the technique. Beyond basic procedural enhancements, a parallel implementation of the DSMC method is developed using the master/slave programming model. This new tool is applied to several example problems. These are used to compare the performance of the parallel code to a comparable scalar implementation and to investigate the dynamic load balancing capabilities and the overall scalability of the parallel software. High parallel performance is demonstrated using up to 512 processors on a Cray T3E parallel supercomputer. (C) 1999 Elsevier Science S.A. All rights reserved.