Parallel scalar optimized direct simulation Monte Carlo method for rarefied flows on the SP-1
Simply transferring codes originally developed for a supercomputer to a cache based workstation architecture will generally result in poor performance. Better data locality is required for good cache management which can generally only be achieved by rewriting the code using a different data structure. Data locality at a higher level is also required if a distributed memory parallel computer is used. A change in the DSMC algorithm is then required.
- 1.Bird, G. A., Molecular Gas Dynamics, Clarendon Press, Oxford, 1976.Google Scholar
- 2.Boyd, I. D.: Vectorization of a Monte Carlo Simulation Scheme for Nonequilibrium Gas Dynamics, J. Comp. Phys. 96, 1991Google Scholar
- 3.Dietrich, S.: An Efficient Computation of Particle Movement in 3-d DSMC Calculations on Structured Body-Fitted Grids, Proc. of the 17th RGD-Symp., 1990Google Scholar
- 4.Dietrich, S.; Boyd, I. D.: A Scalar Optimized Parallel Implementation of the DSMC Method, AIAA 94-0355, Reno, 1994.Google Scholar