Parallelization of a Modular Particle-Continuum Method for Hypersonic, Near Equilibrium Flows

摘要

Extension of a modular particle-continuum (MPC) method is outlined to take advantage of available cluster computer technology and allow future extension to simulation of full three dimensional flow. This method loosely couples an existing direct simulation Monte Carlo (DSMC) code to a Navier-Stokes solver (CFD) while allowing both time-step and cell size to be completely decoupled between each method. This approach allows the solver to maintain the physical accuracy of DSMC in regions where the Navier-Stokes equations break down, while achieving the computational efficiency of CFD in regions that are considered fully continuum. Parallelization techniques that take advantage of the modular implementation are outlined and evaluated using a set of flow problems with various free stream conditions as test cases. In general, linear speedup is realized for small problems but with a slope less than unity, with no effect on the final solution. Overall, the scaled efficiency of larger computations on more processors remains above 80% for the test case examined. Comparisons between solution time requirements for full DSMC and the MPC are made. A moderate speedup of about five is attainable for the higher Knudsen number case, while a speedup of over thirty compared to the full DSMC time is realized for the near equilibrium case.