Parallel Computing For Direct Simulation Monte Carlo

Direct Simulation Monte Carlo (DSMC) method is based on theory of gas molecular dynamics. It is an efficient method to solve gas flow problems. DSMC method can simulate movements and collisions of the gas molecules. So it can be used to give solutions to the questions of propagation characteristic of acoustic wave propagation in gas. Since DSMC has inherently data parallel characteristic, it can use parallel computing method to calculate it which can make the computation process much easier. Currently, researchers worked on the parallelization of DSMC whose simulate conditions, simulate tools and parallel algorithms are different.The thesis firstly analyzes the cause of using DSMC method to solve acoustic wave propagation problem in gas. DSMC method's characteristics are its physical theoretical basis is clear, reflect regular physical pattern, has wide simulate range, tend to program. DSMC is a important way to investigate acoustic attenuation.Then the thesis introduces the theory of parallel computing. Among various kinds of dedicated and conventional parallel computer systems, the principle and features of PC cluster system are highlighted in this thesis. The primary coverage of the theory of parallel computing includes setting up parallel computing module, communications and synchronization between machines, measurement of algorithm performance.Using MATLAB Distributed Computing Toolbox and MATLAB Distributed Computing Engine to carry out the parallel algorithm of DSMC is the method that practiced in this thesis. The configuration and the use method of MDCE are explained in detail. Meanwhile the operating principle of DCT is also introduced here. A parallel algorithm of DSMC which can be brought into effect by MATLAB is designed. It is based on the serial computing algorithm.Lastly, the thesis analyzes the simulate result which comes from the parallel computing by MATLAB. It shows the influence on the simulation efficiency by varying the number of workers in the network. On the other hand, the simulation efficiency changement caused by the hardware differences of the cluster is also discussed. In this thesis the reasons which cause the DSMC algorithm is explained clearly. Some farther improving suggestions are given in the last chapter. It is shown that adoption of the MATLAB parallel algorithm can give a quite satisfying result.