使用二維非結構網格之DSMC程式平行化研究

Abstract

本研究是要報告在分散式記憶體的機器上用四角非結構性網格來作DSMC之平行化。利用區域(Domain Decomposition)來分配每個CPU的工作量。以高速空穴流為基準問題來確認程式平行化的可行性。區域切割法主要使用(1)coordinate 切割，(2)two-step 切割，(3)multi-level 切割，來作靜態範圍的分割。一般所使用的 two-step 和 multi-level 切割圖形之工具像JOSTLE , METIS都是用每個CPU所需計算的分子數目來作切割之衡量。平行化程式之效率結果顯示JOSTLE有比較好的表現，此乃由於每個CPU之間的溝通量比較平衡所致，作到25顆CPU時，還可以有高達60多%的效率。最後，計算2個case (一個為流過圓柱的近連續高超音速流Kn=0.025，另一個為流過15°平板的近連續高超音速流Kn=0.0002)，計算範圍利用JOSTLE來切割成25CPU來證明這個平行化程式的強大功能。結果是以之前所作的數值模擬和實驗數值來作比較。

The parallel implementation of the Direct Simulation Monte Carlo (DSMC) method on memory-distributed machines using quadrilateral unstructured mesh is reported. Physical domain decomposition is used to distribute the workload among multiple processors. High-speed driven cavity flow is used as the benchmark problem for the validation of the parallel implementation. Three partitioning techniques including simple coordinate partitioning, two-step partitioning and multi-level partitioning are used for static domain decomposition, respectively. Graph partitioning tools, JOSTLE and METIS, are used to perform two-step and multi-level partitioning, respectively, by considering the estimated particle weighting in each processor. Results of parallel efficiency show that two-step partitioning using JOSTLE performs the best, with 65% up to 25 processors, due to better balanced-load among the processors. Finally, the powerful computational capability of the parallel implementation is demonstrated by computing the near-continuum, hypersonic flows over a cylinder and a flat plate with a 15°-ramp at low Knudsen number of 0.025 and 0.0002 with 25 processors each. The results compare reasonably well with previous simulated and experimental studies.