基于分块非结构化网格上的SIMPLE算法、区域分解算法及MPI并行编程方法，给出了一种并行计算流体力学实旖方案。提出了“串行粗粒度，并行细粒度”的区域分解及网格划分方法；依据内边界网格界面上的通量守恒原则，推导出了子区域间内界面上的变量传递关系式。在魔方计算机上使用500处理器核完成了两类流动问题计算。研究表明，并行计算结果、串行计算结果与基准解吻合很好，通信时间、cache命中率对并行加速比有显著影响。
A parallel computing strategy for CFD (Computational Fluid Dynamics) was developed based on unstructured grid SIMPLE (Semi-Implicit Method for Pressure Linked Equations) algorithm, domain decomposition method and MPI (Message Passing Interface) parallel programming tool. We proposed a "coarse-grained serialization, fine-grained parallelization" method for the domain decomposition and meshing, and derived the formula of variables transfer across the interface between adjacent sub-domains based on the flux conservation principle. Two typical flow cases were tested with 500 processors at Shanghai Supercomputer Center. The numerical results show that the parallel computing method has the almost same accuracy with the serial one, meanwhile, the communication time and the cache hit rate have significant influence on parallel speedup ratio.
参考文献
| [1] | Roger L.Davis;Jixian Yao .Computational Approach for Predicting Stall Inception in Multistage Axial Compressors[J].Journal of Propulsion & Power,2007(2):257-265. |
| [2] | 王宝潼,张楚华,席光.三维分块非结构化网格上的可压缩流计算方法[J].航空动力学报,2009(10):2319-2325. |
- 下载量()
- 访问量()
- 您的评分：
-
10%
-
20%
-
30%
-
40%
-
50%