The Research And Implementation Of Parallel Computing On Numerical Simulation Of The Explosion And Impact Phenomenon

The numerical simulation program of the explosion and impact phenomenon using dynamic FEM method requires small time step to meet the stability requirements and small element size to meet the accuracy requirements. Since the computing scale is so large, the current single computer cannot meet its demand for computing resources. Parallel computing fitly meets the needs in this area.In parallel computing scheme, SMP(Shared Memory Parallel) can reduce the communication overhead among the large number of cores, and play a powerful calculating effectiveness of internal computing nodes. Distributed parallel computing can break through the hardware performance restrictions of a single computing node, which greatly expands the scale of parallel computing. Combining the advantages of SMP and distributed computing, using shared memory in nodes and using distributed computing among nodes, can reflect the power of high performance computer system. This paper respectively implements the parallel optimization based on OpenMP shared memory multi-threaded and parallel optimization based on MPI distributed parallel computing for EP3 D software, and achieve a hybrid parallel optimization.Against the SMP based on OpenMP, this paper discussed the basic idea and the specific steps of optimization, including distinguishing data types and setting parallel mutex etc. On the basis of achievement for OpenMP preliminary optimization, this paper proposed and implemented further optimization for artificially specified tasks and thread private method.Against the distributed parallel computing based on MPI, this paper proposed and implemented the area zoning method based on Metis software. This method can meet the regionalism principles for load balancing and minimizing the communication among processes. We also proposed and constructed a point to point communication model, and implemented parallel computing in process level. We proposed and implemented a parallel optimization scheme based on regionalism, and achieved parallelism of computing and communication.Combining these two parallel methods, this paper finally achieved the hybrid parallel optimization based on MPI and OpenMP. In this paper, all the optimization parallel schemes were tested, the results showed that each program generates accurate results with high parallel speed-up ratio and scalability.