Research On Computational Fluid Dynamics Simulation Program Parallelization Method

Computational Fluid Dynamics(CFD)has been widely used in aviation,meteorology and other fields to solve complex flow problems by numerical calculation to find out various flow phenomena and laws.Lattice Boltzmann method(LBM),as a numerical method for solving CFD,has the advantages of natural data parallelism,simple boundary processing and high stability.However,as the size of grid increases,the LBM method is used to solve the problem How to make the LBM method have good performance on the high-performance cluster in the rapid development of high-performance computers is the hotspot in this field.This paper mainly studies the LBM method's parallelization method in computational fluid dynamics.By analyzing the LBM method,including the analysis of LBM calculation process and parallelizability,combined with the existing mature parallel framework,this paper proposes a LBM parallelization method based on OpenMP to shorten the execution time of LBM method on a single machine.In view of the extension of LBM method in high performance cluster,this paper proposes an LBM parallelization method based on MPI by dividing the computing grid and using MPI for computing communication.In view of the inefficient execution of the pure MPI parallel method on the cluster,this paper proposes two hybrid programming models,and further proposes an LBM parallelization method based on MPI + OpenMP according to hybrid programming models.This method takes full advantage of the characteristics of the cluster and has good performance.For the excessive use of OpenMP in mixed programming,the creation and destruction of threads and the long communication of MPI result in longer computation time,this paper puts forward an improved LBM parallelization method based on MPI and OpenMP.The experiment proved that the improved method reduced the running time.