| Lean Aerodynamics is a discipline that discusses the laws of mass,energy transfer,and the resulting chemical reactions in the flow of rarefied gas.Due to the extremely high vacuum degree environment and the significant rarefaction effect,the traditional continuity method is no longer applicable.Plasma plume is an important branch of rarefied flow.It is mainly used in the fields of plasma device analysis and plasma thruster.With the development of high-performance computing technology,the study of pulsed vacuum arc plasma plume through numerical simulation has become an important research method and a hot spot.The Direct Simulate Monte Carlo(DSMC)method and the Particle-in-cell(PIC)method have been developed and improved to become important numerical simulation methods for simulating neutral particles and ions/electrons,respectively.This paper combines the actual needs of parallel calculation of plasma plume numerical simulation in scientific research and engineering practice,and conducts large-scale parallel computing and dynamic load balancing of plasma plume DSMC simulation for the In-house DSMC simulation program within our research group.Combined with the charged ion PIC simulation program of our project team,the parallel algorithm design of the DSMC/PIC coupling program was further discussed.Because the motion,collision and chemical reactions of the simulating molecules in each grid unit are handled independently,the DSMC method has perfect intrinsic parallelism and is very suitable for parallel computing.However,due to the uneven distribution caused by the simulation of the molecular motion,the phenomenon of unbalanced load will occur in the process of parallel computing,which seriously limits the efficiency of the DSMC parallel algorithm in large-scale simulation.In this paper,the Metis graph splitting library is applied to divide the load of the computing area.In order to realize the General parallel algorithm for unstructured grid DSMC program,we utilize the MPI distributed parallel model for parallel computing,and the master-slave communication mode to process the simulating molecules information interaction that occurs during the simulating molecules motion procedure between processes.In order to improve the efficiency of parallel computing,we further design and implement a dynamic load balancing strategy based on the weighted simulating molecules to alleviate the load imbalance problem and improve the parallel efficiency of the DSMC parallel program.By the parallel numerical simulation test on a single node of the Tianhe-2high-performance computer platform on a model problem with 90 thousand grid cells,the correctness and effectiveness of the parallel algorithm are verified.Compared with the serial DSMC program,the parallel version of MPI effectively reduces the simulation time and achieves a performance acceleration ratio of 7.54 at 24 processes.After using load balancing,the performance of the parallel DSMC program has been further improved,the performance acceleration ratio increases to 9.23 with higher parallel efficiency.By further expanding the test grid size to 980 thousand grid cells,when the parallel scales from 64 processes to 256 processes,a performance acceleration ratio of 1.48 times is achieved,and the parallel efficiency is maintained at37%,which indicates the parallel DSMC program has moderately good performance scalability.Finally,we design the overall parallel computing process of the DSMC/PIC coupling algorithm,in which the Fortran and C++ mixed programming is used,and analize the parallel difficulties of the coupling program.The octree algorithm is utilized to tackle the mapping of the two grid sets needed by DSMC and PIC.We propose a communication strategy to handle the tidal migration of simulating molecules between two grid sets during the interface of different procedures,and present the parallel computing flowchart of the DSMC/PIC coupling program. |
PDF Full Text Request