Research On Parallel Optimization And Load Balancing Of Particle Simulation

It is more convenient,more direct and less costly to restore and study in physical phenomena in space plasma physics by using modern computer numerical simulation.Particle in Cell(PIC)is a commonly used numerical simulation method in space plasma.This method simulates every particle in three-dimensional space and requires a lot of computation,so it is usually used in parallel computation in high-performance computer clusters.With the increase of simulation space and particle quantity,the selection and optimization of parallel strategy directly affect the parallel efficiency and simulation time.In addition,with the progress of the simulation,the uneven distribution of particles in the space leads to the unbalanced load,resulting in the decrease of parallel efficiency and resource utilization,and overload is easy to occur.This paper studies and develops particle simulation of collision-free plasmas.The main research content is divided into the following three parts:1.Parallel optimization scheme.Analyze and research the parallel theory and common parallel strategies in particle simulation,and design a parallel computing optimization scheme suitable for collision-free plasma.Use spatial partitioning to divide the overall simulation space into two-dimensional sections.The optimized particle simulation reduces communication overhead in parallel computing,improves parallel efficiency,and can support larger-scale simulation calculations.2.Load balancing scheme for parallel computing.First,analyze and research several common load balancing methods in particle simulation.For the load balancing problem caused by uneven particle distribution in the collision-free plasma particle simulation with spatial partitioning parallelism,a dynamic load balancing scheme is proposed.This scheme can efficiently determine the adjusted simulation area boundary by real-time detection of particle load,dynamically adjust the spatial size of the simulation unit,and thus achieve dynamic load balancing.3.Implementation and testing of parallel optimization and load balancing.The parallel algorithm optimization and load balancing algorithm designed in the previous section are implemented on a high-performance cluster,and the data before and after optimization are compared and analyzed.The test results show that after parallel optimization,the parallel efficiency can be increased by 24.67%,and the dynamic load balancing algorithm can increase the parallel efficiency by 33.69% on the basis of parallel optimization.