Communication optimization method and system for dynamic Monte Carlo parallel simulation

Abstract

The invention provides a communication optimization method and system for dynamic Monte Carlo parallel simulation. The method comprises the steps of constructing a communication data structure; constructing a graph topology structure of a neighbor process; starting an out-of-process iteration; selecting a sub-region from multiple sub-regions according to a preset order; conducting data communication between a current selected sub-region of a process and the neighbor process through a constructed communication data structure; ending the iteration. The invention further provides the communication optimization system for the dynamic Monte Carlo parallel simulation. According to the technical scheme, the graph topology structure of the process is constructed, a neighbor set communication function is adopted to replace an original point-to-point communication function, so that the complexity of communication time is reduced; through the optimization of program processes, an optimal sub-region iterative order is selected, partial communication operation in two adjacent consecutive iterations are merged, and therefore partial communication redundant data are reduced, the synchronization waiting time between the processes is shortened, and the load is more balanced.

Description

technical field [0001] The invention relates to the field of communication optimization of parallel region decomposition, in particular to a communication optimization method and system of dynamic Monte Carlo parallel simulation based on SPPARKS parallel framework. Background technique [0002] The Kinetic Monte Carlo (KMC, Kinetic Monte Carlo) method is one of the widely used computer simulation methods for radiation damage to materials. The key is to calculate the occurrence of the interaction potential energy between neighboring atoms by establishing a stochastic process or probability model. Probability can simulate the evolution process of material microstructure on a large scale and for a long time at the mesoscopic scale. Kinetic Monte Carlo parallel simulation is a parallel simulation method formed by combining high-performance computers and Kinetic Monte Carlo methods, which can further break through the existing space-time scale limitations and memory capacity limi...

AI Technical Summary

Problems solved by technology

[0005] In view of this, the purpose of the present invention is to provide a communication optimization method and system for dynamic Monte Carlo parallel simulation, aiming to solve the load imbalance caused by point-to-point communication between neighbor processes in the existing KMC large-scale parallel algorithm question

Images