Optimizing Parallel Kinetic Monte Carlo Simulation by Communication Aggregation and Scheduling

Kinetic Monte Carlo (KMC) algorithm has been widely applied for simulation of radiation damage, grain growth and chemical reactions. To simulate at a large temporal and spatial scale, domain decomposition is commonly used to parallelize the KMC algorithm. However, through experimental analysis, we find that the communication overhead is the main bottleneck which affects the overall performance and limits the scalability of parallel KMC algorithm on large-scale clusters. To alleviate the above problems, we present a communication aggregation approach to reduce the total number of messages and eliminate the communication redundancy, and further utilize neighborhood collective operations to optimize the communication scheduling. Experimental results show that the optimized KMC algorithm exhibits better performance and scalability compared with the well-known open-source library—SPPARKS. On 32-node Xeon E5-2680 cluster (total 640 cores), the optimized algorithm reduces the total execution time by 16 %, reduces the communication time by 50 % on average, and achieves 24 times speedup over the single node (20 cores) execution.