蒙特卡罗固定源计算的并行设计与实现  

Design and Implementation of Parallel Computing on Monte Carlo Fixed Source Calculation

作　　者：陈诚[1,2] 宋婧[1,2] 张宽[1,2] 孙光耀[1,2] 郝丽娟[2] 郑华庆[2] 吴宜灿[2,1] 

机构地区：[1]中国科学技术大学,安徽合肥230027 [2]中国科学院核能安全技术研究所,安徽合肥230031

出　　处：《核科学与工程》2014年第4期500-504,510,共6页Nuclear Science and Engineering

基　　金：国家ITER 973计划(2011GB113006);中国科学院战略性先导科技专项(XDA03040000);国家自然科学基金(91026004);安徽省自然科学基金(1308085QH138)

摘　　要：蒙特卡罗方法在求解粒子输运问题中已得到广泛运用,由于其收敛速度慢,需要耗费大量时间重复模拟以获得准确的结果。本文介绍了一种蒙特卡罗固定源计算的并行方法,并基于消息传递接口MPI在FDS团队自主研发的超级蒙特卡罗计算软件SuperMC2.0中实现了多核及多节点并行。以IAEA-ADS为测试例题在高性能计算集群上以32个计算内核进行测试,获得了86.81%的并行效率,同等测试条件下并行效率高于MCNP,同时保证了串行结果和并行结果的一致。测试结果表明该并行设计能够有效地解决蒙特卡罗固定源计算的耗时问题。Monte Carlo（MC）method has been widely used in solving particle transport problems.Due to its slow convergence speed,MC method costs a lot of time simulating repeatedly to obtain accurate results.A parallel method for MC fixed source calculation of multi-node and multi-core based on message passing interface（MPI） was implemented in Super Monte Carlo Calculation Program for Nuclear and Radiation Process 2.0.Taken IAEA-ADS as the benchmark,the parallel efficiency reached86.81% which was better than MCNP when using 32 cores of high performance computing cluster,while ensuring the consistency of serial and parallel results.The test results demonstrated the time-consuming problem of MC fixed source calculation can be effectively solved by this design.

关 键 词：蒙特卡罗 并行计算 固定源 MPI SuperMC 

分 类 号：O246[理学—计算数学]