计算物理 ›› 2001, Vol. 18 ›› Issue (2): 177-180.

• 论文 • 上一篇    下一篇

MC程序并行设计及提高加速比措施

邓力1,2, 谢仲生1, 黄正丰2, 许海燕2   

  1. 1. 西安交通大学能动学院核能工程系, 陕西 西安 710049;
    2. 北京应用物理与计算数学研究所, 北京 100088
  • 收稿日期:1999-05-19 修回日期:1999-08-23 出版日期:2001-03-25 发布日期:2001-03-25
  • 作者简介:邓力(1960-),男,四川绵竹,副研究员,博士,北京8009信箱16分箱,100088.

THE PARALLEL DESIGN OF MONTE CARLO CODE AND MEASURES OF ENHANCE SPEEDUP

DENG Li1,2, XIE Zhong-sheng1, HUANG Zheng-feng2, XU Hai-yan2   

  1. 1. Xi'an Jiaotong University, Department of Nuclear Energy and Engineer, Xi'an 710049, P R China;
    2. Institute of Applied Physics and Computational Mathematics, Beijing 100088
  • Received:1999-05-19 Revised:1999-08-23 Online:2001-03-25 Published:2001-03-25

摘要: MC程序的并行设计涉及算法及模块划分,它直接关系到并行加速效率的高低.中子γ耦合输运蒙特卡罗程序MCNP经过并行改造,实现了PVM和MPI两种系统下的并行化.由于作了模块化设计,并行加速效率极佳.PVM版和MPI版并行程序在多个处理器下的加速比均呈线性增长.相比PVM,MPI的适应性更强,多数情况下其效率高于PVM,并行MCNP程序的计算结果可靠.MPI并行程序在16、32和64个处理器上的并行效率分别达到了99%、97%和89%.

关键词: MPI, PVM, 并行加速比, Monte Carlo

Abstract: The parallel design of Monte Carlo code involves computational method and module designs,which is crucial to the parallel efficiency.The coupled of neutron and photon transport Monte Carlo code MCNP has been realized the parallelization in PVM and MPI by modifying the serial code.Due to the form having module being optimized, the parallel efficiency is good where the efficiency of MPI code is stronger than that of PVM code and the speedup of MPI code is higher than that of PVM in most cases.The calculated results of parallel code are reasonable.Both the speedups of PVM code and MPI code have been the linear increasing with the processors.The parallel efficiencies are up to 99% in 16-processors,97% in 32-processors and 89% in 64-processors respectively.

Key words: MPI, PVM, parallel speedup, Monte Carlo

中图分类号: