• 中国计算机学会会刊
  • 中国科技核心期刊
  • 中文核心期刊

J4 ›› 2010, Vol. 32 ›› Issue (10): 85-89.doi: 10.3969/j.issn.1007130X.2010.

• 论文 • 上一篇    下一篇

非结构网格粒子输运Sn并行算法

迟利华,刘杰,田平   

  1. (国防科学技术大学计算机学院,湖南 长沙 410073)
  • 收稿日期:2010-03-03 修回日期:2010-06-10 出版日期:2010-09-29 发布日期:2010-09-29
  • 作者简介:迟利华(1970),女,山东威海人,博士,副研究员,研究方向为并行算法和大规模科学与工程数值模拟并行应用软件;刘杰,博士,副研究员,研究方向为大型科学与工程计算中并行算法、并行应用软件和性能评价等;田平,硕士生,研究方向为并行算法。
  • 基金资助:

    国家自然科学基金资助项目(60673150,60970033);国家863计划资助项目(2008AA01Z137)

The Sn Parallel Algorithm for Particle Transporton Unstructured Grids

CHI Lihua,LIU Jie,TIAN Ping   

  1. (School of Computer Science,National University of Defense Technology,Changsha 410073,China)
  • Received:2010-03-03 Revised:2010-06-10 Online:2010-09-29 Published:2010-09-29

摘要:

本文基于网格区域剖分,提出了一种新的非结构网格粒子输运Sn并行算法,实现了多个角方向和多个能群的同时计算,在计算的过程中不用进行优先级计算和优先级队列维护,只需要按照计算队列的次序组织并行计算。综合考虑所有方向和所有网格点的数据依赖关系,结合Blevel优先级,提出了一种优先级计算方法,优先计算需要数据发送的任务,延迟需要接收数据的任务,达到减少处理器等待时间和计算与通信重叠的目的。使用本文的Sn并行算法和优先级队列针对二维粒子输运问题进行的数值实验表明,并行算法具有良好的并行计算加速效果,扩展到1 024个处理机时,相对64个处理机的并行效率达到52%。

关键词: 并行算法, 粒子输运, 优先级算法, Sn方法

Abstract:

Based on the grid decompostion across processors,a new Sn parallel algorithm for particle transport on unstructured grids is presented. The algorithm implements parallel computation for different angle directions and energy groups,according to the orders of the computing queue. During the parallel computation,it is not necessary to compute the priorities and arrange the priority queues. Synthesizing the data dependence of all directions and all grids,and combining with the Blevel priority,a new priority ordering algorithm is presented. In order to overlap the computation and communication,the algorithm gives priority to computing the grids which need not receive the upstream grids’ messages. Using the Sn parallel algorithm and the priority ordering algorithm,we solve a two dimensional particle transport equations on a parallel computer.The performance results show the algorithms have good scalability. The parallel efficiency is 52% when the processor number is extended for 64 to 1024.

Key words: parallel algorithm;particle transport;priority ordering algorithm;Sn method