期刊文献+

一种基于动态分配模式求解三对角线性方程组的并行算法

A Parallel Algorithm for Solving Tridiagonal Linear Systems Based on the Pattern of Dynamic Dispatch
在线阅读 下载PDF
导出
摘要 分析了双向并行分裂DPP算法存在空闲等待、通信阻塞以及冗余计算等方面的不足,并在此基础上提出一种基于动态分配模式求解三对角线性方程组的并行算法.该算法摒弃了DPP算法平均分配方程组的模式和完成向中间通信后必须消除所有下(上)对角元素的方式,而采用基于运算和通信参数的动态分布模式以及仅适量消元的方法,从而在保持通信畅通的前提下,充分利用计算与通信重叠技术,减少处理机空闲等待和冗余计算.最后分析了新算法的理论性能,并在IBMRS60000机群上进行了数值实验.实验结果表明,该算法的效率较DPP算法有较大提高. The deficiencies of DPP algorithm, such as idle time, communication blocking and redundant computation, are analyzed, then a parallel algorithm for solving tridiagonal linear systems on workstation cluseters, which is based on the pattern of dynamic dispatch, is presented. Based on the computation and communication parameters, this algorithm adopts the pattern of dynamicly dispatching equations on each workstation as well as the method of just diminating part of the the subdiagonal (or superdiagonal) dements after communication toward middle direction is finished, instead of the DPP's pattern of even dispatch and its method of diminating all. The algorithm makes full use of overlapping between computation and communication to decrease the amount of workstations' idle time and unnecessary redundant compuation under the precondition of guaranteeing the clear communication. Finally, the theoretic functions of the new algorithm are analyzed, and numerical experiments are conducted on IBM RS60000 Workstation Clusters. The results show that the ef- ficiency of this algorithm is much higher than that of DPP algorithm.
出处 《微电子学与计算机》 CSCD 北大核心 2008年第5期19-23,共5页 Microelectronics & Computer
基金 国防"十一五"预研基金项目(102010302) 国防"十一五"预研基金项目(402040202)
关键词 三对角线性方程组 并行算法 分裂法 机群 tridiagonal linear system parallel algorithm partition method workstation clusters
作者简介 张选平男,(1963-),副教授.研究方向为并行计算与人工智能. 高由兵男,(1982-),硕士研究生.研究方向为并行计算. 崔骏男,(1981-),硕士研究生.研究方向为并行计算. 陈喜伦男,(1980-),硕士研究生,研究方向为并行计算.
  • 相关文献

参考文献5

  • 1迟利华,李晓梅.求解三对角线性方程组的双向并行分裂法[J].计算机工程与设计,1999,20(1):49-55. 被引量:3
  • 2Wang H H. A parallel method for tridiagonal equation.s[J]. ACM Trans on Mathematical Software, 1981,7(2) :170- 183.
  • 3Michielse P, Vander Vorst. Data transport in wang's partition method[J]. Parallel Computing, 1988,7(1):87-95.
  • 4Jose M Badia, Antonio M Vidal. A parallel algorithm to solve ttidiagonal systems on distruted memory multiprocessors [ J ]. IEEE Trans. on Computers,1992,41(3):203-210.
  • 5Hwang K,Xu Z. Scalable parallel computing: technology, architecture, prcgramming[ M]. Boston: WCB/M cGraw - Hill, 1997.

共引文献2

相关作者

内容加载中请稍等...

相关机构

内容加载中请稍等...

相关主题

内容加载中请稍等...

浏览历史

内容加载中请稍等...
;
使用帮助 返回顶部