计算机应用 ›› 2014, Vol. 34 ›› Issue (11): 3234-3240.DOI: 10.11772/j.issn.1001-9081.2014.11.3234

• 先进计算 • 上一篇    下一篇

固定结构约束下并行计算成比例的可扩展方法

吴沧海1,熊焕亮1,2,姜火文2,3,杨文姬1   

  1. 1. 江西农业大学 软件学院,南昌 330045;
    2. 同济大学 计算机科学与技术系,上海 201804
    3. 江西科技师范大学 数学与计算机科学学院,南昌 330038
  • 收稿日期:2014-05-19 修回日期:2014-06-24 出版日期:2014-11-01 发布日期:2014-12-01
  • 通讯作者: 熊焕亮
  • 作者简介:吴沧海(1979-),女,浙江上虞人,副教授,主要研究方向:并行分布式计算、云计算;熊焕亮(1977-),男,江西新建人,博士研究生,CCF会员,主要研究方向:并行分布处理、云计算;姜火文(1974-),男,江西进贤人,副教授,博士研究生,主要研究方向:并行计算、云计算;杨文姬(1984-),女,江西上饶人, 讲师, 博士研究生,主要研究方向:并行计算、云计算。
  • 基金资助:

    国家863计划项目;国家自然科学基金资助项目;江西省教育厅科技计划项目

Proportional extension of parallel computing under fixed structure constraint

WU Canghai1,XIONG Huanliang1,2,JIANG Huowen2,3,YANG Wenji1   

  1. 1. Software College, Jiangxi Agricultural University, Nanchang Jiangxi 330045, China;
    2. Department of Computer Science and Technology, Tongji University, Shanghai 201804, China;
    3. School of Mathematics and Computer Science, Jiangxi Science and Technology Normal University, Nanchang Jiangxi 330038, China
  • Received:2014-05-19 Revised:2014-06-24 Online:2014-11-01 Published:2014-12-01
  • Contact: XIONG Huanliang

摘要:

针对固定结构下并行计算无法通过规模扩展提升计算性能的问题,提出了一种成比例调整图权的并行计算扩展方法。该方法首先分析影响可扩展性的并行任务因素及体系结构因素;然后采用带权图对并行任务及体系结构进行建模;最后,对并行计算图模型中顶点和边的权值进行调整,实现并行计算的扩展。针对并行任务与体系结构是否具有相同的拓扑结构进行了两组实验,结果显示扩展前后的速度效率不变或近似相等。在上述两组实验的基础上,固定并行任务的算法结构及硬件系统的体系结构,仅调整性能参数,从特定的初始状态开始,以相同的比例作连续多次扩展,结果显示随着并行任务的连续扩展,体系结构资源被充分利用,速度效率逐渐提高,但并行任务扩展至一定程度后,速度效率提高缓慢;而如果并行任务及体系结构按一定的比例一同扩展,并行计算的速度效率近似不变。

Abstract:

Aiming at the problem that the performance of parallel computing cannot be improved by extending its scale under the constraint of fixed structure, a method of proportionally adjusting graph weights was proposed to handle such extension problem. The method firstly investigated the factors from architecture and parallel tasks which affected its scalability, and then modeled the architecture as well as parallel tasks by using weighted graph. Finally, it realized an extension in parallel computing by adjusting proportionally the weights of the vertices and edges in the graph model for parallel computing. The experimental results show that the proposed extension method can realize isospeed-efficiency extension for parallel computing under the constraint of fixed structure.

中图分类号: