计算机工程 ›› 2019, Vol. 45 ›› Issue (5): 59-65.doi: 10.19678/j.issn.1000-3428.0051785

• 体系结构与软件技术 • 上一篇    下一篇

基于任务复制与冗余消除的多核调度算法

任良育,赵成萍,严华   

  1. 四川大学 电子信息学院,成都 610065四川大学 电子信息学院,成都 610065
  • 收稿日期:2018-06-11 出版日期:2019-05-15 发布日期:2019-05-15
  • 作者简介:任良育(1992—),男,硕士研究生,主研方向为并行计算;赵成萍,副教授;严华,教授。
  • 基金项目:

    国家重点基础研究发展计划(2013CB328903-2)。

Multi-core scheduling algorithm based on task duplication and redundancy elimination

REN Liangyu,ZHAO Chengping,YAN Hua   

  1. 四川大学 电子信息学院,成都 610065
  • Received:2018-06-11 Online:2019-05-15 Published:2019-05-15

摘要:

在分布式计算中常把任务之间的协同和通信关系转换为任务图模型,而任务调度是决定分布式计算性能的关键因素之一。为解决OSA、TDCS、RECS等传统经典算法处理器个数消耗多且存在大量冗余任务等问题,提出一种改进的任务图调度算法。该算法基于贪心策略复制任务的前驱以及前驱的前驱,减少调度长度和处理器空闲时间,并在不增加调度长度的前提下,通过合并簇及减少冗余任务降低处理器个数和处理器的负载。实验结果表明,该算法在处理器个数、加速比以及冗余任务比率上都有一定程度的优化,能提升分布式计算性能。

关键词: 分布式计算, 任务调度, 任务复制, 冗余消除, 贪心策略

Abstract:

In distributed computing,the coordination and communication relationships between tasks are often converted to a task graph model.Task scheduling is a critical factor in determining the performance of distributed computing.Traditional classical algorithms such as OSA,TDCS,RECS,can lead to a large amount of processor resource consumption and many redundant tasks.To address the problems,this paper proposes an improved task graph scheduling algorithm.The algorithm reduce the scheduling length and idle time of processors by using the greedy strategy to replicate the precursors of tasks and precursors.Without increasing the scheduling length,it merges clusters and reduces redundant tasks to decrease the number and the load of processors.Experimental results show that the algorithm has a certain degree of optimization in the number of processors,the speedup ratio and the redundancy task ratio,and improves the distributed computing performance.

Key words: distributed computing, task scheduling, task duplication, redundancy elimination, greedy strategy

中图分类号: