九种智能算法在科学工作流调度中的应用比较

Analysis of nine intelligent algorithms applied to scientific workflow scheduling

科学工作流常用于描述大规模科学应用,由于涉及大量计算,常部署于网格或云环境中执行.作为科学工作流管理系统的一个重要组成部分,工作流调度算法的好坏对工作流管理系统的性能有着重要的影响.由于智能算法具有很好的全局搜索能力和组合优化的特点,其在工作流管理系统中的应用日益受到关注.在workflowsim平台上针对典型的科学工作流应用,以工作流执行时间为优化目标,对九种智能算法的性能进行了比较.实验结果表明,对于不同的工作流模型和不同的任务量,相对于遗传算法、蚁群算法和粒子群算法等,化学反应算法具有较小的makespan,混合蛙跳算法、蚁群优化算法和粒子群算法的makespan相对较小,但却在不同任务量和不同工作流模型下表现不稳定.

Scientific workflow is often used to describe large scale scientific applications and is usually deployed in grid or cloud systems because of its involving a large number of calculations. As an important part of the scientific workflow man agement system, workflow scheduling algorithm has important impact on the execution performance of the workflow man agement system. Considering intelligent algorithms have a good global search ability and combinatorial optimization charac teristic, its application has attracted much attention for the research field of workflow management system in recent years. In this paper, the performances of the nine intelligent algorithms are analyzed and applied into workflow scheduling based on workflowsim simulation platform. The experimental results showed that, compared with the GA, ACO and PSO, the CNO has best makespan. The makespan of SPFA, ACO and PSO are relatively smaller, but the performances are unstable for dif ferent tasks or different workflow model.