可迁移网格依赖任务重调度模型及算法

Rescheduling model and algorithm of migratory grid dependent tasks

针对可迁移依赖任务的重调度问题,提出了基于约简DAG可迁移任务图的重调度模型,并基于免疫遗传算法实现了以提高应用性能为目标的求解算法.实验表明,与经典的动态调度算法Max-Min和基于启发式的AHEFT静态算法相比较,由于调度目标的一致性,初始调度的性能在重调度过程中被较好地保持,并且由于任务迁移的支持和遗传算法在全局优化上的性能优势,应用性能得到较大提升;又由于任务图的约减过程和免疫因子对算法收敛的作用,提出的IGA算法效率得到显著改善,使资源动态性和异构性的适应能力得到进一步增强.

For migratory grid dependant tasks, a algorithm. The experiment reduced DAG task graph based rescheduling model was oriented rescheduling algorithm was implemented based on immune genetic shows that compared with the classic dynamic Max-Min scheduling algorithm and heuristic static AHEFT algorithm, the initial rescheduling performance is well kept during the rescheduling process due to the consistency of rescheduling objective. The applying performance gets greatly improved due to the support of task migration and the advantage of immune genetic algorithm in general optimization. Moreover, algorithm efficiency of the proposed IGA gets significantly improved due to reduction of task graph and effect of immune factor on algorithm convergence, and the adaptability of resource dynamics and heterogeneity gets further enhanced.