网格依赖任务重调度资源选择算法研究

Research on Resource Selection Algorithm in Grid Dependent Tasks Rescheduling

为网格依赖任务选择稳定资源是减少重调度触发频繁的关键。针对稳定资源选取的问题,文中提出了资源成功执行任务标记方法,依据标记值对资源进行动态分级,在此基础上提出了基于税收机制的成功标记重调度方法。在该方法中引入了税收征收不同税率的思想,对资源的标记值按大小分段,根据标记值来为任务分配当前较稳定的资源。对于不同分段中的标记值,资源在成功完成一项任务或失败时,对标记值的增加或减少程度会不同。实验结果表明,该方法可以提高资源利用率及负载均衡,有效提高资源稳定性判断的准确性,从而减少重调度触发的频率。

Selecting a stable resource for grid-dependent task is the key issues to reduce the frequency of tasks rescheduling. Aiming at this problem,present a marking method that count the number of resources to perform tasks successfully. Based on this tag value classify resource dynamically. Then put forward the improved method based on the tax mechanism successful index rescheduling method. This method uses different tax rates for different tag value range,segmenting the tag values of resources according to the size,allocating the rel-atively stable resources for task according to the current tag values. For the tag value in different segments,when resources are in the suc-cessful completion of a task or fails,the tag value will be different for increasing or decreasing degree of tag values. Experiment results show that the method can make the resources fully utilized and improve load imbalance,effectively improving the accuracy of current sta-bility judgment. The possibility of rescheduling trigger is reduced.