针对网格任务倾向于高估预留时间的事实,提出一种支持叠交预留请求的松弛预留策略,用于降低预留机制对系统性能的负面影响。理论分析给出了违约风险的量化计算方法,并针对工作流的协同预留问题设计了相应的预留接纳算法。实验结果显示,松弛策略在面对较高的预留请求率时能显著降低预留机制对系统性能所造成的负面影响;基于实际工作流的实验结果显示,该策略能有效降低由于高估预留时间而导致的任务执行效率损失。
To address the problem that applications tend to overestimate reservation duration, a novel reservation tech- nique based on overlapping relaxing policy was proposed, so as to mitigate the negative effects of stringent reservation mechanisms on system performance, The analysis of the relaxed policy was represented theoretically and a relaxed ad- mission algorithm was designed for large-scale grid workflows. Experimental results show that relaxed policy can sig- nificantly mitigate the negative effects of classical policies when system is in presence of heavy workload. Experimental results based on a real workflow indicate that the relaxed policy is able to decreasing the efficiency loss due to the over- estimation of reservation durations.