在共享环境下基于时间预测的资源预留

Resource Reservation in Shared Environments Based on the Prediction of Time

资源预留是实现网格作业服务质量保证的重要方法。主要针对共享环境中引入资源预留后导致排队任务平均等待时间剧增的问题,提出了一种基于等待时间预测的资源预留,其中排队任务分为远程任务和本地任务,引入优先级策略,预测不同优先级等待时间来计算花费,基于花费选择资源进行预留。实验结果显示预测相对误差和基于花费的调度算法都具有有效性和实际的潜力。

It is an important method that resource reservation can guarantee QoS of the grid jobs. Introduces resource reservation into shared environments to lead to the problem that queuing jobs mean waiting time increase heavily, then proposes an policy of resource reservation based on the prediction of waiting time. Queuing jobs include remote jobs and local jobs. Introduces priority policy, and predicts the waiting time of different priorities to calculate the cost to choose resource.Experimetal results show that prediction relative error and scheduling algorithms based on the cost are practically effective and have a real potential.