一种基于动态聚合方法的网格任务调度算法

网格中的资源都是动态的,传统的静态任务调度算法不能适应网格的动态特性。通过对资源在未来一段时间内的状态进行预测,可以提高调度算法的性能。文中提出了一种用动态聚合进行调度的算法。首先对处理器的负载进行取样,然后根据网格任务的执行时间,对处理器的取样值进行动态聚合,再利用AR(p)模型进行预测,最后利用预测到的值作为参数对网格任务进行调度,把网格任务分配给每个处理器,使得每个处理器完成子任务的时间都相同,从而使得整个任务的执行时间最短。实验表明,这种算法能很好地适应处理器负载高度变化的情况。

A Dynamic Aggregation-Based Job Scheduling Algorithm for Grid

Resource in the grid environment is dynamic.The traditional static job scheduling algorithm is not suitable for it.The performance of scheduling algorithms can be improved if the status of resource can be predicted in the future.This paper introduces a dynamic aggregation based scheduling algorithm.First,the CPU load is sampled.Then these sample data are aggregated according to the task's execution time.Third,the AR(p) model is used to predict the status of the CPU load in the future.At last,the predicted data are used to guide the job scheduling.The job will be assigned to every CPU,and every CPU will execute the piece of job with the same time,which enables that the whole execution time of the job is smaller.It is proved by experiment that this algorithm suits well the situation that CPU load varies highly.