一种基于预测的反馈网格作业调度模型*

网格计算中的资源是动态和异构的，常规的静态作业调度方法不适宜网格计算环境，对于网格计算中一类并行计算的有效执行有赖于网格资源（CPU和网络带宽等）与作业的有效匹配。提出了一种基于资源预测结果对作业进行调度的策略，首先阐述了网格主机负载预测的研究成果——IAR模型，并提出了一种预测网络带宽的工具——网络性能平面，利用资源预测结果构造了一种反馈作业调度模型并对一类基于时间平衡的作业进行实验。结果表明，该模型在与其他诸多方法比较中，取得了执行时间较短和稳定性较好的效果。

Feedback Job Scheduling Model in Grid Computing Based on Resource Prediction

The resources in grid computing environments is heterogeneous and dynamic.The normal ways of job scheduling are not fit to grid computing.Efficient execution of a kind of parallel computations can require mappings of tasks to grid resources(including CPU load and net bandwidth,etc).The paper proposes a job scheduling model that based on the result of resources prediction.Firstly present techniques: an improved AR model developed in previous work,then address a tool: net surface which can predict the capability of net bandwidth,finally present a feedback model based on the results of resources prediction and experiment it on a kind of time-balancing job,the result of these experiments demonstrate the model can produce execution times that are both significantly faster and less variable than other models.