动态预测的网格调度算法

Prediction Based Dynamic Update Grid Scheduling Approach

针对网格环境下计算节点的自治性、异构性、动态性、分布性等特征,提出了一种基于动态修正预测的调度算法。该调度方法依据历史数据和最近访问过的计算节点的性能、网络通信延迟等信息,形成经验规则并根据其进行计算,预测计算节点的将来性能,并通过使用动态修正有效降低预测误差,将任务提交给轻负载或性能较优的计算节点完成。实验结果表明,该方法不但可以有效减少不必要的延迟,而且在任务响应时间、任务的吞吐率及任务在调度器内等待被调度的时间方面比随机调度等传统算法要优。

Since grid resources are autonomic, heterogeneous , distributed and their status change over time. In this paper, a prediction based dynamic update grid scheduling approach is presented. In this approach, grid scheduler utilizes recent resource performance data, such as computer's load, latency of net communicating to calculate and predict the coming performance of grid resources, then the tasks are submitted to the low load resource. Comparing this dynamic grid scheduling approach with some other research, this approach is more generous than other scheduling approaches. Experimental results demonstrate that this approach diminishes latency and contributes to the overall grid load balancing, and therefore significantly improves resource utilization , response time of tasks and shorten execution time using the performance prediction versus a random selection of resources.