任意可分负载在动态计算资源上的调度算法

DIVISIBLE LOAD SCHEDULING ALGORITHM FOR DYNAMIC COMPUTING RESOURCES

随着网格计算技术的不断发展,如何充分利用网络中广泛分布的动态计算资源越来越受到关注。为了充分提高并行计算中,在多个动态计算资源上的具有任意可分特性的大规模应用的任务响应速度,提出了一种探测缓存式动态调度算法PBDLS(Probing and Buffering Dynamic Load Scheduling)。该算法利用探测技术对动态资源的运行状态进行实时跟踪,根据预测结果自适应调整任务的分发量,并提出任务预存策略,用来最大限度地填补由于网络状态预测偏差导致的计算时间空闲时间,从而全面提高任务执行效率。算法经2000多组仿真表明:在多种动态网络环境下,PBDLS算法的调度效率整体上优于现有的DA1、DA2和DLT算法,并具有较好的稳定性。

With growing development of the grid computing,more and more attentions are paid on the full use of dynamic computing resources widely distributed in internet.To improve the response speed of large-scale applied tasks of multiple dynamic computing resources in parallel computation with divisible workloads property,an algorithm called PBDLS（Probing and Buffering Dynamic Scheduling） is presented.By introducing probing technique,the algorithm can trace the running status of dynamic resources in real time,and self-adaptively adjust the amount of distributed load based on predicted result.Furthermore,by presenting a ＂load pre-buffering＂ strategy,the algorithm can dramatically fill the computing idle time caused by the prediction error of the network status,therefore to improve the overall task execution efficiency.More than 2000 sets of simulation show that the PBDLS algorithm outperforms integrally the traditional algorithm as DA1,DA2 and DLT in scheduling efficiency and with better stability in various dynamic network environments.