网格环境中节点CPU负载的分形预测  

Fractal Prediction of CPU Load in Grid Environment

作　　者：张飞[1] 曾国荪[2] 

机构地区：[1]同济大学计算机科学与技术系,上海200092 [2]国家高性能计算机工程技术中心同济分中心,上海200092

出　　处：《计算机科学》2006年第8期61-63,79,共4页Computer Science

基　　金：国家自然科学基金资助项目(60173026);上海科委重大项目(03DZ15029);上海高校网格技术E-研究院资助(200301-1)

摘　　要：网格环境下,常常需要知道网格资源在未来某一时刻具有什么样的性能,比如,调度器需要该性能估测以便进行高效的资源调度、提供满足要求的QoS以及保证整个网格系统的负载平衡。正如在其他任何计算环境中一样,计算能力是所有网格资源中最为重要的资源,通常用CPU负载来刻画节点主机的忙碌程度、衡量节点所能提供的计算能力。已有的研究表明CPU负载具有自相似性和长相关性,这启发我们使用本文介绍的分形的方法进行CPU负载的预测。实验结果证明该方法具有较高的预测精度,因而具有较好的实用价值。In grid environment, it is in need to know what performance grid resources can afford at some given time in the future. For instance, the scheduler need this performance estimation to make efficient resource scheduling, provide desired QoS and keep the whole grid system optimally load-balanced. Among all kinds of grid resources, computing circles are most valuable in grid system as well as in other computing environments. CPU load is naturally used to depict how busy the host machine is and what performance the host can give to user programs. Former research on CPU load shows that CPU load has self-similarity and long-dependence. This leads to our employing fractal geometry methods to make CPU load prediction described in this paper. The experimental results are encouraging and the accuracy this method yields is also quite satisfying.

关 键 词：网格 资源性能 CPU负载 分形 预测 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]