网格环境中节点CPU负载的分形预测.pdfVIP

2006V ol33 8 CPU * ) ( 同济大学计算机科学与技术系 上海200092) ( 国家高性能计算机工程技术中心同济分中心 上海200092) 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 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 Qo S 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 网格, 资源性能, CPU 负载, 分形, 预测 Fractal Prediction of CPU Load in Grid Environment ZH A N G Fe ZENG GuoSun ( D epart men t of Compu t er Sc ence and T echn ol ogy, T ongj U n vers t y , S han gha 200092) ( T ongj Branch, N at onal Eng neer n g T echnology Cent er of H gh P erformance Comput er, Sh angh a 200092) Abstract In g r d env ronment , t s n need to kno w w hat perfo rmance g r d r eso urces can affor d at some g ven t me n the futur e. Fo r nstance, the scheduler need t h s perfo rmance est mat on to make eff c ent resource schedul ng , pro v de des r ed Q oS and keep the w ho le gr d system opt mally loadbalanced . Amo ng all k nds of g r d resour ces, comput ng c r cles ar e mo st valuable n g r d system as w ell as n o ther comput ng env r onments . CP U lo ad s naturally u sed to dep ct ho w bu sy the ho st mach ne s and w hat perfo rmance the ho st can g v e to user pro gr am s. For mer research on CP U load sho w s that CPU load has selfs m lar t y and longdependence. T h s leads to our emplo y ng fr act al geo met ry methods to make CPU load pr ed ct on descr bed n th s paper . T he ex per mental results are encourag ng and the accuracy th s metho d y elds s also qu te sat sfy ng . Keywords G r d , R eso ur ce perfo rm