计算资源共享平台中基于分类器的可用性预测

Classifier-based availability prediction for computing resource sharing platform

提出与描述了计算资源共享平台中的网络资源的可用性预测方法,该方法使用了基于数据挖掘的分类器算法,采用了分布式数据收集器工具,该工具运用收集数据和跟踪数据的方式来获取的预测的可用性的信息,从而完成资源的使用情况的预测。讨论了实现资源可用性预测的软件框架,描述了资源预测的具体过程。实验结果表明,即使在高挥发性的分布式计算平台中,该资源可用性预测技术能够很好地预测计算资源共享平台中机器的处理器利用率、内存的负载、机器的可用性,为调度器和副本备份提供参考。

The classifier-based availability prediction for computing resource sharing platform is presented and described. It uses the distributed data collector tools and can assess the feasibility and accuracy of predicting desktop resources availability from academic organizations based on the resource usage traces collected from desktop machines. The software framework for automated behavior prediction of this approach is discussed subsequently. The prediction processes are also described in detail. The results show that indeed even such a highly volatile environment allows for meaningful and robust prediction of a variety of metrics, It can provide effective support for better usage of resources in computing resource sharing platform and for improving their dependability.