面向运行时间预测和容错感知的网格资源分配  

Runtime prediction and fault-aware oriented grid resource allocation

作　　者：赵胜[1] 王媛媛[1] 

机构地区：[1]河北北方学院图书馆,河北张家口075000

出　　处：《计算机工程与应用》2011年第16期65-68,共4页Computer Engineering and Applications

摘　　要：提出与描述了一种面向任务运行时间预测和容错感知(Fault-Aware)的网格资源分配策略,采用主动容错的方式,在资源出错之前尽量提前避免它出错或异常的情况发生。该策略把网格中任务的运行时间(runtime)预测和资源的在线时间(uptime)预测结合起来,相对于普通的调度策略具有比较高的资源利用率。在具体的CoBRA网格中间件中实现了该容错感知调度,描述了实现该容错感知调度策略模块的功能。测试过程中选择了睡眠任务技术,划分四种不同的场景进行实验,把该容错感知资源分配与普通的FCFS调度策略进行比较,结果证明在可变化的资源可用性的情况下系统可以加快应用的整体执行时间,具有很小的偏差。A scheduling policy in which can estimate job runtimes and predict resource availability to efficiently distribute workloads for grid resource is presented and described.Instead of just tolerating failures like fault-tolerant scheduling,the technique will actively try to prevent failures from occurring.It is based on the availability of job runtime predictions and resource availability estimates,thus the high usage rate of resource can be obtained.By distributing jobs only to resources available for the full executing time of the job,no CPU cycles are wasted on jobs that will be unable to complete.For the implementation and testing,the CoBRA grid system is used and several software modules are discussed.Experiments are done for a set of sleep jobs,compared to the FCFS scheduling technique,the results show that large reductions in total runtime can be achieved in situations with variable resource availability.

关 键 词：任务运行时间 资源分配 可用性预测 容错感知 网格计算 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]