基于日志关联规则挖掘和模糊推理的集群系统故障预测方法
首发时间:2020-02-13
摘要:随着集群系统规模和复杂性的不断增长,故障的发生已经成为一种常态。故障预测是一种有效的可靠性主动管理和失效预防机制。针对现有基于系统日志的故障预测方法存在的预测召回率低、运行效率差等问题,提出一种新的集群系统故障预测方法。在日志关联规则挖掘阶段,引入加权、数据库划分和模式合并的思想,对经典频繁模式树算法进行改进,保留了更多的故障关联模式,大大减少了时间和空间开销。在故障预测阶段,采用不确定模糊推理的策略增加故障预测的完备性和容错性。基于真实的 IBM BlueGene/L 集群系统日志的实验结果表明,该方法具有较高的预测精准率和召回率,并且表现出良好的时空性能。
For information in English, please click here
Failure prediction method of cluster system based on log association rules mining and fuzzy reasoning
Abstract:With the increasing scale and complexity of cluster systems, the occurrence of failures has become a norm. Failure prediction is an effective reliability proactive management and failure prevention mechanism. Aiming at the shortcomings of the existing log-based failure prediction methods, such as low predictive recall and poor operation efficiency,a new method for cluster system failure prediction is proposed. In the log association rules mining stage, the concepts of weighting, database partitioning, and pattern merging are introduced to improve the classic frequent pattern tree algorithm, which retains more failure association patterns and reduces the time and space overhead. In the failure prediction stage, the strategy of uncertain fuzzy reasoning is used to increase the completeness and fault tolerance of failure prediction. Experimental results based on real IBM BlueGene/L cluster system logs show that the method has high prediction precision and recall, and shows good spatiotemporal performance.
Keywords: Failure prediction Cluster system Association rule mining Fuzzy reasoning
基金:
引用
No.****
动态公开评议
共计0人参与
勘误表
基于日志关联规则挖掘和模糊推理的集群系统故障预测方法
评论
全部评论0/1000