Abstract: With the explosive growth of the data on the Internet,the scale of the server cluster is rapidly expanding.How to carry out large-scale cluster monitoring and analysis becomes a difficult problem in the Internet industry.Therefore,this paper presents a new method for detection and analysis of the monitoring data according to the monitoring jittering data.It adopts pattern-based outlier detection method without setting a threshold,takes the eigenvalues,calculaties the outliers,and obtains the abnormal range and degrees.Experimental results show that the algorithm can extract data features for time sequence of jittering data,and has a higher precision and better applicability than the outlier detection algorithm based on mean-variance.

Key words: outlier detection, onitoring data, tatistics, attern, time sequence

摘要： 随着互联网数据规模的增长,服务器集群的规模快速扩大,对大规模的集群进行监控和分析成为互联网行业运维的难点。为此,根据监控统计数据剧烈波动的特点,提出一种MySQL异常检测分析算法,采用基于模式的异常检测方法,无须设置阈值,分段取模式特征值,计 算异常点、异常区间和异常程度。实验结果表明,该算法对于抖动剧烈监控数据的时序序列可以较好地提取数据特征,与基于均值方差的异常检测算法相比,具有更高的精准度,对监测数据的适用性较强。

关键词: 异常检测, 监控数据, 统计, 模式, 时间序列

CLC Number: 

• TP393