云计算系统中基于噪声模板跳表的日志过滤方法

摘要

提出了一种基于时间序列相似度的日志特征提取方法,通过将噪声采样信息按照特定类型标识建模为时间序列,使用Haar小波变换提取序列特征,基于跳表构造噪声模板库.目标日志时间序列通过与噪声模板进行相似度比较来确定其是否为噪声日志.基于真实云计算平台的实验表明,提出的方法能够有效提高故障特征的有效性.%First model noisy event logs into time series, use Haar wavelet transform to get the approximation time series and construct the noisy template library with skip list Then, the similarity between target log time series and noisy template to filter the noisy logs was compared. Experiment results on a real cloud service shows that successfully filter out the noisy log and increase the validity of fault feature.