A novel technique for long-term anomaly detection in the cloud

摘要: High availability and performance of a web service is key, amongst other factors, to the overall user experience (which in turn directly impacts bottom-line). Exogenic and/or endogenic factors often give rise anomalies that make maintaining high delivering very challenging. Although there exists large body prior research anomaly detection, existing techniques are not suitable for detecting long-term owing predominant underlying trend component time series data. To this end, we developed novel statistical technique automatically detect cloud data. Specifically, employs learning both application as well system metrics. Further, uses robust metrics, viz., median, median absolute deviation (MAD), piecewise approximation accurately even presence intra-day weekly seasonality. We demonstrate efficacy proposed using production data report Precision, Recall, F-measure measure. Multiple teams at Twitter currently on daily basis.