A Computing Efficient Algorithm Specialized for Pseudo-Constant Metrics

AIOps (Artificial Intelligence for IT Operations) improve operation and maintenance efficiency of complex IT systems by quickly locating the root causes of failures through machine learning methods. The key of AIOps is to quickly locate anomaly in huge amount of monitoring metric time series. In this paper, we show that there are a considerable number of Pseudo-Constant metrics in actual large systems, whose values vary little or occasionally change violently. Based on feature of this kind of metric, we propose a Binary Mixed Detection algorithm for Pseudo-Constant metric. We build a model to evaluate its detection time and also analyze its false negative and false positive probability. Finally, we show the performance of BMD on actual data set. The experimental results show that BMD can save 85% of the detection time.