Design And Implementation Of Application Performance Anomaly Monitoring System

With the popularity and continuous development of distributed architecture,how to guarantee the efficient operation of distributed applications draws enterprise's more and more attention,as a consequence,application performance management(APM)comes into being,it can monitor and optimize application performance,improve application service quality.Anomaly detection is one of the core capabilities of APM,recently,basic anomaly detection schema of application performance is fixed threshold based on 3 Sigma or artificiality,the schema is simple and easy to implement but ignores the regularity of the monitored data over time,so missing or misjudged anomalies may easily occurred.Therefore,this dissertation puts forward the anomaly monitoring system of application performance,applies time series prediction model to the regularity of the monitored data over time,and regards values with large scale deviation from the predicted value as anomalies.The system is composed of application performance sniffer subsystem and application performance anomaly detection subsystem,and the main research work can be summarized as follows:Firstly,the dissertation designs the schema of application performance sniffer based on aspect oriented programming(AOP),and implements a transparent and low-consumption sniffer to java applications.Secondly,the dissertation designs and implements automatic optimum schema selection to current monitored series based on prediction accuracy of many predefined time series prediction models,which dispenses with artificial model selection work.Thirdly,the dissertation designs improved calculation schema of relative deviation,which leads to less missing or misjudged anomalies.At last,the dissertation conducts some experiments,low-consumption feature of sniffer and reliability of anomaly recognition based on prediction are proved.The implemented anomaly monitoring system of application performance with the background of APM,the main features of the system are that it can efficiently sniffer application performance data,intelligently recognize the anomalies,so it does good to intelligently monitor enterprise's key business and guarantee the stable operation of Internet applications.