Research On CDN Traffic Anomaly Detection Technology Based On Time Series Analysis

The Content Delivery Network(CDN)plays an increasingly important role in the evolution of an increasingly sophisticated network infrastructure and the growing size of Internet users.CDN relies on the deployment of large-scale distributed infrastructure to bring traffic applications and hotspot content as close as possible to users to improve the quality of experience(QoE).However,network congestion,server anomalies,and even large-area networks are caused by unexpected interruptions in nodes,explosive growth of hot topic access,and malicious intrusions by criminals when CDN is oriented to users' services.Therefore,the CDN should have the ability to quickly and accurately detect abnormal network traffic,which is of great significance for ensuring the normal operation of the network.Based on above background,this paper studies the problem of CDN traffic anomaly detection,which can effectively detect the anomalies in CDN traffic.The research work of this paper is mainly as follows:1.A time series representation model for CDN traffic data characteristics is proposed.Based on the classification and preliminary preprocessing of CDN traffic by node,the aggregation mode of PAA method is improved to aggregate the data according to the variable time window,in order to reduce the size of the data,reduce part of the noise,provide a data source for the following time series anomaly detection system model,and reduce the computational pressure of the anomaly detection model.2.A time series anomaly detection model based on improved Hierarchical Temporal Memory(HTM)network is proposed.The encoder of the HTM is improved for the characteristics of CDN traffic,which achieves flexible coding of traffic data and delivers the encoded data to space.High-sparse representation of data and senior sequence learning in the pool and sequence memory to complete the detection of CDN traffic anomalies.At the same time,the calculation method of abnormal probability is proposed to reduce the false positive rate of the model,improving the accuracy of model detection.3.Based on the above research and combined with ELK+Filebeat log analysis system to design and implement CDN traffic anomaly detection system.We introduce and display the design and implementation process of each module,and attack system node by DDoS to simulate the traffic anomaly in CDN.The results show that the CDN flow anomaly detection system based on time series analysis can effectively detect the traffic anomalies occurring in the CDN.