Research On Real-time Network Traffic Classification Algorithm

The rapid development of Internet witnesses the growth of network trafficdiversity and complexity in recent years, which has posed unprecedented scalingchallenges for many areas such as network QoS and network security safeguard.Network regulators can’t take effective management measures without exactlyknowing about network traffic. So network traffic classification plays an importantrole in network administration, it has great realistic significance to research networktraffic classification algorithm with high level of accuracy and efficiency.At present, the research on network traffic classification mostly stay at the levelof protocol, and there are few targeted research on encrypted traffic classification. Inaddition, the efficiency of existing algorithm is difficult to meet the requirement ofpractical application. In view of these issues, real-time traffic classificationalgorithms on media channels and encrypted network service were studied in thisthesis.Firstly, the formal definition and evaluation criterion of network trafficclassification were presented. Then three traditional network traffic classificationalgorithms were analyzed and compared.Secondly, a peer-to-peer media channel traffic classification algorithm based onsupport vector machine was proposed. The characteristic set consisting of five flowfeatures was constructed through the analysis of UDP media channel traffic. Theexperimental results showed that the algorithm has good generalization ability andefficiency, the overall classification accuracy was96%.Then, an encrypted network service traffic classification algorithm based ondecision tree was proposed. Through the analysis of four HTTPS application, thelength sequence of first eight packets was selected as a flow’s characteristic.Characteristic-Inducing algorithm and Range-Splitting algorithm were presented toimprove the efficiency of model training. The algorithm was proved to be simpleand efficient in the experiment, and the classification accuracy reached above95%Finally, a real-time encrypted network traffic classifier system was designedand implemented in a high speed network environment with large traffic. Thesystem relies on high performance packet capture platform, can identify the targetnetwork service traffic fast and accurately, then make real-time classification ofdifferent applications, has good stability and low resource occupancy rate.