Data stream mining based real-time highspeed traffic classification

Abstract

In current high-speed network, Peer-to-Peer (P2P) applications have overtaken Web applications as the major contribution on the Internet. Thereby, how to identify P2P traffic in real-time accurately and efficiently is a key step for network management. In this paper, we highlight the importance of applying data stream method in traffic classification to achieve real-time P2P traffic identification. We not only introduce a VFDT-based real-time highspeed traffic classification method, but also take thoroughly analysis on how to select a reasonable tie confidence (TieC), minimum gathering flow (MinGF) and category number (CaNum). Meanwhile, analysis has been done to ascertain the packet's interval which is used to calculate flow's real-time attribute. Experiment results have shown that when TieC is less than threshold, the larger TieC is, the better accuracy of identification is; when TieC exceeds threshold, decision trees are the same. Concerning MinGF and CaNum, although the smaller both of them are, the better performance of decision tree is, the value of them must be properly set according to requirements of classification system.