P2P traffic detection on large scale netflow data

P2P traffic detection is one of the hot topics in communication network area. It is meaningful especially for telecommunication operators, because P2P applications could lead to heavy network congestion. To allocate the network bandwidth more reasonably, the first and fundamental step is to identify the P2P traffic. Up to now, some P2P traffic detection methods have already achieved high accuracy. But almost all of them are designed for tracing data in packet granularity, and it is not appropriate for detecting P2P traffic in large scale real world telecommunication network. Furthermore, most existing methods mainly integrate several different detection rules together without any structure, and thus have a limitation on extensibility. In this paper, we design a P2P traffic detection framework over large scale NetFlow data. It can be easily applied to daily network traffic analysis. It is of modular structure and therefore good extensibility. The framework is based on traffic behavior analysis. The accuracy of the detection is tested using real world NetFlow data collected at Shanghai Telecom, and it is shown that the proposed framework is able to achieve relatively high accuracy and flexibility at the same time.