| With the emergence of Peer-to-peer (P2P) technology in the late '90s, it is widely used in recent years and P2P traffic has become the most significant portions of the network traffic due to its advantages in terms of file-sharing, in-depth search and distributed computing, etc.. However, although P2P promotes the internet, it brings the security problems such as network congestion because it occupies huge bandwidth. Therefore, accurate identification of P2P traffic plays an important role on efficient P2P traffic management. In addition, the accurate classification of P2P traffic will also play an important role on traffic prediction, dynamic access control and intrusion detection.This thesis presents some kinds of traditional P2P traffic identification methods, and implements the P2P traffic classification algorithm, which is based on the traffic behavior at the transport layer. However, the accurate rate of this algorithm is relatively low, and it cannot realize the online traffic classification. This algorithm is only useful in off-line traffic records analysis. In this thesis, we mainly study on data mining based P2P traffic classification algorithms. This type of algorithms does not require to access the packet payload, and moreover, it can realize the high-accuracy online traffic classification. In this thesis, reseaches have mainly been done to study the application of different typesdata mining techniques on traffic classification, and meanwhile, the method to evaluate the performance of the algorithm of data mining is proposed. And the results of analysis are presented, too. Meanwhile, we propose the concept of real-time feature subset. As the results show, good performance will be achieved when the real-time feature subset is used. Finally, we apply the algorithm of VFDT (Very Fast Decision Tree) in P2P traffic classification, and evaluate the performance of this algorithm in terms of traffic classification through thorough experiments. |
PDF Full Text Request