Study On P2P Stream Media Traffic Identification Technology Based On Flow Compound Characteristics

In recent years, P2P technology has witnessed a rapid development, and P2P applications has gradually expanded to more areas from the traditional file sharing applications. With the popularization of processing techniques of mulitimedia information on the Internet, streaming media applications based on P2P framework (P2P streaming media) have been born. P2P streaming media provides Internet users with multimedia service of high-quality and high availability, but this entertaining non-critical flow takes up a lot of network resources and bandwidth, which brings serious impact on the critical business (such as e-government) operations. P2P streaming media application requires high quality of service. Therefore, it is necessary to supervise business of the P2P streaming media in order to limit or ensure P2P streaming media flow. P2P streaming media identification is the foundation of monitoring and managing P2P treaming media. Therefore, how to effectively identify P2P streaming media traffic on the Internet turns to an in-depth exploration topic at home and abroad.Aiming at the critical situation that P2P streaming media identification is still on the primary stage and the unicity of traffic identification method, a method of P2P streaming media identification based on compound fearure which combines signature characteristics of application layer with statistical characteristics of flow attributes for P2P streaming media has been proposed in this thesis. Problems that identification technology based on signature characteristics of application layer can not identify unknown P2P streaming media and signature characteristics is invalid with the high speed updating of softwares have been solved by this method. Meanwhile, the problem that identification technology based on statistical characteristics of flow attributes is difficult to identify in real time has been solved by this method. The main research contents of this thesis are as follows:①A few kinds of P2P traffic identification technologies have been explored, and advantages and disadvantages among these methods have been analyzed. The communication and work interactive principle of P2P streaming media is analyzed, and three flow characteristics of P2P streaming media are discussed. P2P streaming media characteristics based on signature characteristics of application layer and statistical characteristics of flow attributes have been presented. ②Using data mining and machine learning, to take decision tree for an example, statistical characteristics of flow attributes has been selected and the method of P2P streaming media identification based on flow level from the perspective of statistical characteristics of flow attributes has been studied. Modified classification module of double layer has designed and implemented for the enhancing of classification precision rate.③Aiming at the packet in P2P streaming media flow, the automatic extraction method of signature characteristics of application layer has been proposed and implemented. This method extracts packets load signatures from every P2P streaming media flow. The old model of finding signature characteristics of application layer by manual analysis has been improved.④Combining with signature characteristics of application layer and statistical characteristics of flow attributes for P2P streaming media, P2P streaming media identification system based on compound characteristics has been designed and implemented. The identification method based on statistical characteristics of flow attributes is used on off-line analysis module. While the identification method based on signature characteristics of application layer is used on online identification module, and the algorithm of linear classification algorithm based on cache concept has been designed and implemented in this module.