P2p Iptv Vod System Measurement And Svm-based Traffic Identification Technology

With the popularization of broadband networks, the improvement of computer performance and the emergency of large capacity storage, the Internet-based television (IPTV) services have been emerging one by one, especially P2P IPTV. The traffic generated by the P2P video applications has overwhelmed the internet traffic. Almost of the P2P IPTV system uses private and not open architecture and protocol, meanwhile some applications even use encryption for data. Therefore, it is particularly important to efficiently identify, to measure, to monitor, to guide, to control the traffic generated by the P2P IPTV systems.First of all, an analysis of the development, classification, characteristics and application domain of P2P network is given in detail. The challenges of the P2P network to network security management are described. The property of P2P IPTV and the difference between P2P file-sharing systems is studied, as well as the method of P2P IPTV system measurement.The VoD systems of the very popular four P2P IPTV applications in domestic, namely PPTV, PPStream, Kankan and QQlive, are the object of study. Respectively, the unrelated traffic is filtered from the traces which are collected with the passive measurement in the test bed deployed. The traffic characteristics and peer behavior characteristics are presented based on the study on the final four traces, while the differences among them are studied.Secondly, the traffic identification methodologies, features and measurement indexes are described. The common workflow of supervised classification is presented, as well as State of the art of characteristics and experiment traces in academia circles. The principle and property of SVM (Support Vector Machine) is provided, the kernel function and multi-classification are discussed. The packet load distribution is introduced, and then the dimension of features is reduced. At the end of the part, the rejection mechanism is introduced to decrease precision of identifying the unknown traffic, so the process of supervised classification is revised.Finally, based on the traces which are collected in the test bed deployed, the optimal kernel function, time of window, condensed features and rejection threshold are chosen. The four kinds of P2P IPTV streaming traffic is classified precisely in fine-grain, while the unknown traffic is done too.