Characterising And Realtime Classifying The P2P Streaming Traffic Based On The Model

The Peer-to-Peer (P2P) network is a flexible logic overlay network built on the physicalcarrier network based on the thought of “All for one and one for all” and is very strongly robust.While the robustness provides huge development space to the different kind of P2P service, italso brings great challenges to the P2P network management. In the distributed andself-organized P2P overlay network, it is almost impossible to pre-control the network from thecontrol and data plane. Afterwards-control based on the node traffic is nowadays one of the mainmeans to P2P network management. However, the deficiency of P2P traffic classificationmethods still makes the P2P traffic out of control which heavily damages the interests of allparties over the Internet.Considering the great potential of Deep Flow Inspection (DFI) techniques in P2P trafficclassification, the dissertation, supported by National High-Tech Research and DevelopmentProgram of China (863program)“Highly Trustworthy Network Management System”, tries toextract the P2P streaming traffic characteristics by answering the following two scientificquestions:“What are the characteristics of P2P service in the overlay network?” and “How thesecharacteristics shows in the carrier network?”. The dissertation analyzes the P2P streamingtraffic characteristics in the carrier network by mapping the characteristics deduced from the P2Pstreaming overlay models and builds the corresponding traffic classification algorithms. Themain research achievements of this dissertation are as follows:Proposing a practicable real-time P2P traffic classification algorithm. Combining thedual roles property of P2P nodes and the flow traffic dissymmetry between two directions ofsessions, the dissertation proposes a practicable real-time P2P traffic classification algorithmR2TC (Role-based Real-time Traffic Classification), which estimates the roles of nodes based onthe long packet proportion in the last N packets of flows and classifies the P2P traffic accordingto the dual roles of nodes. The algorithm just focuses the flows including the long packet, whichreduces its memory consumption, and considers only last N packets, which fulfill the demands ofreal-time P2P traffic classification. The evaluation results show the algorithm is able to classifythe P2P traffic with at least93%byte recall and90%precision.Building the stationary transmission model of P2P live streaming with asynchronousbuffer states and proposing a P2P live streaming traffic classification algorithm based on thefailed sessions. The dissertation builds the P2P live streaming model according to the fact thatthe buffer states of P2P nodes are asynchronous, and extends the model to describe the stationarytransmission according to the success ratio of chunk downloading. The analysis of differentchunk buffer policies between P2P live streaming, P2P Video on Demand (VOD) and P2P filesharing shows that the failed session proportion is higher in P2P live streaming than in the othertwo applications and a Failed-Session-based Traffic classification algorithm for P2P livestreaming (FSTC-live) is proposed, which is able to classify the P2P live streaming traffic with94%recall and91%precision.Analyzing the importance of file fragment for P2P VOD streaming and proposing a P2P VOD traffic classification algorithm based on the file fragment. The dissertation analyzes theimportance of file fragment scheme for P2P VOD streaming system reaching to steady stateresort to present model, and examines its existence by actual P2P streaming traffic. Different filefragment scheme will lead to different packet length distribution. With this viewpoint, aFile-Fragment-based Traffic Classification algorithm for P2P VOD streaming (FFTC-VOD) isproposed, and the trustworthy region is introduced to reduce the misjudgments. The evaluationresults shows that the algorithm is able to guarantee the false positive rate and false negative ratewhen the radius of trustworthy region equals to0.5, and classify the P2P VOD streaming trafficwith at least90%recall.Building the churn model of multi-channel P2P streaming and proposing a P2Pstreaming traffic classification algorithm based on the channel churn of users. The dissertationbuilds the churn model of multi-channel P2P streaming to analyze the influence of channel churnon peer number in each channel, aggregate sojourn time, aggregate residual sojourn time andneighborhood update, which shows that the channel churn of users will lead to the burst ofsuccessful session of nodes. Considering the strong correlation among P2P streaming, channelchurn and burst of successful session, a Channel-Churn-based Traffic Classification algorithmfor P2P streaming (CCTC-streaming) is proposed, and an united streaming traffic classificationmethod is proposed by integrating the algorithms of this dissertation, the evaluation resultsshows that the CCTC-streaming can prompt the performance of P2P streaming trafficclassification to95%recall and97%precision,...