Research On P2P Streaming Traffic Identification

The video service has become a basic application of the current network applications. Thevideo based on P2P technology has become an important part of the network video service.Cisco’s report in2011showed that P2P streaming had became the first major application of thenetwork with an annual growth rate of48%.As the development of P2P streaming traffic, it takesa lot of bandwidth and affects other applications.It is diffic ult to manage the P2P streamingtraffic becacuse of the openness of P2P technology and discrete of P2P nodes. It leads to theproliferation of P2P streaming traffic. Nowdays, content regulation has been a matter of lifeand death to the development of P2P streaming media. In order to ensure that the P2P streamingmedia content regulation Identifying P2P streaming media effectively and quickly has been moreand more important.Supported by "Common Security and Control Framework in Tri-Network Convergence"project of National High Technology Research and Development Program of China (863Program).This article distinguish P2P live and P2P on-demand to fit the actual needs. A methodof identifying P2P streaming traffic based on the decision tree of CART and a method ofidentifying every application in P2P streaming traffic based on the reduction type of SVM aresuppoted to resolve the problem it is diffcult to identify P2P streaming traffic.Then this papercombines these two methods with a method of identifying P2P traffic based on the connenctionof nodes and designes a syetem to identify P2P streaming traffic accurately and timely.The mainwork is as follows:1The behavioral characteristics, traffic characteristics of P2P live streaming, P2Pon-demand streaming and P2P file sharing, are compared.This paper finds their difference andextracts7features.Then it combines these features with the CART algorithm and achieves toidentify the class of the P2P streaming. Experiment results show that: the method does not onlydistinguish P2P l streaming traffic and peer-to-peer file sharing stream, but also drop the P2Plive streaming and on-demand streaming. This method has accurate and take very short time.2The flow characteristics which are extracted to identifying the class of P2P streamingtraffic are further subdivided.Then they are used to identify P2P streaming applications with themethod of SVM.To solve the problem that Multi-class SVM takes a lot of time on trainingsamples and identifying the type of application, this paper propose a reduction type of SVM.Itreduce the number of samples on the time of training, and there is a pre-recognition recognitionbefore SVM. The experiments show that the reduction-SVM has a lower training and recognitiontime.It has a higher accurate and a shorter time to identify P2P streaming applications comparedwith Abacus.3On the basis of the above two methods, combined with the the P2P identification methodbased on the connectiong of nodes, this paper designes a system of identifying P2P streamingtraffic baseded on the connection of super-nodes in the P2P streaming network. The systemare tested, and the results show the accuracy rate of identify P2P streaming traffic based onCART is more than95%.The accuracy rate of identifying P2P streaming applications based on the reduction-SVM is more than90%.At the same time, at least25%of the nodes in the networkcan be identified through the connection of super nodes.The system uses these methods incombination to identify P2P streaming traffic and achieves the need of " Common Security andControl Framework in Tri-Network Convergence".