Research On P2P Traffic Identification Technology Based On Wavelet SVM

With the rapid development of the P2P technology, the application of P2P has taken up 60%-80% of the total Internet traffic .And it seriously affects normal network operation, aroused congestion of network and reduces the performance of other operations. At present, some existing methods of P2P traffic identification, such as port scanning, application signature matching and characterization of traffic, can't identify increasing P2P with dynamic port and encrypted, which make the identification of P2P traffic becomes more and more difficult. There is a need of a P2P traffic identification algorithm that facilities the deployment of a network traffic.SVM has special advantages with avoiding local optimum, overcoming dimension disaster, resolving small samples and high dimension for P2P classification problems, a new way of solve the problem P2P traffic identification.In the paper, we begin with the operating principle of P2P traffic identification and analyze the problem in the identification of P2P traffic and the needed technology to realize the effective P2P identification scheme, and propose a novel model of wavelet function and support vector machine for P2P traffic identification. Focus on the work of the paper has been done is as follows:First, selection of feature vector: according to the node flow shows a different behavior characteristic, from the data packet, network flow, node connectivity of three levels of feature vectors, selected with the behavioral characteristics of the three-dimensional feature vector and as the support vector machine input vector.Second, structure of Kernel function: through the wavelet analysis combined with the SVM method of compact, introduced wavelet functions of Mexican hat to construct SVM kernel function, propose a novel model of wavelet function and support vector machine for P2P traffic identification.Third, training algorithm: propose of a wavelet support vector machine Boosting iteration algorithm is applied to P2P traffic identification. Focus on the learning process by training the wrong sub-samples to improve the generalization ability of learning machine, Reduce of false positive. Final, we verified that the model for P2P traffic identification, acquisition of real P2P network traffic and designed and implemented a SVM for traffic recognizer in the popular modeling software LIBSVM in MATLAB, experiment from the false positive rate and false negative rates were measured and With RBF kernel function SVM recognition model were compared. Experimental results show that the algorithm can effectively improve the accuracy of P2P network traffic identification.