Research On The Approach Of P2P Network Traffic Classification Based On Network Trafifc Characteristics

P2P technology has been developing rapidly. But meanwhile, it also brings aboutsome problems including network congestion, the issues of network security andintellectual property protection and so on. So as to promote network managementperformance and to avoid the above problems, research on P2P flow identification hasbecome the most important problem.Through comparison of basic principles of sereval popular machine learnings andapproaches of different P2P flows classification. The approach of P2P flowclassification based on P2P network traffic characteristics is intensively discussed, andthe pivotal traffic characteristics affecting traffic classification are explicated.Introduction of the support vector machine technique is beneficial to constructing modelof P2P flow classification based on specific traffic characteristics, the model issimulated in the LIBSVM tools. The main works of the thesis are as follows:(1) The basic principles and methods of supervised machine learning algorithms,semi-supervised machine learning algorithms and unsupervised machine learningalgorithms are studied, the approaches of P2P flow classification on ports, session,deep packet inspection and network flow behavior characteristics are analyzed. andfinally the approach of P2P flow classification based on network traffic characteristics isintroduced, it brings in a monitoring process on netwrok traffic characteristics andfocuses on the definition and generation of network flow including online packetscapture, and the key characteristics working upon traffic classification is illustrated.(2) The approach of P2P flow classification based on the SVM is proposed andexpounded, and it emphasizes how to choose the SVM kernel function and the SVMalgorithm properly to do flow classification. The classification process based on theSVM and the Multi-classification model named DAGSVM model based on P2Pnetwork traffic characteristics are provided including the corresponding algorithm. Thepenalty parameter value and the kernel parameter value of SVM are proposed to getthrough Grid search and k-flod Cross-validation.(3) Simulation is implemented with the aid of the LIBSVM tools. To meetexperimental accuracy, the experimental data called Auckland Set are acquired fromUniversity of Auckland. Make full use of the characteristics of little samples,nonlinearity, high dimensions of support vector machine, the flow classification modelis used to predict the input data. Compared with the traditional classification approaches based on the SVM, and its performance are discussed and analyzed, clearly it is veryexcellent.