Traffic Classification Based On T-distribution Mixture Model For Multimedia Services

With the development and popularity of Internet in recent years,there are various new types of network services emerging.However,the growth of network services brings great difficulties to network traffic classification and management work.Multimedia services(such as audio services,video services and web browsing)is one of the most important network services,which occupies the largest part of network traffic in the Internet.Therefore,it is important to have a research on traffic classification for multimedia services,which can help ISP(Internet Service Provider)to allocate appropriate resources for users and ensure the quality of Internet services.This thesis focuses on six kinds of multimedia services: online audio services(instant voice communication and network music services),online video services(network live TV and web video streaming)and web browsing services(search engine and network news),and performs a fine-grained analysis.In this thesis,some significant QoS(Quality of Service)related features of multimedia services are analyzed and modeling works of multimedia traffic samples are carried out.The main work is as follows:Traditional Gaussian mixture model(GMM)is susceptible to the influence of edge values and outliers,therefore,student's t-distribution mixture model(TMM)is adopted to improve GMM.This thesis uses the EM(Expectation Maximization)algorithm to build classification model for six multimedia services.Then,limited t-distribution mixture model(LTMM)is presented to reduce the number of iterations for EM algorithm,which is able to cut down the running time of modeling work effectively.Moreover,this thesis makes a comparison among proposed mixture models,traditional K-Means and GMM by theoretical analysis and experiments,which demonstrates the reasonablity of two proposed models.This thesis uses an improved semi-supervised traffic classification method to classify the multimedia services combined with the two proposed mixture models.The method can be divided into three parts: data preprocessing,data modeling and data classification.The clustering center and the fine-grained classification algorithm of the model are discussed in detail.The experiment results show that the semi-supervised classification method can gain a higher accuracy for multimedia services.