Traffic Identification And Measurement Of VoIP And P2P IPTV Applications

In recent years, VoIP and P2P IPTV have become the popular streaming media applications in the Internet. Due to the proprietary of their protocols, the random port of their sessions and the complexity of their networks, it is a challenge for identification and measurement of their traffic. This paper proposed the efficient algorithms and carried out the measurement and analysis for VoIP and P2P IPTV traffic.A traffic identification algorithm has been proposed for QQ voice application, named B&C, which is based on Bayesian and Chi-square statistic methods. Based on the analysis of source model of QQ voice application, the Bayesian algorithm was applied to identify the voice traffic of the QQ voice application. Based on the analysis of the non-voice traffic of QQ voice application, the heartbeat packet is defined and the Chi-square statistic algorithm is applied to identify the heartbeat packet. The experiment results show that the B&C algorithm is very effective for identification of QQ voice traffic. Its fault negatives are less than 2% and the fault positives are less than 0.1%.The identification object is further extended to all VoIP traffic from the only QQ voice traffic. A traffic identification algorithm is proposed for VoIP application, named HBFBA, which is based on the host behavior and flow behavior analysis. First, in the host behavior analysis, the difference between the port numbers of the communication hosts is calculated to distinguish the VoIP traffic from the traditional traffic. And then in the flow behavior analysis, the entropy of the packet size is calculated to analyze disorder degree of the packet size and self-adaptive estimation values are calculated to analyze the relativity of the adjacent inter-packet times. The experiment results show that HBFBA could identify traffic of many VoIP applications with the low values of fault negatives and fault positives. Moreover HBFBA could maintain its validity for new VoIP applications.A new identification algorithm is proposed for P2P IPTV applications, named IE-NNRBF, which is based on the improved entropy and neural network of radial basis function. It could automatically extract the feature of the packet size and the head format using the improved entropy as the dynamic traffic signature. Then it identifies the traffic by nonlinear classification using the neural network of radial basis function as the classifier. The experiment results show that IE-NNRBF could obtain the accurate rate of more than 98%.The vedio on demand overlay network of P2P IPTV, spanned by PPLive application, is investigated with active measurements. A piece of software, named as VoDCrawler, is designed to actively measure the overlay network and a new model is proposed to solve the problem of the convergent condition of the measurement process. The topologies and the churns of the overlay networks are studied according to the result analysis of the active measurement. It revealed that the overlay network has exhibited the"small-world"property; the peers in the overlay network are impatient and the peer current uptime generally is an indicator of remaining uptime. Network management could be improved further by the optimization and control of these properties and parameters.