Research Of User Behavior And System Capability Of P2P Application

Recently, as widespread adoption of broadband residential accesses and entertainment taking more important role of network applications, VoD (Video on demand) system is getting a tremendous growth. For example, PPLive, a popular P2P streaming system, had100million global installed bases,2million simultaneous online users, and accounted for10%traffic in China's internet in2009. For afford user large amount high definition videos while decrease the load of image servers, service provider often use P2P mechanism which use a small storage contributed by every peer cooperates with image server. In the P2P transmission mode, the user is not only who the service provider to but also provide services for. Their behaviors also have great influence on the performance of system. Therefore, The research of user behavior is not only can help to understand user demand but also have important on improving the overall system performance, reduce server load and provide high quality services.In this paper, we firstly analyze user viewing behavior from many perspectives by measured the actual VoD system. Then, we focus on the impact of user watching behavior on system performance of one common channel. After that a mathematical model is proposed to analyze the relationship between user watching behavior, distribution of chunks, download rate and pre-fetch strategy. The validity of this model is verified by simulation. It provides a more accurate way to study user watching behaviors. It also gives some helps for user behavior study and new pre-fetch strategy design for other P2P applications. By using this model, we show that the difference between user watching progress and downloading progress can reach20%of the entire length of film in reality.Secondly, our research moves forward to the entire VoD system with a huge number of channels. Throw analyze a large amount of user behavior log data which collected by P2P VoD service provider, we find the characteristics of user watching behavior in each time of all channels and then use a model to describe it. Then, an estimate method based on the running rules of population in each channel was proposed. By using practical data, the result show that the relative error of this predictive algorithm of the most popular channels (account for75%population) was nearly11.9%. After that, the model is introduced into the analysis of system performance. The definition of resouce hiting rate is be made to evaluate the system capbility. Further, the relationship among the population in VoD system, server load, total rescources and qulity of service is decuced.At the end of this paper, we study another major application of P2P technology-file sharing system. The research focus on three main factors affecting downloading rate——ATD (availability to demand ratio), the number of concurrent downloaders, the number of simultaneous download tasks. Then, an adaptive resource scheduling mechanism concerning about the characteristics of user behavior and downloading rate was proposed. Experiment results on the practical operating system demonstrated that the proposed mechanism can increases the downloading rate by20%to30%, decrease the server load by20%. Currently, it was become a useful management tool for a well-known Internet service provider.