Research On P2P Traffics Prediction Models At Application Layer In Metro Area Network

With the rapid development of network communication technology, the Internet begins to carry more and more application services, it presents very high demands for network quality of service, traffic control and network management. Analysis and forecast of flow are the foundation of network management and performance analysis: the operating rules and characterizations of the network are obtained through analyzing flow; modeling based on the flow characteristics, can not only forecast flow patterns of behavior, but also be applied to other fields such as congestion control and quality of services.At present, the scale of P2P is becoming larger and larger, it not only changing the composition of the current network traffic and patterns of behavior, but also consuming a network bandwidth highly. Therefore, it has seriously affected the performance of traditional Internet services. So, the models and forecast methods of P2P traffic are studied in this paper based on the characteristic of P2P traffic.The main contributions of this paper are as follows:(1) The paper summarizes the achievement and the current situation about studying network traffic at home and abroad, and systematically analysis the various characteristics of network traffic and the features of estimative means of Hurst parameter, and the self-similar characteristic of the P2P network data is validated. At the same time, it pointed out that the traffic prediction model should be based on flow characteristics Analysis, offering the foundation for the following researches.(2) The nonstationarity of the network has big influence on modeling traffic accurately. In this thesis, a method of stabilizing traffic is proposed. It can eliminate the influence that the nonstationarity reacted on modeling traffic in some extent.(3) Because of the periodicity nature of P2P flow, this paper presents a method base on the exponential smoothing method for correcting the final forecast data. The experiments show that this method can eliminate the influence to some extend that singular points reacted on forecast data, improving prediction accuracy.(4) In order to establish the model accord with actual traffic characteristic, including the nature of self-similar and periodicity, a hybrid P2P network flow forecast model that combined with wavelet technology and time series analysis is proposed and the future behavior trend of true traffic data by using the forecast algorithm of this model is forecasted and analyzed. The simulation results show that the hybrid model can not only fully describe and characterize the flow properties, but also forecast P2P traffic behavior more accurate than the traditional methods.