Research On Key Technologies Of Peer-to-Peer Network Traffic Control Management

As a new model of Internet applications, Peer-to-Peer (P2P) technology has rapidly become a hot topic in the world with its traffic having occupied more than 70% of the total internet traffic. Meanwhile,the huge P2P random traffic has not only broken the original network operator's operation and business model but also brings more challenge for ISPs to achieve more effective traffic management.The future development direction is to provide a P2P foundation platform which is operatable and managable, and can coordinate all benefits of users, P2P serve providers as well as network providers to promote the healthy development of the entire P2P industry chain. In order to deal with the challenge of huge P2P traffic, on the one hand the operator expands the investment and increases the band width. On the other hand, they are looking for appropriate technique to reliaze effective traffic management which can limit and guide the P2P traffic so as to avoid the excessive band comsumption by the P2P traffic. Therefore in order to realize reliable data transmission and reasonable network taffic resource distribution, it is impretive to establish corresponding control management mechanism for P2P taffic detection, forecast and control model according to the behavioral characteristics of P2P network traffic.Based on the deep research of P2P traffic control management product, current p2p protocols as well as the relevant techqiues, the thesis studies the key mechanisms for P2P control management including P2P taffic detection technology, P2P traffic forecast algorithm and its control mechanism.The main contribution of this thesis is as follows:(1) Aiming at the existing deficiencies of P2P traffic identification methods, and absorbing the advantages of existing methods, the thesis proposes a P2P traffic classification based on feature selection and flow properties of traffic. A large number of experiments show that SVM algorithm based on FCBF method does well in P2P traffic identification. To improve the sensitivity and accuracy of SVM algorithm , a P2P traffic detection model based on loss function of SVM is also proposed, which uses the concept of fuzzy theory and loss function of SVM algorithm to explain the results of SVM algorithm. Analysis shows that the improved algorithm is more accurate and can be applied to identify those unknown, new P2P protocol or those which adopt encrypted transmission. However, due to the P2P traffic identification method based on feature selection is slow which can only be used for offline recognition, the thesis presents a lightweight stream characteristics based online identification method which is fast and efficient in real-time P2P traffic identification. Experimental results show that the proposed method is feasible.(2) A novel P2P traffic forecasting model based on wavelet analysis is proposed. Combined with Kalman filter and wavelet analysis techniques, the proposed model uses wavelet reconstruction to update and estimate the various components of the smooth part of a class of non-stationary random process, meanwhile the Kalman filtering method is used to achieve a dynamic multi-step prediction . The method uses all their previous information to predict the traffic in a cycle. Compared with the wavelet analysis, the proposed method is real-time and recursive and can be used for estimation and analysis of dynamic real-time network traffic. Simulations show that it achieves good tracking and forecasting performance for its high degree of overlap with the real network traffic and its small forecasting errors. Besides, compared with the traditional non-linear prediction models such as neural networks, the proposed methods has the low complexity and is suitable for real time prediction, which can meet the traffic management requirements of P2P network operators.(3) To realize P2P traffic localization, two methods are thoroughly studied, namely , a P2P caching system and a lightweight node distance measurement based on the CDN routing. The former scheme can not only provide users within the network the control of P2P services and ensure their P2P applications experience, but also can reduce the backbone bandwidth pressure. Therefore the structure and deployment of P2P caching system as well as its measurement in the campus network environment is thoroughly studied in the thesis.The latter node selection mechanism combined the logical distance with the actual geographical distance which is conducive to flow localization. Finally, some simulation results show its significance in P2P traffic localization.(4) In order to verify the effectiveness of the proposed algorithms and methods, we implemented a mobile agent-based P2P network traffic management system. In the P2P traffic detection module of the realized system, we combine the stream-based real-time transmission characteristics and the traditional method of DPI for on-line identification of P2P traffic. In its P2P traffic control module, it is mainly a combination of the traditional methods. The realized system includes two modules which complements each other, and can be used separately for network administrators. By applying research results to the realisitic management system of P2P traffic, the thesis exhibits the significance in practical application.