With the increasing development of P2P(Peer to Peer) applications,There are much P2P traffic of file sharing, streaming media, instant message and many other types in the network.P2P not only fully utilizes the network bandwidth, but also consumes the network resource excessively.The rapid growth of various P2P application causes a great burden to the router and makes a large number of data packet loss and network delay.The network congestion becomes more serious,so accurate identification of P2P traffic makes great sense to efficient network management and reasonable utility of network resources.Because of the different QoS(Quality of Service) requirements of different P2P application, the researches on P2P traffic control mainly focus on how to allocate the network resources reasonably.This thesis analyzes the characteristics of P2P traffic, the typical identification technology of P2P traffic and the traffic manage strategy, then a network traffic classification method based on improved binary tree SVM and a network traffic control method based on classification are proposed.The main contributions are:(1) The generation of P2P, technical characteristics and flow characteristics are introduced. This thesis summarizes the common identification technology of P2P traffic and analyzes the typical queue management and queue scheduling algorithm.(2) A P2P traffic classification method based on improved binary tree SVM is proposed. In order to reduce the "errors accumulation" and improve the accuracy of classification, a new metrics which synthesizes the distance of classes centers and mixed degree to construct the binary tree is proposed. Meanwhile, both the single attribute and attribute subset selection are used to improve the efficiency of classification. Finally, The results of experiments with P2P traffic show that our method is superior to the classic"1-v-1"and"1-v-r"SVM methods in both classification accuracy and efficiency.(3) A network traffic control solution for the different P2P applications is proposed. The solution manages network congestion based on both the management and the scheduling of queue. the BLRE algorithm which is based on packet loss ratio to realize the management of queue is proposed, The BLRE combines the event probability adjustment in BLUE and the buffer partition strategy in RED to determine the loss ratio. It can keep the stability of the queue and adjust well for the burst traffic. Because of the introduction of the ratio of the average queue length in the calculation of the packet loss rate, The adjustment of the packet loss rate is more fair and reasonable. For the scheduling of queue, this thesis proposes the BAQS algorithm which dynamically adjusts the bandwidth weights based on the average queue length, The algorithm allocates different bandwidth to various applications according to the current status of traffic. In order to reflect the fairness, the adjustments ensure that high priority queue in priority to meet the requirements and also take into account the fairness of the low priority queue. The theoretical analysis and simulation experiment proves the effectiveness of the solution of the network traffic control. |