Research On The Network Traffic Classification And Control Technologies Based On Hidden Markov Model

As the Internet applications enriches, the contradiction between the demand ofnetwork traffic and network bandwidth is increasing continuously, so it becomes veryimportant to apply network traffic management. In the service model of “best effort”,most of the bandwidth may be consumed by non-critical network applications such asP2P file transfer, which leads to the lack of guarantee about the quality of service forkey network applications. To help solving the problem, a possible way is classifying thedifferent application flows among the network traffic, and then applies traffic control asneeded.The general idea of the paper is: firstly combine the time sequence characteristics(syntactic structures) and statistical characteristics in network flow’s interaction usingHidden Markov Model (HMM), which is taken as the basis of traffic classification.Secondly, for the traffic of classified key applications, an improved traffic shapingalgorithm based on token bucket will be applied to guarantee their quality of service.The main works done in this dissertation are as the following:①The method of constructing and application of the HMM based network trafficclassification model is studied. By analyzing the interaction process of the network flow,the approaches of extracting its syntactic structures and statistical characteristic isproposed, which can be used to establish the flow’s HMM. The classical forwardalgorithm and Baum-Welch algorithm are also modified so that they can be applied inthe classification and learning process of network flow HMM.②Technologies of traffic control and shaping based on token bucket algorithmunder Netfilter framework are studied as well. First, the classical token bucketalgorithm is improved by two aspects, one is control accuracy and the other is the logicof data retransmission. Second, considering the differences between UDP and TCPprotocol, the buffer management mechanism is optimized by respectively adopting thebuffer size estimating and Window-sizing. Finally, the bandwidth guarantee mechanismis proposed by applying the closed-loop negative feedback control theory.③The prototype system of network traffic management based on the HMMclassifier and traffic controller is designed and implemented. The implementation of theonline traffic classifier is done by using Java tools. Meanwhile, the Netfilter basedtraffic controller is implemented in the base of the key data structures of Linux network stack and kernel timer. The experiments proved that the function of the prototypesystem is correct and effective.