| With the rapid development of Internet, the current traffic is much more complex than the past, while a large number of application have been widely used, for example the field of search engines, instant messaging, online shopping, banking, game, and so on. While the development of Internet business greatly facilitates and enriches people’s lives, study and work, it brings enormous challenges to the ISPs. In order to providing QoS, there are two problems that have to be solved. One is how to achieve traffic identification, the other is how to route different traffic according to their categories. This paper focus on these two problems, and the main work is concluded as follows:(1) studies on the traffic identification algorithms, including the techniques using well-known TCP/UDP port number, deep packet inspection, well-know protocol behavior and machine learning algorithms, and analyses the advantages and disadvantages, then concludes that the algorithms based on supervised machine learning is the best. And more research is done to prove SVM(Support Vector Machine) could achieve the more excellent performance than other supervised machine learning algorithms.(2) In order to improve the capability of SVM in real-time traffic identification, the algorithm Adaboost is introduced. And a new algorithm based on SVM and Adaboost is proposed. Through the real-time traffic identification experiment, it is proved that this algorithm can achieve moderate accuracy and performs better than initial SVM.(3) Analyses the current technologies for QoS and their disadvantage, and designs a new routing control architecture on which data plane and control plane are separated.(4) Based on the real-time traffic identification and routing control architecture that are both proposed in this paper, a system of traffic identification and routing control technology is implemented. Finally, the test indicts that our system is available. |
PDF Full Text Request