Research On Key Techniques Of Traffic Identification In Broadband Access Network

Nowadays the Internet has developed dramatically, many new technologies, such as the Internet of things and cloud computing, are emerging constantly, and broadband access has already become one of the international development trends. The network traffic has become increasingly complex and dynamic with the increase of network bandwidth and diversification of application types. Accurate and timely identification of network traffic according to application types plays important roles in many areas, such as traffic engineering, QoS and network security management. Therefore traffic identification has been one of the hottest topics in the Internet research field.Traffic of access network have characteristics of high-speed, dynamic, complexity, continuity and large amount of data, concept drift, unbalance distribution among different application types. So traffic identification methods should achieve rapid response and adapt to the concept drifting in the limited storage space and time. The traditional and existing traffic identification methods based on static data set are difficult to adapt to high speed traffic identification in broadband access network.Aiming at the access network traffic identification, we conduct a series of studies, including three traffic identification algorithms and a decision fusion prototype system based on uncertainty evidence reasoning. The main contributions of this thesis are as follow in detail:1. We propose a classification algorithm, named as Adaptive Grading Slide Window Decision Tree (AGSW-DT).AGSW-DT algorithm is based on Hoeffding decision tree. It realizes the different rate adaptive detection of concept and decision tree update by detecting concept drifting according to the Information Gain Ratio of nodes, adjusting concept slide windows and training example set dynamically in accordance with the detection results. Comparing to the experiment results of C4.5and CVFDT, AGSW-DT algorithm overcomes the problem of concept updating incompletely caused by the skewing distribution of data concept. The proposed algorithm can improve the efficiency of concept drifting detection and can obtain more balanced classification accuracy among different application types. The algorithm can apply to the network traffic engineering and bandwidth management fields in terms of known application types.2. Put forward online clustering algorithm OL-DBSCAN based on density and traffic identification scheme.OL-DBSCAN algorithm clusters traffic flows based on sub-flow statistical features instead of full flows for the demands of early traffic identification, and solves the parameter selection problem of clustering algorithm by introducing the Q value. The proposed scheme based on the OL-DBSCAN combines with DPI method. The experimental results show that the scheme is capable of identifying the encrypted traffic, extracting characteristics of new application types, and can adapt to the traffic characteristics change over time. The scheme can apply to network security management, which extracts characteristics of unknown application types for managers to make further analysis and processing. 3. Propose Transport-layer Connection Topological Pattern (TCTP) traffic classification algorithm.TCTP algorithm makes use of the characteristics of different application types showed in the transport layer. Firstly, the algorithm extracts typical patterns of application types to generate the mapping between application types and grid of clustering. And then creates service type pools of nodes, which can be verified by DPI technique. Finally, the traffic would be classified according to the mapping and service type pools. This algorithm is not depends on traffic time statistical information, so it has not time sensitive issues of statistical features as classification and clustering algorithm, and has the high real-time and reliability. TCTP algorithm is a good complementary for classification and clustering methods.4. Propose a multiple classifier decision fusion prototype based on uncertainty evidence reasoning theory.Each traffic identification algorithm has its own advantages. To utilize different algorithms results effectively, the decision information fusion is need to implement. We propose a decision fusion framework based on uncertainty evidence reasoning, and experiments illustrate that the decision fusion can improve the precision of traffic classification substantially, and reduce the rejected rate and the error rate simultaneously. Comprehensive identification results show that the decision fusion prototype is superior to each single classifier in many performance indexes, and gives play to the advantages of each classifier.