Research On Hierarchical Traffic Classification System For Campus Network

With the popularity of the Internet, the growing network traffic and emerging variety ofnew applications has greatly increased the complexity and management difficulty of thenetwork, which seriously threatens the network quality of service and network security, andpresents a huge challenge to network management. In this case, the network trafficclassification has gradually become an academic research hotspot, because it has a good abilityto identify network applications, and can provide information to support network management.Due to real-time, perishable and non-reproducibility of network traffic, traffic classificationmust have a capacity of identify traffic timely, thus play a more realistic significance of networkmanagement. Therefore, network traffic online classification becomes a focus of the researchtraffic classification.Through study of the principles and status of various network traffic classification, werealize that the most current network traffic classification methods are single-point mode,which focuses on improve classification accuracy, but ignores the problem that the growingexport traffic cannot be classified timely because of lack of high performance. In order toachieve online network traffic classification, we study the distributed technology, break theprevious single-point mode, and finally propose a hierarchical network traffic classificationmethod. The method deploys multiple classification nodes to low-speed links for online trafficclassification, and summaries their results by the center to achieve the whole network trafficonline classification. The method has two features. First, it avoids directly classifying exporttraffic of campus network and overcomes single-point’s performance defect of online trafficclassification; Second, classifier dynamic loading mechanism can achieve new applicationsidentification, classifier training and dynamically update, so it can respond effectively to thephenomenon of declining accuracy caused by concept drift. Hierarchical network trafficclassification is an effective solution for online traffic classification.Based above method, a hierarchical network traffic classification model is designed,including the central node, the classification node and Web integration management. Centralnode and classification nodes complementary perform campus network traffic identificationtask. Web integration management is Web implementation of classification system management. In the implementation process of classification model, a hierarchical networkinteractive protocol and a classifier distribution management method are designed; unknowntraffic find and identification function based on semi-supervised learning and classifier trainingfunction are implemented; LIBPCAP platform, visualization techniques are used to captureonline traffic and classification results display. Finally, the experiment and test illustrate thefeasibility of hierarchical online traffic classification and the practicability of the Webmanagement.