Research On Parallel Network Traffic Classification Technique

The task of network traffic classification is to be associated traffic flow with its corresponding application type or protocol type. Accurate identification of network traffic contributes to monitor and manage network effectively for network administrator, which can improve service quality, reduce redundant bandwidth, detect malicious intrusions etc. For new internet traffic flow,traditional network traffic classification technology can't solve the network traffic classification task effectively. The port-based method has a high accuracy rate when there are a few of applications on the internet, but it can't be applied on the new applications which utilize dynamic port or hide themselves port under the well-known port. The method based on payload is more effective than port-based method, but it has very high time and space complexity and it can't identify encrypted traffic flow. Behavior-based method is not suitable for real-time traffic classification. With the extensive application of machine learning, researchers apply the machine learning method based on statistical characteristics of the applications to the network traffic classification. And this method can obtain very good performance. However, many machine learning methods are lack of training dataset for training model and require a lot of computing resources and time costs during the training phase. Moreover, the classification throughput couldn't meet the demand of real-time traffic classification for the exponential growth of traffic.Therefore, researching on a method to solve the real-time network traffic classification issue effectively is more important for new high-speed internet traffic management.Modern computing resources have the powerful computing capabilities with the development of the parallel computing technology. If the computing capabilities are applied in the field of network traffic classification sufficiently, it will be helpful for the real-time network traffic classification. Thus, this paperfocuses on studying parallel network traffic classification techniques based on popular parallel computing resources and machine learning. Firstly, the popular parallel computing technology were studied in this paper. And we adopt two kinds of widely used computing methods, including the framework of Map Reduce and GPU(Graphics Processing Units, GPU) parallel computing respectively, to carry out the parallel optimization of the machine learning methods for studying the parallel network traffic classification. Then, Many experiments were carried out to verify the effectiveness of the parallel network traffic classification methods, which is very much practical significance for the real-time classification to the high-speed internet traffic flows. Finally, this paper designs and realizes a network traffic collection and analysis system called NTCAS. The system could collect the traffic from different clients and monitor them. Also, it provides the offline traffic analysis operations, including annotating the application class to traffic flow, feature extraction, traffic classification etc. In addition, NTCAS extends the function of the traffic analysis and integrates more traffic classification methods. It provides a system platform support for studying on the parallel and real-time network traffic classification.