Research Of Network Traffic Classification Based On Sampling

Network traffic classification, as an important and basic network monitoring tool, is widely used in QoS guarantee, network security testing, billing and other network activities. With high-speed network technology continues to evolve, especially G-bit and T-bit network technologies, the data rate of network is getting more and more fast. The traditional measurement method becomes less effective owing to the fact that a huge overhead would be caused to the router and interface cards. Therefore, more attention is paid to the sampling-based traffic monitoring technology. The research of sampling techniques must focus not only network traffic classification accuracy, but should consider its high efficiency, high-throughput, real-time and other performance parameters more. To achieve this, we studied the network traffic classification technology based on sampling.The network traffic classification methods are introduced at first in the thesis. Then several machine learning algorithms in Weka software are described. After that, the sampling technology is discussed from the aspects of categories and current applications in detail.Based on the above research, the fast locating, sampling and machine learning modules are implemented in our TACS (Traffic Analysis and Classification System).The design of these modules, as well as the details of implementation process of machine learning using Weka source, are described. Combined analysis of the characteristics of network traffic, the measured results of the main applications, especially P2P streaming, are analyzed. Finally, the characteristics of the system are summarized and the performance need to be improved next step is given.