The Research Of Network Traffic Classification And Its Algorithms

Network applications are more complex and have more kinds with the fast development of network. Many kinds of network applications legal and lawless not only lick up more network resource but also threaten both security and QoS consideration. Because of the advantages on load proportion and using resource, traffic engineering has attracted a great deal of interest, on which traffic classification pays important roles in many areas such as network management, traffic inspection, service class mapping, inspection of security and errors, network charging.There are several methods which are used to traffic classification, such as based on port, based signifiture and BLINK. But these methods have some disadvantages. The use of statistical techniques to detect network applications recently has received a great deal of interest. This method rely on features of the traffic statistics e.g. packet size distribution, packet interval time etc, which don't detect the payload of packet. In fact there are two kinds of traffic classification using machine learning: supervised and unsupervised. For example, Bayse and EM are supervised. K-means is unsupervised. Supervised have the mentor signal but unsupervised don't. This method has gotten high accuracy.Supervised depends on the prior manual analysis which is infeasible to cope with the fast growing number of new applications. One of the challenging issues for existing detection schemes is that we need not only identify known traffic but also detect unknown traffic. We use Self-organizing Mapping (SOM) to classify the traffic in this paper, which doesn't need prior manual analysis and carries on the study of data through self-organizing.In order to avoid the affection of network condition, we use the concept of the change ratio of time gap to select appropriate traffic features. We carry out the model of traffic classification based on SOM and achieve the aim of detecting new traffic.