Classifying and Clustering the Internet Tra?c by Kohonen Network

摘要

In this paper we report progress done in the development towards an automated classifier for Internet core-network traffic patterns. The traffic in the Internet has been growing exponentially during the last ten years. The traffic engineering requires more substantial information on the traffic patterns and classes in the actual network. One of the most promising methods we have been studying is to use unsu-pervised clustering methods, such as self-organising maps a.k.a. Kohonen networks to search traffic classes from the measured network traffic. The self-oganised map is well suited for traffic classification, because it does not require any preclassified training data. We describe our basic methods, data analysis and preliminary results achieved. Results show that the described method is promising enough for further studies.