Traffic modeling and classification using packet train length and packet train size

Abstract

Traffic modeling and classification finds importance in many areas such as bandwidth management, traffic analysis, traffic prediction, network planning, Quality of Service provisioning and anomalous traffic detection. Network traffic exhibits some statistically invariant properties. Earlier works show that it is possible to identify traffic based on its statistical characteristics. In this paper, an attempt is made to identify the statistically invariant properties of different traffic classes using multiple parameters, namely packet train length and packet train size. Models generated using these parameters are found to be highly accurate in classifying different traffic classes. The parameters are also useful in revealing different classes of services within different traffic classes. © Springer-Verlag Berlin Heidelberg 2006.