Classification and Modeling of Internet Applications

Abstract

The classification of Internet traffic is an active research topic due to its applicability in the areas like differentiated services and network security. The introduction of voice, video and other real-time applications to the Internet has resulted in increasing demand for service differentiation and has triggered the need for change in traffic handling on the Internet. Traditionally, such a classification is done using the packet header field of 'port number,' which is a unique number associated with the application that generated the packet. In addition to adding complexity and extra computation in traffic handling, certain recent developments in networking techniques have rendered port numbers unreliable for this purpose. This motivates our scheme of classification that uses the distribution of packet sizes in a buffer or collected during a short time interval at a switch or a router. We demonstrate that the applications can be classified by these distributions. This 'implicit' classification builds a foundation for estimation and prediction of traffic mix, which is a long-term goal of this research project.