Misclassification of game traffic based on port numbers: a case study using Enemy Territory

The identification of game traffic in the Internet is very useful for a number of tasks. For trend analysis it is important to find out how much game traffic is in the Internet and how much traffic certain games contribute. To provide better than best effort QoS for game traffic in the network it is necessary to identify game traffic before it can be prioritised. Traditionally, network applications have been classified based on port numbers. It has been argued that purely relying on port numbers does result in significant number of unidentified flows for applications such as peer-to-peer file sharing and game traffic. While this has already been shown for peer-to-peer traffic no such studies exist for game traffic. In this paper we focus on one particular game and estimate how much of the traffic cannot be identified when solely relying on port number based identification. We find that the number of game flows using non-default port numbers is significant. Our evaluation is based on real traffic captured at clients and public game servers.