Traffic modelling of on-line first person shooter games

Over the past few decades, there has been a great deal of attention paid to modelling the traffic generated by First-person Shooter (FPS) games. This interest has been motivated by the large growth in on-line gaming, which is now a multi-billion dollar industry. The FPS genre is of particular importance to the network engineering community due to its highly interactive nature. The interest in this genre is based not only on the effectsof the network on playability and user experience of the game, but also on the effects that traffic produced by this genre of games may have on other traffic flowing in the network. In order that network simulation can provide more insightful and useful results for studying the interaction between this game traffic and other classes of traffic, good traffic models are required to be developed. This thesis makes a number of contributions to the state of the art in game traffic modelling and synthesis. We show that currently used models are incomplete in that they do not scale with the number of users , nor do they capture the autocorrelated nature of the traffic, and they do not take into account behaviour of individual FPS games. We address each of these matters. In particular, a general model for FPS game traffic during the game-play phase is presented. The development of this traffic model was based on measurements of traffic produced by seven popular FPS games. This model, which is based on the Ex-Gaussian and Gamma Modified Gaussian mixture distributions, is novel and computationally simple and it can be scaled to model scenarios with a large number of game participants without having to conduct and collect data from large numbers of controlled experiments. Since this is a computationally simple model, it is ideal for use as the basis of either a hardware or software based traffic generator that can be used to synthesise representative FPS game traffic. This model is an improvement over previous models, as it also incorporates the serial autocorrelation that exists in the packet traffic that flows in the server-to-client direction by using a simple ARMA time-series approach. A traffic generator based on this model has also been developed for use with the OMNet++ simulation software. This traffic model can be used by Internet Service Providers (ISPs) and the network engineering community to investigate network performance issues, arising from scenarios where FPS games and other applications interact and what techniques are likely to be successful in minimizing their impact on each other.