AbstractInternet delay is caused by transmission medium and also by routing and queuing. Although delay is often tolerable in many applications such as file transfer or Web, it does affect performances for delay-sensitive applications such as IP phone and video streaming. For First Person Shooter (FPS) games, where delay is even more important, Internet delay can determine who wins or loses a game. The previous research showed that player at a distance was disadvantaged due to larger delay. Network traffic for such games was analysed using packet size and inter-packet time metrics fitting a statistical model. However, no research explored the relation between “clean” and “interfered” traffic, only Joyce (2000), Bangun (2000), Jehaes (2003) and Carrig (2005) explored the interactions between game and other traffic, and the effect of other traffic on the game server-client delay was overlooked. This dissertation explores the traffic generated by an FPS game, using Quake III as an example, taking into account the effects of network delays.
An FPS game was played both in an isolated local network with no other traffic and was repeated across the Internet to explore the difference between “clean” test-bed traffic and realistic Internet “interfered” traffic. The metrics of packet size and interpacket time of empirical traffic were analysed. A simple statistical approach was applied using goodness-of-fit test based on the probability calculation to produce a mathematical model. The results were used to set up a simulator experiment to further explore the behaviour of game traffic under varying degrees and also different types of network traffic.
The approximate statistical distributions for “clean” test-bed game traffic were based on packet size and inter-packet time. As to packet-size, server traffic had extreme probability distribution, while client traffic was simply distributed in a short interval. As to inter-packet time, both server and client had deterministic distribution. Interarrival time in “interfered” network was totally different from inter-send time and was network dependent, and this was confirmed in a simulation experiment. The difference with results in previous research was also explained by the simulation results. The game delay was not only affected by volume but also by inter-packet time behaviour of network traffic, because simulation showed that traffic streams with same parameters but different inter-packet time affected game traffic delay differently. The pseudo game traffic generated by analytical distribution can replace the realistic game traffic in simulation experiment.
|Date of Award||Jun 2008|