Prediction of online game performance degradation under network impairments

It is known that network impairments cause degradation in the online playing experience. Awareness of this degradation can enable game servers to take adaptive action that can mitigate or alleviate the game degradation quickly before it causes a player to leave the game in frustration. In this paper, we focus on a first person shooter game and determine the impact of network impairments on game performance using experimentation with player bots. We analyze game metrics such as the affected player´s score, accuracy and effectiveness in shooting and taking evasive action. We show the use of statistical and machine learning techniques to determine the set of game metrics that can be used to discriminate between game states in near real-time. Our results indicate that the game state classifiers were very accurate in detecting high levels of impairments and were also reasonably accurate down to the time scale of 20-second intervals. These prediction techniques can be incorporated into gaming middleware to enable the mitigation of network-caused impairments.