Modeling player session times of on-line games

One of the most important aspects in determining the global traffic characteristics of on-line games is to model the traffic behavior of the client. While modeling the client ON-OFF times in web traffic has allowed researchers to generate accurate fractal models of aggregate web traffic, similar work has not yet been done for the domain of games. Previous work in client traffic behavior has shown that clients consume a nearly constant amount of resources when they are on. However, the distribution of the length of ON times for clients has not yet been established [1]. In this paper, we study the player session time distribution over a one-week trace of a popular on-line game server. Our results indicate that the session ON times of game clients steeply decays over time, with a knee at about 15--30 minutes. In addition, we show that the session time PDF can be fitted accurately with a Weibull distribution with parameters (β = 0.5, η = 20, and γ = 0), a result that diverges from previous studies of player session times [2].