A Packet Arrival Model for Wolfenstein Enemy Territory Online Server Discovery Traffic

Abstract

Clients for online multiplayer first person shooter (FPS) games typically discover game servers through a two-step process. Clients initially query a well-known master server for a list of currently registered game servers, and then sequentially probe each game server in the order they were returned by the master server. The starting and stopping of clients over time creates a 24-hour cycle of 'background noise' (probe traffic) impacting on registered game servers, independent of a given server's actual popularity with players. Based on over 10 million probe packets from two topologically distinct Wolfenstein enemy territory servers in 2006, this paper shows that probe arrivals are uncorrelated and exhibit exponentially distributed inter-probe intervals during both busiest and least-busy hours of the 24-hour cycle. A modified Laplace curve is then shown to be a reasonable estimator of lambda for the exponentially distributed probe arrivals during any hour of the day. The ability to easily synthesise probe traffic patterns will augment existing approaches to modeling the IP traffic loads experienced by game servers and network devices attached to game servers.