Monitoring and Analyzing a Game Server Scenario

Abstract

Today, most literature about services in system administration is about conventional services like email servers. How could one monitor and analyze a scenario where the service in question is a game server? As these two services are technologically different, conventional monitoring tools may miss vital information in the context of game servers.

This thesis focuses on developing a monitoring system for a game server in order to learn and understand the characteristics of a game server process in a production environment. An experiment is carried out to control some of the influencing variables, like the number of players and game server instances, and to observe the system under the different conditions. Results show that the number of instances did not affect the overall performance in that way we expected. The concurrent players on the server dominates the CPU load. We find that a strong linear relationship exist between these two variables. When it comes to memory usage, players barely affect this resource in our experiment, but the number of game server instances (Team Fortress 2 dedicated server) does. The server process allocates most of its needed memory in the beginning. The amount allocated depends on which map is set when the game server is executed.

This study has shown that the game server, in this case Team Fortress 2, is a predictable service in terms of resources.