Optimizing the Loads of Multi-Player Online Game Servers Using Markov Chains

Abstract

Multiplayer online games are on the rise with millions of registered player and hundreds of thousands concurrent players. Current state of the art servers achieve scalability by splitting the game world into linked mini worlds that can be hosted on separate servers. One of the problems is that, players are flocking to one area, resulting in an overloaded server with decreasing Quality of Service(QoS) for the players. A number of approaches were developed to address these issues, by balancing loads between the over-loaded and under-loaded servers. This paper investigates a new dimension that is created due to the load balancing of servers. Load balancing among servers is sensitive to correct status information. The Markov based load prediction was introduced in this paper to predict load of under-loaded servers, based on arrival (μ) and departure (λ) rates of players. The prediction based algorithm is proposed to minimize the effect of out dated status information. This approach was compared with normal load status information exchange algorithm. The model presented in this paper does not deal directly with optimizing the load balancing of the server but rather tries to present a new insight that need to be considered when developing load balancing algorithm, that is the reliability of the information that is shared. Simulation results show that Markov based prediction of load information performed better from the normal load status information sharing.