Masters Thesis

Mirrored Arbiter Architecture: a network architecture for large scale Multiplayer Game

Multiplayer games become increasingly popular; mostly because players rather interact with other humans than artificial intelligence. Multiplayer games are implemented on top of different network architectures based on the characteristics of the games. Typically, multiplayer games run on Client-Server (CS), Peer-to-Peer (PP), Client-Multi-Server (CMS), or Peer-to-Peer with Central Arbiter (PP-CA) architectures. In this thesis, we first examine these architectures from two perspectives: bandwidth requirement at the server and clients, and latency. Client-Server architecture is not suitable for a game that needs fast response from players and requires high bandwidth making it hard to scale when the number of players increases. Peer-to-Peer architecture has short network latency but the bandwidth requirements at clients are too high. Client- Multi-Server architecture resolves the bandwidth requirement problem of CS but it does not resolve the latency problem of CS. Peer-to-Peer with Central Arbiter architecture has short latency and small bandwidth requirement at the server but the traffic load at clients is high and the computational power at the server can create a bottleneck problem. We then purpose Mirrored-Arbiter architecture (MA) that combines the merits of CMS and PP-CA. This architecture has all the benefits of PP-CA, and also resolves two main problems of PP-CA by using interest management techniques and multicast. We implemented a multiplayer game called “TankWar” to valid the design of Mirrored-Arbiter architecture. Both analysis and experiments were conducted to show the effectiveness of Mirrored-Arbiter architecture.