Referee-based architectures for massively multiplayer online games

Abstract

Network computer games are played amongst players on different hosts across the Internet. Massively Multiplayer Online Games (MMOG) are network games in which thousands of players participate simultaneously in each instance of the virtual world. Current commercial MMOG use a Client/Server (C/S) architecture in which the server simulates and validates the game, and notifies players about the current game state. While C/S is very popular, it has several limitations: (i) C/S has poor scalability as the server is a bandwidth and processing bottleneck; (ii) all updates must be routed through the server, reducing responsiveness; (iii) players with lower client-to-server delay than their opponents have an unfair advantage as they can respond to game events faster; and (iv) the server is a single point of failure.The Mirrored Server (MS) architecture uses multiple mirrored servers connected via a private network. MS achieves better scalability, responsiveness, fairness, and reliability than C/S; however, as updates are still routed through the mirrored servers the problems are not eliminated. P2P network game architectures allow players to exchange updates directly, maximising scalability, responsiveness, and fairness, while removing the single point of failure. However, P2P games are vulnerable to cheating. Several P2P architectures have been proposed to detect and/or prevent game cheating. Nevertheless, they only address a subset of cheating methods. Further, these solutions require costly distributed validation algorithms that increase game delay and bandwidth, and prevent players with high latency from participating.In this thesis we propose a new cheat classification that reflects the levels in which the cheats occur: game, application, protocol, or infrastructure. We also propose three network game architectures: the Referee Anti-Cheat Scheme (RACS), the Mirrored Referee Anti-Cheat Scheme (MRACS), and the Distributed Referee Anti-Cheat Scheme (DRACS); which maximise game scalability, responsiveness, and fairness, while maintaining cheat detection/prevention equal to that in C/S. Each proposed architecture utilises one or more trusted referees to validate the game simulation - similar to the server in C/S - while allowing players to exchange updates directly - similar to peers in P2P.RACS is a hybrid C/S and P2P architecture that improves C/S by using a referee in the server. RACS allows honest players to exchange updates directly between each other, with a copy sent to the referee for validation. By allowing P2P communication RACS has better responsiveness and fairness than C/S. Further, as the referee is not required to forward updates it has better bandwidth and processing scalability. The RACS protocol could be applied to any existing C/S game. Compared to P2P protocols RACS has lower delay, and allows players with high delay to participate. Like in many P2P architectures, RACS divides time into rounds. We have proposed two efficient solutions to find the optimal round length such that the total system delay is minimised.MRACS combines the RACS and MS architectures. A referee is used at each mirror to validate player updates, while allowing players to exchange updates directly. By using multiple mirrored referees the bandwidth required by each referee, and the player-to mirror delays, are reduced; improving the scalability, responsiveness and fairness of RACS, while removing its single point of failure. Direct communication MRACS improves MS in terms of its responsiveness, fairness, and scalability. To maximise responsiveness, we have defined and solved the Client-to-Mirror Assignment (CMA) problem to assign clients to mirrors such that the total delay is minimised, and no mirror is overloaded. We have proposed two sets of efficient solutions: the optimal J-SA/L-SA and the faster heuristic J-Greedy/L-Greedy to solve CMA.DRACS uses referees distributed to player hosts to minimise the publisher / developer infrastructure, and maximise responsiveness and/or fairness. To prevent colluding players cheating DRACS requires every update to be validated by multiple unaffiliated referees, providing cheat detection / prevention equal to that in C/S. We have formally defined the Referee Selection Problem (RSP) to select a set of referees from the untrusted peers such that responsiveness and/or fairness are maximised, while ensuring the probability of the majority of referees colluding is below a pre-defined threshold. We have proposed two efficient algorithms, SRS-1 and SRS-2, to solve the problem.We have evaluated the performances of RACS, MRACS, and DRACS analytically and using simulations. We have shown analytically that RACS, MRACS and DRACS have cheat detection/prevention equivalent to that in C/S. Our analysis shows that RACS has better scalability and responsiveness than C/S; and that MRACS has better scalability and responsiveness than C/S, RACS, and MS. As there is currently no publicly available traces from MMOG we have constructed artificial and realistic inputs. We have used these inputs on all simulations in this thesis to show the benefits of our proposed architectures and algorithms.