Real-time, interactive, multi-user (RIM) applications are networked applications that allow users to collaborate and interact with each other over the Internet for work, education and training, or entertainment purposes. Multiplayer games, distance learning applications, collaborative whiteboards, immersive educational and training simulations, and distributed interactive simulations are examples of these applications. Of these RIM applications, multiplayer games are an important class for research due to their widespread deployment and popularity on the Internet. Research with multiplayer games will have a direct impact on all RIM applications.

While large-scale multiplayer games have typically used a client/server architecture for network communication, we propose using a peer-to-peer architecture to solve the scalability problems inherent in centralized systems. Past research and actual deployments of peer-to-peer networks show that they can scale to millions of users. However, these prior peer-to-peer networks do not meet the low latency and interactive requirements that multiplayer games need. Indeed, the fundamental problem of maintaining consistency between all nodes in the face of failures, delays, and malicious attacks has to be solved to make a peer-to-peer networks a viable solution.

We propose solving the consistency problem through secure and scalable event ordering. While traditional event ordering requires all-to-all message passing and at least two rounds of communication, we argue that multiplayer games lend themselves naturally to a hierarchical decomposition of their state space so that we can reduce the communication cost of event ordering. We also argue that by using cryptography, a discrete view of time, and majority voting, we can totally order events in a real-time setting. By applying these two concepts, we can scale multiplayer games to millions of players.

We develop our solution in two parts: a cheat-proof and real-time event ordering protocol and a scalable, hierarchical structure that organizes peers in a tree according to their scope of interest in the game. Our work represents the first, complete solution to this problem and we show through both proofs and simulations that our protocols allow the creation of large-scale, peer-to-peer games that are resistant to cheating while maintaining real-time responsiveness in the system.