Uncoordinated Load Balancing and Congestion Games in P2P Systems

In P2P systems, users often have many choices of peers from whom to download their data. Each user cares primarily about its own response time, which depends on how many other users also choose that same peer. This interaction is best modeled as a game among self-interested agents, which we call uncoordinated load balancing. The players in this game are the rational and strategic users who are free to act in their own self-interest. We describe some of our recent work on this problem, and propose several new research directions, including analyzing Nash equilibria under general latency functions, a cost to switch servers, settings where user groups are dynamic, as well as the complexity of finding Nash solutions, and incentives for peers to be truthful in revealing their load.