Incentive Issues in Networks


Scott Shenker
Affiliation: Xerox PARC
Abstract: The traditional network algorithm design paradigm assumes that users will cooperate (i.e., use the specified algorithm). While this approach has been tremendously successful so far, I don't think we can continue to neglect incentive issues when we design future networks algorithms; users may not cooperate if it is not in their best interest to do so. Assuming that users are selfish rather than cooperative leads naturally to a game-theoretic approach to networks. In this talk I will give an overview of some recent game theory research as it applies to resource allocation in networks and other distributed systems. Much of my research focuses on the following question: in an environment where users are selfish, which social policies can be achieved through the appropriate choice of allocation policies? We are far from a complete answer to this question, but I will share with you our current state of ignorance. Much of the work reported on here was done jointly with Eric Friedman (Econ, Rutgers) and Herve' Moulin (Econ, Duke).