Design of Price Mechanisms for Network Resource Allocation Via Price of Anarchy

Abstract

We study the design of price mechanisms for communication network problems in which a user's utility depends on the amount of flow she sends through the network, and the congestion on each link depends on the total traffic flows over it. The price mechanisms are characterized by a set of axioms that have been adopted in the cost-sharing games, and we search for the price mechanisms that provide the minimum price of anarchy. We show that, given the non-decreasing and concave utilities of users and the convex quadratic congestion costs in each link, if the price mechanism cannot depend on utility functions, the best achievable price of anarchy is 31.4%. Thus, the popular marginal cost pricing with price of anarchy less than 33.3% is nearly optimal. We also investigate the scenario in which the price mechanisms can be made contingent on the users' preference profile while such information is available.