Harvard Theory of Computation Seminar

Flow Allocation Games: Pricing, Equilibria and Fast Convergence

John Byers, Boston University



Place and Time: Monday 4/24, Refreshments at 2:30, talk at 2:45. Maxwell Dworkin 319.

ABSTRACT

In this talk, I will begin with a quick tour of the theory and networking practice of distributed flow allocation problems: given a fixed capacitated network in which a set of long-lived flows is given and the routes these flows use are pre-ordained and fixed throughout, assign rates to the flows to achieve a global objective using a distributed algorithm.

I will then focus on a game-theoretic version of flow allocation in which flows play an iterative game running in a set of distributed rounds. In each round, each flow bids for network resources by distributing a fixed set of tokens across links in its path, and link bandwidth is then apportioned in proportion to the number of tokens a flow has placed on the link. In subsequent rounds, the flows may then reallocate their set of tokens across links to improve their allocation. We characterize the Nash equilibria of this game, relate them to the equilibria in the TCP Vegas analysis, and describe strategies which converge to the equilibria.

Joint work with Danny Raz (Technion).