B-A Cassell, T Alperovich, MP Wellman, and B Noble

Sixth Workshop on the Economics of Networks, Systems, and Computation (NetEcon), June 2011.


Users of wireless networks increasingly face a choice among multiple available access points. Clients generally make this decision with limited information about the access points or traffic trends in the system. We examine the strategic implications of an emerging wireless technology: utilizing multiple access points (APs) simultaneously. Clients using this technology require up-to-date information about the expected delays at available APs, which can be obtained through active probing. We model this scenario as a load balancing game, augmented to incorporate abstractions of these two technologies. Using techniques of empirical game-theoretic analysis, we evaluate a range of plausible strategies through simulation. We find that variants of the Hedge algorithms, previously shown effective at the single unit load balancing game under the bulletin board model, remain promising for scheduling multiple jobs per period; however, when delay information can only be obtained from using or probing an access point, all variants of the Hedge algorithm we examined were outperformed by simple decision-theoretic optimization policies.