Adaptive distributed b-matching in overlays with preferences

Rapport, 2012

An important function of overlay networks is the facilitation of connection, interaction and resource sharing between peers. The peers may maintain some private notion of how a "desirable" peer should look like and they share their bounded resources with peers that they prefer better than others. Recent research proposed that this problem can be modeled and studied analytically as a many-to-many matching problem with preferences. The solutions suggested by the latter proposal guarantee both algorithmic convergence and stabilization, however they address static networks with specific properties, where no node joining or leaving is considered. In this paper we present an adaptive, distributed algorithm for the many-to-many matching problem with preferences that works over any network, provides a guaranteed approximation for the total satisfaction in the network and guarantees convergence. In addition, we provide a detailed experimental study of the algorithm that focuses on the levels of achieved satisfaction as well as convergence and reconvergence speed. Finally, we improve, both for static and dynamic networks, the previous known approximation ratio.