Decentralized algorithms for flow control in networks

Abstract

Within the context of flow control problems in data communication networks, we consider decentralized algorithms which optimize a criterion reflecting the trade-off between the throughput and the total delay in a virtual circuit. We show that the "greedy" algorithm which provides individually optimum solutions for the users converges when the criterion is the power of the virtual circuit. We also present an algorithm executed using decentralized information which converges to a set of Pareto-optimum message rates among the users. Both of the algorithms make use of the local information available to end users and no exchange of information is needed.