Distributed Iterative Optimal Resource Allocation With Concurrent Updates of Routing and Flow Control Variables

Consider a set of active elastic sessions over a network. Session traffic is routed at each hop (potentially through multiple network paths) based only on its destination. Each session is associated with a concave increasing utility function of its transfer rate. The transfer rates of all sessions and the routing policy define the operating point of the network. We construct a metric f of the goodness of this operating point. f is an increasing function of the session utilities and a decreasing function of the extent of congestion in the network. We define ldquogoodrdquo operating points as those that maximize f, subject to the capacity constraints in the network. This paper presents a distributed, iterative algorithm for adapting the session rates and the routing policy across the network so as to converge asymptotically to the set of ldquogoodrdquo operating points. The algorithm updates session rates and routing variables concurrently and is, therefore, amenable to distributed online implementation. The convergence of the concurrent update scheme is proved rigorously.