Fast reconfigurable communication networks

Two distributed channel allocation algorithms are presented for a reconfigurable communication network with a multiple-channel link architecture. In this architecture, every switching node can dynamically allocate transmitters and receivers among links to match traffic fluctuations and hence improve the average delay and throughput performance over that of a fixed-link architecture. The authors show that the first algorithm always converges to an optimum configuration while assuming that link may never be saturated and requiring global information on channel allocation. The second algorithm needs local information only and poses no limitation on link capacity but may terminate at suboptimum configurations. The conditions under which the local search algorithm may fail to converge to an optimum configuration is also discussed.<>