Multicommodity flow problems with non-linear and convex costs often arise in traffic allocation of communication networks, as the performance measures are mainly based on average transmission delays due to congestion. The physical network forms an open queuing network where the commodities should be simultaneously routed from their origins to their destinations. Exact performance evaluation methods are available when the arrival and service processes of the network are considered as Poisson processes. In these cases, the network can be decomposed and each network arc is analyzed separately as an M/M/1 queuing system. However, in many other cases these assumptions of Poisson processes are not verified in practical situations. Approximate decomposition methods have been developed to evaluate the performance of queuing networks with general probability distributions, given that for these networks in general there are no known exact methods, or they are of difficult computation. In this work we propose an algorithm that combines multi-flow routing methods and approximate decomposition methods to deal with multicommodity flow problems in general open queuing networks.
multicommodity flow problems; open queuing networks; cycle cancelling algorithms; approximate decomposition methods
