Optimal flow control of multiclass queueing networks with partial information

Structural results and explicit solutions for the optimal flow control problem of multiclass queueing networks with decentralized information are given. Two criteria are investigated: the network (respectively, user) optimization criterion maximizes the average network (user) throughput subject to an average network (user) time delay constraint. It is shown that these problems can be analyzed in terms of an equivalent network by using the generalized Norton's equivalent. The structure of the network (user) optimization problem is exploited to obtain further structural results, namely a representation (separation) theorem. The optimal flow control under both criteria is solved using a linear programming formulation. The structure of the optimal control is shown to be of a window type in both cases. For load-balanced networks, the optimal flow control is found explicitly in terms of the given system parameters.< >