Fluid Network Models: Linear Programs for Control and Performance Bounds

It has been shown recently that uid limit models provide the means to substantially simplify the stability analysis of multiclass queueing networks. This paper begins to address the most natural second question: do uid models provide a tool for developing high performance scheduling policies for queueing networks? A linear program is constructed to give upper and lower bounds on achievable performance over all work conserving policies. The bounds are tested through simulations using Simulink. Stochastic translations of the control policies are tested through simulations using UltraSAN on a stochastic-discrete model of the network.