Design of dynamic control policies for stochastic processing networks via fluid models

In this paper we propose a methodology for the design of dynamic policies for scheduling multiclass queueing networks. First, given the solution to a fluid optimization problem, a discrete-review policy is described for translating the fluid optimal control policy into an implementable policy for the stochastic network. Such a policy has been proved to be stable and achieve asymptotically optimal performance under fluid scaling. Using this translation mechanism one can proceed in formulating a fluid optimal control problem which incorporates diverse design and performance specifications, as it is typical in realistic applications. Finally, a simple approximation algorithm for the value function of fluid optimal control problems for a general class of convex performance criteria is described.