Robust Fluid Control of Multiclass Queueing Networks

This thesis applies recent advances in the field of robust optimization to the optimal control of multiclass queueing networks. We develop models that take into account the uncertainty of interarrival and service time in multiclass queueing network problems without assuming a specific probability distribution, while remaining highly tractable and providing insight into the corresponding optimal control policy. Our approach also allows us to adjust the level of robustness of the solution to trade off performance and protection against uncertainty. We apply robust optimization to both open and closed queueing networks. For open queueing networks, we study control problems that involve sequencing, routing and input control decision, and optimize the total holding cost. For closed queueing networks, we focus on the sequencing problem and optimize the throughput. We compare the robust solutions to those derived by fluid control, dynamic programming and stochastic input control. We show that the robust control policy leads to better performance. Robust optimization emerges as a promising methodology to address a wide range of multiclass queueing networks subject to uncertainty, as it leads to representations of randomness that make few assumptions on the underlying probabilities. It also remains numerically tractable, and provides theoretical insights into the structure of the optimal control policy. Thesis Supervisor: Dimitris J. Bertsimas Title: Boeing Professor of Operations Research