9610486 Bertsimas Multiclass queuing networks give rise to a rich class of models for manufacturing, and computer-communication systems. The practical needs to evaluate and improve the performance of such systems have motivated extensive research efforts on the analysis, optimization and stability of these models. This research will make contributions to the development of solution methods for several open problems in the analysis of multiclass queueing networks. Specifically, the investigator will study the effects of variability, changeover times, and general holding-time distributions in developing improved bounds on the optimal performance of these networks. The research draws on a mathematical programming-based approach (the achievable region approach) to performance optimization of stochastic systems. The overall objective of the research is to derive heuristic scheduling policies with sub-optimality guarantees based on relaxations derived from the achievable region approach. Multiclass queueing networks have proved quite useful in the computer and communications field, where they have become the models of choice for design, analysis, and optimization of telecommunications systems. While their use as models for manufacturing systems has thus far been impeded by certain technical limitations, this research specifically addresses issues that can potentially overcome these limitations. The potential is high that this research will lead to significant breakthroughs on some extremely difficult problems, thereby enabling the successful application of these models in manufacturing systems, most notably in the area of semiconductor manufacturing.
