The Throughput Analysis of Product-Form Queueing Networks

Product-form closed queueing networks are widely-used analytical models for the performance analysis of computer and communication systems. Since the publication in 1963 of Jackson’s theorem, the first significant development in the theory of queueing networks, hundreds of papers have been devoted to the topic. Nevertheless, several important theoretical issues remain open. Among these, the efficient analysis of closed networks, which monopolized the attention of performance analysts for over twenty years. This dissertation presents new results concerning the exact and approximate analysis of closed queueing network models. We propose new algorithms and formulas for computing throughputs and other performance measures, and which have reduced computational requirements with respect to existing methods. The main contributes of this work are: 1. the development of a new algorithm for the exact analysis of multiclass queueing networks that is efficient also with models that have large population sizes 2. the development of new exact load-dependent algorithms for single and multiclass models 3. the definition of a geometric technique for identifying bottlenecks in both open and closed systems 4. the development of a new bounding technique of performance measures that improves on the popular Balanced Job Bounds (BJB) Each chapter presents numerical results that illustrate either the efficiency or the increased accuracy of the proposed techniques with respect to state-of-the-art methods.