Speeds: A New Technique for the Analysis and Optimization of Queueing Networks

摘要

The sample path perturbation extrapolation of discrete event systems (SPEEDS), a time domain-based approach which views a queueing network as a stochastic system evolving in time and which observes the sample realization of its trajectory, is dIscussed. This approach differs from the simulation approach to queueing systems because analysis is used to derive answers to the question: What will happen if the sample trajectory is repeated exactly except for a small perturbation at some time t. The SPEEDS approach is shown to combine the advantages of both the theoretical and the simulation approaches while minimizing their disadvantages. Experimental results validating the approach to a general queueing network with a single customer class are provided and both classical and nonclassical networks are treated. Perturbation analysis is discussed in terms of infinitesimal perturbations and analysis and experimental validation of perturbation rules are provide for finite perturbations.