Modeling service distributions in queueing network simulation

The importance of accurate models for the input processes of simulation is generally recognized. However, the dependencies between the input processes have, at least in queueing network simulation, been paid little attention. In this paper, modeling the joint distribution of the input processes required in queueing network simulation is considered. Particularly, the paper proposes a method for modeling the dependencies between service demands. In addition, it proposes a method for dividing the service demand into service times per visit. These methods are important in simulation modeling of a computer system when accounting files provide the identification data. The experiments reported indicate that the proposed methods produce accurate estimates for quantiles of the system response-time distribution.