Simplifying Layered Queuing Network Models

The amount of detail to include in a performance model is usually regarded as a judgment to be made by an expert modeler and the question “how much detail is necessary?” is seldom asked and difficult to answer. However, if a simpler model gives essentially the same performance predictions, it may be more useful than a detailed model. It may solve more quickly, for instance. Or a model for a complex sub-system such as a database server may be usefully simplified so it can be included in larger system models. This paper describes an aggregation process for layered queuing models that reduces the number of queues (called tasks and processors, in layered models) while preserving the total execution demand and the bottleneck characteristics of the detailed model. It demonstrates that this process can greatly reduce the number of tasks and processors with a very small relative error.