Comparing Layered Queuing and Historical Performance Models of a Distributed Enterprise Application

Two common approaches for predicting the response times of distributed enterprise applications on new server architectures are solving queuing models and extrapolating from previously gathered performance data. The dynamic recalibration of a layered queuing model and a historical model is investigated experimentally using the IBM Websphere Performance Sample benchmark. It is found that these models can make predictions for new server architectures at a low recalibration overhead with accuracies of 84% and 83%, respectively. The methods are evaluated considering: model recalibration, the responsiveness of predictions, the systems which can be modelled, and ease of use given a minimal level of performance modelling expertise.