Often, at the time when a performance model is built, some of its input parameters are not known exactly. This can be for a variety of reasons, such as the system itself not having been completed, a benchmarking process not yet having been carried out, or that precise estimates for the parameters are very difficult to obtain. In these situations, interval values can be used to express the uncertainties of the input parameters and sensitivity studies of performance models have the potential to show the influence of input parameter uncertainties on the resulting performance measures. In this paper we study the sensitivities of an Enterprise JavaBeans performance model. This model is built using the standard method of model decomposition which gives various submodels. Solutions obtained for these submodels have been adapted to interval parameters and hence interval-based sensitivities are studied in the so-called container submodel and then the whole model. Using these techniques, parameters that require to be characterised more carefully than others can be identified.
pubs.doc.ic.ac.uk: built & maintained by Ashok Argent-Katwala.