Software usage models are the basis for statistical testing. They derive th
eir structure from specifications and their probabilities from evolving kno
wledge about the intended use of the software product. The evolving knowled
ge comes from developers, customers and testers of the software system in t
he form of relationships that should hold among the parameters of a model.
When software usage models are encoded as Markov chains, their structure ca
n be represented by a system of linear constraints, and many of the evolvin
g relationships among model parameters can be represented by convex constra
ints. Given a Markov chain usage model as a system of convex constraints, m
athematical programming can be used to generate the Markov chain transition
probabilities that represent a specific software usage model. (C) 2000 Els
evier Science B.V. All rights reserved.