A Polyhedron Approach to Calculate Probability Distributions for Markov Chain Usage Models

Statistical usage testing of hardware/software systems is based in the main on a Markov chain usage model. This kind of model represents the expected use of the system by a usage profile, i.e. appropriate probability values that are attached to the state transitions. In this paper we present a constraint-based polyhedron approach to calculate the probability distribution for the MCUM from a given set of usage constraints. Comparing the computed probability distributions of our polyhedron approach with the maximum entropy technique shows that our result is much closer to the intented constraint semantics. Using the polyhedron method, customer profiles can be calculated so that they reflect the intended system usage of different customers or customer types much better. In order to demonstrate the applicability of our approach, workflow testing of a complex RIS/PACS system in the medical domain was carried through and yielded very promising results.