A Theory of Testing for Markovian Processes

We present a testing theory for Markovian processes in order to formalize a notion of efficiency which may be useful for the analysis of soft real time systems. Our Markovian testing theory is shown to enjoy close connections with the classical testing theory of De NicolaHennessy and the probabilistic testing theory of Cleaveland-Smolka et al. The Markovian testing equivalence is also shown to be coarser than the Markovian bisimulation equivalence. A fully abstract characterization is developed to ease the task of establishing testing related relationships between Markovian processes. It is then demonstrated that our Markovian testing equivalence, which is based on the (easier to work with) probability of executing a successful computation whose average duration is not greater than a given amount of time, coincides with the Markovian testing equivalence based on the (more intuitive) probability of reaching success within a given amount of time. Finally, it is shown that it is not possible to define a Markovian preorder which is consistent with reward based performance measures, thus justifying why a generic notion of efficiency has been considered.