Bounded model checking for GSMP models of stochastic real-time systems

Model checking is a popular algorithmic verification technique for checking temporal requirements of mathematical models of systems. In this paper, we consider the problem of verifying bounded reachability properties of stochastic real-time systems modeled as generalized semi-Markov processes (GSMP). While GSMPs is a rich model for stochastic systems widely used in performance evaluation, existing model checking algorithms are applicable only to subclasses such as discrete-time or continuous-time Markov chains. The main contribution of the paper is an algorithm to compute the probability that a given GSMP satisfies a property of the form "can the system reach a target before time T within k discrete events, while staying within a set of safe states". For this, we show that the probability density function for the remaining firing times of different events in a GSMP after k discrete events can be effectively partitioned into finitely many regions and represented by exponentials and polynomials. We report on illustrative examples and their analysis using our techniques.