In the field of model checking, abstraction refinement has proved to be an extremely successful methodology for combating the state-space explosion problem. However, little practical progress has been made in the setting of probabilistic verification. In this paper we present a novel abstraction-refinement framework for Markov decision processes (MDPs), which are widely used for modelling and verifying systems that exhibit both probabilistic and nondeterministic behaviour. Our framework comprises an abstraction approach based on stochastic two-player games, two refinement methods and an efficient algorithm for an abstraction-refinement loop. The key idea behind the abstraction approach is to maintain a separation between nondeterminism present in the original MDP and nondeterminism introduced during the abstraction process, each type being represented by a different player in the game. Crucially, this allows lower and upper bounds to be computed for the values of reachability properties of the MDP. These give a quantitative measure of the quality of the abstraction and form the basis of the corresponding refinement methods. We describe a prototype implementation of our framework and present experimental results demonstrating automatic generation of compact, yet precise, abstractions for a large selection of real-world case studies.

Abstract. We present a methodology and implementation for verifying ANSI-C programs that exhibit probabilistic behaviour, such as failures or randomisation. We use abstraction-refinement techniques that represent probabilistic programs as Markov decision processes and their abstractions as stochastic two-player games. Our techniques target quantitative properties of software such as "the maximum probability of file-transfer failure" or "the minimum expected number of loop iterations" and the abstractions we construct yield lower and upper bounds on these properties, which then guide the refinement process. We build upon stateof-the-art techniques and tools, using SAT-based predicate abstraction, symbolic implementations of probabilistic model checking and components from GOTO-CC, SATABS and PRISM. Experimental results show that our approach performs very well in practice, successfully verifying actual networking software whose complexity is significantly beyond the scope of existing probabilistic verification tools.

No abstract

No abstract