Parallel algorithms for verification on large systems

The model-checking problem is the question whether a given system model satisfies a property. The property is usually given as formula of a temporal logic, and the system model as labelled transition system. However, the well-known state-space explosion effect is responsible for yielding transition systems of exponential size when compared to their description, and common sequential algorithms often are not capable to solve the model-checking problem with resources available on a single computer. In this thesis, we develop parallel and, in particular, distributed algorithms which exploit the combined resources of a network of commodity workstations to solve problem instances which are beyond the capabilities of today’s sequential algorithms. Specifically, our algorithms solve the model-checking problem for two important fragments of the μ-calculus which subsume many well-known temporal logics (CTL, LTL, CTL∗). We describe our algorithms based on a characterization of the problem at hand in terms of two-player games. The underlying data structure, the game graph, is colored according to the player who has a winning strategy from the current game configuration. Finally, the color of the initial configuration tells who is the winner of the game, and thus whether the transition system satisfies the property or not. Through experimentation, we found that our algorithms scale well, and are able to solve the largest problem instances of the VLTS benchmark suite. In a second part, we investigate ways to efficiently generate (low-level) transition systems suitable for many verification tools from compact high-level descriptions of the input model. We propose a virtual-machine based approach, which uses an intermediate format to break the translation from high-level to low-level representations of a model into two steps. This well-known compiler technique simplifies the translation and still is very fast in practice. We show the practicality of our approach through the example of a compiler for the PROMELA modelling language which targets our intermediate language—the virtualmachine’s byte-code. With a comparison of benchmarks, we show that our approach is competitive to state-of-the-art tools like SPIN in speed, with additional advantages, like easier reusability, and application as component in distributed model-checking algorithms like the ones we proposed earlier.