Systems composed of many identical processes can sometimes be verified inductively using a network invariant, but systems whose component processes vary in some systematic way are not amenable to direct application of that method. We describe how variations in behavior can be "factored out" into additional processes, thus enabling induction over the number of processes. The process is semi-automatic: The designer must choose from among a set of idiomatic transformations, but each transformation is applied and checked automatically.

Automated finite-state verification techniques have matured considerably in the past several years, but state-space explosion remains an obstacle to their use. Theoretical lower bounds on complexity imply that all of the techniques that have been developed to avoid or mitigate state-space explosion depend on models that are "well-formed" in some way, and will usually fail for other models. This further implies that, when analysis is applied to models derived from designs or implementations of actual software systems, a model of the system "as built" is unlikely to be suitable for automated analysis. In particular, compositional, hierarchical analysis (where state-space explosion is avoided by simplifying models of subsystems at several levels of abstraction) depend on the modular structure of the model to be analyzed. We describe how as-built finite-state models can be refactored for compositional state-space analysis, applying a series of transformations to produce an equivalent model whose structure exhibits suitable modularity. The process is supported by a parser which can parse a subset of Promela syntax and transform Promela code into refactored state graphs.

No abstract

No abstract

No abstract