Mining Sequential Patterns to Explain Concurrent Counterexamples

摘要

Concurrent systems are often modeled using an interleaving semantics. Since system designers tend to think sequentially, it is highly probable that they do not foresee some interleavings that their model encompasses. As a consequence, one of the main sources of failure in concurrent systems is unforeseen interleavings. In this paper, we devise an automated method for revealing unforeseen interleavings in the form of sequences of actions derived from counterexamples obtained by explicit state model checking. In order to extract such sequences we use a data mining technique called sequential pattern mining. Our method is based on contrasting the patterns of a set of counterexamples with the patterns of a set of correct traces that do not violate a desired property. We first argue that mining sequential patterns from the dataset of counterexamples fails due to the inherent complexity of the problem. We then propose a reduction technique designed to reduce the length of the execution traces in order to make the problem more tractable. We finally demonstrate the effectiveness of our approach by applying it to a number of sample case studies.