Model-based stochastic analysis with probabilistic graph query evaluation

Abstract Uncertainty and probability often play a key role in the rigorous evaluation of extra-functional requirements, such as reliability, availability, and performability, on architecture models. For that purpose, stochastic models (like fault trees, continuous time Markov chains) are often derived automatically by complex model transformations to carry out a computationally expensive stochastic analysis. As such, the automated synthesis of stochastic models requires highly specialized expertise in the problem domain, the model transformation language and the target stochastic analysis formalism, which is often lacking in practice. Furthermore, while backend analysis tools may be repeatedly called during model-based architecture synthesis problems, those analysis tools are not optimized for such usage profiles resulting in severe scalability problems. To mitigate these challenges, our paper aims to lift stochastic analysis to a model-level by introducing the formal concepts of probabilistic graph queries for basic and compound events as a lightweight extension of a popular model query language. Moreover, we provide a framework to evaluate such probabilistic queries directly over system models. We evaluate our approach against two baselines on three external case studies of model-based analysis including a complex interferometry mission planning case introduced by NASA researchers.