Vérification efficace des automates temporisés paramétrés
Hoang Gia Nguyen · HAL (Le Centre pour la Communication Scientifique Directe) · 2018
Critical real-time systems are becoming ubiquitous and are playing a vital role in our world. To provide guarantees that the system is behaving correctly, the correctness of these systems need to be verified before running. Besides functional checking, the timed behavior checking is also crucial. Indeed, the correctness of the systems also depends on the timing values or delays of internal operations that can be affected by the environment. Verification techniques assure that software or hardware systems fully satisfy all their expected requirements. Most formal verification methods for timed systems guarantee the correctness of a timed system for the predefined timing values in its blueprint, but not for other undefined timing values which might occur by the environment change and lead to undesired system behaviors. Unfortunately, verifying such system for various timing values can be an obstacle and time-consuming. Therefore, by abstracting these specific timing values with parameters, many timing values of a system can be easily synthesized and checked at the same time : this technique is also known as parameter synthesis. As a huge challenge for the verification, parameter synthesis techniques also suffer from the “state space explosion” problem, which is the explosion of the number of possible states while verifying a system formally. Firstofall, we are interested in taking advantage of the capabilities of current distributed architectures, and parameter synthesis algorithms should be redefined and adapted to the distributed case. We propose in the thesis several distribution schemes that can accelerate our parameter synthesis procedures. We also focus on studying the techniques such as symbolic verification, zone subsumption, etc. and how they affect the state space explosion problem. Then we introduce several smart state exploration techniques with some heuristics, in order to reduce the state space explosion. These techniques and heuristics are integrated into our new synthesis algorithms, and one of these algorithms is also extended in a distributed manner which gives an impressive performance in our benchmarks. Furthermore, to achieve a reliable result we present an approach for detecting timed systems doing an infinite amount of actions in a finite time, which is known as the Zeno phenomenon in theory. In reality, it is infeasible and such counterexamples should always be avoided. Additionally, to detect the non-Zeno phenomenon on a large scale network model, we also distribute our approach on clusters. In the end, we introduce an algorithm to detect non-Zeno runs and its distributed version of it for large-scale models. At the time of writing this thesis, this is also the first work on non-Zeno parameter synthesis.