Abstract: The main obstruction to automatic verification of Finite State Systems is the huge amount of memory required to complete the verification task (state explosion). This motivates research on distributed as well as disk based verification algorithms. In this paper we present a disk based Breadth First Explicit State Space Exploration algorithm as well as an implementation of it within the Mur$\varphi$ verifier. Our algorithm exploits transition locality (i.e. the statistical fact that most transitions lead to unvisited states or to recently visited states) to decrease disk read accesses thus reducing the time overhead due to disk usage. A disk based verification algorithm for Mur$\varphi$ has been already proposed in the literature. To measure the time speed up due to locality exploitation we compared our algorithm with such previously proposed algorithm. Our experimental results show that our disk based verification algorithm is typically more than 10 times faster than such previously proposed disk based verification algorithm. To measure the time overhead due to disk usage we compared our algorithm with RAM based verification using the (standard) Mur$\varphi$ verifier with enough memory to complete the verification task. Our experimental results show that even when using 1/10 of the RAM needed to complete verification, our disk based algorithm is only between 1.4 and 5.3 times (3 times on average) slower than (RAM) Mur$\varphi$ with enough RAM memory to complete the verification task at hand. Using our disk based Mur$\varphi$ we were able to complete verification of a protocol with about $10^9$ reachable states. This would require more than 5 gigabytes of RAM using RAM based Mur$\varphi$.

Abstract: The main barrier to automatic verification of concurrent systems is the huge amount of memory required to complete the verification task (state explosion). In this paper we present a probabilistic algorithm for automatic verification via model checking. Our algorithm trades space with time. In particular, when memory is full because of state explosion our algorithm does not give up verification. Instead it just proceeds at a lower speed and its results will only hold with some arbitrarily small error probability. Our preliminary experimental results show that by using our probabilistic algorithm we can typically save more than 30% of RAM with an average time penalty of about 100% w.r.t. a deterministic state space exploration with enough memory to complete the verification task. This is better than giving up the verification task because of lack of memory.

Abstract: A railway interlocking system (RIS) is an embedded system (namely a supervisory control system) that ensures the safe, operation of the devices in a railway station. RIS is a safety critical system. We explore the possibility of integrating automatic formal verification methods in a given industry RIS design flow. The main obstructions to be overcome in our work are: selecting a formal verification tool that is efficient enough to solve the verification problems at hand; and devising a cost effective integration strategy for such tool. We were able to devise a successful integration strategy meeting the above constraints without requiring major modification in the pre-existent design flow nor retraining of personnel. We run verification experiments for a RIS designed for the Singapore Subway. The experiments show that the RIS design flow obtained from our integration strategy is able to automatically verify real life RIS designs.

Abstract: We present a case study on automatic synthesis of control software from formal specifications for an industrial automation control system. Our aim is to compare the effectiveness (i.e. design effort and controller quality) of automatic controller synthesis from closed loop formal specifications with that of manual controller design followed by automatic verification. The system to be controlled (plant) models a metal processing facility near Karlsruhe. We succeeded in automatically generating C code implementing a (correct by construction) embedded controller for such a plant from closed loop formal specifications. Our experimental results show that for industrial automation control systems automatic synthesis is a viable and profitable (especially as far as design effort is concerned) alternative to manual design followed by automatic verification.

Abstract: Many safety critical reactive systems are indeed embedded control systems. Usually a control system can be partitioned into two main subsystems: a controller and a plant. Roughly speaking: the controller observes the state of the plant and sends commands (stimulus) to the plant to achieve predefined goals. We show that when the plant can be modeled as a deterministic finite state system (FSS) it is possible to effectively use formal methods to automatically synthesize the program implementing the controller from the plant model and the given formal specifications for the closed loop system (plant+controller). This guarantees that the controller program is correct by construction. To the best of our knowledge there is no previously published effective algorithm to extract executable code for the controller from closed loop formal specifications. We show practical usefulness of our techniques by giving experimental results on their use to synthesize C programs implementing optimal controllers (OCs) for plants with more than 109 states.