We describe a dynamic partitioning scheme usable by explicit state space exploration techniques that divide the state space into partitions, such as many external memory and distributed model checking algorithms. The goal of the scheme is to reduce the number of transitions that link states belonging to different partitions, and thereby limit the amount of disk access and network communication. We report on several experiments made with our verification platform ASAP that implements the dynamic partitioning scheme proposed in this paper. The experiments demonstrate the importance of exploiting locality to reduce cross transitions and IO operations, and using informed heuristics when choosing components to be used as a basis for partition refinement. The experiments furthermore show that the proposed approach improves on earlier techniques and significantly outperforms these when provided with access to the same amount of bounded RAM.

In order to manage very large distributed databases such as those used for banking and e-government applications, and thus to han- dle sensitive data, an original peer-to-peer transaction protocol, called NEO, was proposed. To ensure its effective operation, it is necessary to check a set of critical properties. The most important ones are related to availability of data that must be guaranteed by the system. Thus, our objective aims at verifying critical properties of the NEO protocol so as to guarantee such properties are satisfied. The model is obtained by reverse-engineering from the source code and then formal verification is performed. We focus in this article on the two phases of the NEO protocol occurring at the initialisation of the system. The first one, the election phase, aims at designating a special node that will pilot the over- all system. The bootstrap protocol, triggered at the end of the election, ensures that the system will enter its operational state in a coherent way. Therefore, the correctness of these two phases is mandatory for the reliability of the system.

Partial order reduction limits the state explosion problem that arises in model checking by limiting the exploration of redundant interleavings. A state space search algorithm based on this principle may ignore some interleavings by delaying the execution of some actions provided that an equivalent interleaving is explored. However, if one does not choose postponed actions carefully, some of these may be infinitely delayed. This pathological situation is commonly referred to as the ignoring problem. The prevention of this phenomenon is not mandatory if one wants to verify if the system halts but it must be resolved for more elaborate properties like, for example, safety or liveness properties. We present in this work some solutions to this problem. In order to assess the quality of our propositions, we included them in our model checker Helena. We report the result of some experiments which show that our algorithms yield better reductions than state of the art algorithms like those implemented in the Spin tool.

State caching is a memory reduction technique used by model checkers to alleviate the state explosion problem. It has traditionally been coupled with a depth-first search to ensure termination. We propose and experimentally evaluate an extension of the state caching method for general state exploring algorithms that are independent of the search order (i.e., search algorithms that partition the state space into closed (visited) states, open (to visit) states and unmet states).

The ComBack method is a memory reduction technique for explicit state space search algorithms. It enhances hash compaction with state reconstruction to resolve hash conflicts on-the-fly thereby ensuring full coverage of the state space. In this paper we provide two means to lower the run-time penalty induced by state reconstructions: a set of strategies to implement the caching method proposed in [20], and an extension through delayed duplicate detection that allows to group reconstructions together to save redundant work.

Learning concurrency paradigms is necessary but it is not sufficient since the choice of run-time semantics may introduce subtle programming errors. It is the aim of this paper to exemplify the importance of process queuing and awaking policies resulting from possible choices of the monitor concept implementation. The first part of the paper compares the behaviour of concurrent processes sharing a unique waiting queue for condition synchronization when implemented in Java or in Ada. A particular solution of the dining philosophers paradigm will be used to show how the difference in the monitor semantics may lead or not to deadlock. This comparison provides insight for deriving a correct Java implementation. The second part of the paper shows how the implementation can be refined when using Ada entry families and requeue with requeue once restriction. The result is elegant, safe and fair, and deterministic. This paper ends with quantitative comparisons of concurrency complexity and of concurrency effectiveness. We conclude that Java and C# multithreading need defensive concurrent programming while Ada allows more latitude for developing correct concurrent programs.

The sweep-line method allows explicit state model checkers to delete states from memory on-the-fly during state space exploration thereby lowering the memory demands of the verification procedure. The sweep-line method is based on a least progress-first search order that prohibits the immediate use of standard on-the-fly LTL model checking algorithms that rely on a depth-first search order. This paper proposes and experimentally evaluates an algorithm for LTL model checking compatible with the search order prescribed by the sweep-line method.

The sweep-line method is an explicit-state model checking technique that uses a notion of progress to delete states from internal memory during state space exploration and thereby reduce peak memory usage. The sweep-line algorithm relies on the use of a priority queue where the progress value assigned to a state determines the priority of the state. In earlier implementations of the sweep-line method the progress priority queue is kept in internal memory together with the current layer of states being explored. In this paper we investigate a scheme where the current layer is stored in internal memory while the priority queue is stored in external memory. From the perspective of the sweep-line method, we show that this combination can yield a significant reduction in peak memory usage compared to a pure internal memory implementation. On an average of 60 example instances, this combination reduced peak memory usage by a factor of 25 at the cost of an increase in execution time by a factor of 2.5. From the perspective of external memory state space exploration, we demonstrate experimentally that the state deletion performed by the sweep-line method may reduce the I/O overhead induced by duplicate detection compared to a pure external memory state space exploration method.

This paper presents CNDFS, a tight integration of two earlier multi-core nested depth-first search (NDFS) algorithms for LTL model checking. CNDFS combines the different strengths and avoids some weaknesses of its predecessors. We compare CNDFS to an earlier ad-hoc combination of those two algorithms and show several benefits: It has shorter and simpler code and a simpler correctness proof. It exhibits more stable performance and scalability, while at the same time reducing memory requirements.

The algorithm has been implemented in the multi-core backend of the LTSmin model checker, which is now benchmarked for the first time on a 48 core machine (previously 16). The experiments demonstrate better scalability than other parallel LTL model checking algorithms, but we also investigate apparent bottlenecks. Finally, we noticed that the multi-core NDFS algorithms produce shorter counterexamples, surprisingly often shorter than their BFS-based counterparts.

Even though the well-known nested-depth first search algorithm for LTL model checking provides good performance, it cannot benefit from the re- cent advent of multi-core computers. This paper proposes a new version of this algorithm, adapted to multi-core architectures with a shared memory. It can ex- hibit good speed-ups as supported by a series of experiments.

In this article we focus on the modelization of the BonjourGrid protocol which is based on the Publish- Subscribe (Pub-Sub) paradigm, a paradigm for asynchronous communication that is useful for implementing some approaches in distributed programming. The aim of this paper is to isolate the generic mechanisms of construction for the publish-subscribe approach then to model and verify, based on those mechanisms, the BonjourGrid protocol that allows the coordination of multiple instances of desktop grid middleware. We produce models using colored Petri nets in order to describe a specific modeling approach for the Pub-Sub paradigm. Such models are important, first, to formally verify the adequacy of BonjourGrid in the coordination of resources in desktop grids - for example by proving the absence of a deadlock in the BonjourGrid protocol, and second, to offer a 'composition' mechanism for integrating any protocol based on the Pub-Sub paradigm. These ideas are illustrated along the BonjourGrid case study and they constitute a methodology of building Pub- Sub systems.

This paper presents the modelling process and first analysis results carried out within the NEOPPOD project. A protocol, NEO, has been designed in order to manage very large distributed databases such as those used for banking and e-government applications, and thus to handle sensitive data. Security of data is therefore a critical issue that must be ensured before the software can be released on the market.

Our project aims at verifying essential properties of the protocol so as to guarantee such properties are satisfied. The model was designed by reverse-engineering from the source code, and then initial verification was performed. This modelling work requires choices of adequate abstraction levels both at the modelling and verification stages. In particular, the overall system is so large that the model should be carefully built in order to make verification possible without getting too far from the actual protocol implementation.

We describe a dynamic partitioning scheme usable by model checking techniques that divide the state space into partitions, such as most external memory and distributed model checking algorithms. The goal of the scheme is to reduce the number of transitions that link states belonging to different partitions, and thereby limit the amount of disk access and network communication. We report on several experiments made with our verification platform ASAP that implements the dynamic partitioning scheme proposed in this paper.

The ASCoVeCo State space Analysis Platform (ASAP) is a tool for performing explicit state space analysis of coloured Petri nets (CPNs) and other formalisms. ASAP supports a wide range of state space reduction techniques and is intended to be easy to extend and to use, making it a suitable tool for students, researchers, and industrial users that would like to analyze protocols and/or experiment with different algorithms. This paper presents ASAP from these two perspectives.

Duplicate detection is an expensive operation of disk-based model checkers. It consists of comparing some potentially new states, the candidate states, to previous visited states. We propose a new approach to this technique called dynamic delayed duplicate detection. This one exploits some typical properties of states spaces, and adapts itself to the structure of the state space to dynamically decide when duplicate detection must be conducted. We implemented this method in a new algorithm and found out that it greatly cuts down the cost of duplicate detection. On some classes of models, it performs significantly better than some previously published algorithms

Positive flows provide very useful informations that can be used to perform efficient analysis of a model. Although algorithms computing (a generative family of) positive flows in ordinary Petri nets are well known, computing a generative family of positive flows in colored net remains an open problem. We propose in this paper a pragmatic approach that allows us to define an algorithm that computes a generative family of particular but useful positive flows in a large subclass of colored nets: the simple well-formed nets.

The ignoring problem refers to the fact that some actions may be infinitely postponed by a state space search algorithm that makes use of partial order reduction (POR). The prevention of this phenomenon is mandatory if one wants to verify more elaborate properties than the deadlock freeness, e.g., safety or liveness properties. We present in this work some solutions to this problem. In order to assess the quality of our propositions, we included them in our model checker Helena. We report the result of some experiments which show that our algorithms yield better reductions than state of the art algorithms like those implemented in the Spin tool.

Valmari's Stubborn Sets method is a member of the so-called partial order methods. These techniques are usually based on a selective search algorithm: at each state processed during the search, a stubborn set is calculated and only the enabled transitions of this set are used to generate the successors of the state. The computation of stubborn sets requires to detect dependencies between transitions in terms of conflict and causality. In colored Petri nets these dependencies are difficult to detect because of the color mappings present on the arcs: conflicts and causality connections depend on the structure of the net but also on these mappings. Thus, tools that implement this technique usually unfold the net before exploring the state space, an operation that is often untractable in practice. We present in this work an alternative method which avoids the cost of unfolding the net. To allow this, we use a syntactically restricted class of colored nets. Note that this class still enables wide modeling facilities since it is the one used in our model checker Helena which has been designed to support the verification of software specifications. The algorithm presented has been implemented and several experiments which show the benefits of our approach are reported. For several models we obtain a reduction close or even equal to the one obtained after an unfolding of the net. We were also able to efficiently reduce the state spaces of several models obtained by an automatic translation of concurrent software.

This paper presents the high level Petri nets analyzer Helena. Helena can be used for the on-the-fly verification of state properties, i.e., properties that must hold in all the reachable states of the system, and deadlock freeness. Some features of Helena make it particularly efficient in terms of memory management. Structural abstractions techniques, mainly transitions agglomerations, are used to tackle the state explosion problem. Benchmarks are presented which compare our tool to Maria. Helena is developed in portable Ada and is freely available under the conditions of the GNU General Public License.

The inclusion of dynamic tasks modelisation in QUASAR, a tool for automatic analysis of concurrent programs, extends its applicative usefulness. However this extension leads to large size models whose processing has to face combinatory explosion of modeling states. This paper presents briefly Ada dynamic tasks semantic and dependences and then it explains the choice of an efficient generic modeling pattern. This implies to consider the naming, the hierarchy, the master retrieval, the termination of dynamic tasks and their synchronization dependences successively. The adequacy of both this modeling and the QUASAR techniques is highlighted by the analysis of two non-trivial Ada programs. The large reduction factor between the initial and final state numbers of these program models shows that the state explosion can be limited, making automatic validation of dynamic concurrent programs feasible.

In this paper, we develop a syntactical version of elaborated reductions for high-level Petri nets. These reductions simplify the model by merging some sequential transitions into an atomic one. Their conditions combine local structural ones (e.g. related to the actions of a thread) and global algebraic ones (e.g. related to the threads synchronization). We show that these conditions are performed in a syntactical way, when a syntax of the color mappings is given. We show also how our method outperforms previous ones on a recent case study with regard both to the reduction ratio and the automatization of their application.

The limited amount of memory is the major bottleneck in model checking tools based on an explicit states enumeration. In this context, techniques allowing an efficient representation of the states are precious. We present in this paper a novel approach which enables to store the state space in a compact way. Though it belongs to the family of explicit storage methods, we qualify it as semi-explicit since all states are not explicitly represented in the state space. Our experiments report a memory reduction ratio up to 95time in the worst case.

Structural model abstraction is a powerful technique for reducing the complexity of a state based enumeration analysis.We present in this paper accurate reductions for high-level Petri nets based on new ordinary Petri nets reductions. These reductions involve only structural and algebraical conditions. They preserve the liveness of the net and any LTL formula that does not observe the reduced transitions of the net. The mixed use of structural and algebraical conditions signicantly enlarges their application area. Furthermore the specication of the transformation is parametric with respect to the cardinalities of coloured domains.

Concurrency introduces a high degree of combinatory which may be the source of subtle mistakes. We present a new tool, Quasar, which is based on ASIS and which uses fully the concept of patterns. The analysis of a concurrent Ada program by our tool proceeds in four steps: automatic extraction of the concurrent part of the program; translation of the simplified program into a formal model using predefined patterns that are combined by substitution and merging constructors; analysis of the model both by structural techniques and model-checking techniques; reporting deadlock or starvation results. We demonstrate the usefulness of Quasar by analyzing several variations of a non trivial concurrent program.

In this paper we present an original and useful way for specifying and verifying temporal properties of concurrent programs with our tool named Quasar. Quasar is based on ASIS and uses formal methods (model checking). Properties that can be checked are either general, like deadlock or fairness, or more context specific, referring to tasks states or to value of variables; properties are then expressed in temporal logic. In order to simplify the expression of these properties, we define some templates that can be instantiated with specific items of the programs. We demonstrate the usefulness of these templates by verifying subtle variations of the Peterson algorithm. Thus, although Quasar uses up-to-date formal methods it remains accessible to a large class of practitioners.

State caching is a memory reduction technique used by model checkers to alleviate the state explosion problem. It has traditionally been coupled with a depth-first search to ensure termination.We propose and experimentally evaluate an extension of the state caching method for general state exploring algorithms that are independent of the search order (i.e., search algorithms that partition the state space into closed (visited) states, open (to visit) states and unmet states).

The ComBack method is a memory reduction technique for explicit state space search algorithms. It enhances hash compaction with state reconstruction to resolve hash conflicts on-the-fly thereby ensuring full coverage of the state space. In this paper we provide two means to lower the run-time penalty induced by state reconstructions: a set of strategies to implement the caching method proposed in [18], and an extension through delayed duplicate detection that allows to group reconstructions together to save redundant work.

Model checking and simulation tools based on the colored Petri nets formalism spend a significant amount of time in performing enabling tests. This consists in taking into account the color mappings of the net to determine valid transitions variables assignments at a given marking. This work proposes an algorithm for the enabling test problem. It implements the relations of conflict and causality between transitions to efficiently maintain a set of enabled transitions. This set is updated during the search algorithm according to the transitions fired (or unfired). However, in most cases this approach is not sufficient to compute the set of enabled transition bindings, and has to be followed by a unification algorithm. This is the objective of the second part of this work.

L'utilisation croissante de l'informatique dans des systèmes à caractère critique pose le problème aigu de leur validation. Le model checking est une technique de vérification des systèmes informatiques basée sur l'exploration exhaustive de leurs comportements à la recherche de comportements déviants de la spécification. Elle offre de nombreux avantages par rapport à d'autres outils de validation mais se heurte au problème bien connu de l'explosion combinatoire : les états du système peuvent être bien trop nombreux pour être énumérés en un temps raisonnable, et ce, même avec la puissance de calcul des machines dont nous disposons actuellement. Plusieurs techniques ont été définies durant les dernières décennies afin de lutter contre ce problème. En conséquence, le model checking s'est répandu dans l'industrie, et est maintenant couramment utilisé pour valider des protocoles ou des composants matériel. Plus récemment le model checking de programmes a émergé comme un nouveau thème de recherche. Le principe n'est plus de vérifier qu'un modèle du système est conforme aux attentes mais de confronter directement le programme à la spécification.
Cette thèse s'inscrit dans le cadre de la vérification automatique de programmes concurrents basée sur un modèle formel intermédiaire, les réseaux de Petri colorés. Nous nous attachons particulièrement à définir, ou adapter, des méthodes qui visent à lutter contre le phénomène d'explosion combinatoire induit par les algorithmes d'exploration du graphe d'accessibilité. Nous œuvrons pour cela à deux niveaux : au niveau structurel, par des techniques d'abstraction du modèle, et au niveau sémantique, par des techniques de réduction du graphe d'accessiblité du système. Afin de valider l'intérêt pratique des techniques proposées nous les avons implantées dans deux outils : Helena un model checker pour les réseaux de Petri de haut niveau et Quasar une plate-forme pour la validation de programmes Ada concurrents.