Alexandre Hamez

Building efficient model checkers using hierarchical set decision diagrams and automatic saturation

Abstract

Shared decision diagram representations of a state-space provide efficient solutions for model-checking of large systems. However, decision diagram manipulation is tricky, as the construction procedure is liable to produce intractable intermediate structures (a.k.a peak effect). The definition of the so-called saturation method has empirically been shown to mostly avoid this peak effect, and allows verification of much larger systems. However, applying this algorithm currently requires deep knowledge of the decision diagram data-structures. Hierarchical Set Decision Diagrams (SDD) are decision diagrams in which arcs of the structure are labeled with sets, themselves stored as SDD. This data structure offers an elegant and very efficient way of encoding structured specifications using decision diagram technology. It also offers, through the concept of inductive homomorphisms, flexibility to a user defining a transition relation. We show in this paper how, with very limited user input, the SDD library is able to optimize evaluation of a transition relation to produce a saturation effect at runtime. We build as an example an SDD model-checker for a compositional formalism: Instantiable Petri Nets (IPN). IPN define a type as an abstract contract. Labeled P/T nets are used as an elementary type. A composite type is defined to hierarchically contain instances (of elementary or composite type). To compose behaviors, IPN use classic label synchronization semantics from process calculi. With a particular recursive folding SDD are able to offer solutions for symmetric systems in logarithmic complexity with respect to other DD. Even in less regular cases, the use of hierarchy in the specification is shown to be well supported by SDD. Experimentations and performances are reported on some well known examples.

Continue reading

Hierarchical set decision diagrams and automatic saturation

By Alexandre Hamez, Yann Thierry-Mieg, Fabrice Kordon

2008-03-01

In Petri nets and other models of concurrency –ICATPN 2008

Abstract

Shared decision diagram representations of a state-space have been shown to provide efficient solutions for model-checking of large systems. However, decision diagram manipulation is tricky, as the construction procedure is liable to produce intractable intermediate structures (a.k.a peak effect). The definition of the so-called saturation method has empirically been shown to mostly avoid this peak effect, and allows verification of much larger systems. However, applying this algorithm currently requires deep knowledge of the decision diagram data-structures, of the model or formalism manipulated, and a level of interaction that is not offered by the API of public DD packages. Hierarchical Set Decision Diagrams (SDD) are decision diagrams in which arcs of the structure are labeled with sets, themselves stored as SDD. This data structure offers an elegant and very efficient way of encoding structured specifications using decision diagram technology. It also offers, through the concept of inductive homomorphisms, unprecedented freedom to the user when defining the transition relation. Finally, with very limited user input, the SDD library is able to optimize evaluation of a transition relation to produce a saturation effect at runtime. We further show that using recursive folding, SDD are able to offer solutions in logarithmic complexity with respect to other DD. We conclude with some performances on well known examples.

Continue reading

libDMC: A library to operate efficient distributed model checking

By Alexandre Hamez, Fabrice Kordon, Yann Thierry-Mieg

2007-03-13

In Workshop on performance optimization for high-level languages and libraries — associated to IPDPS’2007

Abstract

Model checking is a formal verification technique that allows to automatically prove that a system’s behavior is correct. However it is often prohibitively expensive in time and memory complexity, due to the so-called state space explosion problem. We present a generic multi-threaded and distributed infrastructure library designed to allow distribution of the model checking procedure over a cluster of machines. This library is generic, and is designed to allow encapsulation of any model checker in order to make it distributed. Performance evaluations are reported and clearly show the advantages of multi-threading to occupy processors while waiting for the network, with linear speedup over the number of processors.

Continue reading