Thèse de doctorat

Distributed message passing applications are in the mainstream of information technology since they exploit the power of parallel computer systems to produce higher performance. Designing distributed programs remains challenging because developers have to reason about concurrency, non-determinism, data distribution… that are main characteristics of distributed programs. Besides, it is virtually impossible to ensure the correctness of such programs via classical testing approaches since one may never successfully reach the execution that leads to unwanted behaviors in the programs. There is thus a need for more powerful verification techniques. Model-checking is one of the formal methods that allows to verify automatically and effectively some properties on models of computer systems by exploring all possible behaviors (states and transitions) of the system model. However, state spaces increase exponentially with the number of concurrent processes, leading to “state space explosion”. Unfolding-based Dynamic Partial Order Reduction (UDPOR) is a recent technique mixing Dynamic Partial Order Reduction (DPOR) with concepts of concurrency theory such as unfoldings to efficiently mitigate state space explosion in model-checking of concurrent programs. It is optimal in the sense that each Mazurkiewicz trace, i.e. a class of interleavings equivalent by commuting adjacent independent actions, is explored exactly once. And it is applicable to running programs, not only models of programs. The thesis aims at adapting UDPOR to verify asynchronous distributed programs (e.g. MPI programs) in the setting of the SIMGRID simulator of distributed applications. To do so, an abstract programming model of asynchronous distributed programs is defined and formalized in the TLA+ language, allowing to precisely define an independence relation, a main ingredient of the concurrency semantics. Then, the adaptation of UDPOR, involving the construction of an unfolding, is made efficient by a precise analysis of dependencies in the programming model, allowing efficient computations of usually costly operation. A prototype implementation of UDPOR adapted to distributed asynchronous programs has been developed, giving promising experimental results on a significant set of benchmarks.