Habilitation thesis of Frédéric Gava

Some problems require performance that only massively parallel and distributed machines can offer.However, programming such architectures is a difficult task. Several reasons can be highlighted for that, suchas the intrinsic complexity of these architectures (possibility of deadlocks and data-races), a lack of training ofthe programmers and a lack of tools to test and to prove the correctness of parallel programs. But, first andforemost, high-level structures such as algorithmic skeletons (patterns of parallel computations) are too oftenunused, because they are not known by most programmers. However, they facilitate the writing of parallelprograms while ensuring them a greater safety and, potentially, their correctness. In addition, access to thesesupercomputers is performed using unsecured networks. Ensure the security of exchanges between users anddata-centers is as important as being able to verify the correctness of programs.In this work, we investigate a model of parallel computations, called bsp, which allows a structured executionof programs. The manuscript is organized in three parts, each corresponding to a doctoral thesis supervised bythe author, within the framework of research projects.In a first part, we study a functional language for bsp programming, called bsml, including its primitives, itsimplementation, the type system of bsml programs that guarantees their safety of execution, a methodology forproving programs in the coq theorem prover and various extensions such as parallel exception handling. Severalapplications are given, including the implementation of various algorithmic skeletons.In a second part, we describe the development of a tool for deductive verification of bsp algorithms, that isto say, to annotate the code with logical assertions and to generate proof obligations that validate or not theaforesaid assertions. This tool is an extension of an already existing deductive verification tool, called why.bsp-why consists in transforming the bsp parallel code (and logical assertions) into purely sequential why one.Operational semantics, developed using the coq theorem prover, are also described. Examples are also given totest the reliability of the approach.In the last part, we exploit the structured nature of many security protocols and match it to the bsp model,in order to obtain efficient parallel algorithms to model-check security protocols. We consider the problem of theverification of logical (temporal) formulas ltl and ctl∗ on security protocols. State-space bsp algorithms willalso be formally checked using the bsp-why tool described in the previous part of the manuscript.