Formalisation of asynchronous interactions

• Main Author: Chevrou, Florent
• Format: Others
• Published: 2017
• Online Access: http://oatao.univ-toulouse.fr/19493/1/CHEVROU_Florent.pdf; Chevrou, Florent <http://www.idref.fr/223784222>. Formalisation of asynchronous interactions. PhD, Sûreté de Logiciel et Calcul à Haute Performance, Institut National Polytechnique de Toulouse, 2017

Large computing systems are generally built by connecting several distributed subsystems. The way these entities communicate is crucial to the proper functioning of the overall composed system. An in-depth study of these interactions makes sense in the context of the formal development and verification of such systems. The interactions fall in two categories: synchronous and asynchronous communication. In synchronous communication, the transmission of a piece of information - the message - is instantaneous. Asynchronous communication, on the other hand, splits the transmission in a send operation and a receive operation. This make the interleaving of other events possible and lead to new behaviours that may or may not be desirable. The asynchronous world is often viewed as a monolithic counterpart of the synchronous world. It actually comes in multiple models that provide a wide range of properties that can be studied and compared. This thesis focuses on communication models that order the delivery of messages: for instance, the "FIFO" models ensure that some messages are received in the order of their emission. We consider classic communication models from the literature as well as a few variations. We highlight the differences that are sometimes overlooked. First, we propose an abstract, logical, and homogeneous formalisation of the communication models and we establish a hierarchy that extends existing results. Second, we provide an operational approach with a tool that verifies the compatibility of compositions of peers. We mechanise this tool with the TLA+ specification language and its model checker TLC. The tool is designed in a modular fashion: the commmunicating peers, the temporal compatibility properties, and the communication models are specified independently. We rely on a set of uniform operational specifications of the communication models that are based on the concept of message history. We identify and prove the conditions under which they conform to the logical definitions and thus show the tool is trustworthy. Third, we consider concrete specifications of the communication models that are often found in the literature. Thus, the models are classified in terms of ordering properties and according to the level of abstraction of the different specifications. The concept of refinement covers these two aspects. Thus, we model asynchronous point-to-point communication along several levels of refinement and then, with the Event-B method, we establish and prove all the refinements between the communication models and the alternative specifications of each given model. This work results in a detailed map one can use to develop a new model or find the one that best fits given needs. Eventually we explore ways to extend our work to multicast communication that consists in sending messages to several recipients at once. In particular, we highlight the differences in the hierarchy of the models and how we modify our verification tool to handle this communication paradigm.