Components based systems with probabilities

Doron Peled / PERSYVAL-Lab

In the design of reliable complicated systems, one can benefit from a high level formalism that is close enough to the specification, and, on the other hand, automatically and efficiently translatable into the physical layer. Component-based systems can represent a collection of units working together, with an emphasis on the interactions, choices and concurrency among their actions. Adding the ability of making probabilistic choices in component-based systems is nontrivial since each component has its own view, which is partial to the global view of the entire system; maintaining a global view would just eliminate the concurrency. A solution for that is to precede each probabilistic choice by a phase where the local view of a component is locked in order to prepare for such a decision. In this lecture, I would present the mathematical concepts related to concurrent probabilistic systems, based on Markov Decision Processes. Then, a solution based on concurrent artifacts such as independence and confusion will be given. A simple and elegant implementation of component-based systems with probabilistic choices, using semaphores, will be described.