Refining Distributed Systems using the B Method

This paper makes a proposal for introducing into the B Method [3] a new refinement operator dedicated to take into account distribution fro m the very specification level through all stepwise refinement process of B abstract machines towards actual implementations. Our operator expresses how a whole system composed of many concurrent and communicating processes can be designed by refinement. Such distributed refinements allow to progressively introducing some degree of parallelism and communications into the operations of abstract machines. The proposed operator is compatible with classical algorithmic and data refinements found in the B Method, and we show how it can be used also to derive sequential programs from distributed refinements whenever the implementation will require.