Data Abstraction in Coordination Constraints

This paper studies complex coordination mechanisms based on constraint satisfaction. In particular, it focuses on data-sensitive connectors from the Reo coordination language. These connectors restrict how and where data can flow between loosely-coupled components taking into account the data being exchanged. Existing engines for Reo provide a very limited support for data-sensitive connectors, even though data constraints are captured by the original semantic models for Reo.

When executing data-sensitive connectors, coordination constraints are not exhaustively solved at compile time but at runtime on a per-need basis, powered by an existing SMT (satisfiability modulo theories) solver. To deal with a wider range of data types and operations, we abstract data and reduce the original constraint satisfaction problem to a SAT problem, based on a variation of predicate abstraction. We show soundness and completeness of the abstraction mechanism for well-defined constraints, and validate our approach by evaluating the performance of a prototype implementation with different test cases, with and without abstraction.