Exploring Design Alternatives for RAMP Transactions Through Statistical Model Checking

Arriving at a mature distributed system design through implementation and experimental validation is a labor-intensive task. This limits the number of design alternatives that can be explored in practice. In this work we use formal modeling with probabilistic rewrite rules and statistical model checking to explore and extend the design space of the RAMP (Read Atomic Multi-Partition) transaction system for large-scale partitioned data stores. Specifically, we formally model in Maude eight RAMP designs, only two of which were previously implemented and evaluated by the RAMP developers; and we analyze their key consistency and performance properties by statistical model checking. Our results: (i) are consistent with the experimental evaluations of the two implemented designs; (ii) are also consistent with conjectures made by the RAMP developers for other unimplemented designs; and (iii) uncover some promising new designs that seem attractive for some applications.