Facilitating Consistency Check between Specification and Implementation with MapReduce Framework

We often need well-formed specifications in order to properly maintain or extend a system by members who were not in charge of the original development. In contrast to our expectation, formal specifications and related documents may not be maintained, or not developed in real projects. We are trying to build a framework to develop specifications from a working implementation. Testability of specifications is important in our framework, and we develop executable, or testable, formal specifications in model-oriented formal specification languages such as VDM-SL. We figure out a formal specification, check it with the corresponding implementation by testing, and modify it if necessary. While the specific level of rigor depends on the aim of the project, millions of tests may be performed in developing highly reliable specifications. In this paper, we discuss our approach to reducing the cost of specification test. We use Hadoop, which is an implementation of the MapReduce framework, so that we can expect the scalability in testing specifications. We can automatically distribute the generation of test cases from a property, the interpretation of the executable specification and the execution of its corresponding implementation code for each test data using Hadoop. While straightforward sequential execution for large data set is expensive, we observed scalability in the performance in our approaches.