An incremental refinement approach to a development of a flash-based file system in Event-B

Nowadays, many formal methods are used in the area of software development accompanied by a number of advanced theories and tools. However, more experiments are still required in order to provide significant evidence that will convince and encourage users to use, and gain more benefits from, those theories and tools. Event-B is a formalism used for specifying and reasoning about systems. Rodin is an open and extensible tool for Event-B specification, refinement and proof. The flash file system is a complex system. Such systems are a challenge to specify and verify at this moment in time. This system was chosen as a case study for our experiments, carried out using Event-B and the Rodin tool. The experiments were aimed at developing a rigorous model of flash-based file system; including implementation of the model, providing useful evidence and guidelines to developers and the software industry. We believe that these would convince users and make formal methods more accessible. An incremental refinement was chosen as a strategy in our development. The refinement was used for two different purposes: feature augmentation and structural refinement (covering event and machine decomposition). Several techniques and styles of modelling were investigated and compared; to produce some useful guidelines for modelling, refinement and proof. The model of the flash-based file system we have completed covers three main issues: fault-tolerance, concurrency and wear-levelling process. Our model can deal with concurrent read/write operations and other processes such as block relocation and block erasure. The model tolerates faults that may occur during reading/writing of files. We believe our development acts as an exemplar that other developers can learn from. We also provide systematic rules for translation of Event-B models into Java code. However, more work is required to make these rules more applicable and useful in the future.