Computational grids have become popular for constructing large scale distributed systems. Grid applications typically run in a very heterogeneous environment and fault tolerance is therefore very important for their correctness. Since the construction of correct distributed systems is difficult with traditional development methods we propose the use of formal methods. We use Event B as our formal framework, which we extend with new constructs such as remote procedures and notifications for reasoning about grid systems. The extended language, called Distributed B, ensures that the application can handle both node and network failures. Furthermore, the new constructs in Distributed B enable straightforward implementation of the specifications, as well as automatic generation of the needed proof obligations.
|Namn||TUCS Technical Reports|