Specification and proof of a distributed recovery algorithm

Graph reduction, a computational model which supports the parallel execution of functional languages, is discussed. An MIMD (multiple instruction/multiple data) machine, Flagship, which supports the graph reduction model has been built. The authors investigate the formal specification and proof of an algorithm which can ensure the successful execution of a functional program in the presence of the failure of a processing element (PE) of the Flagship machine. The specifications of the algorithm, the graph reduction model, and the augmented graph reduction model, which can tolerate the failure of a PE, are described using CSP (communicating sequential processes) notation. The algebraic transformation rules of CSP are used to prove that, in the presence of PE failure, the fault-tolerant graph reduction model behaves correctly.<>