A fault-tolerant software algorithm for a network of transputers

Abstract

This thesis presents a software algorithm that resends work packages to processors when one or more of the worker processors fails or when the link with one or more processors fails. There are two resend criteria used in this algorithm: "resend at end of initial assignment" and "resend at time out". The work, divided into several packages in order to run on several processors in parallel, will be completed as long as at least one work processor remains working and communicating with the main processor. This algorithm could add some fault-tolerance to computer processing in embedded systems.