On the feasibility of a spaceborne fault-tolerant hypercube

The feasibility of implementing a fault-tolerant hypercube architecture for space applications is discussed. Node-level architectures and designs are considered and a first-order reliability model is presented. It is shown how error recovery can be implemented using program rollback or roll-forward techniques. Shared memory augmentations to the message-passing structure can be used to get around the inefficiencies of multicomputers to provide efficient use of hardware to achieve the needed reliabilities while maintaining performance.