Efficient implementation strategies for the DRB approach in fault-tolerant hypercubes

The distributed recovery block (DRB) is an efficient approach for the uniform treatment of hardware and software faults in real-time applications. Full DRB mapping assigns each task of an application tasks graph to run on a pair of processor nodes. The primary cost factors that determine its efficiency are dilation bound, expansion factor and congestion. In this paper, two alternative schemes are described that improves the DRB approach. The first describes a mapping of application tasks such that each node in the hypercube executes one or more application tasks in the lifetime of a task graph. The second scheme incorporates spare processors to provide an efficient architecture for implementing the DRB approach