Distributed recovery block based fault-tolerant routing in hypercube networks

This paper presents a fault-tolerant routing algorithm that employs a modified distributed recovery block (DRB) approach. The section of a parallel or distributed system spanning between the source and destination nodes is partitioned into a series of overlapping DRB groups. Each DRB group consists of three nodes: a current node and two successor nodes. Primary successor executes the primary try while alternate successor executes an alternate try. The primary successor node delivers the message, whereas the alternate is ready to take over if the primary fails. The successful successor in an active DRB group becomes the current node of the next DRB group on the routing path. A prototype version of the routing method is implemented for a hypercube topology and its performance is compared with adaptive routing techniques based on backtracking.