Concurrent rollback for crash recovery in extended hypercube networks

Recovering from processor failures is an important problem in the design and development of reliable systems. We present a concurrent rollback algorithm in extended hypercube networks to recover from crash failures which involves small message and time complexities. The network of an extended hypercube is a hierarchical, low diameter, recursive structure. By appending only O(1) additional information to each message, we use less than O(Nlog N) message exchanges and O(log² N) time elapsed for recovery work where N is the number of processors of the extended hypercube network. The algorithms can be used to recover from the failure of an arbitrary number of processors