Fast Leader (Full) Recovery Despite Dynamic Faults

Abstract

We give a leader recovery protocol that recovers a legitimate configuration where a single leader exists, after at most k arbitrary memory corruptions hit the system. That is, if a leader is elected before state corruptions, the same leader is elected after recovery. Our protocol works in any anonymous bidirectional, yet oriented, ring of size n, and does not require that processes know n, although the knowledge of k is assumed. If n ≥ 18k + 1, our protocol recovers the leader in $O(k^{{{\scriptscriptstyle2}}})$ rounds using O(logk) bits per process, assuming unfair scheduling. Our protocol handles dynamic faults in the sense that memory corruption may still occur while the network has started recovering the leader.

See www-verimag.imag.fr/Technical-Reports,264.html?lang=en&number=TR-2012-18 for the full version.