A Distributed Election Protocol for Unreliable Networks

This paper presents a broadcast-based election protocol for distributed systems implemented on unreliable networks. The protocol is highly robust, tolerating failures such as message loss and network partitioning. Our protocol is different from others in that it effectively detects events that may possibly make the election results inconsistent, aborts the election, and causes the nodes to execute another election round. When the election completes, there is guaranteed to be agreement on a single leader. When the network is partitioned, groups with a consistent view and a single leader are formed in each partition. Compared to other broadcast-based election protocols, which use a group-merging mechanism to fix the problem of several groups independently electing different leaders, the number of election rounds needed to arrive at a single leader is on average considerably fewer.