Self-stabilizing Leader Election in Dynamic Networks

Abstract

Three silent self-stabilizing asynchronous distributed algorithms are given for the leader election problem in a dynamic network with unique IDs, using the composite model of computation. A leader is elected for each connected component of the network. A BFS tree is also constructed in each component, rooted at the leader. This election takes O(Diam) rounds, where Diam is the maximum diameter of any component. Links and processes can be added or deleted, and data can be corrupted. After each such topological change or data corruption, the leader and BFS tree are recomputed if necessary. All three algorithms work under the unfair daemon.The three algorithms differ in their leadership stability. The first algorithm, which is the fastest in the worst case, chooses an arbitrary process as the leader. The second algorithm chooses the process of highest priority in each component, where priority can be defined in a variety of ways. The third algorithm has the strictest leadership stability. If the configuration is legitimate, and then any number of topological faults occur at the same time but no variables are corrupted, the third algorithm will converge to a new legitimate state in such a manner that no process changes its choice of leader more than once, and each component will elect a process which was a leader before the fault, provided there is at least one former leader in that component.