The promise, and limitations, of gossip protocols

Recent years have seen a surge of interest in gossip protocols, with proposals to apply them for purposes ranging from autonomic self-management, repair of inconsistencies, reliable multicast and distributed search. Yet the field of distributed computing is littered with technologies that had initial promise, but were ultimately rejected by the industry. Researchers who measure their work through its impact need to ask some tough, basic questions. What are the uses to which gossip is particularly well-matched, and what are its limitations? What alternatives are there to gossip-based solutions, and when would we be better-off using a non-gossip protocol? When, in effect, is gossip the technology of choice?