Survey on Scalable Failure Detectors

Maintaining a timely view of the current system status is essential to the performance and functionality of distributed systems. Failure detectors have long been essential to distributed systems. In this paper, we evaluate two failure detection algorithms specifically aimed at large-scale systems. Both assume fail-stop (non-Byzantine) models but the similarities end there. Dynamo’s failure detector relies on pinging with a weak eventual completeness model based on randomization. On the other hand, the classic gossip-protocol is based instead on heartbeats with a strong completeness model. Our simulations test the conclusions of Dynamo’s failure detector in order to evaluate its gains from the traditional gossip heartbeat style approaches. We end with remarks on the advantages of both algorithms and the systems that are best suited for each.