A hybrid approach for building eventually accurate failure detectors

Abstract

Unreliable failure detectors introduced by Chandra and Toueg are abstract mechanisms that provide information about process crashes. On the one hand, failure detectors allow a statement of the minimal requirements on process failures that allow solutions to problems that cannot otherwise be solved in purely asynchronous systems. However, on the other hand, they cannot be implemented in such systems: their implementation requires that the underlying distributed system be enriched with additional assumptions. Classic failure detector implementations rely on additional synchrony assumptions such as partial synchrony. More recently, a new approach for implementing failure detectors has been proposed: it relies on behavioral properties on the flow of messages exchanged. This shows that these approaches are not antagonistic and can be advantageously combined. A hybrid protocol (the first to our knowledge) implementing failure detectors with eventual accuracy properties is presented. Interestingly, this protocol benefits from the best of both worlds in the sense that it converges (i.e., provides the required failure detector) as soon as either the system behaves synchronously or the required message exchange pattern is satisfied. This shows that, to expedite convergence, it can be interesting to consider that the underlying system can satisfy several alternative assumptions.