How to Implement a Time-Free Perfect Failure Detector in Partially Synchronous Systems

This paper introduces our partially synchronous Θ-Model, which is essentially the well-known FLP model augmented with a bound Θ on the ratio between the maximum and minimum end-to-end delays of messages simultaneously in transit between correct processes. We show that the Θ-Model admits the implementation of a perfect failure detector, which allows to employ classic solutions for solving important distributed computing problems like consensus. Since Θ may still hold when some assumed upper bound on the maximum delay is violated, those solutions work correctly in scenarios where synchronous implementations would fail. We show that our algorithm, which employs heartbeat messages and a timer-free timeout mechanism based upon synchronized heartbeat rounds, indeed satisfies the properties of a perfect failure detector and provides a number of attractive additional features.