Abstract: The Heartbeat Failure Detector of the present invention, is a new, simple method and device that has several identical modules, and each module is attached to a processor in the system. Roughly speaking, the module of a processor x maintains a heartbeat counter for every other process. For each process y, the counter of y at x (called the heartbeat of y at x) periodically increases while y is alive and in the same network partition, and the counter stops increasing after y crashes or becomes partitioned away. Using such a device, x can solve the communication problem above by resending m only if the heartbeat of y at x increases. Note that if y is crashed (or partitioned away from x), its heartbeat at x stops almost immediately, and so x will also immediately stop sending copies of m to y.

Abstract: New failure detector mechanisms particularly suitable for use in asynchronous distributed computing systems in which processes may crash and recover, and two crash-recovery consensus mechanisms, one requiring stable storage and the other not requiring it. Both consensus mechanisms tolerate link failures and are particularly efficient in the common runs with no failures or failure detector mistakes. Consensus is achieved in such runs within 3□ time and with 4n messages, where □ is the maximum message delay and n is the number of processes in the system.