In an asynchronous network it cannot be exact. A short timeout detects real crashes quickly but suspects healthy nodes; a long timeout does the opposite.
Failure detector
A component that guesses whether a remote node is down, usually from missing heartbeats within a timeout.