This paper shows how to implement a time-free perfect
failure detector in a partially synchronous distributed system.
Rather than upper and lower bounds on the minimum and maximum
end-to-end computation + transmission delays between correct
processors, our algorithm needs to know only an upper bound on their
ratio. Since this bound may still hold when an assumed bound on the
maximum delay is violated, our perfect failure detector may still
work correctly in situations where synchronous ones fail, i.e., has
higher coverage.  We prove that our solution, 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 can be generalized in several ways.