Unreliable failure detectors enrich asynchronous distributed systems
with semantics that are sufficient to solve consensus in the presence of
crash failures. Implementing unreliable failure detectors
requires a system that provides at least some synchrony, however,
typically a known upper bound upon the end-to-end delays of messages.
Recently, we introduced an implementation of a perfect failure detector
in a novel partially synchronous model, referred to as the Theta-Model,
where only the ratio Theta of maximum vs. minimum end-to-end delay
between correct processes must be known a priori. In this paper, we
present an alternative perfect failure detector algorithm, which is
based upon a clock synchronization algorithm for the Theta-Model.
It not only surpasses our first implementation with respect to failure
detection time, but provides the semantics of an eventually perfect
failure detector during system booting as well.