Thus, the guard occasionally does not really favor a
lower-priority process over a higher-priority process, but it
may cause a further scheduling latency until the highestpriority
process will be run. That latency has an upper
bound and depends on the maximal possible number
of pending epilogues at a time. The scheduling latency
can be avoided only if guard and scheduler are one thing,
i.e. if the guard conforms to the strategy of the real-time
scheduler. However, this requires the management of an
interrupt-transparent queue other than according to FIFO—
which tends to increase overhead and somewhat limits the
real-time properties. The approach followed by the current
PURE implementation therefore is to only register the
scheduling decisions during epilogue propagation and then
have a single, finishing scheduler epilogue that performs
preemption. This measure reduces the scheduling latency
by avoiding “false preemption” that might be caused by the
epilogues.