>While the requirement for timestamping and the internal implementation
>of skalibs favours the wall clock [1][2], I think it does not preclude
>programs from using the monotonic clock solely for internal
>timekeeping,
>which might make the clock skew issues less serious.
It's definitely a possibility, but it would require some heavy
rewriting
- typically, implementing a second set of iopause() primitives around
CLOCK_MONOTONIC. (And defaulting to CLOCK_REALTIME on systems that do
not provide CLOCK_MONOTONIC, I guess.) And using them, and juggling
both wallclock and stopwatch times.
When you use several clocks in one process, you don't have a unique
view of what "the current time" is anymore. This is a bit annoying,
because
a lot of the skalibs functions assume that there's one true time source,
i.e. tain_now(). Generally it's a correct assumption. I'm not sure
whether the benefits of handling the special case of booting a machine
that doesn't know the time yet (so will require one time jump at some
point) are enough to balance the inconvenience of redesigning and
rewriting so much stuff.
--
Laurent
Received on Sat Oct 29 2016 - 15:39:41 UTC