Subject: re: Proposed changes to cc_microset().
To: matthew green <email@example.com>
From: Frederick Bruckman <firstname.lastname@example.org>
Date: 07/08/2004 11:13:43
On Fri, 9 Jul 2004, matthew green wrote:
> > considering you can end up at less than 1/2 the clock that means you'll
> > be more than half a second out by the next second. seems pretty severe,
> > ie, it needs to be handled correctly... at 10 second interval, you'd
> > end up all over the place...
> Remember, this isn't overall timekeeping, just sub-second
> timekeeping. Running at half-speed for 10 seconds just means you'll be
> off by as much as a second during that interval, not that you'll
> accumulate overall drift.
> i understand that. i just don't see why it's even remotely
> acceptable... acceptable in that we say "don't fix" rather than
> "fix, but it's not really a big problem."
> (i'm not sure i understand your 10 second comment. if the cycle
> calibration is going to be increased from 1 second to 10, won't
> eg a banias 1.4ghz running a 600mhz be about 6 seconds wrong at
> the extreme?)
No, because the cycle counter only interpolates between seconds,
currently. (With my proposed change, it would be something like 1/10
second minimum.) So the most clock_gettime() could be off would be 1/2
second. I think we should attempt to correct it, nonetheless, but
there are other challenges to getting ntpd to work properly with a
laptop that can change frequency or even go to sleep, so I wouldn't