On 9/21/2011 11:28 AM, Dave Hart wrote:
On Wed, Sep 21, 2011 at 15:06, Marco Marongiu<[email protected]> wrote:
Il 21/09/2011 16:51, Brian Utterback ha scritto:
Define "the right way".
> From my original message, and D.Hart's reply, I'd say the right way
would be that the system support either to have a second 60 after the
second 59 (leap insertion) or that second 00 follows 58.
But this would break POSIX...
Arguably this is what is done by ntpd now, within the important limit
that no platform on which ntpd runs supports a second numbered outside
[0-59]. So either the second numbered 59 is followed by another
second numbered 59, breaking the monotonically increasing requirement,
or similar with the second numbered 0, both of which amount to
stepping the clock back one second, or you choose to slew or "smear"
the one second adjustment over some longer period of time, possibly
stepping the visible clock frequency twice to do so, or using the more
sophisticated google cosine-shaped smear to limit the rate of change
of the leapsec-induced frequency adjustment. The slewing approach has
the upside of no risky backward step, and the downside of up to 1s
offset vs. UTC during the slew.
Cheers,
Dave Hart
It's unfortunate that the earth DOES NOT rotate exactly 360 degrees in
exactly 24.000000000000 hours. This bit of poor design causes all sorts
of problems. Leap seconds are just one of the symptoms!
_______________________________________________
questions mailing list
[email protected]
http://lists.ntp.org/listinfo/questions
