[ntp:hackers] smearing the leap second

Miroslav Lichvar mlichvar at redhat.com
Thu Jun 25 10:51:06 UTC 2015


On Wed, Jun 24, 2015 at 10:59:44AM -0700, John Stultz wrote:
> Yea. Realistically, I think any of the short-slews are far too likely
> to break applications.

With the two-second slew some applications would probably break, but
I'd guess the number would be significantly smaller than the number of
applications that break with the current stepping approach. I think it
would make a lot of people happier if it was the default approach.

> > I think ideally each system clock could be "unlocked" from others and
> > either controlled via clock_adjtime() as an independent clock or its
> > frequency offset could be set relative to another clock. This would
> > allow us to minimize the phase error of CLOCK_REALTIME, minimize the
> > frequency error of CLOCK_MONOTONIC, and keep CLOCK_TAI accurate when
> > CLOCK_REALTIME is slewing to correct a leap second.
> 
> Right, however keeping N independent clocks has performance costs that
> have to be balanced since it makes things in hotpaths like timer
> expiration more difficult to calculate.

Wouldn't that be a problem with the UTC_SLS clock(s) too?

-- 
Miroslav Lichvar


More information about the hackers mailing list