[ntp:hackers] Is 24 hours too tight a window to arm leap second?

Dave Hart davehart at gmail.com
Fri Mar 20 22:27:37 UTC 2009


Come the end of the month there will be reports of bogus leap seconds
as always.  -dev has dramatic improvments to leap second handling
compared to -stable, but I question one aspect.  I haven't looked at
the case of a local leap seconds file, only network behavior.  At
least in that case, the leap bit is not armed until one day before the
scheduled leap second insertion.  This has the beneficial effect of
limiting the spread of bogus or stale leap bits through the ntp
heirarchy, but my concern is it will also mean operators may check up
on their NTP servers more than 24 hours before the end of the month,
see it showing no leap bits and assume this is one end of quarter
their ntpd will behave, then be surprised to find a leap second
inserted.

I know it's a trade-off, but my hunch is 24 hours is too little
advance notice of a leap insertion for operators, despite the benefit
of damping false leap bits longer.

Cheers,
Dave Hart


More information about the hackers mailing list