[ntp:hackers] Leap second support in versions prior to 4.2.6
Brian Utterback
brian.utterback at oracle.com
Sat Jun 2 18:36:47 UTC 2012
On 6/2/2012 2:29 PM, Dave Hart wrote:
> On Sat, Jun 2, 2012 at 5:23 PM, Brian Utterback
> <brian.utterback at oracle.com> wrote:
>> I know that. The point I was making is that the NTP on Mars project allows
>> for very long round trip times and very long times between updates. With
>> that it is not feasible to propagate a leap second indicator to all servers
>> at all levels withing 24 hours. Or at least, not reliably.
> There's also the ACTS dialup refclock, which can be configured to poll
> as infrequently as 2**17s or 36.4h. Nonetheless, ntpd doesn't light
> the pending leap insertion bit until the last day. We know there are
> kernels out there that will insert a requested leap second without
> regard to date (specifically, on other than the 4 possible insertion
> dates each year). Signalling the leap all month (or after the first
> day, to avoid boundary conditions with the prior month rollover) would
> help Mars and long-poll dialup, but hurt many more.
>
> Cheers,
> Dave Hart
Another area where the protocol is weak. We can look back and see the
need to squeeze as much meaning as possible into every bit, but that is
past. In terms of actual waste though, it was wasteful to dedicate even
a bit to something that is only going to be set for one day every year
or so. The leap indicator screams out to be an extension field.
Brian Utterback
More information about the hackers
mailing list