Re: [Ntp] Leap second draft

[Martin Burnicki <[e-mail]>]

Tony Finch wrote:
> Warner Losh <[e-mail]> wrote:
>>
>> And it's less obscure than the crazy notion that leap seconds could happen
>> any time other than June and December as a practical matter.
> 
> RFC 5905 allows for monthly leap seconds. There's a chance we will need
> more than two leap seconds in a year before the end of this century.
> 
> Some of the problems with leap second signalling are because NTP clients
> only get a two bit warning signal, which isn't enough especially if the
> server is setting the bits using assumptions about how leap seconds are
> currently administered (at most every 6 months), and the client is
> interpreting it according to RFC 5905 and TF.460 (at most every month) and
> the server incorrectly provides more than one month warning.
> 
> I think if the server explicitly tells the client when the leap second(s)
> occur, then there's less scope for problems caused by unwarranted
> chumminess with the IERS's bulletin C schedule.

>From what I've seen in the past there are only very few leap second
problems due to NTP itself, neither due to the protocol, nor due to the
implementation, ntpd.

If a GPS receiver emits a wrong leap second announcement, e.g. at the
end of September when in fact the leap second occurs at the end of
December, then it wouldn't even help if the full (wrong) date was passed
to clients instead of just an announcement flag.

If an OS kernel just steps the time back by 1s to insert a leap second
then this will happen, too, if a full date is specified for the leap
second, with the same effects on applications that get confused due to
the step.

If an OS kernel has a bug and the kernel hogs up the CPU (happened with
some older Linux kernels) when the kernel itself tries to insert a leap
second then there's no big difference whether the kernel has received
the leap second announcement from an NTP implementation evaluating just
a flag, or a full date, or from some PTP slave software, or from any
other source.

If an OS doesn't support leap seconds at all (Windows before *very*
current versions) then you can only try to implement a workaround.
Similar with smearing.

Please note I absolutely find it useful to be able to specify the full
date for which a leap second has been scheduled, but keep in mind that
doing so doesn't prevent from the kind of problems mentioned above,
which are not even related to NTP.


Martin
-- 
Martin Burnicki

Senior Software Engineer

MEINBERG Funkuhren GmbH & Co. KG
Email: [e-mail]
Phone: [phone]
Linkedin: https://www.linkedin.com/in/martinburnicki/

Lange Wand 9, 31812 Bad Pyrmont, Germany
Amtsgericht Hannover 17HRA 100322
Geschäftsführer/Managing Directors: Günter Meinberg, Werner Meinberg,
Andre Hartmann, Heiko Gerstung
Websites: https://www.meinberg.de  https://www.meinbergglobal.com
Training: https://www.meinberg.academy