Leap seconds

Paul Theodoropoulos paul at anastrophe.com
Sun Jan 14 21:16:31 UTC 2018

On 1/14/2018 12:51 PM, Gary E. Miller wrote:
>> I'm running a Raspberry Pi 1 Raspian Stretch Stratum one server,
>> built to the spec of the Stratum-1-Microserver HOWTO on the
>> ntpsec.org site.
> Cool!

It was a lot of fun building it last summer. As a proud papa, I feel the 
need to share photos:

> From the ntpleapfetch man page:
>         "This can be run as a cron job. As the file is rarely updated, and
>         leap seconds are announced at least one month in advance (usually
>         longer), it need not be run more frequently than about once every
>         three weeks."
Ah, I'd missed that.

> Sadly, the concensus on best-practices does not exist. For example,
> Google does leap smearing:
> https://developers.google.com/time/smear
> And, just to add confusion, they used to smear over 24 hours, but
> recently they smeared over 20 hours.
> Alternatively, Amazon and Microsoft still smear over 24 hours.  Fun!
> Now you can avoid those idiots, but do you know which of your chimers
> are using those idiots as chimers?
> So, if exact time around the leap second moment is important to you, best
> to run a local GPS, with current leapfile and stay up for the New Year.
> Or, just do not do anything that requires accurate time for 24 hours
> before and after the leap secnd moment.  Go party!  :-)

That's ultimately kind of discouraging/sad. Clock discipline is a 
crucial service - we know that by the very existence and need for 
NTPsec! You would expect that players - whether big or small - would 
come to consensus (is that a  pun?) on how best to deal with leapsecs 
before just starting to smear them wily-nily. Geez.

I for one am going to pull down my own leapfile and check for a new one 
every 27 days. :)

Thanks very much for the reply.

Paul Theodoropoulos

More information about the users mailing list