Re: Seeking clarifaction of tai64nlocal and leap seconds

Michael Sierchio <[email protected]> Thu, 20 Sep 2012 10:59:31 -0700
Newsgroups gmane.comp.djb.syslog
Message-ID <CAHu1Y70y7-cPy1NRD7FexAG+jt60a4nNq56MShLH1CYfaGaLRw@mail.gmail.com>
On Thu, Sep 20, 2012 at 8:56 AM, Andy Bradford
<[email protected]> wrote:
> Thus said Michael Sierchio on Thu, 20 Sep 2012 08:33:29 PDT:
>
>> You  would be  unique. What time  synchronization mechanism  would you
>> use?
>
> Why, taiclock and taiclockd of course. It comes with the clockspeed
> package. :-)
>
> Really, it depends on  what you intend by it. If you mean  how do I keep
> my clock synchronized with the world, then it's simple; I run clockspeed
> and I train clockspeed using an external SNTP source.
>
> $ sntpclock 1.2.3.4 > /usr/local/clockspeed/adjust

These are NOT serious timekeeping mechanisms - not sufficient to keep
a group of computers within 1ms of each other locally, or within 10ms
of actual TAI (not what your computer's clock says).

ntp stratum 1 servers often have a timesource (GPS) and a reliable PPS
signal generated by a Rubidium or Cesium clock.  DJB's timekeeping
stuff doesn't reflect a deep interest in or understanding of
timekeeping. I'm not saying ntp isn't showing its age, or that TAI is
not a more useful base than UTC (though TAI is equivalent to GPS time,
with a permanently fixed offset).

> This could be run  from a cronjob once a month  (or even less frequently
> really). When running clockspeed, however,  you only need to synchronize
> if your clock's frequency changes.

Your clock's frequency changes with the temperature such that it makes
a fairly accurate thermometer, and there is also noise (jitter).
Unless you have a reliable PPS mechanism and a kernel that can
discipline its clock with it, don't take it too seriously. ;-)

I have previously referenced the Wikipedia page on TAI - there are
some useful pages on timekeeping at USNO and elsewhere.  See this, too
http://phk.freebsd.dk/soekris/pps/

- M