[LinuxPPS] Why patch refclocks?

[Cirilo Bernardo]

On Sun, Jun 1, 2008 at 1:28 PM, Mark A. Day <techwhiz at embarqmail.com> wrote:
> On 06/01/2008 08:56:14 AM, Mark A. Day wrote:
>> A bit off-topic, but why are we patching ntpd refclock drivers when the
>> atom-pps driver already exists? If I have multiple gps sources attached to
>> my host, wouldn't I want the most accurate pps source to service all of
>> them?

Actually, no.  You don't want a single PPS source synchronizing
everything no matter how good you think it is, because if something
goes wrong you're screwed.  So even though other sources have fairly
large jitter, they're still useful; comparisons will show if your
super-precise clock is failing.

>
> Attempting to answer my own question... No, each pps source is to be
> associated with a time source to which it refers. Then what of the atom-pps
> driver? He is reserved for pps sources which have no associated time source?
>

Drivers like the 'atom-pps' have a very special use - for example, you
can use a cheap low-power portable rubidium vapor clock to provide a
precision of  better than 1ppb (better than 1 second in 30 years).
However, although the ticking has incredible precision, it is not
actually synchronized with any time source.  Using other far less
precise time sources, NTPD can determine the offset of the atomic
clock and after that, use the atomic clock to precisely synchronize
the other time signals.  If you simply took the output of a GPS and
used the output of an atomic clock as your reference for the start of
each second, you will actually be pushing NTPD to set the wrong time
to somewhere within +/- 0.5 seconds.