QNX (was: Re: Raspberry Pi NTP config with fudge factors

[Eric S. Raymond]

Gary E. Miller <gem at rellim.com>:
> > It sounds like this has implications for how I should write the HOWTO
> > recipe.  Are you saying that with refclock 20 the source can get
> > marked as a falseticker if it loses PPS for a while?
> 
> Yup.  Easy to test, just disconnect the PPS signal, if the #20 acts
> like the #48, it will start tracking the NMEA time, which will be offset
> and jittery.  When the PPS returns ntpd will rule out that clock due to
> the high jitter.  Eventually it sorts itself out, as long as the GPS
> stays locked.

Well, shit. That's a strong argument for deprecating #20 and eventually
shooting it through the head.  Hal, I hope you're paying attention to this.

> I'm busy today, but I'll run some tests later to verify that is the
> current behavior.  At a minimum #2 makes it very hhard to see the NMEA
> time and fix a good offset.  

Please do that.

> Last I looked at #48 it was bad in the default configuration due to the
> same NMEA/PPS mashing.
> 
> > If that's so, then maybe we do want to base the build on refclock 28
> > and what you call gpsd-lite.
> 
> With the QNX thing, I wonder how portable #28 is.  Maybe the thing to do is
> fix #48 or #20.  But I fear we'll start Yack Shearing.

OK, this is a policy call.  I just checked with Mark on IRC and he concurs.

QNX is firmly out of scope.  We'll handle it later if there is demand
and resources backing the demand.

We *will* assume POSIX-compliant shared memory on our target systems.

My intention is to officially deprecate refclocks 20 and 48 in favor
of 28 (SHM), explaining that the way 1PPS and in-band information is
mingled produces bad behavior on 1PPS dropouts.   Actual deprecation will
wait on confirmation from Gary's tests.

The HOWTO will be rewritten to use a stripped gpsd-lite version feeding
SHM.

Thanks for bashing your way through this, Frank and Gary.  You've
clarified several murky issues.
-- 
		<a href="http://www.catb.org/~esr/">Eric S. Raymond</a>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 811 bytes
Desc: Digital signature
URL: <http://lists.ntpsec.org/pipermail/devel/attachments/20160506/6503c1fd/attachment.bin>