[ntp:questions] Stratum 0 refclocks and PPS

Dave Hart hart at ntp.org
Wed May 9 01:07:28 UTC 2012


On Tue, May 8, 2012 at 10:57 PM, Mike S <mikes at flatsurface.com> wrote:
> I've got a couple of Soekris NET4501s, with Moto M12+ GPS timing receivers.
> They've been running pretty well with nanoBSD, using the special high
> resolution elan-mmcr PPS driver.
>
> But, there doesn't seem to be any optimal refclock driver combination. (ntpd
> 4.2.7p236)
>
> If I use the Atom refclock driven by elan-mmcr for PPS, and the Oncore
> driver for TOD and PPS, the PPS sources seem to conflict. NTP appears to
> alternate which PPS is used, causing quite a bit of jitter (Atom is very
> stable, Oncore is subject to interrupt latencies). Not linking
> /dev/oncore.pps.0 to /dev/cuau1 causes the Oncore driver to not work at all.

The Oncore driver is designed to require PPS to operate.

> I've played with using "prefer" (prefer Oncore, prefer Atom, prefer both),
> but that doesn't seem to make things any better.
>
> Using the Oncore driver alone would make sense, but to take advantage of the
> accuracy provided by the elan-mmcr driver, I'd have to point
> /dev/oncore.pps.0 to /dev/elan-mmcr. When I do that, the Oncore driver
> doesn't work. I'm guessing that's because elan-mmcr only signals asserts,
> and never clears, but maybe it's something else.
>
> Things get better with just the Atom driver and an external prefer peer, but
> that means it can't be a stand-alone NTP server.

I'm not familiar with the elan-mmcr driver.  From your description it
appears you have a single hardware PPS signal which is exposed to
software via two different means (perhaps because it's connected to
both a RS-232 input pin and another PPS input?).  The obvious solution
in that case is to point /dev/oncore.pps.0 to /dev/elan-mmcr as you
tried.  I would focus on debugging that configuration, with no
separate PPS/ATOM driver configured.

There may be a bug lurking in the Oncore driver your configuration
exposes, or the problem may be with the elan-mmcr PPSAPI
implementation.  We don't have enough information yet.

Cheers,
Dave Hart


More information about the questions mailing list