Stratum-1 NTP from Clock Ensembe

Derek Hare

unread,

Mar 14, 2002, 12:34:48 PM3/14/02

to

I'm trawling for some idea's, and advice...

I have two Frequency Standard Systems, each with three Caesium HP5071A
Primary Standards. The Cs clocks are steered to UTC using an HP58503 GPSR
by a custom controller from TSC (running OS2). So, in theory at least, each
of the PPS outputs are synchronised to within a few nanoseconds of UTC.

Other useful facts, each of the two systems are geographically seperated by
some seven miles. Each has a server [for other system housekeeping duties]
running NT4 Server.

What I need to do, at zero to mimimal cost, is to provide Stratum-1 NTP
services from each of the two systems. I have 1PPS and GPS-time available -
and a choice of two OSs.

Guru's - any suggestions?

Ulrich Windl

unread,

Mar 15, 2002, 2:44:28 AM3/15/02

to

OS/2 and WIndows/NT 4.0 are basically the same if you talk about OS
clock quality. I'd set up a Linux or FreeBSD machine that gets the
system clock with rdate or something primitive enough for OS/2 or
Windows, and then feed the PPS output to that machine to trim the
clock to the ultimate accuracy.

Ulrich

Terje Mathisen

unread,

Mar 15, 2002, 3:23:37 AM3/15/02

to

Recover an old Pentium from storage, install FreeBSD on it, enable PPS
processing, and run the latest NTP version with the GPS (NMEA?) driver.

The result should be a _very_ good Stratum 1 server!

Terje
--
- <Terje.M...@hda.hydro.com>
"almost all programming can be viewed as an exercise in caching"

David L. Mills

unread,

Mar 15, 2002, 12:48:20 PM3/15/02

to Derek Hare

Derek,

First, rule out the NT4 server completely. Old Intel PC hulks actually
make rather good NTP servers, but never in Windwoes. Crank up a couple
with late FreeBSD (not Linux). I assume you have local means (via the
GPS receiver?) to resolve seconds numbering and that there is a driver
for the GPS receiver. If not, drivers are not difficult to fabricate.

Next, set up NTP servers at both locations with reference clock driver
and atom driver (for PPS) and configure each location with a peer
pointing at the other location for backup. In fact, you might not want
to do that, as in the case of NIST and USNO servers. If one site goes
down with dead GPS, dependent clients will switch to the other site, not
depend on second-hand service from the deprived site.

A 200-MHz Pentium running FreeBSD 4.1 and PPS via a parallel port keeps
the clock to within a microsecond or two. No fuss, no bother.

Dave

Carl R. Friend

unread,

Mar 19, 2002, 9:46:01 PM3/19/02

to

"David L. Mills" wrote:
>
> First, rule out the NT4 server completely. Old Intel PC hulks actually
> make rather good NTP servers, but never in Windwoes. Crank up a couple
> with late FreeBSD (not Linux). I assume you have local means (via the
> GPS receiver?) to resolve seconds numbering and that there is a driver
> for the GPS receiver. If not, drivers are not difficult to fabricate.

I'm not trying to start any OS flamewars here, but I am curious
about the BSD recomendation over Linux. I just (this past weekend)
fired up a GPS clock at home on a Linux box, and according to the
output of ntpq it stays nicely within the single-digit range of
microseconds (it may be lying) once I shut off the POSN3D option.

Terje Mathisen

unread,

Mar 20, 2002, 2:29:53 AM3/20/02

to

"Carl R. Friend" wrote:
>
> "David L. Mills" wrote:
> >
> > First, rule out the NT4 server completely. Old Intel PC hulks actually
> > make rather good NTP servers, but never in Windwoes. Crank up a couple
> > with late FreeBSD (not Linux). I assume you have local means (via the
> > GPS receiver?) to resolve seconds numbering and that there is a driver
> > for the GPS receiver. If not, drivers are not difficult to fabricate.
>
> I'm not trying to start any OS flamewars here, but I am curious
> about the BSD recomendation over Linux. I just (this past weekend)
> fired up a GPS clock at home on a Linux box, and according to the
> output of ntpq it stays nicely within the single-digit range of
> microseconds (it may be lying) once I shut off the POSN3D option.

On FreeBSD, with stock kernel (i.e. no PPS-Kit) you get nanoseconds, and
usually see performance down in the fractional microsecs with a good
GPS.

Yes, when exhanging timestamps on a LAN with fractional milliseconds
ping times, this really doesn't matter that much, but it's nice to know
that your sever is as accurate as possible, right? :-)

Ulrich Windl

unread,

Mar 20, 2002, 6:24:19 AM3/20/02

to

"David L. Mills" <mi...@udel.edu> writes:

> Derek,
>
> First, rule out the NT4 server completely. Old Intel PC hulks actually
> make rather good NTP servers, but never in Windwoes. Crank up a couple
> with late FreeBSD (not Linux). I assume you have local means (via the
> GPS receiver?) to resolve seconds numbering and that there is a driver
> for the GPS receiver. If not, drivers are not difficult to fabricate.
>
> Next, set up NTP servers at both locations with reference clock driver
> and atom driver (for PPS) and configure each location with a peer
> pointing at the other location for backup. In fact, you might not want
> to do that, as in the case of NIST and USNO servers. If one site goes
> down with dead GPS, dependent clients will switch to the other site, not
> depend on second-hand service from the deprived site.
>
> A 200-MHz Pentium running FreeBSD 4.1 and PPS via a parallel port keeps
> the clock to within a microsecond or two. No fuss, no bother.

Just because you mentioned Linux: I see no reason why the Linux kernel
should behave worse: For our Pentium-100 with ISA-SCSI card getting
PPS from a DCF77 PZF clock I see: 4 microseconds offset, stability
0.01PPM, jitter 1.6 microseconds.

BTW: A non-NANO-compiled ntpdc failes to print correct values for a
NANO-compiled ntpd...

Ulrich