[ntp:questions] Re: Comparing offset of multiple refclock/servers
David L. Mills
mills at udel.edu
Mon Aug 15 18:26:06 UTC 2005
I've done a lot of experiments like yours. First, decide whether the
goal is to compare the clocks against each other (open loop) or the to
assess the performance of the synchronized system (closed loop). If open
loop, say "disable ntp" on all three systems so they run open loop. Then
collect the peerstats on all the machines for each other and the clocks
for a few days and write a matlab program to plot the differences. The
program will assume the local system clock of each machine is simply a
flywheel to compare the clock of one machine against the clock of
another. You can of course connect all three clocks to one machine and
watch them babble.
For closed loop performance, consider one of the GPS clocks a reference
and calibrate the others from it, keeping track of the little wiggles
that do occur between the individual referenc clocks and system clocks.
We usually keep our public servers within 100 microseconds closed-loop.
From my experience the GPS receivers are much better than WWVB
receivers, certainly in my case and possibly in yours due to cochannel
RFI. Both of my 8170 receivers were modified at the factory many years
ago to conform with the behavior of the Netclock/1 which I also have
online. So far as cochannel RFI performance, the original and modified
receivers are about the same.
Twenty five years ago when the FCC was doing its job the WWVB
performance was reliably within the millisecond. As time went on and the
FCC did not enforce the limits specified in Part 15 of their rules,
things have deteriorated. Once upon a time the power of an unintentional
radiator was limited to 15 microvolts per meter at a distance lambda / 2
PI, where lambda is the wavelength in meters. I know that very well, as
I was called as consultant in a case where a campus carrier-current
radio station was observed much above that limit and I had to measure
the field strength and correct the problem. The Ann Arbor 560-kHz signal
had leaked on the high voltage lines and was heard 59 in Ypsilanti,
several miles away.
I just broke out a recent version of Part 15 and am astonished by what
the special interests have done to make holes for wifi, vehicle
location, RF tags and whatnot. The limits for all special interest
devices have increased easily tenfold over the years, with current FCC
proposals on broadband over power line (BPL) an absolute disaster for
emergency, broadcast and amateur operations.
Consider unintentional radiators, say due to conducted RFI leaking into
the power grid and the wires acting as antenna. Current Part 15.209 says
300 microvolts per meter at a distance 2400 / frequency (khz) in meters.
That's 300 microvolts per meter at 40 meters. Current WWVB signals here
are in the order of one microvolt per meter at the receiver. Do the
sums. A radiator like that would clobber WWVB if less than 4.6 miles
from the receiver. The original limit was 15 microvolts per meter at 796
meters which is about one microvolt per meter at 3.8 miles. This is huge.
Consider the case of conducted RFI, which was originally targeted at CB
receiver local oscillators leaking into the power line. Consider also
the case where some high power welding machine at 60 kHz backs up into
the power grid. FCC 15.207 says not more than a millivolt can leak into
the power line. You're telling me a 500 kW welding machine leaks less
than a millivolt at 560-V three-phase AC source? What say the Chrysler
management if I show up at their front door with a field strength meter
pegged at 60 kHz? Heck, the FCC consultants told them BPL wouldn't work,
but nobody was listening.
The bottom line is that WWVB should be considered an endangered species
and probably a hazard for DCF77 and HBG in Europe as well. I don't know
why the FCC didn't make an exception at 60 kHz. They did make exception
for 5 kHz either side of the WWV carrier frequencies, but then allowed
50-kW redneck broadcasters in Arkansas with carrier frequencies right at
the stopband edges. Sideband splatter kills WWV 5 MHz every night. Gotta
forgive the FCC, since the engineers are long gone and replaced by
John Ackermann N8UR wrote:
> As part of my timekeeping fixation, I just put three FreeBSD-based NTP
> servers online on my home network, each driven by its own refclock (two
> HP Z3801A GPSDOs, and one Spectracom 8170 WWVB clock).
> In addition to plotting the internal offset and jistter performance
> (which I have on-line at http://www.febo.com/time-freq/ntp/stats), I'd
> like to compare and plot the relative offset of the three systems
> against one another. It seems like that should be easy, but I'm having
> some trouble figuring out the best way to go about it, since even the
> PPS sources may show a slight offset in the ntpq -p results, and of
> course if one of the refclocks has an offset (e.g., right now my WWVB
> clock has about 3ms of systemic offset), the server that relies on it
> won't directly show that -- it won't know that the time it's syncing to
> is a little bit off..
> One option might be to go to a fourth machine that is looking at these
> three, and use it as a sort of transfer standard -- looking at the
> relative offset of other machines, and then subtracting them against
> each other to obtain the differences. But that seems sort of klunky.
> Am I missing something obvious? Any suggestions would be appreciated.
> questions mailing list
> questions at lists.ntp.isc.org
More information about the questions