John Pettitt wrote:
>A man with two clocks never knows the right time.
>
>I’ve been trying to track down a persistent offset in my stratum 1
>clock and I’ve reached a point where I’m stuck and open to suggestions.
>
>The system runs FreeBSD 5.4BETA1 on Soekris 4801 hardware with a
>Garmin GPS18LVC as a refclock and pps provider using the NMEA driver.
> I have this GPS hooked to two machines with a daisy chain cable. The
>second box is a Celeron 2.9Ghz on an Intel chipset also running
>FreeBSD 5.4BETA1 (upgrading from 5.4 to 5.4 didn’t change any of the
>symptoms). Both machines show time within 50us of each other.
>
>The symptom is this – my clock is persistently 2ms off from a basket
>of other stratum 1 clocks. The spec on the GPS18LVC says the pps
>output is +/- 1us so at least for the moment I’m not blaming the GPS
>(although I could be persuaded).
>
>My first assumption was that the asymmetry in my DSL line was causing
>the offset. However I no longer believe that for two reasons.
>First the calculated asymmetry is about 1.2ms not 2. Second I slaved
>machine in a server farm at my ISP to my stratum 1 and it tracks
>within 100us (and show the same offset to the basket of S1’s).
>
>My next thought was that I had a weirdness in the de-glitch in the
>serial port that was messing with the PPS (2ms =~ 1 char @ 4800bps) –
>however changing the baud rate of the GPS to 38400 didn’t change the
>apparent offset.
>
>So where I’m at now is either
>
>1) my GPS is off by 2ms (why?) or
>
>2) my IPS (sonic.net) has a sufficiently asymmetric connection to the
>rest of the world that it skews all external reference clocks.
>
>How do I tell if my clock is wrong or if everybody just appears wrong?
>Any thoughts, short of buying another GPS, on how to prove one or the
>other would be most welcome.
>
>John
>
>
I would be inclined to believe the GPS! I have a Motorola M12+ Timing
receiver connected to a Sun Ultra 10 running Solaris 8. I also
configure several network servers on the same machine. The better of
the network servers show offsets from the GPS. During daylight hours
when networks are busy the offsets tend to bounce around quite rapidly
but at night they smooth down to a consistent offset for each server. A
couple are plus and a couple are minus. All are within plus/minus five
milliseconds. The worser network servers wander all over the place,
from plus twenty-five milliseconds to minus twenty-five and back again a
full cycle taking an hour or two.
My conclusion, from watching my GPS and Ultra 10 for the last year or
so, is that timing over the internet should not be relied on for
accuracy greater than plus/minus ten milliseconds.
I have another Ultra 10 connected to a PST/Traconex WWV Receiver. That
one has given me hints as to why some of the network servers wander so
badly. At night when I get a good signal from WWV this server agrees
with GPS to within a couple of milliseconds. During the daylight hours
signal quality deteriorates badly and so does the accuracy, to
plus/minus twenty-five milliseconds.
You can have asymmetry anywhere between your site and the servers you
use. For sure you have some in the ADSL connection to your IP.
Requests and replies are not guaranteed the same network path going and
coming. Indeed, it is quite possible to send a packet from Philadelphia
to Los Angeles direct and have the reply take a detour through Dallas.
Your servers have connections to their own IPs and you don't know if
they are T1, T3, cable modem, ADSL or tin cans and string!!!
If you really NEED microsecond accuracy, get another GPS or two or maybe
even a cesium clock of your very own! |