[ntp:questions] Re: Small fixed offset

Martin Burnicki martin.burnicki at meinberg.de
Fri Oct 28 10:15:19 UTC 2005


ariel.burbaickij at gmail.com wrote:
> Hello dear newsgroup participants,
> as I see from skeaming through the newsgroup the fixed offset topic is
> one of the most favorite ones, well I also need it, the only difference
> being the fact that
> the delay needed is small indeed -- around 150 milliseconds. Why is it
> needed?
> We need to compensate for the network latency for the signals
> backhauled from
> central part of USA to Central Europe for different accounting
> purposes. How the setup of time synchronization is intended? We plan to
> set up separate ntp server with GPS clock connected to it that DOES NOT
> communicate with any peers on Internet (we can set also 2 or 3 of them
> for redundancy purposes). Fudge of +/- approx. 150 milliseocnds will be
> set up on this server /these servers. Pieces of equipment to which
> signals are backhauled to will have this server /these server
> as their synchronisation clock. Will this approach work?

I'm not quite sure why you think you'd have to manually compensate for the
network latency. It's one of NTP's built-in features to determine the
network latency and compensate it.

I'm located in Germany, and as an example I've temporarily added 2 external
servers in the U.S. to my NTP setup which uses a GPS PCI card as primary

# ntpq -p -c as
     remote        refid      st t when poll reach   delay   offset  jitter
*GENERIC(0)      .GPSi.        0 l   40   64  377    0.000    0.017   0.027
+wan-server-1    .----.        1 u   37   64  275  153.336    5.475   3.651
+wan-server-2    .----.        1 u   32   64  377  154.290    6.403   2.881

In the example above I've masked the names and refids of those servers, but
you can see that the delay due to the network latency is about 150
milliseconds, whereas the time offset which is determined by NTP is just 5
or 6 milliseconds. 

I'm sure the offset would be even smaller if I my local connection to the
internet was via a line with the same speed for upload and download.
However, I've an asymmetric connection where the downlink speed is a
multiple of the uplink speed. This asymmetry is hard to determine and
compensate for NTP.

Hope this helps.

Martin Burnicki

Meinberg Funkuhren
Bad Pyrmont

More information about the questions mailing list