[ntp:questions] What accuracy should I expect on a LAN?

skip at pobox.com skip at pobox.com
Wed Feb 8 15:34:30 UTC 2006

Our LAN has a mixture of Solaris and Linux machines.  Most are Intel
boxes, but a few are SPARCs.  We use xntpd across the board.
Connection speeds are 100Mbit or Gigabit ethernet.  We have three
internal NTP servers that sync with the outside world.

How close should we expect NTP to keep the clocks on the computers
synchronized?  Looking at the output of "ntpq -pn" on one machine I

     remote           refid      st t when poll reach   delay   offset
*     3 u  870 1024  377     0.40   -0.813
+    3 u  286 1024  377     0.50   -0.566
+     3 u  867 1024  377     0.34    3.490

Does that indicate that the machine is within 3.5ms of the time of one
of our servers and within 1ms of the time of the other two (offset

Another machine shows quite different offsets:

     remote           refid      st t when poll reach   delay   offset
*     3 u   30   64  377     0.44   67.072
+    3 u   63   64  377     0.44   70.433
+     3 u   62   64  377     0.40   75.967

If the offset column is really the computed difference in the local
and remote clocks 60+ms offsets seem huge to me for a LAN.

Maybe it's that I'm having trouble interpreting the last three columns
of ntpq's output.  The Solaris 10 man page says:

    The current estimated delay offset and dispersion of the peer, all
    in milliseconds.

but doesn't define "delay", "offset" or "dispersion".  I can guess at
the first two, but the third means absolutely nothing to me.

Can someone provide or point me to some more readable/complete
description of what this all means?


Skip Montanaro
skip at pobox.com

More information about the questions mailing list