[ntp:questions] Testing Sync Across Several Systems

Ulrich Windl Ulrich.Windl at RZ.Uni-Regensburg.DE
Thu Aug 13 09:25:21 UTC 2009


T <g4173c at motorola.com> writes:

> Greetings:
>
> We have about 50 Linux/Solaris/Windows boxes running ntpd at several
> different sites. Some of the systems from time to time go out of sync.
> My question is there a way to test ntpd machines are all in sync with
> the master
> server?
>
> I was thinking of using ssh to get on to each machine to do a date and
> then go back to the master and do a date and compare, but this seems
> problematic at best. What do people do to check that all machines are
> in sync?

Hi!

Recently I've started a completely different approach: Instead of
looking at phase and frequency offsets, I defined a set of "sanity
conditions" to check ntp servers for. At the moment I have 3 conditions
for the system status, 11 conditions for the "peer status", and 2
conditions for the clock status. I compute a value between 0 and 1 of
those, feeding them into rrdtool.

This works amazingly nice: I don't care about a specific offset, I only
want to know whether a ntpd looks as if it should have a reliable time.
That way I found a few servers with mis-configured peers, and I found
that the NTP daemon in HP-UX 11.31 and Solaris 10 both report a "freq"
value through a mode 6 query that is by a factor of 1000 too large
(i.e. by that factor larger than the value witten into loopstats). I only
noticed, because one server had a frequency error of over "4000
displayed PPM"...

The scripts to do the stuff are very new, so I don't wnat to publish
them (in case you would ask), but we can discuss the approach here or in
private e-mail.

Regards,
Ulrich




More information about the questions mailing list