[ntp:questions] Re: Why do clients reject win 2003 ntp server?

Danny Mayer mayer at gis.net
Fri Jul 15 12:34:23 UTC 2005


Brian Utterback wrote:
> We just finished debugging a problem from a customer that may be
> related to this problem.
> 
> The customer had many, many NTP clients. They mostly worked just
> fine, but some would not sync with the 2 NTP servers they had.
> If they restarted the NTP daemon they would sync fine, but without
> restarting, the daemon would refuse to sync and there wasn't anything
> obviously wrong.
> 

Well the first thing wrong is exactly 2 servers. You need at least 3
just to decide if any two are close enough to each other to make a
decision on which to prefer. With just two, there's no way of making any
decision.

> Further investigation showed that these clients were rejecting the
> servers because of the root dispersion being too large. "This is
> bizarre" sez I, why would some clients think that the root dispersion
> was too large and some did not, and this state persist for days?
> 
> I did a packet trace and made an amazing discovery. These servers
> (both running Windows 2000, SP 4) did not correctly set the root
> dispersion field in their NTP responses to the clients. Instead,
> they sent back exactly what the client had sent them!
> 
This is not an unusual situation when people don't fully understand not
just the spec but the technical reasons behind it. You could consider
that a deficiency of the spec since proper interoperatability is the
goal of the spec.

Danny




More information about the questions mailing list