[ntp:questions] Re: Server Sync Problem

Eddy not.here at ntp.isc.org
Sat Dec 17 12:38:39 UTC 2005


Michael wrote:
..
> 3) Problem 1
> If I tell these servers to sync directly to the GPS server as clients, they 
> do suceed, but then
> they lose sync for a few minutes and then find it again etc etc. NTP reports 
> in /var/log/messages
> that "No servers reachable", but I have sat there with tcpdump running 
> watching the UDP
> request/responses ticking away with no problems whatsoever. No packets are 
> lost at all.
> But NTP just doesnt seem to like whats going on.
> However, if I make the client "burst" to the server, it never seems to lose 
> sync at all. Why I need
> to use burst on a LAN when I cant see any packet loss, I dont know but it 
> seems to fix the problem.
> 
> 4) Problem 2
> Now, if I sync a server lets call it A, off the GPS server, then try to sync 
> the other servers off A, none
> of them will sync at all even though NTP can see server A and ntpq reports a 
> reach of 377, the offset is
> < 1ms and the jitter is < .01ms  and tcpdump sees all the UDP 
> request/response packets without any problems.
> So communication with the server appears to be fine, but NTP just doesnt 
> want to sync up at all.
> 
> Anybody got any ideas as to what is happening as I am going slowly mad.

Going mad will not help at all :).

It might be iptables trying to handle the UDP packets in a "statefull"
way. The longer ntpd is running the larger the polling intervall becomes
and iptables decides that it is "not related" traffic and it might get
dropped.

Just my 2 cents.

Eddy




More information about the questions mailing list