[ntp:questions] Re: ntpd going wild

Richard B. Gilbert rgilbert88 at comcast.net
Tue Feb 14 18:17:46 UTC 2006


Hermann Hastig wrote:

> Hi,
> 
> could anybody explain what caused the following messages from my ntpd
> around February 1?
> 
> Jan 20 00:43:32 [ntpd] time reset -0.158476 s
> Jan 20 00:43:32 [ntpd] kernel time sync disabled 0001
> Jan 20 01:04:54 [ntpd] time reset +0.156179 s
> Jan 31 17:32:58 [ntpd] time reset +0.280824 s
> Jan 31 17:32:58 [ntpd] kernel time sync disabled 0001
> Jan 31 18:11:23 [ntpd] time reset -0.291079 s
> Jan 31 19:24:02 [ntpd] time reset -0.168920 s
> Jan 31 23:23:15 [ntpd] time reset -0.176246 s
> Jan 31 23:44:41 [ntpd] time reset +0.551415 s
> Feb  1 00:23:07 [ntpd] time reset -1.160321 s
> Feb  1 00:23:07 [ntpd] frequency error -503 PPM exceeds tolerance 500 PPM
> Feb  1 01:01:30 [ntpd] time reset -0.328978 s
> Feb  1 01:39:59 [ntpd] time reset +0.284085 s
> Feb  1 02:56:48 [ntpd] time reset -0.144699 s
> Feb  1 03:35:13 [ntpd] frequency error -554 PPM exceeds tolerance 500 PPM
> Feb  1 04:09:21 [ntpd] time reset +0.486346 s
> Feb  1 04:09:21 [ntpd] frequency error 528 PPM exceeds tolerance 500 PPM
> Feb  1 04:30:46 [ntpd] time reset +1.114938 s
> Feb  1 05:09:13 [ntpd] time reset -1.170027 s
> Feb  1 05:09:13 [ntpd] frequency error -570 PPM exceeds tolerance 500 PPM
> Feb  1 07:55:54 [ntpd] time reset +0.217560 s
> Feb  1 08:17:16 [ntpd] time reset +0.625292 s
> Feb  1 08:55:43 [ntpd] time reset -1.170204 s
> Feb  1 08:55:43 [ntpd] frequency error -507 PPM exceeds tolerance 500 PPM
> Feb  1 09:34:10 [ntpd] frequency error -32659 PPM exceeds tolerance 500 PPM
> Feb  1 10:12:32 [ntpd] time reset +0.230971 s
> Feb  1 10:51:03 [ntpd] time reset +0.227721 s
> Feb  1 11:29:28 [ntpd] frequency error 532 PPM exceeds tolerance 500 PPM
> Feb  1 12:03:43 [ntpd] no servers reachable
> Feb  1 12:46:24 [ntpd] time reset -1.731346 s
> Feb  1 12:46:24 [ntpd] frequency error 512 PPM exceeds tolerance 500 PPM
> 
> Even though I have set maxpoll to 16 in order to cause as few traffic as
> possible to the pool.ntp.org servers I'm a bit surprised that ntpd had
> _that_ much problems to keep the clock stable late January 31.
> 
> So, what might have caused this, and is it my fault or the three
> upstream servers'?
> I don't see why ntpd completely lost control with a setting that worked
> flawlessly 24/7 for a year, on the other hand I can't believe that all
> of the three configured servers had different times the same day. I
> mean, one, ok, but all three?
> 
> Thanks, Hermann

Did something happen to your environment?  It looks as if your local 
clock is, or was, on the edge of the +/-500 PPM tolerance range.  A 
temperature change could have pushed a marginal clock over the edge.

Another possibility would be network convulsions of some sort affecting 
the path to one or more of the servers you use.  If; e.g. you are in New 
York, using a server in Los Angeles and suddenly all the replys are 
being relayed via Brazil, the time ntpd extracts from that server is 
going to off by a good many milliseconds.

I have seen normally good servers suddenly develop very large offsets 
(twenty to forty milliseconds) with respect to my GPS reference clock. 
I have no reason to believe that the servers themselves suddenly jump 
twenty to forty milliseconds!  I have no reason to suspect my GPS clock 
because other network servers continue to maintain their normal and 
expected offsets.

Also, it is not generally a good idea to change the default values of 
MINPOLL and MAXPOLL.  A long poll interval such as 2^16 seconds makes it 
very difficult to correct large errors.  With the default poll intervals 
and everything working properly, your system should settle down with a 
poll interval that might be as high as 1024 seconds; this is not going 
to overload anybody's server.  What kills servers is the broken or 
misconfigured software that's polling every second!




More information about the questions mailing list