[ntp:questions] Offset is Slowly Increasing Over Time

Michael L. Semon mlsemon33 at verizon.net
Wed Mar 28 06:37:46 UTC 2007


Jason C. Wells wrote:
> If the dispersion on a client machine is spiking at 8 seconds does that 
> indicate that my NTP server wants to be a rodeo cowboy?  But seriously...
> 
> I gathered up the loop and peer stats per the FAQ, posted here:
> 
> http://www.stradamotorsports.com/~jcw/NTP_Stats.xls
> 
> If I restart all the NTP processes on my network I can get my clocks to 
> sync for a while.  At some later time the offset creeps up to over one 
> second and the clients will reject my server.
> 
> The dispersion graph is interesting.  The offset doesn't seem to 
> misbehave in a pattern resembling dispersion.
> 
> About the only thing that I can tell from the graphs is that they are 
> pretty.  I was hoping that some good natured soul would look at them and 
> say, "Here's your problem."
> 
> My local NTP server syncs with my ISP just fine.
> 
> Thanks,
> Jason C. Wells

On the PC where you recorded your peerstats, did you reboot it and/or 
otherwise cause an intermittent connection with the NTP server(s) upstream?

There's one period of nearly 18 days between syncs and two other gaps of 
nearly 2 days.  The other spikes are where your offset exceeded a 
certain window, and ntpd stepped the clock instead of slewing it. 
That's why your offset goes, for instance, from -1.4 seconds to nearly 
zero in one step.

The numbers in your dispersion column seem fine.  If there's a big 
offset--such as the ones after rebooting--the number goes up, and ntpd 
whittles that down as it gets closer to the one correct time.

You might post the server lines and maybe the filegen lines in your 
ntp.conf file.  If I had to make one based on the results, it might look 
like this:

server 192.168.1.204 minpoll 14 maxpoll 17 iburst

But then again, maybe your data is a snippet from a longer file, and I 
didn't see where the local NTP server started with polling intervals 
like 32s and 64s.

Michael




More information about the questions mailing list