[ntp:questions] Unable to synchronise meinberg ntp server to clients.

unruh unruh at wormhole.physics.ubc.ca
Fri Sep 16 01:26:24 UTC 2011


On 2011-09-16, Richard B. Gilbert <rgilbert88 at comcast.net> wrote:
> On 9/15/2011 5:23 PM, Chuck Swiger wrote:
>> On Sep 15, 2011, at 1:32 PM, David Woolley wrote:
>>>> server 1.fi.pool.ntp.org minpoll 10 iburst
>>>
>>> The mistake most people make is setting minpoll too low.  This is too high.
>>
>> What's wrong with a minpoll of 10?
>>
>> I recall OS vendors shipping an ntp.conf using "minpoll 12 maxpoll 17"...
>>
>> Regards,
>
> Until you have read and understood Dave Mills' book on the subject, just 
> consider the defaults to be mandatory!
>
> David L. Mills, "Computer Network Time Synchronization: The Network Time 
> Protocol on Earth and in Space, Second Edition"
>
> The useful poll intervals range is, for most purposes, from four to ten, 
> which being translated, means from 16 to 1024 seconds!  The software 
> will select a value in that range and adjust the value as conditions change.
>
> I think of the low poll ranges as correcting errors quickly while the 
> long poll intervals correct small errors very accurately.

Long intervals correct rate errors accurately, while short intervals
correct offset errors. Thus, if you are concerned that your clock
accurately tick one second per second, then long intervals are better.
If you are concerned that your clock's time be as close as possible to
UTC, then short intervals are better. Also long intervals make it hard
to keep up if the rate changes suddenly for some reason (eg you start
using your machine, which heats it up and changes the clock rate). Thus
most computers have a 24 hour cycle-- hot during the day cooler at
night with the attendant changes in clock rate. A long poll interval (eg
an hour) will make it hard to keep up with those changes. increasing the
error in both the rate and the offset of the clock. 

. 





More information about the questions mailing list