[ntp:questions] Synchronize clocks without a reference clock

Spoon devnull at localhost.com
Mon Mar 19 10:12:45 UTC 2007

Maarten Wiltink wrote:

> Spoon wrote:
>> Consider a cluster of computers on a LAN. These computers have no
>> acess to a reference clock (no Internet, no GPS, etc).
>> One could still want to have all their clocks synchronized, even if
>> this means they are all slowly drifting away from the "real" time.
>> I've read a FAQ which seems to describe this use case.
>> http://www.ntp.org/ntpfaq/NTP-s-refclk.htm#AEN4264
>> All I need to do would be to add these lines:
>> server # LCL, local clock
>> fudge stratum 12 # increase stratum
>> to my configuration file? Did I understand correctly?
> It would work, most likely.
> You'd be following a single server's eccentricities, though, and there
> are two cases where this could be bad. First, if you've chosen a server
> that wanders around rather a lot (say, next to a door), your entire
> network will get time that also wanders around a lot. Not a very big
> deal, usually.

When you say that the clock in a server next to a door will wander a 
lot, are you referring to the effects of the sudden temperature swing 
when a cold draft comes in the server room?

> The other potential problem is that there is a maximum drift that can be
> corrected. If the difference between your server and any other host is
> larger than that, the client won't be able to follow the server, while
> both would be able to follow UTC. This is unlikely but possible.

I was wondering whether it is possible to reduce the maximum rate of 
change? Is that what the tinker dispersion option is for?


"The argument becomes the new value for the dispersion increase rate, 
normally .000015 s/s."

Does this mean that the maximum rate of change is 15 µs per second?

Could I safely change this value to 5 µs per second?

> Orphan mode may be better. It's in the docs, too.

Found it.


Thanks for the pointer. I will read it carefully.


More information about the questions mailing list