[ntp:questions] Synchronize clocks without a reference clock

Per Hedeland per at hedeland.org
Sat Mar 31 10:50:19 UTC 2007


In article <l7qdnTKi8ZvLmpPbnZ2dnUVZ_ufinZ2d at megapath.net>
hal-usenet at ip-64-139-1-69.sjc.megapath.net (Hal Murray) writes:
>
>>Well, I was mainly thinking of the case where you have orphan mode set
>>up as a fallback to external servers - then the drift value that has
>>been calculated while external servers were reachable is a good "quality
>>indicator". But I assume the non-"masters" in an orphan-only setup
>>calculate a drift relative to the "master" - in that case, if you're
>>happy with the drift of the "master", you will be least unhappy if the
>>host with the lowest drift relative to the departed "master" is chosen
>>as the new "master".
>
>The value of drift doesn't have any correlation to goodness.

Well, in the case where you have a bunch of hosts synchronized to a
server that only uses its local clock, it is obviously good if that
server has a low drift relative to "real" time - both for keeping the
time as close to the "real" time as possible, and for reducing the risk
of the particular problem mentioned earlier in the thread, that the
difference in drift between the server and some client exceeds 500 ppm.

On the other hand, the method I suggested, looking at the calculated
(whether based on "real" time or on previous orphan-"master" time) drift
when making the choice of a new orphan-"master", can't actually work of
course: If you have a calculated drift for server X, compensation for
that drift will continue to be applied if server X is chosen as new
"master", thus all servers are equally "good" in that case.

The case where it *would* be useful to know the drift is in the initial
selection of a "master" in a group that has never had the opportunity to
calculate drift relative to "real" time (and hence the chosen "master"'s
clock will be used without any drift compensation) - and in that case
you by definition don't have any values to base the choice on. (Maybe
that was what Steve was trying to tell me.:-)

--Per Hedeland
per at hedeland.org




More information about the questions mailing list