[ntp:questions] Re: Problem with good synchronization.

Harlan Stenn stenn at maccarony.ntp.org
Mon Oct 18 05:04:24 UTC 2004


Dave,

Your "interesting case" is the one I see the most.  I have 1 path to the
'net and no local refclocks.

If for no other reason than to keep NFS happy the machines on my net
need to agree with each other.

So I have my primary machine listen to its local clock at a stratum 1 lower
than where it ordinarily runs.

I have another machine listening to ist local clock at a stratum 2 lower
than the primary.  It also tries to listen to servers on the 'net.

When all is well, the primary is believed by everybody.

When the outside servers are unreachable, the primary machine drops one
stratum, and everybody still listens to the primary.

If for some reason the primary fails, everybody listens to the secondary
machine.  If the secondary machine cannot reach the servers on the net,
it will eventually lower its stratum by no more than 3 notches (because
it will listen to its locak refclock).

Yes, there is a bit of a dance while different machines at different poll
intervals settle on "who knows best", but there is amost always only a single
leader with this approach.

If the 2 bulls run at the same stratum, in my experience, when the network
servers are unreachable there is no other machine available to "break the tie", so each bull heads off on its own.  The cows follow one bull or the other
and the cliques head off in different directions.  Bad for me.

Am I missing something?

H



More information about the questions mailing list