| when time changed, my guess is that the TDF got screwed up on either the
one node with the time provider or on the three nodes without the time
provider.
If this were the case, the displayed system time would be correct, but the
UTC time that DTSS uses would have been off by an hour. Since this would
have not intersected with the current time interval, there would have been
a fault. The way that faults is resolved is by looking for the smallest
interval that includes the maximum number of systems (and the local system
doesn't have a higher weight than any of the others, so the local server
can be declared faulty). With four servers, it would require that three
servers agree.
Now there is an exception to this "majority rule" rule: the Time Provider
is King. This means that the system with the external TP will not pay
any attention to the other servers as far as setting the time is concerned.
However, all of this will be noisily reported in the DECnet event log
as time server faults and non-intersecting times. If you enable the
event logger and look at the output after 24 hours, I think that you
will see where the disagreement lies. As long as you have control over
the systems, its easy to resolve the problem.
(The problem I have is figuring out shich time server is faulty when they are
scattered around the world; if I can issue a "ncl sho node foo dtss all" then
I can pretty much track down the source of a time fault in a global environment,
but the event logging is what points out the problem and provides the starting
point.)
|