| I'm replying long after the note since it appears that my seen map got blown
and so I hadn't seen my unseen notes for months.
You can set a minimum time between syncs so that this will limit the
traffic over the wan, which is particularly important if using ISDN, etc.
This says
Synchronization Hold Down = +0-00:05:00.000Iinf
don't synchronize more often than 5 minutes, but if the inaccuracy is
still low, it will wait longer.
A clerk will synch with a minimum of the minimum required servers.
It will use "all" of the servers on a local LAN and then if that
doesn't meet the minimum then it will use as many more wan servers
as required. It will randomize the list of wan servers and go down
the list until it gets the required minimum or it reaches the end of
the list.
If it can't reach at least the minimum then it will report a fault
and then continue increasing the inaccuracy and otherwise operating
as it usually does. It will try again after the sync hold down time
and when it succeeds then it will adjust the time to bring it into
synchrony.
By randomly chosing from the list of servers, it effectively synchs
with all, but over a longer period of time. The overhead of a time
server is usually not very high, I'm using a VAX 3100 workstation
for one of less than a dozen global servers within Digital. This is
one of three registered global servers in my set; they all get pretty
close to the same number of requests, in this case, 1300 in about a week.
However, most sites now have servers on their LANs so this has reduced
the load significantly. At one point the vax 3100 was processing about
one request every 3 seconds. The time serving didn't make any noticable
impact on the system's responsiveness.
|