[1794] in Release_7.7_team
update times
daemon@ATHENA.MIT.EDU (Jonathon Weiss)
Thu May 20 19:02:27 1999
Message-Id: <199905202302.TAA03537@speaker-for-the-dead.mit.edu>
From: Jonathon Weiss <jweiss@MIT.EDU>
To: release-team@MIT.EDU
Date: Thu, 20 May 1999 19:02:17 EDT
It took approximately 5 hours for my indy to update from 8.2 to 8.3
(plus half an hour of mkserv).
The nightly hesiod update happens at around 1:15am
There are 2 hours of named caching for hesiod records, which means a
machine will notice new hesiond info between 1:15am and 3:15am
assuming it is unused.
There is a 4 hour desync period for starting updates, which means that
assuming no one is logged in a machine will start its update between
1:15am and 7:15am
This means a public indy will finish its update between 6:15am and
12:15pm.
As was demonstrated last year, this will cause mass confusion in the
morning.
If we're clever, and ops is willing (which I beleive we are) there are
a couple of possibilities.
Assuming we want to target 7:00am (this is a little arbitrary, but it
leaves some time for mkserv to run if it is a private machine, and for
some people to come in before 9am) as an update completion time for a
machine that isn't used overnight (eg a machine in someone's office)
then we can run a manual hesiod update at 8PM. This seems like it is
a little early, we might be able to improve it a little, but not a
whole lot. It is possible that we can bring the hesiod named caching
down to 5 minutes earlier in the day, this however is more effort than
running the manual update which we'd have to do anyway.
Thouhts?
On a related note, I was thinking about some of the problems we had
last year, and I was going to suggest that we update each platform on
it's own night to reduce resource contention. I'd suggest that we
split up the suns too, but I'm not convinced that's necessary, since
they aren't doing an os update.
Thoughts?
