Troy Dawson said...

|May I advise against an 'everything' install from a different 
|perspective, other than time.

Certainly.

8^)

However...

|DON"T DO AN EVERYTHING INSTALL ON ANYTHING PRODUCTION
|
|I could be wrong, but I have yet to know anyone who actually does know 
|the in's and out's of every single package in Scientific Linux.  This 
|includes me and Connie.
|
|A good example is someone who complained that our LVM was broken.  But 
|it seems they installed 'everything' which includes the cluster LVM, and 
|since they hasn't setup a cluster, LVM couldn't connect to the cluster, 
|so it wouldn't run correctly.

Installing doesn't mean turning on.

Part of our install process turns off every last
thing we don't *need* running on every single box.

Then if we need extra stuff on certain systems,
we'll add that.  For classes of systems that need
the same things turned on, we have scripts to turn
just those things back on.

Our install reboots twice, and involves one manual
intervention to set the system name since we don't
use DHCP/BOOTP to assign names, so we're probably
a bit odd that way.  But otherwise it's all scripted,
and we end up with everything installed, but very
minimal systems in terms of what's running on them.

So our systems have LVM, squid, spamassassin,
bgpd, canna, hpoj, and irda, even though not none
of the desktops or simfarm systems (or even our
internal infrastructure systems) need it.  But
they're all turned off automatically before they
hit production ststus.