[SAGE] Linux automated installation/bare-metal recovery?

[Jason Antman <jason@xxxxxxxxxxxxxxx>]

Hi,

$University is going to be replacing a few older Red Hat boxes with new 
CentOS5. Most of our older stuff was built as one-off boxes 
(hand-configured, etc.) and I'm trying hard to move everything new over 
to something more stable. Specifically, my group runs about 2 dozen core 
boxes, mostly unique (little duplication between 
services/configurations, at most 3 sets of primary/secondary boxes). At 
the moment (all inherited), most of the "backups" are done as manual 
tarballs of /etc and other relevant directories. So, since I'm going to 
be rebuilding a few of them, it's time to rollout some realistic plan 
for recovery. As we've mostly standardized on one physical platform 
(SunFire x4100), binary compatibility isn't a big concern, and we have a 
few spare identical machines.

What are you guys doing for installation/bare metal recovery? The idea 
of config management (Puppet/Cfengine/etc.) has already been vetoed by 
management given the small number of boxes we run (though our services 
support ~60k users).

So far, the two running theories are:
1) Use Kickstart for installation (and recovery for new boxes) and 
backup configs/non-package files (Bacula).
2) Use a gold master Kickstart file as a base, hand-configure machines 
past that, backup *everything* in Bacula and make use of Bacula's Bare 
Metal Recovery method. Disk capacity is cheap, and we just purchased 6TB 
for backups.

Given the number of users we support, our main criterion is time from 
failure to recovery on replacement hardware, with ease of automation (at 
least easy enough that anyone can do it with a simple HowTo document) as 
a close second.

Any other ideas to throw out there that I may have missed?

Thanks for any suggestions,
Jason

PS - We're a 100% OSS shop with a pretty low software budget, so 
anything that's not both Free and free is out of the running.