Something that I forgot to address earlier...
pretty much identical. A DEC Alpha, x86-32, x86-64, Sun Sparc, they are all identical with slight hardware changes. I can push out a config to all of these machines and be reasonably confident that it will work on all of them.
How many non-x86-32 machines do you admin? For interests sake, does anyone here actually have a "production Linux server" (other than "Look, our flat has a SGI coffee table") running something other than x86? I'd like to see the BSD tiers applied to Debian, so it goes stable on i386, and then the developers for the 1% of platforms can sort it out without holding back the other 99%. By admining a non-x86 machine you've already demonstrated what hopefully equates to "more than a clue"...
I once saw the command that Debian should rename their "versions" from:
Stable -> Enterprise Testing -> Desktop Unstable -> Developer
Enterprise doesn't just imply "stability" (More often than not, it doesn't imply that at all! Enterprises rarely ever have one machine that does anything; failure of any part is fine as long as the cluster stays intact, how else do you think you get Windows uptimes longer than security patch releases?). Enterprise implies, to me, support for enterprise class hardware and situations, working with enterprise vendors. It would mean you need a relative amount of commerciality about your distribution, or at the very least, people who are working with companies that sell hardware and software. While as a sysadmin who would rather run Debian, it bugs me that commercial Linux-friendly hardware vendors say "SuSE & Red Hat" these days, I see the point. For a Linux vendor to certify their product on Debian, even on Stable, means 13 different platforms, for 1% of the admins..etc I'm told that commercial products on Red Hat don't even get support if you've recompiled the kernel; that doesn't mean you can't do it, but that if you want to go back to the company with an issue, they expect you to at least make sure that the issue isn't caused by your kernel changes. Testing, we go back to what Daniel said in a recent email; packages automatically go from unstable to testing in 10 days with no bugs. This means testing isn't much more than an old unstable, with just as many chances for there to be a random bug which your desktop user might find after a month. I feel bad about trusting servers to Fedora; we all moved away from Red Hat to Debian once upon a time anyway (hey, it was sold to our resident RHCE!), but the dissent is getting louder. I'd love to use White Box Enterprise, the SRPM-recompile of RHEL3, but then you get the same "it's as unsupported as Debian is" issue. Craig