Subject: Re: Rototil of sysinst partitioning code
To: NetBSD-current Discussion List <current-users@netbsd.org>
From: Greywolf <greywolf@starwolf.com>
List: current-users
Date: 06/05/2003 13:51:46
Thus spake Greg A. Woods ("GAW> ") sometime Today...

GAW> > I have my POV, you have yours, both at extreme ends of the scale.
GAW>
GAW> The problem is you keep justifying your P.O.V. with baseless and/or
GAW> incorrect and/or disparaging arguments.

Beg pardon?  You see them as baseless because they don't apply to your
own sensibilities; thus, also, do you see them as being incorrect, because,
since they don't apply to your own sensibilities, "they shouldn't appear
sane to anyone with any ounce of sense."

As far as disparaging, well, if I disparage something, in my own eyes,
that something has truly earned it.  I try hard not to disparage people,
and if I have affronted or disparaged you, I apologise.  I will not
apologise, however, for being in vehement disagreement with you :).

GAW> Whenever you give good technical arguments they don't apply to /usr
GAW> specifically -- and in fact they are usually arguments for including the
GAW> base /usr stuff on the root filesystem!

You know, this almost makes sense, but I still don't want it to be assumed
that /usr will be available if booted to single-user.  I've covered my
reasons, but if you really want me to (and I don't think that everyone
does), I will do so again.

GAW> > But to put everything under / is not a bright move.
GAW>
GAW> We agree on this 101%, but "this" != "putting /usr on root"!

I beg to differ only slightly:  It's part of it.  It's just becoming
less of an issue.  The other stuff is more important, so I'll gloss
over this facet for the time being.

I apologise if this seems to be a senseless discussion, but to me, it
warrants airing.  The ability to configure things explicitly at install
time, and to have reasonable defaults offered, is invaluable.  I have
used sysinst a precious few times (usually during a default install
for i386), and I got very frustrated at not being able to split my
partitions among multiple spindles/controllers for a more balanced
access.

Food for thought, and I'm not sure where to put this right now,
so I'll just toss it out here -- if anyone wants to respond to this,
it should probably be retitled in the Subject: line.

	- interleaved or concatenated dump devices (so that the kernel
	  core can be dumped across multiple partitions).  This would
	  save us from needing to allocate hw.physmem bytes worth of
	  space on a single disk if we had multiple disks (ccd could,
	  of course, help this);

	- compression-on-dump (so that we don't use up quite as much space,
	  except in worst case);

	- dumping only relevant kernel memory (so that we don't use up
	  quite as much space, except in worst case).

				--*greywolf;
--
"Windows/NT - From the people who brought you EDLIN".