Subject: Re: RFC: root on raidframe howto
To: Martti Kuparinen <martti.kuparinen@iki.fi>
From: Greg A. Woods <woods@weird.com>
List: port-i386
Date: 08/07/2003 15:22:36
[ On Thursday, August 7, 2003 at 09:24:32 (+0300), Martti Kuparinen wrote: ]
> Subject: Re: RFC: root on raidframe howto
>
> I have updated the draft, please take a look at
> 
> http://www.piuha.net/~martti/tmp/raid

One of the problems with all these types of guides is that they get
bogged down in the details and can be so over-burdened with information
specific to one particular example configuration that even experts can
have a great deal of trouble following them and translating them into a
generic description from which other system specific configurations can
be concocted.

What we really need is an overview, or a "Theory of Operation".  I.e. a
document that outlines the goals and rationale of what's being
attempted, and which lays down the basic rules of thumb for designing a
specific configuration.

Once you've got that then a walk-through of a specific example, perhaps
one for each unique type of platform, relating back to how each specific
parameter reflects the rules of thumb, will help illuminate all the
details which might be missed otherwise (e.g. specific command-line
options, examples showing command output, etc.)

I think there's still a massive gaping hole in the NetBSD documentation
outlining the "theory of operation" for specific system boot processes,
with (of the one's I've read) the i386 docs (and mbr(8) in particular)
being most detailed but all-in-all still sadly lacking a really
comprehensive outline all in one place of how it works, how different
devices, disk geometries, etc., are handled, and what this all implies
to how one must partition one's disks and how one writes software to
take over from whatever the system firmware can do.  David Laight did
recently post a much better (more lucid and more detailed) description
of both the old and new i386 boot procedures.  It doesn't help either
that raid(4) doesn't describe exactly where the component labels are
stored and how much space is reserved for them and how this is all
reflected in both the outer hard disk's partition label as well as in
the virtual partition label constructed for the virtual RAID device.

-- 
						Greg A. Woods

+1 416 218-0098                  VE3TCP            RoboHack <woods@robohack.ca>
Planix, Inc. <woods@planix.com>          Secrets of the Weird <woods@weird.com>