[large cc lists trimmed, i think everyone must be on one of the two
lists]

On Wed 2 Oct, 1996, matthew green <mrg@eterna.com.au> wrote:
>i also like the model of ODS (as under solaris 2):
>	- a metadevice acts like a normal disk partition
>	- a metadevice can be composed of any number of real partitions
>	  or metadevices, either concatenated or striped, or mirrored.

I (having been playing with IRIX 6.2 a lot recently) like IRIX's
XFS/XLV setup. It's pretty similar to Sun's ODS/SDS, but you don't
have to change md.conf files and reinit the metadevice when you
move disks around controllers and change their ids.

At boot time, an 'assemble' operation checks all disks on the
system, and finds details of which logical volume they belong to,
if any. (Someone's already said that they think it useful to be able
to easily reconfigure your disks, and I concur - when you're dealing
with large numbers of disks and large numbers of controllers you
really don't want to have to be writing down numbers on pieces of
paper.)

Each logical volume has a name which is used when referring to the
logical volume device (/dev/{r,}dsk/xlv/volumename), and consists
of log, data, and real-time data sub-volumes. Each sub-volume
consists of a number of volume elements (each being a disk partition),
which can be concatenated or striped together.  A sub-volume can
contain up to 4 plexes (SGI's term for mirrors), each being up to
128 volume elements.

Of course, SGI have the advantage of not having tried to build the
system around UFS/FFS, so you grow the filesystem whilst it is
mounted (the instructions specifically say you *must* mount before
growing!), and it's far better in crash situations.

<URL:http://www.sgi.com/Technology/TechPubs/dynaweb_bin/0620/bin/nph-dynaweb.cgi/SGI_Admin/IA_DiskFiles/1.toc> 

for some pointers to XFS/XLV concepts.

-- jrg.