On Wed, 16 Jun 1999, Bruno Prior wrote:
> > Anyway, I just did a clean install of RedHat 6.0 on a new server, and
> > attempted a software RAID setup on it. Everything seems to work fine, the
> > /dev/md0, 1 & 2 mount and read/write properly, but the autodetection fails
> > on startup.
> [Snipped]
> > kmod: failed to exec /sbin/modprobe -s -k md-personality-3, errno = 2
>
> Don't you just love RedHat? In two successive releases they have managed to
> create confusion that dominates the list for a while. [...]
Yes, the initrd image was buggy for RAID - this has been fixed meanwhile.
I'm actually happy that it's not _that_ easy to get a new RAID install
done, it's still too easy to mess data during a not careful enough RAID
upgrade. Red Hat actually payed for getting the latest ~1 year of RAID
improvements done - i'm rather sure we'd still use mdtools and user-space
ckdisk if i had another job ...
back on topic: a 2.2.10 and 2.0.37 RAID release will happen Soon, the
resync bug has been finally fixed. There is one more bug i'd like to see
tracked down - though it's probably not a RAID related one. The new
release will also include Martin's failed-disk changes.
future plans/directions for the 1.0 RAID code include:
- integrating the PIII-checksum changes into the RAID code - it
greatly improves RAID5 performance on PIIIs. (this was developed
by Doug Ledford, Red Hat)
- supporting many disks per array (now the limit is 12). I
actually have this mostly coded, the code supports up to ~250
disks, should be enough for a few years. There are migration
issues to be sorted out.
- SMP-threaded RAID code - this is partly done already.
- [maybe integrating the RAID code with the 2.3 pagecache - this
is a rather big project but will finally get rid of some of the
copying and memory overhead within the RAID5 code.]
-- mingo
