|
I earlier posted a message describing my difficulties with
booting root Raid0. From the start I was uneasy about the vulnerabilty of my
suggested setup, so I later decided to get another 1Gig disk and go for Raid5.
This seems to be the most popular mode and therefore well supported by
experience from other users on this mailing list. It proved to be pretty much
plain sailing, so I thought I'd share my experience with others who may be not
as lucky as I was!
Firstly I read Jakob Oestergard's latest HOWTO, who's clear
and concise description gave me confidence to proceed.
I hung 3 1Gig SCSI Quantum Fireball drives on an
Adaptec 1542 controller and fdisk'ed all with 33 meg swap partitions (however -
more about that later); for the remaining space on the others, I fdisk'ed one
Linux native and the 2 others Linux raid.
I installed Slackware 7 on the disk with the Linux native
partition, patched the source with the latest raid patches (ignoring 2 rejects),
and recompiled a kernel with Raid5 support built-in. I then created a raidtab
for the two partitions tagged as Raid, installed Lilo and
rebooted. Whilst rebooting I noted that the raid partition /dev/md0 was
detected and I was subseqently able to mount it on /mnt.
Next I copied all the contents of the partition I'd installed
Linux on, onto the Raid partition mounted on /mnt, entered
the original installation partition as a failed-disk in the
/mnt/etc/raidtab and ammended the /mnt/etc/fstab to show /dev/md0
mounted as root. I also entered the other swap partitions in the
/mnt/etc/fstab.
I then tried rebooting, but had Lilo report that it didn't
recognise device 0x900 (or similar). At this point I decided to abandon
Lilo and simply boot from a kernel image on a floppy (as a previous
posting advised). So I used rdev to tell the kernel image that /dev/md0 was root
- rdev /zImage /dev/md0 - and wrote the new image to a floppy - dd if=/zImage
of=/dev/fd0. On rebooting off the floppy, everything came up like a charm, with
the new /dev/md0 mounted as / running in degraded mode (on 2 cylinders!).
I took out the failed-disk entry in the raidtab, fdisk'ed the
unused partition as Linux Raid and hotadded it to /dev/md0 and was rewarded with
the sound of the disks working in chorus rebuilding the raid
partition.
It's all worked so well that I'm resetting it all up with swap
on a raid5 partition (thanks - Bruno Prior!) and a spare disk. I'm also worried
about the performance of the ISA AHA1542 controller, so I'm going to rob another
of my machines of its PCI NCR53C810 and use that instead. (I've also considered
fitting an AHA1540 I have in addition to the AHA1542 and splitting the disks
between them - I'm not sure what's best. Comments welcomed!)
I hope the forgoing - whilst repeating much of what's in Jakob
Oestergard's HOWTO, is helpfull!
Regards: Jim Ford
|
