Marco -
I just built my system yesterday using stock RedHat 6.0. The stock kernel
apparently has raid support built in as modules, so you can't raid your
root directory immediately, and you need to "raidstart -a" in order to
access the devices after a reboot, but it has enough to allow you to
build them.
Basically, I built a smallish installation (no X-windows) on a partition
I plan to use as a backup boot environment. Next I created, using this
stock installation, all the mirrored partitions I wanted for a production
kernel (mirrored /, /tmp, /usr, /var, /home, swap - not /boot). After
syncronization (which takes awhile) I made efs file systems on the mirrored
areas, and mounted my new production tree on /mnt. This resulted in
/ (my backup system), /mnt, /mnt/tmp, /mnt/usr, /mnt/var, /mnt/home, and
/mnt/boot. Using a "cd /; find . -xdev | cpio -pm /mnt" I copied my
smallish installation onto the new areas. This resulted in an ALMOST
usable environment.
Next I "chroot /mnt; mount -o remount -a". The remount freshens up /etc/mnt
abit - not perfect, but it was good enough to keep things like "df" from
complaining.
In the chroot environment I went ahead and downloaded the 2.2.11 kernel,
raid0145 patch, and matching raidtools. I think you need 2.2.10 to get
kernel autodetect capability - in any case, I thought it was wise to
get the latest integrated set of code (all were dated 08/24/99 which
seem pretty current <grin>). Built a new kernel with raid support
included (NOT as modules), tweaked /etc/fstab to reference the md devices
instead of partitions, added new entries in /etc/lilo.conf to do the
same, ran lilo and rebooted.
Thats pretty much it. After coming up and testing a few things I read
the current HOWTO that came with the 2.2.11 package. That document
indicated there were problems with using a raid device for swapping due
to some memory allocation race conditions. I havn't had a problem
yet, but suspect I should back off there for now. Don't forget to
set you mirrored partition types from 83 (linux native) to fd via
fdisk in order for the kernel to autodetect them.
For what its worth:
I'm using a Abit BP6 Dual Celeron motherboard with 256mb of memory,
(2) Maxtor 10.4gb IDE disk - one on each of the motherboards IDE
controllers setup as master devices, and an IDE CDROM setup as a
slave on the second motherboard IDE controller. My goal was to
create a totally mirrored system environment to protect myself from
data loss due to a disk outage. The partition layout is:
Primary Secondary
1) /boot 16MB DOS partition for motherboard firmware updates
2) / 704MB / Backup root, Primary is pure Redhat, Secondary
is where I started customizing.
3) swap 64MB swap - non-mirrored swap for use by backup roots
4) Extended partition
5) / 128MB / MD0 - mirrored production root
6) /tmp 128MB /tmp MD1 - mirrored production /tmp
7) swap 256MB swap MD2 - mirrored production swap
8) /usr 1536MB /usr MD3 - mirrored production /usr
9) /var 1024MB /var MD4 - mirrored production /var
10) /home ~6GB /home MD5 - rest of disk, mirrored /home
Currently I have 3 problems:
1 - /boot is exposed. If I lose the first drive (hda), I'm not bootable.
2 - Based on the latest HOWTO, I should be using partition 3 (non-mirrored
swap). If I lose a disk, I'll go down. If its disk 2, I can come
back up without data lose though.
3 - I want to do a full install with X-windows via the install tool, but the
CD kernel doesn't recognize the MD areas. Looks like I'll have to
piece the rest of the environment together with RPM.
Hope that helps. I read e-mail frequently so feel free to ask for
clarification.
Kevin C.
Marco Shaw write:
>
> Is there a way to search this list? Or does the stock RedHat 6.0 (2.2.5
> kernel) have the necessary patches for using s/w raid?
>
> Thanks,
> Marco
>
--
Kevin Carpenter
Kevin's Home Page: http://www.monrou.com/kevinc
(Expressing his comments from home in St. Louis, where this message originated)
