I am trying to build a File Server with the current stable release of Debian
and the 2.6 kernel. I have 2 20 GB IDE drives to boot from and 9 45GB Ultra
Wide SCSI drives on 2 Adaptec 2940UW controllers.

I want to build a Debian file server with a software RAID array of SCSI
drives and a software RAID array of IDE drives. The IDEs are for booting and
OS partitions while the SCSI array will be for users files. The IDE array
will be RAID1 and the SCSI array will be RAID 5.

I've tried several installs and I always run into something that prevents
this from working. I have no problem with the boot RAID 1 array, but the
SCSI array is just refusing to work.

I managed to get 2 RAID 1 partitions for / and /home on the IDE drives. The
system boots up happy and all. My problem is with the 9 45 GB drives. I have
tried:

Installing the OS with out the SCSI drives attached. Once the OS was all
updated and loaded I tried to use mdadm to add the drives. mdadm would not
create /dev/md2. Out of desperation I used mknod to create /dev/md2 and
built a RAID 5 array with a etx3 filesystem, but upon reboot the device
disappeared. Making the device and mounting the fs failed.

I then reinstalled the whole thing and built all 3 RAID partition in the
installer. I now have a /dev/md2 device that persists, but the filesystem
does not. I can use mdadm to build the array and mkfs to make the
filesystem, but on reboot the automount fails because it tried before the
SCSI drives are found. Trying to mount by hand after booting fails to find a
valid OS.

Do I have to find 2 small SCSI drive just to boot to so the SCSI drives are
found at the necessary time? This is maddening! I've got a very similar Red
Hat 9 system working and it did not give me this much grief.

Any suggestion is appretiated.

Thanks,

Mark C

[ Post a follow-up to this message ]

On Mon, 2006-09-04 at 21:08 +0000, Mark C wrote:
> I am trying to build a File Server with the current stable release of Debi
an
> and the 2.6 kernel. I have 2 20 GB IDE drives to boot from and 9 45GB Ultr
a
> Wide SCSI drives on 2 Adaptec 2940UW controllers.
>
> I want to build a Debian file server with a software RAID array of SCSI
> drives and a software RAID array of IDE drives. The IDEs are for booting a
nd
> OS partitions while the SCSI array will be for users files. The IDE array
> will be RAID1 and the SCSI array will be RAID 5.
>
> I've tried several installs and I always run into something that prevents
> this from working. I have no problem with the boot RAID 1 array, but the
> SCSI array is just refusing to work.
>
> I managed to get 2 RAID 1 partitions for / and /home on the IDE drives. Th
e
> system boots up happy and all. My problem is with the 9 45 GB drives. I ha
ve
> tried:
>
> Installing the OS with out the SCSI drives attached. Once the OS was all
> updated and loaded I tried to use mdadm to add the drives. mdadm would not
> create /dev/md2. Out of desperation I used mknod to create /dev/md2 and
> built a RAID 5 array with a etx3 filesystem, but upon reboot the device
> disappeared. Making the device and mounting the fs failed.
>
> I then reinstalled the whole thing and built all 3 RAID partition in the
> installer. I now have a /dev/md2 device that persists, but the filesystem
> does not. I can use mdadm to build the array and mkfs to make the
> filesystem, but on reboot the automount fails because it tried before the
> SCSI drives are found. Trying to mount by hand after booting fails to find
 a
> valid OS.
>
> Do I have to find 2 small SCSI drive just to boot to so the SCSI drives ar
e
> found at the necessary time? This is maddening! I've got a very similar Re
d
> Hat 9 system working and it did not give me this much grief.
>
> Any suggestion is appretiated.
>
Using software RAID is flawed.
Get a hardware SCSI card (dropping the IDE would also be good).

--
Jeremy Boden

[ Post a follow-up to this message ]

> Using software RAID is flawed.
> Get a hardware SCSI card (dropping the IDE would also be good).
>
> --
> Jeremy Boden
>
>
Agreed, but the budget does not permit this option.  Yes this is less than
perfect, but it should be able to be done.

[ Post a follow-up to this message ]

Jeremy Boden wrote:

> Using software RAID is flawed.
> Get a hardware SCSI card (dropping the IDE would also be good).

Hmmmmm - I am actually not sure I would agree on that.  I have several
times found myself losing data on a hardware raid system that failed in
some spectacular way that was impossible to fix with the available tools
(the mere mention of the word "Promise" doubles my blood pressure).  On
the other hand - I have yet to lose any data EVER using Linux software
raid-1.  The great thing as I see it is that using for example lvm or
ext3 on top of raid-1 each of the disks is a valid stand-alone disk in
case the raid screws up - not that I have ever seen that happen either.
So in case of failure I can move one of the disks to another system and
always recover the files even without ever getting the raid up again.
Even performance is reasonably impressive (read: faster than a single
disk for reads and about the same for writes).

I admit I lost some data once on a Linux software raid-5, so since then
I've only used raid-1.  I just like the fact that you can read the disk
even without getting raid up and running.

--
Lars

[ Post a follow-up to this message ]

"Lars Boegild Thomsen" <lth@cow.dk> wrote in message
news:1157438449.671124@x330.netcompartner.com...
> Jeremy Boden wrote:
> 
>
> Hmmmmm - I am actually not sure I would agree on that.  I have several
> times found myself losing data on a hardware raid system that failed in
> some spectacular way that was impossible to fix with the available tools
> (the mere mention of the word "Promise" doubles my blood pressure).  On
> the other hand - I have yet to lose any data EVER using Linux software
> raid-1.  The great thing as I see it is that using for example lvm or
> ext3 on top of raid-1 each of the disks is a valid stand-alone disk in
> case the raid screws up - not that I have ever seen that happen either.
> So in case of failure I can move one of the disks to another system and
> always recover the files even without ever getting the raid up again.
> Even performance is reasonably impressive (read: faster than a single
> disk for reads and about the same for writes).
>
> I admit I lost some data once on a Linux software raid-5, so since then
> I've only used raid-1.  I just like the fact that you can read the disk
> even without getting raid up and running.
>
> --
> Lars

Hmmm.  I have enough of these 45GB drives around that I could do a whole
bunch of RAID1 arrays too, but first I have to get past the initial problem.
I would need to use lvm to span the disks into a single partition which I
suspect opens me up to a bit more risk, but less than an insane array.  The
only RAID data loss I ever had was a RAID5 array with a hardware controller.
Software RAID5 has been very good to me under Red Hat 9.

Mark

[ Post a follow-up to this message ]

Mark C wrote:

> Hmmm. _I have enough of these 45GB drives around that I could do a whole
> bunch of RAID1 arrays too, but first I have to get past the initial
> problem.

Well - I haven't installed stable for a while, but "Etch/testing" installs
clean with lvm on top of raid-1 - just make sure you got a /boot
partition that is not part of the lvm (I usually leave / off the lvm
too).  Using Debian/Testing installer, my usual method is:

1st partition: 512 MB Physical partition for raid-1
2nd partition: 512 MB SWAP
3rd partition: the rest - physical partition for raid-1

Once all drives has been partitioned I create the arrays needed - first
the boot one (which becomes md0) next the ones used for lvm.

Once raid's have been created I assign the first 512MB one as a normal
partition with ext3 mounted on "/".  The other raids I mark as "physical
volume for logical volume manager".

Then I configure lvm - in this case one volume group that includes all the
raids not used for root partition.

That's about it.  This have worked flawlessly in all the testing's I've
installed over the past - say 2 years at least.

Only thing that needs a bit of manual work is to get GRUB installed on
more than one disk so that the system can boot even if the first disk
trashes itself.

--
Lars

[ Post a follow-up to this message ]