[lug] Adding 2nd SCSI card
jkarns at csd.net
Wed Oct 9 20:13:55 MDT 2002
On Wed, 9 Oct 2002, Gary Hodges said:
> I'll try and keep it short and sweet.
> We have:
> Tyan dual Athlon w/two 2000+ installed
> Adaptec 29160 card installed in a 64 bit slot
> RH v7.3 2.4.18-10smp kernel
> The machine seems to be very stable. I ran kernel compiles over a
> weekend (~2000) without any trouble. A few weeks ago we installed an
> IDE RAID tower, and about one week ago we finally got around to
IDE RAID w/ SCSI controller?
> configuring it. I wasn't intimately involved in this step so bear with
> me. There were problems encountered when mkfs was run and when we tried
> to make a directory on the tower. It is configured as RAID5. The
> machine would be brought to its knees when mkfs of mkdir was executed.
> Mouse wouldn't move, etc. OK, I'm trying to keep it short and sweet.
Ok, I'll belabor the obvious. I'm not familiar with Tyan mobos in
particular, but with the mobos I've been using lately, it's been a bit of
a trick getting cards situated in such a way to avoid IRQ sharing. I
assume that newer kernels handle IRQ sharing more eloquently than in the
past, but my preference is to avoid it if I can.
Many of these mobos are designed so that certain slots share the same IRQ.
In the case of having the two cards installed simultaneously, this type of
situation could cause a problem, if the cards are using the same IRQ.
> In an attempt to figure out if the problem was RAID tower or
> machine/card related, we thought we would pick up another SCSI adapter
> (LSI Logic U160), plug the tower into that and see if we got the same
> errors. With the LSI card in the 1st slot and the Adaptec in the 2nd
> slot I saw both SCSI BIOS' during the pre-boot stuff, but I couldn't
> get the machine to boot up.
Of course, the above scenario doesn't explain the LSI card failure when
the 29160 is not present. I have run into issues with kernel versions
though. Specifically, I recently upgraded to an MSI Sis 745 mobo, and
can't get the module to load for my Tekram SCSI controller when I boot
from a SuSE 7.3 install CD. It does load ok when booting from a SuSE 7.1
install CD though - and both CD's use a 2.4 kernel, although not the same
An interesting (to me anyway) side note is that according to hdparm, the
i/o bw increased 66% from 15 MB/s to 25 MB/s - and I had thought that my
bottleneck was the controller + disk.
And finally FWIW, mobo timing issues can play a part here too. After
upgrading to the faster mobo, my 3Com 3c905 NIC no longer works - module
loads with no errors, but file x-fer attempts yield massively corrupted
files - I replaced it with a garden variety RTL8139 which avoids the
corruption problems. Perhaps playing with bus timings would give some
John Karns jkarns at csd.net
More information about the LUG