Very Computer

by **thatseattle..** » Tue, 05 Jul 2005 04:15:37

A client has a 3-disk SCSI RAID array attached to an old Mylex DAC960
controller. About six months ago, they lost one of the disks, which was
fine, except there was no alarm and they didn't notice... Last
Wednesday at 9:03AM, they lost a second disk - and believe me, *that*
got their attention. :}

They replaced the two failed drives (older 10K IBM models) with newer
Seagates in the same drive caddies, leaving the remaining working IBM
drive in place (don't ask...). All drives are 18GB SCSI. After the
array was rebuild and yours truly resurrected things from backup tapes,
we noticed that the system was slow and there were huge "gaps" in
activity where for 5-8 seconds seemingly nothing was getting processed
and no disk activity was happening. Long story short we found that the
SCSI bus was undergoing a full reset every minute or so (see dump
below).

So here's the question: assuming termination and cables are all fine
(as the old drive caddies were reused in their same positions, and they
look OK), could the fact that we've got two different types of drives
in the same array be causing the resets? I've heard tale tell the
DAC960 is pretty finicky about the types of drives it likes to talk
to. And I've never tried to do hardware RAID with dissimilar drives (of
different generations, no less) before.

Thoughts/ideas welcome.

-jr-

$ more /proc/rd/c0/current_status
***** DAC960 RAID Driver Version 2.4.11 of 11 October 2001 *****
Configuring Mylex DAC960PG PCI RAID Controller
Firmware Version: 4.06-0-08, Channels: 2, Memory Size: 8MB
PCI Bus: 0, Device: 11, Function: 1, I/O Address: Unassigned
PCI Address: 0xF4104000 mapped at 0xD0800000, IRQ Channel: 18
Controller Queue Depth: 64, Maximum Blocks per Command: 128
Driver Queue Depth: 63, Scatter/Gather Limit: 33 of 33 Segments
Stripe Size: 64KB, Segment Size: 8KB, BIOS Geometry: 255/63
Physical Devices:
0:0 Vendor: SEAGATE Model: ST318406LW Revision: 0108
Serial Number: 3FE0HNRH00007222F4Q5
Disk Status: Online, 35842048 blocks, 730 resets
0:1 Vendor: IBM Model: DDYS-T18350N Revision: S80D
Serial Number: VEL7C714
Disk Status: Online, 35842048 blocks, 730 resets
0:2 Vendor: SEAGATE Model: ST318406LW Revision: 0108
Serial Number: 3FE00FL000007222A9KS
Disk Status: Online, 35842048 blocks, 730 resets
Logical Drives:
/dev/rd/c0d0: RAID-5, Online, 35844096 blocks, Write Thru
/dev/rd/c0d1: RAID-5, Online, 35840000 blocks, Write Thru

by **kermi** » Tue, 05 Jul 2005 05:48:55

> A client has a 3-disk SCSI RAID array attached to an old Mylex DAC960
> controller. About six months ago, they lost one of the disks, which was
> fine, except there was no alarm and they didn't notice... Last
> Wednesday at 9:03AM, they lost a second disk - and believe me, *that*
> got their attention. :}

> They replaced the two failed drives (older 10K IBM models) with newer
> Seagates in the same drive caddies, leaving the remaining working IBM
> drive in place (don't ask...). All drives are 18GB SCSI. After the
> array was rebuild and yours truly resurrected things from backup tapes,
> we noticed that the system was slow and there were huge "gaps" in
> activity where for 5-8 seconds seemingly nothing was getting processed
> and no disk activity was happening. Long story short we found that the
> SCSI bus was undergoing a full reset every minute or so (see dump
> below).

> So here's the question: assuming termination and cables are all fine
> (as the old drive caddies were reused in their same positions, and they
> look OK), could the fact that we've got two different types of drives
> in the same array be causing the resets?

It is not that much different types of drives; rather it is just different
drives. We have a problem when a customer could not (or did not want) to
wait for delivery of qualified drives and they bought the same drive type
from another supplier; the only difference was FW. The result was - every
time they tried to replace these drives online they got corrupted
database :) Another customer got the whole system down (all drives "defect"
and LUNs offline) attempting to add drives online in similar case.

As long as drives are qualified for a given controller it should be possible
to mix them; unfortunately you never (rare) get open information about
compatibility.

You may try to remove IBM and see if running with Seagate still has the same
issue. If yes - it is just that 960 has some problems with them.

Of course you cannot rule out bad drive - one of the new ones.

=arvi=

- Show quoted text -

Quote:> I've heard tale tell the
> DAC960 is pretty finicky about the types of drives it likes to talk
> to. And I've never tried to do hardware RAID with dissimilar drives (of
> different generations, no less) before.

> Thoughts/ideas welcome.

> -jr-

> $ more /proc/rd/c0/current_status
> ***** DAC960 RAID Driver Version 2.4.11 of 11 October 2001 *****
> Configuring Mylex DAC960PG PCI RAID Controller
> Firmware Version: 4.06-0-08, Channels: 2, Memory Size: 8MB
> PCI Bus: 0, Device: 11, Function: 1, I/O Address: Unassigned
> PCI Address: 0xF4104000 mapped at 0xD0800000, IRQ Channel: 18
> Controller Queue Depth: 64, Maximum Blocks per Command: 128
> Driver Queue Depth: 63, Scatter/Gather Limit: 33 of 33 Segments
> Stripe Size: 64KB, Segment Size: 8KB, BIOS Geometry: 255/63
> Physical Devices:
> 0:0 Vendor: SEAGATE Model: ST318406LW Revision: 0108
> Serial Number: 3FE0HNRH00007222F4Q5
> Disk Status: Online, 35842048 blocks, 730 resets
> 0:1 Vendor: IBM Model: DDYS-T18350N Revision: S80D
> Serial Number: VEL7C714
> Disk Status: Online, 35842048 blocks, 730 resets
> 0:2 Vendor: SEAGATE Model: ST318406LW Revision: 0108
> Serial Number: 3FE00FL000007222A9KS
> Disk Status: Online, 35842048 blocks, 730 resets
> Logical Drives:
> /dev/rd/c0d0: RAID-5, Online, 35844096 blocks, Write Thru
> /dev/rd/c0d1: RAID-5, Online, 35840000 blocks, Write Thru

by **Ulrich.Teich..** » Wed, 06 Jul 2005 03:59:47

>A client has a 3-disk SCSI RAID array attached to an old Mylex DAC960
>controller. About six months ago, they lost one of the disks, which was
>fine, except there was no alarm and they didn't notice... Last
>Wednesday at 9:03AM, they lost a second disk - and believe me, *that*
>got their attention. :}
>They replaced the two failed drives (older 10K IBM models) with newer
>Seagates in the same drive caddies, leaving the remaining working IBM
>drive in place (don't ask...). All drives are 18GB SCSI. After the
>array was rebuild and yours truly resurrected things from backup tapes,
>we noticed that the system was slow and there were huge "gaps" in
>activity where for 5-8 seconds seemingly nothing was getting processed
>and no disk activity was happening. Long story short we found that the
>SCSI bus was undergoing a full reset every minute or so (see dump
>below).

That's not a dump of a SCSI bus reset. This would reveal what drive caused
the reset (reset on behalf of....).

Quote:>So here's the question: assuming termination and cables are all fine
>(as the old drive caddies were reused in their same positions, and they
>look OK), could the fact that we've got two different types of drives
>in the same array be causing the resets? I've heard tale tell the
>DAC960 is pretty finicky about the types of drives it likes to talk
>to. And I've never tried to do hardware RAID with dissimilar drives (of
>different generations, no less) before.

Well, all hardware RAID controlers are picky about drive FW and drives
of different types. It would be interesting to see what really collided
in the log.

[del]

Quote:> Logical Drives:
> /dev/rd/c0d0: RAID-5, Online, 35844096 blocks, Write Thru
> /dev/rd/c0d1: RAID-5, Online, 35840000 blocks, Write Thru

2 RAID 5's? On 3 drives? Are you sure that's right?

HTH,
Uli
--

Stormweg 24 |listening to: Suicide Drive (The Deep Eynde)
24539 Neumuenster, Germany|Public * (Interpol) Cauchemar (Opration S)

Very Computer

SCSI bus resets w/Linux 2.4.x and Mylex DAC960 RAID

SCSI bus resets w/Linux 2.4.x and Mylex DAC960 RAID

SCSI bus resets w/Linux 2.4.x and Mylex DAC960 RAID

SCSI bus resets w/Linux 2.4.x and Mylex DAC960 RAID