I've got a system I need to slow down disk access on it would seem. The
syerver, a dual 1.5Ghz Athalon has two 3ware IDE controllers, 8 disks
each for a total of sixteen 180 Gig disks. This system is laid out in
four RAID5 arrays that it shares out via NFS. kernel 2.4.19-ac4, ext3
file systems. I've got 7 of these.
This works great and provides a LOT of cheap disk that we use for
staging backups before cloning off to tape.
The problem is that when the array's get pounded HARD, such as when
legato is cleaning it's file devices (nsrstage -C) the machine will lock
up and spew errors to my console:
3w-xxxx: scsi1: Command failed: status = 0xc7, flags = 0x1b, unit #7.
3w-xxxx: scsi1: Command failed: status = 0xc7, flags = 0x1b, unit #7.
3w-xxxx: scsi1: Command failed: status = 0xc7, flags = 0x1b, unit #7.
3w-xxxx: scsi1: Command failed: status = 0xc7, flags = 0x1b, unit #7.
3w-xxxx: scsi1: Command failed: status = 0xc7, flags = 0x1b, unit #7.
3w-xxxx: scsi1: Command failed: status = 0xc7, flags = 0x1b, unit #7.
3w-xxxx: scsi1: Command failed: status = 0xc7, flags = 0x1b, unit #7.
3w-xxxx: scsi1: Command failed: status = 0xc7, flags = 0x1b, unit #7.
3w-xxxx: scsi1: Command failed: status = 0xc7, flags = 0x1b, unit #7.
3w-xxxx: scsi1: AEN: WARNING: ATA port timeout: Port #7.
3w-xxxx: scsi1: AEN: WARNING: ATA port timeout: Port #7.
3w-xxxx: scsi1: AEN: WARNING: ATA port timeout: Port #7.
3w-xxxx: scsi1: AEN: WARNING: ATA port timeout: Port #7.
3w-xxxx: scsi1: AEN: WARNING: ATA port timeout: Port #7.
3w-xxxx: scsi1: AEN: WARNING: ATA port timeout: Port #7.
3w-xxxx: scsi1: AEN: WARNING: ATA port timeout: Port #7.
3w-xxxx: scsi1: AEN: WARNING: ATA port timeout: Port #7.
This requires a hard reboot. If I use the magic keys and reset it then
it goes down but doesn't come back up properly, usually dieing around
the LILO area, I believe the array or disks are left in an odd state.
Is there a way to actually slow down the disk access/read/write other
than re-making the filesystems? On the other systems the chunk size is
at either 32K or 128K. On this one in particular the chunk size is
1024K which was determined by running a number of tests (bonnie, nfs
reads/writes) and was found to be about the fastest for our money's
worth. The disk hangups didn't appear until just recently. If there's
no other choise I can leave the disks as read only until the data rolls
off then remake them in 32 or 64K.
Any other thoughts?
Robert
:wq!
---------------------------------------------------------------------------
Robert L. Harris | PGP Key ID: E344DA3B
@ x-hkp://pgp.mit.edu
DISCLAIMER:
These are MY OPINIONS ALONE. I speak for no-one else.
Diagnosis: witzelsucht
IPv6 = [email protected] http://ipv6.rdlg.net
IPv4 = [email protected] http://www.rdlg.net
I have nearly an identical system as yours. Dual 1.6 Athlons,
two 3ware 7850's, 8 disks on each but with 160 gig drives.
I have kernel 2.4.19 on it with ext3. My RAID is arranged
slightly different however. I have each controller do a RAID 5
and then I software mirror the two. This box gets hammered
pretty hard at times over NFS and other services but I have
never seen this type of behaviour come out of it. Are you
sure one of your drives isn't going bad, it looks the error
is always coming from unit #7. Do all of them do this?
Steve
Robert L. Harris wrote:
> I've got a system I need to slow down disk access on it would seem. The
>syerver, a dual 1.5Ghz Athalon has two 3ware IDE controllers, 8 disks
>each for a total of sixteen 180 Gig disks. This system is laid out in
>four RAID5 arrays that it shares out via NFS. kernel 2.4.19-ac4, ext3
>file systems. I've got 7 of these.
>
> This works great and provides a LOT of cheap disk that we use for
>staging backups before cloning off to tape.
>
> The problem is that when the array's get pounded HARD, such as when
>legato is cleaning it's file devices (nsrstage -C) the machine will lock
>up and spew errors to my console:
>
>3w-xxxx: scsi1: Command failed: status = 0xc7, flags = 0x1b, unit #7.
>3w-xxxx: scsi1: Command failed: status = 0xc7, flags = 0x1b, unit #7.
>3w-xxxx: scsi1: Command failed: status = 0xc7, flags = 0x1b, unit #7.
>3w-xxxx: scsi1: Command failed: status = 0xc7, flags = 0x1b, unit #7.
>3w-xxxx: scsi1: Command failed: status = 0xc7, flags = 0x1b, unit #7.
>3w-xxxx: scsi1: Command failed: status = 0xc7, flags = 0x1b, unit #7.
>3w-xxxx: scsi1: Command failed: status = 0xc7, flags = 0x1b, unit #7.
>3w-xxxx: scsi1: Command failed: status = 0xc7, flags = 0x1b, unit #7.
>3w-xxxx: scsi1: Command failed: status = 0xc7, flags = 0x1b, unit #7.
>3w-xxxx: scsi1: AEN: WARNING: ATA port timeout: Port #7.
>3w-xxxx: scsi1: AEN: WARNING: ATA port timeout: Port #7.
>3w-xxxx: scsi1: AEN: WARNING: ATA port timeout: Port #7.
>3w-xxxx: scsi1: AEN: WARNING: ATA port timeout: Port #7.
>3w-xxxx: scsi1: AEN: WARNING: ATA port timeout: Port #7.
>3w-xxxx: scsi1: AEN: WARNING: ATA port timeout: Port #7.
>3w-xxxx: scsi1: AEN: WARNING: ATA port timeout: Port #7.
>3w-xxxx: scsi1: AEN: WARNING: ATA port timeout: Port #7.
>
> This requires a hard reboot. If I use the magic keys and reset it then
>it goes down but doesn't come back up properly, usually dieing around
>the LILO area, I believe the array or disks are left in an odd state.
>
> Is there a way to actually slow down the disk access/read/write other
>than re-making the filesystems? On the other systems the chunk size is
>at either 32K or 128K. On this one in particular the chunk size is
>1024K which was determined by running a number of tests (bonnie, nfs
>reads/writes) and was found to be about the fastest for our money's
>worth. The disk hangups didn't appear until just recently. If there's
>no other choise I can leave the disks as read only until the data rolls
>off then remake them in 32 or 64K.
>
>Any other thoughts?
> Robert
>
>
>
>:wq!
>---------------------------------------------------------------------------
>Robert L. Harris | PGP Key ID: E344DA3B
> @ x-hkp://pgp.mit.edu
>DISCLAIMER:
> These are MY OPINIONS ALONE. I speak for no-one else.
>
>Diagnosis: witzelsucht
>
>IPv6 = [email protected] http://ipv6.rdlg.net
>IPv4 = [email protected] http://www.rdlg.net
>
>