SB700 AHCI causes disk power cycle

SB700 AHCI causes disk power cycle

[wessels]

Hi,

I also bought an ASUS M3A78 mATX board with a SB700 to see if it might
solve my problem with the SB600. But it doesn't.
The SB700 shows the same behaviour as the SB600, it periodically
throws an error and the power cycles the disk.
But it always happens on one channel not both. Perhaps a sudden burst
in disk IO causes this but I'm unable to
reproduce this with a synthetic load. Only when vmware is running this
happens. I don't see it when the machine is idling.
The load on the machine is very light both on cpu and disk. Also I
didn't see the error with a single seagate disk attached.

The error and configuration is the same as mentioned in my earlier
post regarding the "SB600 AHCI causes disk power cycle".

What I didn't mentioned then was the filesystem layout
two disks in a mirror
/dev/sda & /dev/sdb
no LVM just a software raid via md
md0 /boot ext3
md1 / ext3
md2 swap
md3 /usr/local/virtmach XFS

Please let me know what else I need to log and/or test.

Thank you for your time and interest,

Wessels

Re: SB700 AHCI causes disk power cycle

[Tejun Heo]

Hello,

wessels wrote:
> Hi,
> 
> I also bought an ASUS M3A78 mATX board with a SB700 to see if it might
> solve my problem with the SB600. But it doesn't.
> The SB700 shows the same behaviour as the SB600, it periodically
> throws an error and the power cycles the disk.
> But it always happens on one channel not both. Perhaps a sudden burst
> in disk IO causes this but I'm unable to
> reproduce this with a synthetic load. Only when vmware is running this
> happens. I don't see it when the machine is idling.
> The load on the machine is very light both on cpu and disk. Also I
> didn't see the error with a single seagate disk attached.
> 
> The error and configuration is the same as mentioned in my earlier
> post regarding the "SB600 AHCI causes disk power cycle".
> 
> What I didn't mentioned then was the filesystem layout
> two disks in a mirror
> /dev/sda & /dev/sdb
> no LVM just a software raid via md
> md0 /boot ext3
> md1 / ext3
> md2 swap
> md3 /usr/local/virtmach XFS
> 
> Please let me know what else I need to log and/or test.

>From your report on sb600.

kernel: ata4: SError: { RecovComm Persist PHYRdyChg 10B8B }
kernel: ata4.00: cmd ea/00:00:00:00:00/00:00:00:00:00/a0 tag 0
kernel:          res 40/00:3c:0d:bd:c9/00:00:0b:00:00/40 Emask 0x10 (ATA bus error)

This combined with the incremented start stop count looks very much
like the harddisk went offline briefly and came back for whatever
reason.  I doubt the ahci driver even with its all might can do that.

Can you please try to hook up the faulting harddrive to a separate
power supply?

  http://modtown.co.uk/mt/article2.php?id=psumod

-- 
tejun

Re: SB700 AHCI causes disk power cycle

[wessels]

> From your report on sb600.
>
> kernel: ata4: SError: { RecovComm Persist PHYRdyChg 10B8B }
> kernel: ata4.00: cmd ea/00:00:00:00:00/00:00:00:00:00/a0 tag 0
> kernel:          res 40/00:3c:0d:bd:c9/00:00:0b:00:00/40 Emask 0x10 (ATA bus error)
>
> This combined with the incremented start stop count looks very much
> like the harddisk went offline briefly and came back for whatever
> reason.  I doubt the ahci driver even with its all might can do that.
>
> Can you please try to hook up the faulting harddrive to a separate
> power supply?
>
>  http://modtown.co.uk/mt/article2.php?id=psumod
>
> --
> tejun
>

Thank you for your interest,

I'll test the system with a bigger power supply. But under full load
(cpu, disk and network) these systems consume about 70 watts.
They are powered by either an antec earthwatt 380W or a
seasonic something energy efficient. The antec also powers an opensolaris
server with 7 disks on a sb700 without a problem.
There's one other thing, I did mentioned, the error occurs on both disks.
Although one channel gets more errors than the other. This behaviour is
the same on all three machines.
I'll arrange a couple of seagates to test if it's limited to the
western digital disks.
Currently I'm running the vm's on a nfs share, keep my fingers crossed, no
disk power cycles. But I'll have to test that a bit longer.

I'll be away till next tuesday so please be patient for my results.

Regards,

Wessels