Re: Slow RAID5 build on ata_piix vs. fast on sata_mv

I'll reply to myself again, since I've solved the issue and figure it
might as well go into archives somewhere. :)

After discovering Adaptec acknowledged an issue these (and related)
drives on their controllers and provides updated drive firmware on their
website, I guess-timated a drive issue.  I obtained the new SN05
firmware directly from Seagate and the drives are performing much faster
on the Intel controller.

For good measure, I flashed the other three drives (on the Marvell
chipset controller), and they're still fine too.

Richard

On Mon, Sep 08, 2008 at 05:32:23PM -0400, Richard Michael wrote:

Top-posting myself, I changed the motherboard BIOS (from IDE to AHCI)
and initrd (adding ahci.ko), and the output of lspci makes a bit more
sense:

SATA controller [0106]: Intel Corporation 82801IR/IO/IH (ICH9R/DO/DH)
  6 port SATA AHCI Controller [8086:2922] (rev 02)
Kernel driver in use: ahci
Kernel modules: ahci


However, now the rebuild is even slower, at ~12MB/sec.

A disk on the ICH9R onboard controller (ahci):

# hdparm -tT /dev/sdd

/dev/sdd:
 Timing cached reads:   8030 MB in  2.00 seconds = 4022.76 MB/sec
 Timing buffered disk reads:  166 MB in  3.02 seconds =  54.98 MB/sec


A disk on the PCI-X (sata_mv) controller:

# hdparm -tT /dev/sda

/dev/sda:
 Timing cached reads:   8094 MB in  2.00 seconds = 4054.45 MB/sec
 Timing buffered disk reads:  324 MB in  3.01 seconds = 107.73 MB/sec


Googling reveals other people with slow AHCI disks on new (> 2.6.23)
kernels.  Would someone using disks on an AHCI controller and a new
kernel please send me timed reads?

This doesn't feel like a linux-raid issue anymore, perhaps I should take
this to the lkml?

Perhaps I've missed something else in the BIOS..

Thanks,
Richard



On Mon, Sep 08, 2008 at 01:50:21PM -0400, Richard Michael wrote:

quoted

On kernel 2.6.26.3 with mdadm 2.6.7, I'm building two RAID5 arrays of
three drives each from six Seagate 1TB drives (ST31000340NS) on two
different controllers.

Initial creation is rather slow for the array of disks on onboard
controller, about ~17MB/sec.  The creation of the array on the PCI-X
controller was much faster, about ~90MB/sec.  (Note: the creation is
sequential, I'm not creating these two arrays simultaneously.)

Three drives on one array are connected to the onboard SATA controller of
an Asus P5EWS Pro motherboard.  lspci shows me two IDE interfaces, one
four port and the other two port which I suppose correspond, in total,
to the six onboard ports:

Intel Corporation 82801IR/IO/IH (ICH9R/DO/DH)
  4 port SATA IDE Controller (rev 02)
Intel Corporation 82801I (ICH9 Family)
  2 port SATA IDE Controller (rev 02)

They both use the ata_piix kernel module.


The other controller is an eight port SuperMicro PCI-X controller,
AOC-SAT2-MV8, which has been mentioned on the list in the past:

Marvell Technology Group Ltd. 88SE6145 SATA II PCI-E controller (rev a1)

It uses the sata_mv kernel module.


I've created both RAID5 arrays with default parameters, using simply:
  mdadm --create /dev/md2 --raid-devices=3 --level=raid5
    /dev/sdd2 /dev/sde2 /dev/sdf2


hdparm shows me all drives are using udma6 (e.g. "hdparm -i /dev/sda"),
so I don't think it's a DMA issue (anyway, I've read all SATA drives use
DMA.. ?).

It seems like a controller issue.  Perhaps a different driver is
available for the onboard Intel controller; or some tunables in libata?

Any suggestions?

Thanks,
Richard
--
To unsubscribe from this list: send the line "unsubscribe linux-raid" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

--
To unsubscribe from this list: send the line "unsubscribe linux-raid" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html