lkml.org 
[lkml]   [2005]   [Apr]   [16]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: Poor I/O Performance with MegaRaid SATA 150-4; bug or feature?
Matt M. Valites wrote:
> Hail List,
>
> I've been banging my head against this for a few days, and I wanted to
> see if anyone here could lend a hand.
>
> I have the following configuration:
> P4 3.x Ghz
> 2GB Ram;
> 2 x 36GB WD Raptors; in a RAID1 (sda)
> 2 x 74GB WD Raptor (those 10K RPM SATA drives) in a RAID1(sdb)
> Two free PCI-X slots, one of which occupied by a LSI MegaRaid SATA 150-4.
>
> The problem is I/O on either one of these RAID devices seems to
> be less than half what I'm expecting. The file system used in my testing is
> XFS, and I'm running kernel 2.6.11.6.
>
> The test I'm doing is a simple:
> # time dd if=/dev/zero of=./crap.file bs=1024 count=209715
> Which results in a runtime of about ~53s, in the best case, with all the
> scary write cache enabled. I've tried with deadline, and
> anticipatory. I've also tried several kernels, namely a recent 2.4, so
> I could test megaraid and megaraid2, similar results.
>
> On my desktop box, with one of these drives connected via SATA, i get
> ~25s, also XFS. (2.6.11-gentoo-r6 x86_64).
>
> Is this an expected result? I'm seeing much higher numbers posted around the
> 'Net. Most of those results are from Windows boxes.
>
> I've uploaded my kernel config, lspci -v, and two opreports of a bonnie++ run
> to: http://www.muixa.com/lkml/

I also have one of those cards, at home. I've come to the conclusion
that they're just too old. No NCQ and such other fancy features (for
gods sake, the controllers on the card are sil 3112's!). It's probably
not even PCI-X native.

The only thing that can bring its performance reseanably up to speed is
using write-back instead of write-through on the array. Also try
enabling the write-cache on the drives (all doable in the cards bios
config, not sure if this is what you meant with "with all the scary
write cache enabled"). Doing this is on the other hand not very good for
your data integrity, not good at all.

If only NCQ/TCQ was in, it would have a chance of having decent
performance using write-through. A cool experiment would be setting up
the drives as invidual drives on the card, and use md software raid over it.

Next time I'll probably just use md software raid over a 3ware 9xxx
(JBOD-mode) or AHCI controller. I'm feeling quite uneasy about vendor
lock in nowadays. Groan.

--
Cheers,
André Tomt
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

\
 
 \ /
  Last update: 2005-04-17 01:01    [W:0.045 / U:0.076 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site