linux-ide.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
From: Denys Dmytriyenko <denis@denix.org>
To: Tejun Heo <htejun@gmail.com>
Cc: Gabor FUNK <FUNK.Gabor@hunetkft.hu>,
	linux-ide@vger.kernel.org, Jim Paris <jim@jtan.com>,
	Mark Lord <liml@rtr.ca>
Subject: Re: sata_sil24 stability and performance
Date: Sat, 15 Mar 2008 17:43:47 -0400	[thread overview]
Message-ID: <20080315214347.GA1511@denix.org> (raw)
In-Reply-To: <47CF9880.4080900@gmail.com>

On Thu, Mar 06, 2008 at 04:08:48PM +0900, Tejun Heo wrote:
> Denys Dmytriyenko wrote:
> >>> Also, few months ago instead of sata_sil24 I had my drives connected to
> >>> sata_mv (Supermicro 8-port) and performance was normal...
> >> Everything seems okay.  I wonder where the difference is.  Does "dd
> >> if=/dev/zero of=file oflags=direct bs=1M" make any difference?  And can
> >> you vacate a raw partition and try it on there?
> > 
> > oflag=direct has no effect - same speed. Tried it on a raw /dev/sda device
> > and still no difference. It maybe slighly better, giving me 18 MB/s.
> > 
> > Can it be something with the way SiI 3124 controller is configured in the
> > system?
> 
> Hmmm... That's strange you did specify the "bs" parameter, right?  It
> should essentially give the same performance as "hdparm -t".  I wonder
> where the difference comes from.

Ok, after countless night-hours trying different configurations and peripheral 
combinations, it seems I can get good write performance (55 MB/s) from SATA 
card sitting in PCI-X slot, but bad write performance (18 MB/s) when it sits 
in PCI slot. And it is far from the PCI bandwidth limit.

I did some research and found out that apparently it is a known issue of 
AMD-760 MPX chipset. PCI-X bus is on the AMD-762 north bridge, while PCI bus 
is on the AMD-768 south bridge, which has a write bandwidth limit of 25 MB/s 
"bug", acknowledged by AMD. See these discussion threads:

http://forums.2cpu.com/showthread.php?s=c8040a4e9c9b6390dd389f1b3cca32de&threadid=31211
http://episteme.arstechnica.com/eve/forums/a/tpc/f/77909774/m/1160910035
http://forums.2cpu.com/showthread.php?s=66da493f719e8e64d15dc974cd567192&threadid=23379

Now, I have 2 options:

1. Keep existing system, but utilize 2 PCI-X slots for SATA controllers. This 
requires using 8-port adapters, like the one I already have Supermicro Marvell 
based. The question is - how stable sata_mv these days? It still says HIGHLY 
EXPERIMENTAL...

2. Replace MoBo+CPU+RAM (at least) and keep using SiI3124 based 4-port SATA 
adapters, as sata_sil24 is supposedly the most stable solution I can get.

I'm leaning towards the second option, but it would cost me more, compared 
to getting the second Marvell based PCI-X SATA card. Can you please advise? 
Thanks in advance.

Regards,
Denys

  reply	other threads:[~2008-03-15 21:44 UTC|newest]

Thread overview: 30+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2008-02-19  2:09 sata_sil24 stability and performance Denys Dmytriyenko
2008-02-19  4:36 ` Jim Paris
2008-02-19  6:39   ` Denys Dmytriyenko
2008-02-19 15:32   ` Mark Lord
2008-03-02  6:14     ` Denys Dmytriyenko
2008-03-02  9:39       ` Gabor FUNK
2008-03-04  0:02         ` Tejun Heo
2008-03-04  0:22           ` Denys Dmytriyenko
2008-03-04  3:28             ` Tejun Heo
2008-03-04  6:29               ` Denys Dmytriyenko
2008-03-05  8:11                 ` Tejun Heo
2008-03-06  4:14                   ` Denys Dmytriyenko
2008-03-06  4:25                     ` Tejun Heo
2008-03-06  6:55                       ` Denys Dmytriyenko
2008-03-06  7:08                         ` Tejun Heo
2008-03-15 21:43                           ` Denys Dmytriyenko [this message]
2008-03-17  3:09                             ` Mark Lord
2008-03-18  0:15                               ` Denys Dmytriyenko
2008-03-18  4:09                                 ` Tejun Heo
2008-03-18  4:53                                   ` Denys Dmytriyenko
2008-03-18  6:40                                     ` Tejun Heo
2008-03-20 22:37                                       ` Denys Dmytriyenko
2008-03-21  0:18                                         ` Tejun Heo
2008-04-14  1:19                                           ` Denys Dmytriyenko
2008-04-14  2:49                                             ` Tejun Heo
2008-04-14 10:55                                             ` Gabor FUNK
2008-03-18  9:14                                     ` Gabor FUNK
2008-03-18 13:06                                       ` Gabor FUNK
2008-03-18 20:05                                   ` Mark Lord
2008-03-18 20:06                                     ` Mark Lord

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20080315214347.GA1511@denix.org \
    --to=denis@denix.org \
    --cc=FUNK.Gabor@hunetkft.hu \
    --cc=htejun@gmail.com \
    --cc=jim@jtan.com \
    --cc=liml@rtr.ca \
    --cc=linux-ide@vger.kernel.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).