ICH8, DaniS506, Sysbench, Seagate - not playing nice together

The 915P chipset P4 2.8G motherboard in my eCS system died a month ago, at
least in part due to 1 or more bad caps, at age 40 months on a 36 month
warranty. I ordered
[url]http://www.intel.com/products/desktop/motherboards/DP965LT/DP965LT-overview.htm[/url]
with G965 chipset and onboard video to replace it using the old RAM & CPU.

I found my existing eCS install would not boot on it. I installed rc5 with
little trouble, ran Sysbench on it a bunch of times with various settings
alterations. I wound up doing a BIOS update that helped some, but not enough,
and ultimately decided it was defective and returned it to the vendor.

In part due to the large consumption of my time with the above and shipping
delay, I ordered another P4 and 2 2G DDR2 sticks, and 2 other motherboards:
[url]http://www.ecs.com.tw/ECSWebSite/Products/ProductsDetail.aspx?detailid=681&DetailName=Bios&MenuID=44&LanID=9[/url]
[url]http://www.foxconnchannel.com/product/motherboards/detail_spec.aspx?ID=en-us0000297[/url]

The Foxconn arrived first. It behaved much better than the Intel. The
original eCS booted up without objection. I did more benchmarking with both
new and old CPU and RAM, then boxed it up.

Next day I ran similar tests with the eCS, which behaved similarly to the
Foxconn. It's still in the case pending my determination of what if anything
to do next.

The problem is that on the Intel & Foxconn boards, Seagate SATA2 drives (I
have 4: 2 160G, 1 320G, 1 500G, all with speed limit jumpers removed after
performing most tests) perform adequately on the Sysbench File I/O & Disk I/O
tests, but perform dismally on the simultaneous I/O tests, roughly 20% or
less of expected performance.

On the eCS board the problem is the same, but only on the Intel & Foxconn is
there a workaround. That is, if I put all HDs on sequential ports on the
Intel or Foxconn, then the simultaneous I/O performance seems OK. IOW, drives
on ports 1 & 2 ("master" & "slave") both perform adequatly on all tests, but
if the drives are on any of 1, 3 or 5 ("master") only and none on 2, 4 or 6
("slave"), then performance on simulIO suffers severely.

The eCS board has no SATA port "2" or "4". It has template spaces on the
board where those ports would have gone if they existed, and the BIOS assumes
they exist. What is has are masters 1, 3, 5 & 7, and slave 6. Wierd. A
JMicron chip exists to host IDE, normal mode or (sucky single cable) BIOS
RAID, and #"7" SATA port.

When I boot into maintenance mode (older 14.100c_W4 kernel and 1.7.10
DaniS506) with 2 SATA + IDE CD, ibms506$ reports port "1" on controller 2,
port "3" (logical 2) on controller 3, & IDE on controller 4. When I boot
normally into eCS with 14.104 kernel and 1.8.5 DaniS506, port 1 is controller
0, port 3 is controller 1, etc. as shown on
[url]http://fm.no-ip.com/PC/bench/sysbench/p965t-danis506b.txt[/url] .

[url]http://fm.no-ip.com/PC/bench/sysbench/p965t-danis506c.txt[/url] shows the
additional DaniS506.ADD parameters I tried to improved performance, which
didn't help. [url]http://fm.no-ip.com/PC/bench/sysbench/resul562.html[/url] shows the
nearly last Sysbench test results, where I put only a single SATA2 device on
each of the ICH8 & JMicron chips. [url]http://fm.no-ip.com/PC/bench/sysbench/[/url] has
all my various recent results.

I also tried a Seagate PATA as sole JMicron device with only 1 SATA2 Seagate
on ICH8, but simulIO on both was again bad. A WD on PATA, like the Samsung
and Hitachi on SATA, performed respectably on simulIO. Putting an only
Seagate on the only "slave" SATA port didn't help.

[url]http://fm.no-ip.com/PC/bench/sysbench/resul560.html[/url] was done on rc5, with
ACPI enabled and MP kernel, instead of 1.14 with neither ACPI nor MP kernel,
but there was no material difference in disk benchmarks. The eCS motherboard
BIOS has no disable ACPI option, while the Foxconn does, and the Intel did.

Anyone have similar experience, or know some other way to get the Seagates to
perform like the Hitachi & Samsung on simulIO? Are the Seagates surfacing a
flaw in the Sysbench simulIO test? I'm having a hard time understanding how
in any other way it's possible for the simulIO test to be so out of line with
the other disk performance tests.

Is there any practical real world way to simulate the simulIO test to confirm
or deny the Sysbench test's validity?
--
"Love is not easily angered. Love does not demand
its own way." 1 Corinthians 13:5 NIV

Team OS/2 ** Reg. Linux User #211409

Felix Miata *** [url]http://fm.no-ip.com/[/url]

On Thu, 7 Aug 2008 23:43:01 UTC in comp.os.os2.setup.storage, Felix Miata
<UgaddaBkidding.due2UCE@dev.nul> wrote:
[color=blue]
> Is there any practical real world way to simulate the simulIO test to confirm
> or deny the Sysbench test's validity?[/color]

Run sysbench with the /debug switch and pipe the output to a file. The
individual outputs from the disk tests should be in that log. I wonder if the 4
results are being added up and wrap round at some power of 2.

Yep, there is. It misreports the numbers when they hit 65535KB/sec and they
wrap to zero.

--
Trevor Hemsley, Brighton, UK
Trevor dot Hemsley at ntlworld dot com

On Thu, 7 Aug 2008 23:43:01 UTC in comp.os.os2.setup.storage, Felix Miata
<UgaddaBkidding.due2UCE@dev.nul> wrote:
[color=blue]
> Is there any practical real world way to simulate the simulIO test to confirm
> or deny the Sysbench test's validity?[/color]

Fix here - [url]http://www.os2warp.org/sysbench/sysb094d.zip[/url]

--
Trevor Hemsley, Brighton, UK
Trevor dot Hemsley at ntlworld dot com

On 2008/08/07 20:29 (GMT-0400) Trevor Hemsley apparently typed:
[color=blue]
> On Thu, 7 Aug 2008 23:43:01 UTC in comp.os.os2.setup.storage, Felix Miata wrote:[/color]
[color=blue][color=green]
>> Is there any practical real world way to simulate the simulIO test to confirm
>> or deny the Sysbench test's validity?[/color][/color]
[color=blue]
> Fix here - [url]http://www.os2warp.org/sysbench/sysb094d.zip[/url][/color]

A 10 year old version of Sysbench? O_O
--
"Love is not easily angered. Love does not demand
its own way." 1 Corinthians 13:5 NIV

Team OS/2 ** Reg. Linux User #211409

Felix Miata *** [url]http://fm.no-ip.com/[/url]

On Fri, 8 Aug 2008 07:38:06 UTC in comp.os.os2.setup.storage, Felix Miata
<UgaddaBkidding.due2UCE@dev.nul> wrote:
[color=blue]
> On 2008/08/07 20:29 (GMT-0400) Trevor Hemsley apparently typed:
>[color=green]
> > On Thu, 7 Aug 2008 23:43:01 UTC in comp.os.os2.setup.storage, Felix Miata wrote:[/color]
>[color=green][color=darkred]
> >> Is there any practical real world way to simulate the simulIO test to confirm
> >> or deny the Sysbench test's validity?[/color][/color]
>[color=green]
> > Fix here - [url]http://www.os2warp.org/sysbench/sysb094d.zip[/url][/color]
>
> A 10 year old version of Sysbench? O_O[/color]

A typo, should've been [url]http://www.os2warp.org/sysbench/sysb095d.zip[/url]

--
Trevor Hemsley, Brighton, UK
Trevor dot Hemsley at ntlworld dot com

On 2008/08/08 02:56 (GMT-0500) Trevor Hemsley apparently typed:
[color=blue]
> On Fri, 8 Aug 2008 07:38:06 UTC in comp.os.os2.setup.storage, Felix Miata wrote:[/color]
[color=blue][color=green]
>> On 2008/08/07 20:29 (GMT-0400) Trevor Hemsley apparently typed:[/color][/color]
[color=blue][color=green][color=darkred]
>> > Fix here - [url]http://www.os2warp.org/sysbench/sysb094d.zip[/url][/color][/color][/color]
[color=blue][color=green]
>> A 10 year old version of Sysbench? O_O[/color][/color]
[color=blue]
> A typo, should've been [url]http://www.os2warp.org/sysbench/sysb095d.zip[/url][/color]

I thought so. I tried that before writing but got a 404. Thanks for quick fix
for the URL and the app. Much better now, 74.219M/s instead of 10.whatever on
a ST160. :-)
--
"Love is not easily angered. Love does not demand
its own way." 1 Corinthians 13:5 NIV

Team OS/2 ** Reg. Linux User #211409

Felix Miata *** [url]http://fm.no-ip.com/[/url]