Poor raid 0 performance with the Intel ICH9 raid chip

[adampd]

Hello all,

I have a Gigabyte DS3P V2.0 motherboard which has two raid chipets onboard, the Gigabyte and Intel ones
The intel one is the ICH9 Raid chipset

I've setup two Western digital WD2500JS drives (each being 250gb) in a RAID 0 configuration, but am seeing rather poor performance
The strip size is 128kb
And this is the performance that Sandra shows for the RAID drive.

Benchmark Results
Drive Index : 61.46MB/s
Results Interpretation : Higher index values are better.
Random Access Time : 17.52ms
Results Interpretation : Lower index values are better.

And yet my single WDC5000AKS drive performs a fair bit better.

Benchmark Results
Drive Index : 96.66MB/s
Results Interpretation : Higher index values are better.
Random Access Time : 12.8ms
Results Interpretation : Lower index values are better.

If anyone could offer any advice, I would greatly appreciate it.

 

[bilbat]

Jeez - that does seem abysmal - this, too, is off an ICH9R:

'couse it's one of two pairs of VR150's on a fast quad... What are your processors specs? It is, after all, software RAID, being done strictly by the CPU...

 

[adampd]

Heh nice specs

My CPU - Intel Core 2 Quad CPU (2.66Ghz)
4Gb of Memory (Corsair 4GB Kit (2x2GB) DDR2 800MHz/PC2-6400 XMS2)
Motherboard - Gigabyte GA-P35-DS3P iP35 Socket 775 8-channel audio ATX

No idea why the specs were so poor, but it was dissapointing to say the least, heh

 

[adampd]

Using one of the drives by itself, with the same operating system setup, I'm getting...

Benchmark Results
Drive Index : 49.11MB/s
Results Interpretation : Higher index values are better.
Random Access Time : 18ms
Results Interpretation : Lower index values are better.

and

[img=http://img194.imageshack.us/img194/1865/hddrive.png]

From HDTach

So I have no idea why there's such a bottleneck with that Intel controller, seems odd.

 

[bilbat]

Which northbridge do you have? (or post the full board p/n...) Many of the later BIOS contain newer ICH code, which, along with a driver update, speeds things up somewhat... Also, if you'd like, I can walk you through, say, a 20-25% low-effort, easy-on-the-hardware overclock that will also give the whole system a bit of a boost...

Bill

 

[MRFS]

The WD2500JS is aging a bit: it's not in the current WD Product list:

http://www.wdc.com/en/products/productcatalog.asp?language=en


Note the date on this review:

http://www.retrevo.com/s/Western-Digital-WD2500JS-Storage-review-manual/id/17432ag304/t/1-2/


It's also deactivated at Newegg:

http://www.newegg.com/Product/Product.aspx?Item=N82E16822144417


It's also not in their RAID Edition ("RE") series,
which supports Time-Limited Error Recovery.


This performance was about right for that HDD era:

http://img194.imageshack.us/i/hddrive.png/


Try checking the jumper setting for the
interface speed, and launch the Intel Matrix
Storage Console, under "Current Serial ATA
Transfer Mode".

The transfer mode may be "Generation 1"
which means 150MB/second, NOT 300MB/second.


MRFS

 

[MRFS]

Here's an HDTach run on a WD1600YS wired to ICH7R with AHCI enabled;
the WD1600YS is about 18 months newer than your problem drive:

http://www.supremelaw.org/systems/ich7r/WD1600YS.HDTach.JPG

... pretty close, given the technology in your WD HDD.


You should be getting HDTach results closer to this measurement,
which used 2 x WD5000YS RE2 HDDs on a Marvell controller, i975X chipset:

http://www.supremelaw.org/systems/io.tests/Marvell.2xWD5000YS.JPG


You might also want to see if NCQ is set wrong:
with some benchmark software, enabling NCQ can
add measurable computational overhead.


p.s. I don't think your WD HDD supports NCQ, however.


MRFS

 

[adampd]

Thanks all for the information
I double checked all the settings, in the bios and on the drive (I actually have TWO of the Wd2500JS's, they came out of an external buffalo drive station)
Neither have any jumpers set, so should be working in full Generation 2 mode
The one I've got installed now is showing as Generation 2
Earlier it was showing "UDMA 5 Selected" and "UDMA 6 Supported" in HDDScan
I enabled ACHI in the bios and rebooted, it now shows as UDMA 6 Selected instead
The benchmark is the same regardless of IDE/ACHI mode, which is a shame, heh

(WD5000AAKS scores in at 100.4 since the change to ACHI)

To Billbat - Did you see the motherboard model I posted earlier? If not, it's the Gigabyte GA-P35-DS3P (Version 2.0)
I'm running the latest bios (F12)
Board URL - http://www.giga-byte.com/Support/Motherboard/BIOS_Model.aspx?ProductID=2625

As for overclocking, Sure, any tips are welcome

 

[adampd]

One thing I meant to ask
I always use the drivers from the Intel site, as opposed to the gigabyte site
I assume that's ok?
The ones on the Gigabyte site are somewhat older than intel's ones.

 

[MRFS]

I'm a little confused: you "enabled AHCI".

On Intel's ICH I/O controller hubs, "AHCI" and "RAID"
are mutually exclusive.

Also, RAID is a superset of AHCI.


It sounds to me as if your "slow" WDs are running
in JBOD mode (Just a Bunch Of Disks = NON-RAID).


In your BIOS, if you switch back to RAID
you should also be able to enable
the Intel RAID "Option ROM"
(Read Only Memory chip).

The latter contains a POST program which
you invoke with CTRL-I (at the right moment).

Once inside that Option ROM, you can
initialize your RAID 0 array e.g. by assigning
the component hard drives.

Then, after booting into Windows,
you then need to partition and format
that RAID 0 array like any other hard drive.


All of the latter should be documented in
your motherboard's User Manual.


I hope this helps.


MRFS

 

[adampd]

Right now I've got 2 drives
The WD2500JS as my C/Master drive and the WD5000AAKS as my D drive (For storage)
In the bios there's 3 options for the Sata ports, IDE, ACHI and RAID
I have no need for RAID, as it was underperforming, as per the original post.
I read suggestions that switching from standard IDE mode to ACHI, would improve performance for disks.

 

[bilbat]

My own testing has shown no advantage for using AHCI, BUT - I've recently learned that I'm running the wrong tests; in addition to providing hot-plug (if you have a use for it) AHCI's NCQ abilities appear to give a sizeable advantage if your useage mix consists of a lot of rapid-fire, small, (especially write) requests; if you are, like I am, mostly handling large files that ask for sustained access to successive sectors, AHCI won't show up at all... But then, neither can it hurt anything

 

[MRFS]

Thanks for clarifying, AdamPD:

Intel recommends that RAID be set initially in the BIOS,
even if all disks start out as JBOD. I tend to agree
with this recommendation, for reasons that are specific
to our ASUS motherboards (e.g. ease of transition
from JBOD disks to RAID arrays).

Because RAID logic is a superset of AHCI logic,
RAID mode should also exploit NCQ and
support hot-plug for disks in JBOD mode.

The performance of your individual WD2500JS HDDs
appears normal to me.

For RAID 0 performance, I always recommend that
users go with WD's RAID Edition ("RE") HDDs,
because they are specifically engineered for
parallel operation in RAID arrays.

For example, read up on "Time-Limited Error Recovery"
at Western Digital's website.


I hope this helps.


p.s. Why don't you upgrade your C: drive
to a VelociRaptor? $190 after MIR at Newegg:

http://www.newegg.com/Product/Product.aspx?Item=N82E16822136322
(sale ends today, however)

There is also a less expensive 150GB VelociRaptor:

http://www.newegg.com/Product/Product.aspx?Item=N82E16822136296


The VR interface runs at 300MB/second, and
tests show the VR to be a vast improvement over
earlier WD Raptors. The cache is also 16MB
instead of the 8MB on your WD2500JS.


MRFS

 

[bilbat]

Intel recommends that RAID be set initially in the BIOS, even if all disks start out as JBOD. I tend to agree with this recommendation, for reasons that are specific to our ASUS motherboards (e.g. ease of transition from JBOD disks to RAID arrays)

I'll third that second - I always recommend that both RAID 'pre-load' drivers be installed whenever dropping in an OS, for the Gigabytes, too - if you have the drivers loaded, they certainly do not require that you use the RAIDs, but then, should your plans or finances change in the future, your options are open, without a PITA kludge to get them into an existing OS. I also can swear by the VR 150s; I use two pair, RAID0'd, alternating swaps and OS's, so the relevant swap file is always on the opposite pair from the OS - makes the system fly...

 

[adampd]

Heh, ironically, the reason I'm using this WD2500JS drive, is because my original Raptor (the original 74gb model) died on me a few weeks back, after 5+ years of service.
If I could afford a replacement, I certainly would
But I did figure out why windows was taking so long to load though! My acoustic settings on both drives was set to the half way point, changing it to "Loud/Fast" has speeded up windows a fair bit, heh
Just a shame the Raid setup never worked right.
I was looking at getting a decent raid card, but being that I'm off work with long term sick leave, I can't justify paying for that, or a raptor
hehe
Thanks for all your help though!

 

[bilbat]

Just for my personal, future 'knowledgebase', where is this set? Is it a WD utility, a driver-based setting, or what? It certainly makes all the sense in the world that, if you're somehow trying to 'keep down the clatter' of the steppers 'seeking' the heads, that it would make a dent in seek times, as well as access...

Take a peek at this one - think it's the same?
http://www.tomshardware.com/forum/261360-30-ep45-ud3r-slow-access-times

 

[MRFS]

I'm not sure I understand your question, but let me try:

An "Intel chipset" is a set of integrated circuits ("chips")
that are purchased by motherboard manufacturers
like ASUS and Gigabyte, in order to render those
motherboards compatible with Intel CPUs, third-party RAM,
and industry-standard peripherals like SATA and PATA HDDs,
video cards, optical drives, USB devices and the like.

In addition to manufacturing those circuits into "chips"
that are soldered onto printed circuit boards,
Intel also develops software called "chipset drivers"
which must be installed along with the system
software that constitutes an operating system.

The same occurs whenever a motherboard manufacturer
integrates RAID controllers made by third-parties like Marvell:
device drivers must also be loaded along with the OS
before those controllers will operate correctly.

Normally, chipset driver software e.g. for the P45 and
X58 chipsets, comes on a "Support CD" provided
by the motherboard manufacturer. However,
those drivers are occasionally updated with bug fixes
and improvements -- by Intel at Intel for Intel chipsets --
and Intel makes those updates available on its Internet website.

Thus, the chipset drivers that one should be using
are the very lastest versions as published at
Intel's website, and NOT the version that comes
on the Support CD sold with any given P45 or
X58 motherboard (if the two are different).

Another source of faults is the BIOS that comes
on an EPROM soldered to each motherboard:
BIOS updates are likewise made available
at the motherboard manufacturer's website.

If, on the other hand, one has already installed
the very latest chipset drivers e.g. after downloading
same from Intel's website, AND if that software is
not performing to industry standard specifications
using peripherals like Western Digital's SATA/3G
hard drives, it is most probable that the motherboard
manufacturer has allowed a defect to creep into its
implementation of that particular chipset --
either in the hardware and/or in the BIOS software.

Further proof of this hypothesis can be obtained,
as users have already reported, by moving the
HDD in question to a different motherboard
and running the same measurement software
on that other system, perhaps by also using
cables that are already known to be free of
any defects.

Clearly, a hardware or BIOS bug could very easily
be responsible for a timing or synchronization
delay in servicing routine I/O requests
to and from a standard SATA/3G HDD.


Forgive me for being long-winded:
I wanted to cover all these points
to increase the chances of answering
your question.


MRFS

 

[pparks1]

Just for my personal, future 'knowledgebase', where is this set? Is it a WD utility, a driver-based setting, or what? It certainly makes all the sense in the world that, if you're somehow trying to 'keep down the clatter' of the steppers 'seeking' the heads, that it would make a dent in seek times, as well as access...

Take a peek at this one - think it's the same?
http://www.tomshardware.com/forum/261360-30-ep45-ud3r-slow-access-times

Well, I did find an interesting thread on the Seagate 7200.12's and lower acoustic level settings on the firmware that were suspect for causing increased seek times;
http://www.silentpcreview.com/forums/viewtopic.php?p=472546&sid=2352d4fb1719f2842e6e1747d4fd4787

In the above case, it seemed rather hard (if not impossible) to make acoustic setting changes on the Seagate drives older than the 7200.6 series. And even when this poster was able to finally force the setting, the change didn't improve the issue.

And in my particular case, when this same hard drive was moved to another machine....it didn't exhibit the same results. So, if this was a hardware firmware issue on the drive itself...the performance issue should follow the drive around. The only difference when I tested in another computer was that the Seagate drive was a secondary drive to the OS drive....so it wasn't being utilized at all to run the OS or index, etc.

 

[adampd]

The program I used to change the acoustic settings on both my WD drives, was WinAAM

http://www.brothersoft.com/winaam-188688.html

 

[pparks1]

The program I used to change the acoustic settings on both my WD drives, was WinAAM

http://www.brothersoft.com/winaam-188688.html

According to the thread above....the user tried WinAAM and it didn't work with this particular drive. Ah well...also considering just buying another hard drive and turning this Seagate drive into an external drive in a USB enclousure. It wasn't terribly expensive and I could use the extra capacity as backup storage for sure.

 

[adampd]

Oops, I missed that post, apologies.

Another couple you can try is

http://www.techpowerup.com/systool/
and
http://hddguru.com/content/en/software/2006.01.22-HDDScan/

Systool didn't work for me, but HDDscan did.