Slow RAID1 read on ICH10R

A

af22

Gawd
Joined
Jun 21, 2010
Messages
610
Hello,

I have some weird performance issues on my raid array on my new x58 i7 pc.

I have 2x 500GB 7200 Hitachi Drives in Raid 1 on my ICH10R:

Sequential Read: 44meg/sec (bad)
Sequential Write: 75meg/sec (normal)

I used to run the same hard drive on my intel p35 q6600 on the ICH9? in raid 0,
there i got 150meg/sec on both sequential read and write.

I thought raid1 was supposed to give me faster read and slower writes? my writes are just about the same as a single drive, read are twice as slow?

any ideas?
 
Did you use CrystalDiskMark or AS SSD to test the performance? If not try these utilities first.

RAID1 generally gives you the same read performance as a single disk and the same write performance of a single disk. Some advanced implementations give you extra performance when reading; but i don't think these implementations are available on Windows OS. The most advanced RAID1 implementation i've seen in FreeBSD's geom_mirror driver with 'load' algorithm.
 
yes, i am using crystaldiskmark and as ssd. also, the machine is on windows 7 with the latest intel RIST utility.

it's weird how my raid 0 performance was fine on my older ICH9. i wonder if one of the drives are failing
 
But it's proprietary; and their tests showing more than 100% performance increase are suspicious. The theoretical maximum for striping alone is 100%; not more.

But they do claim an increase in both sequential and random reads on RAID1; that is exactly what the 'load' algorithm on geom_mirror does. But it can handle different performance disks; such as one 15k rpm disk which would get 75% of the requests and one 7200rpm disk which gets 25%; just guessing these values but the load algorithm can prevent a slower device from slowing down the faster device.
 
2x 500GB Hitachi Drives Raid1 ICH10R (via x58 mobo):

--------------------------------------------------
CrystalDiskMark 2.2 (C) 2007-2008 hiyohiyo
Crystal Dew World : http://crystalmark.info/
--------------------------------------------------

Sequential Read : 46.247 MB/s
Sequential Write : 85.878 MB/s
Random Read 512KB : 46.804 MB/s
Random Write 512KB : 57.820 MB/s
Random Read 4KB : 0.761 MB/s
Random Write 4KB : 1.102 MB/s

Test Size : 100 MB
Date : 2010/09/11 15:07:23


Here's my work laptop with a WD 250GB Scorpio Black on SATA150:
Sequential Read : 73.481 MB/s
Sequential Write : 69.076 MB/s
Random Read 512KB : 37.492 MB/s
Random Write 512KB : 56.698 MB/s
Random Read 4KB : 0.707 MB/s
Random Write 4KB : 1.791 MB/s

Test Size : 100 MB
Date : 2010/02/12 16:30:36


Performance is depressing =(
 
Could you search for an option called 'Enable write caching' in the Intel drivers? Try Device Manager and search for the intel controller's preferences; i think it's there.

If you enable it, do your reads get any better? the CrystalDiskMark looks about right; only sequential read is too low. That might be because it is trying to read from both drives without proper buffering. The write caching option can correct this, however it would mean unsafe writes as they are buffered in your RAM. But at least you can test if this affects your performance.
 
that is exactly what the 'load' algorithm on geom_mirror does. But it can handle different performance disks; such as one 15k rpm disk which would get 75% of the requests and one 7200rpm disk which gets 25%; just guessing these values but the load algorithm can prevent a slower device from slowing down the faster device.
Click to expand...

Cool!
 
i enabled the write caching in the RIST, now the writes are slower!

i am going to break the raid array and bench each drive individually to see if one is the problem

--------------------------------------------------
CrystalDiskMark 2.2 (C) 2007-2008 hiyohiyo
Crystal Dew World : http://crystalmark.info/
--------------------------------------------------

Sequential Read : 46.329 MB/s
Sequential Write : 69.627 MB/s
Random Read 512KB : 44.316 MB/s
Random Write 512KB : 34.321 MB/s
Random Read 4KB : 0.799 MB/s
Random Write 4KB : 1.811 MB/s

Test Size : 100 MB
Date : 2010/09/12 11:04:24
 
Does the test change if you use 1000MB test file instead of 100MB?

Not really sure why you get only ~45MB/s sequential read; it should be higher.
 
no difference with 1000 mb. starting to think one of the drives has problems.

--------------------------------------------------
CrystalDiskMark 2.2 (C) 2007-2008 hiyohiyo
Crystal Dew World : http://crystalmark.info/
--------------------------------------------------

Sequential Read : 42.244 MB/s
Sequential Write : 76.149 MB/s


Test Size : 1000 MB
Date : 2010/09/12 12:41:28
 
If you can break the RAID and benchmark each drive separately with CrystalDiskMark. You could be right that one of the disks may have some issues lowering performance. But if these turn up good then i wouldn't really know what you could try. Perhaps compare performance with a software RAID1 instead.
 
here's an update:

- i unplugged one of the drives in the raid1 array, benched it with crystal disk, sequential read/write was: 40/60 meg/sec. switched to the other drive, got the same results

- switched the bios back to AHCI mode, benched one drive, got 90/90 meg/sec (normal speed)

Looks like when i use RAID mode in the bios, drives just get slow. No idea why.

Now, i am keeping it on AHCI and use windows raid instead. The best part about using windows raid is that i have both a mirror partition and a stripped partition on the same two hard drives!
 
af22 said:
here's an update:

The best part about using windows raid is that i have both a mirror partition and a stripped partition on the same two hard drives!
Click to expand...

That's interesting. I never thought of doing that. Can you show some benchmarks ?
 
Benchmarks:

2x 500gb 7200 Hitachi Drives (3 years old): Dynamic Disk mode using Windows 7 Software Raid with AHCI mode set in bios. (I have an OCZ 60GB Agility on my C:)

Partition D: 292 GB Mirror (Raid 1)
Parition E: 345 GB Strip (Raid 0)

So half of the hard drive is mirrored across the second drive, the other half is stripped across the second drive

Partition D: 292 GB Mirror (Raid 1):
--------------------------------------------------
CrystalDiskMark 2.2 (C) 2007-2008 hiyohiyo
Crystal Dew World : http://crystalmark.info/
--------------------------------------------------

Sequential Read : 81.830 MB/s
Sequential Write : 77.022 MB/s
Random Read 512KB : 42.118 MB/s
Random Write 512KB : 47.172 MB/s
Random Read 4KB : 0.617 MB/s
Random Write 4KB : 0.960 MB/s

Test Size : 100 MB
Date : 2010/09/18 14:48:51


Parition E: 345 GB Strip (Raid 0):
--------------------------------------------------
CrystalDiskMark 2.2 (C) 2007-2008 hiyohiyo
Crystal Dew World : http://crystalmark.info/
--------------------------------------------------

Sequential Read : 151.626 MB/s
Sequential Write : 149.820 MB/s
Random Read 512KB : 59.850 MB/s
Random Write 512KB : 69.230 MB/s
Random Read 4KB : 0.856 MB/s
Random Write 4KB : 1.624 MB/s

Test Size : 100 MB
Date : 2010/09/18 14:45:20


I even tried briefly doing a 3 drive raid 0 setup. i got sequential read / writes in the 240 meg/second =) but didn't want to include my backup drive in the array....

Note: this is on a x58 / i7 930, so i don't really notice the cpu utilization on using windows raid.

windows7raid.png
 
I have the same problem.

Motherboard: Asus P5Q Deluxe (ICH10R)
Hard Disks: 1 SSD Samsung 128GB (Windows 7), 2x2TB Seagate LP Green

The two HDD Seagate are set as Raid1.

Read Sequential: 52MB/s
Write Sequential: 132MB/s

If the two HDD are set as normal disk:

Read Sequential: 142MB/s
Write Sequential: 135MB/s

The read speed in Raid1 mode is too slow. It could be like that in normal mode.

How can I solve this?

Thank you! :)
 
This sounds like either a firmware bug or the drivers need to be updated/configured.

FakeRAID with RAID1 should give a large increase in read speed and the write should be the equivalent to a single disk.
 
Red Falcon said:
This sounds like either a firmware bug or the drivers need to be updated/configured.

FakeRAID with RAID1 should give a large increase in read speed and the write should be the equivalent to a single disk.
Click to expand...

Do you have any test to do?

Thanks!
 
You must log in or register to reply here.
Back
Top