ICH8R RAID5 write performance

Hi

Apologies if this is posted in the wrong group, if so could you point
me in the right direction.

I've just built a new home server; operating system is Server 2003 R2
standard. Motherboard is an Intel DQ965GF with a Core2Duo 1.86 and 2GB DDR2 Kingston RAM.

I've been doing some testing before posting so hopefully you'll be able to help me out.

In the system I have 3x500GB Seagate SATA drives and 1x320GB Maxtor
SATA drive. Initially I created a RAID 5 with the 3x500's. I partitioned that into a 30GB for the system and that left somewhere over 900 for the data.

Once it was up and running I updated the drivers and then tested the
drive performance. I used YAPT. Read performance on both partitions
was as I had hoped, it was around 130meg/sec. Write performance on the data partition was 90meg/sec. However write performance on the system partition was only 10meg/sec. I've checked the net and can't find
anything and I've confirmed write cache is on.

The next thing I did was to brake the RAID and have the drives as stand alone. I installed Server on the 320GB maxtor and then tested performance. The 3 Seagates all returned consistent 72meg/sec read and write. The maxtor returned 62meg/sec read and write.

I've now recreated the RAID 5 and installed Server again but all as 1
big partition. Read performance is back up to 130's,eg/sec but write
is still 10meg/sec.

I've tested performance of the RAID when booting off the maxtor and it's 10meg/sec W and i've also installed XP on the maxtor and the performance is trhe same.

Help

Many Thanks

Dan
16 answers Last reply
More about ich8r raid5 write performance
  1. If you are doing RAID 5 get a controller card! It's the only way to do it fast.
  2. I'll happily buy one if this is expected and normal performance form the ICH8R. But I can't believe it is that poor.
  3. On-board RAID 5 write performance is typically poor, which is a typical reason for avoiding them for RAID 5. However, this is worse than I would have expected -- I'd have guessed around 1/2 of a single drive's performance (considering that a write could take a read + write of parity).

    I suggest trying different stripe sizes. I presume you've already tried 64k. I'd suggest 64k / (number of data drives) = 64k / (number of drives in RAID 5 - 1).

    You could also try a different benchmark program or simple large file copies -- Maxtor to RAID.
  4. Yes the RAID is on 64k now. Could you suggest a write test harddrive benchmarking program? I've tried sandra and it needs the drive to be empty and i'm getting fed up with reinstallations :-)
  5. Quote:
    or simple large file copies -- Maxtor to RAID.


    OK, 19.3gig avi file:

    From Raid to Maxtor it took 5mins 20sec so that is 61.76meg/sec
    and
    From Maxtor to Raid it took 5mins 15sec so that is 62.74meg/sec

    Those are the speeds that the Maxtor is capable of. I guess the benchmarking program is not reporting properly. Has anybody got a suggestion for something else to benchmark write performance with?

    Thanks

    Dan
  6. heres a small hdd read/write performance testing utility i had downloaded a few years ago, Raptest (its under 100KB), its pretty accurate... after searching google, i found it here (towards the bottom of the page):

    http://www.dslreports.com/faq/3498

    hope this helps to resolve the problem :)
  7. Cheers, that gives some better results.

    1st run was 90r and 77w. After that I guess it's in the cache as the results are usually 110+r and 130w!!
  8. cool :), yeah, the first run after a restart gives the most accurate results, as nothing is cached to memory really yet
  9. Quote:

    OK, 19.3gig avi file:

    From Raid to Maxtor it took 5mins 20sec so that is 61.76meg/sec
    and
    From Maxtor to Raid it took 5mins 15sec so that is 62.74meg/sec

    Those are the speeds that the Maxtor is capable of. I guess the benchmarking program is not reporting properly. Has anybody got a suggestion for something else to benchmark write performance with?


    These are good numbers, and I'd generally take actual file transfer performance over synthetic benchmarks.

    You could try Iometer. Here's one suggestion on how to use it.

    http://www.infrant.com/forum/viewtopic.php?t=265

    I'd change them to use smaller access sizes (64K) and even larger files (e.g. 10 GB).

    Or you could use Iozone, as Tom's did:

    http://www.smallnetbuilder.com/nas/nas-how-to/how_we_test_nas/

    I'd change these to use 64K access size and at least 1 GB file size.
  10. IOmeter

    64k block size
    1gig test file
    Write: 122.872026 MBps
    Read: 137.795535 MBps

    Kick Ass :D

    Doing a 10gig file now
  11. These are awesome numbers, so... better check they're right.

    Are you sure this is RAID 5? :)

    \iobw.tst is ~ 10 GB now?

    1 GB ~ 2 million sectors
    10 GB ~ 20 million sectors

    # of Outstanding I/Os = 1?

    Normal execution for at least 1 minute?
  12. Here are my result files:

    RAID5
    http://www.islandpcservices.co.uk/dan/misc/raid5.htm

    Standalone Maxtor
    http://www.islandpcservices.co.uk/dan/misc/maxtor.htm

    Those were with a 1gig file and 2 mins run time. The speeds with bigger tst files is consistent.

    The RAID has some 300gig of data on it now so the test file would have been created near the middle of the disc. The Maxtor only has 10gig on it. I'd imagine if both drives were empty the speeds would be slightly faster.

    Let me know if you'd like me to do other tests.
  13. These are really great numbers for on-board RAID 5. Thanks for the post. I'm going to think about getting one for myself and then dropping this "you must get an add-on controller for good RAID 5 writes" business.

    The read performance test is not entirely reliable if you have a ton of RAM. At even around 60 MB/s, reading for 2 minutes = 7.2 GB read, which exceeds your test size significantly. So if the some or all of the data fit in cache, you'd get incorrect results.

    But it's harder to fake/cache out a write performance test (assuming that the data is changing / cache is not that smart to detect unchanged data for writes), and as the numbers roughly match a 2-drive RAID 0 expected performance, I'm starting to believe them.
  14. Cool, if you're happy then i'm happy :D

    Here is the 10gig test, for 10 mins, on the RAID after installing Exchange and a reboot.

    http://www.islandpcservices.co.uk/dan/misc/10gigraid.htm
  15. Livelee,

    Did you happen to enable write back cache on your 05 raid array? After reading Toms review on the ICH8R, I was really skeptical, because the highest I have seen before that is around 10-20 MB/s
    Here is Tom's Review:

    http://www.tomshardware.com/2007/01/03/the_southbridge_battle/index.html

    I noticed this one a couple of months ago, so you can compare:

    http://www.gamepc.com/labs/view_content.asp?id=raid505&page=1

    It's the only one I could see. The nvidia benchmark is consistent with tom's hardware scores, but the ICH7r is really off.

    You can check for write-back by opening up the intel raid software and right clicking on the volume.

    Here's a picture to show you where to look:
    http://www.xtremesystems.org/forums/showpost.php?p=1884737&postcount=21

    Have you had any other problems with the raid05 array??

    I would appriciate any information that you can offer.

    Thanks
  16. for the tests I had in on, but it's turned off now. No problems with the setup so far.
Ask a new question

Read More

Performance NAS / RAID Storage