2

I have 8 hard drives in my PC, assigned as storage.

  • Three of them are attached to the mainboard.
  • Four of them are attached to a controller [ SiL 3114 ]
  • One of them attached to an another controller [ SiL 3114 #2 ]

These hard drives are the very same, Samsung Spinpoint models. Same RPM, same size.
I wanted to make a raid from ALL of them, using Windows Server 2008R2. However, the performance is just outright terrible. It's hardly faster than one hard drive.

I checked the SMART values, they are all correct.
Any ideas, what might be the problem?

Ps.: There is a spare 320gb WD disk in the machine, and a 320gb WD as system drive.

Improve this question
  • 3
    Can you quantify "horrible" with some benchmark results? Also, this might be better suited for superuser as software raid over a bunch of controllers is something you might find more frequently in a enthusiast rig than a production server. Also what RAID (i.e. Raid 5 or 10) are you using? – Kyle Brandt May 1 '12 at 12:27
  • I tried to ask a raid specific question there (just about how to measure performance). But received on upvotes or answers. So I thought Serverfault may be more experienced with RAID builds. – Apache May 1 '12 at 12:29
  • Performance test: i.imgur.com/iEqSY.png | Same test, on two WD Black Caviar drives in RAID0: i.imgur.com/aq2Zl.jpg – Apache May 1 '12 at 12:31
  • 2
    How big is the stripe size on your array - a 64k stripe size on 8 7200RPM disks would give you a theoretical sequential read performance of around 230MB/sec if you get one stripe per revolution of the disks. SW RAID performance on windows is pretty crap at the best of times, so it's certainly within the bounds of possibility that this is a legit result for the platform. – ConcernedOfTunbridgeWells May 1 '12 at 13:39
  • As a corollary, H/W RAID controllers can be had quite cheaply on ebay. If you want more performance, try looking at what an adaptec, LSI or 3Ware controller would cost. Adaptec ones will go up to 1MB stripe sizes. – ConcernedOfTunbridgeWells May 1 '12 at 13:43
5

You clearly have a bus bottleneck problem. Your motherboard can't manage more than one transfer at once.
You can test it:
- boot a Linux lice-cd
- run a dd test againt 2 disks

sync; echo 3 > /proc/sys/vm/drop_caches
(dd if=/dev/sda of=/dev/null bs=1M count=5000 &); dd if=/dev/sdb of=/dev/null bs=1M count=5000

You probably will see a cumulated transfer of about 120 Mb/s.

Improve this answer
  • 2
    Exactly. Half the I/O is to the controller with half the drives on it. That controller is on a 32-bit, 33MHz bus with a maximum theoretical rate of 133MB/s and a realistic transfer rate of 100MB/s. So 200MB/s is his I/O limit. (Worse if the two SIL controllers share a PCI bus, which they probably do.) – David Schwartz May 1 '12 at 20:20

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.