Find the answer to your Linux question:
Results 1 to 3 of 3
Enjoy an ad free experience by logging in. Not a member yet? Register.
  1. #1

    Software raid 5 write performance

    I wasn't sure where to post this question so administrators, feel free to move it.

    I have a media server I set up running Ubuntu 10.4 Server (I posted here because I don't think the distro is very relevant), and I set up a software raid 5 using 5 Western Digital Caviar Green 2TB 7200RPM 64MB drives. Individually they benchmark (using the Ubuntu's mdadm GUI (pali?somthing...) at about 100-120mb/s read write.

    I set the raid 5 up with a stripe size of 256kb, and then I waited the 20 hours it took to synchronize. My read speeds in raid are up to 480mb/s, but my write max is just under 60mb/s. I knew my write performance would be quite a bit lower than my read, but I was also expecting at least single drive performance. I have seen other people online with better results in software, but have been unable to achieve the results they have gotten.

    My bonnie++ results are more or less identical (I used mkfs.ext4 and set the stride and stripe-width).

    The PC has 2048mb of RAM and a 2.93Ghz Dual Core Pentium (Core 2 Architecture), so I doubt think that's the bottle neck. These drives are on the P55 South Bridge SATA controller.

    Anyone know what I am doing wrong or have any suggestions?


  2. #2
    Linux Guru Rubberman's Avatar
    Join Date
    Apr 2009
    I can be found either 40 miles west of Chicago, in Chicago, or in a galaxy far, far away.
    Actually, you are seeing what is about the correct write speed on this setup. The best you can expect is about 4/5 (80%) of nominal single-disc speed with a RAID-5 configuration, and if the RAID parity generation and such is done in software, then you can expect about another 20% performance hit, depending upon your system hardware and memory. FWIW, RAID-5 is infamous for lousy write performance, which is why it is only recommended for read-mostly applications.
    Sometimes, real fast is almost as good as real time.
    Just remember, Semper Gumbi - always be flexible!

  3. #3
    I figured it out. The problem was Palimpsest (the GUI program). It partitions the drives and creates an array out of the partitions. mdadm from command line by default does not. On a test array I was fiddling with using Palimpsest I got write of 35mb/s, using mdadm I got 90mb/s. So the bottom line is, don't use palimpsest...

  4. $spacer_open

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts