RAID version performance comparisons

Discussion in 'PC Technical Talk' started by Graham J, Jul 3, 2009.

  1. Graham J

    Graham J Guest

    I've read this up on the web and want to clarify my ideas - can I run this
    past you for comment, please?

    This is RAID for use in a server where a UPS guarantees power and an orderly
    shutdown. I assume a hardware RAID controller. Write performance I assume
    to be asynchronous, i.e. totally managed by the controller so in general it
    won't be affected by the number of disks or the RAID style.

    What I'm interested in is read performance - I imagine this will have the
    most effect on the performance as perceived by users.

    Data is stored in stripes across all the disks. I don't know whether a
    stripe contains a block from each disk, or a track, or a cylinder; but I
    presume a cylinder. It follows that performance is in any case only
    improved where the data volume read exceeds one stripe - for smaller volumes
    the disk rotational latency will be the overriding consideration.

    What I'm after is a general rule of thumb comparing RAID1 and RAID 5, as
    follows:

    RAID 1 improves disk performance by up to a factor of 2 since a stripe is
    completely read after the rotational latency of the second disk.

    RAID 5 improves disk performance roughly by a factor of the number of disks
    less 1. As above, the stripe contains data from all the disks, and is again
    read after the rotational latency of the last disk. But the descriptions
    I've seen show that within the stripe, data from 1 disk is parity (so that
    no further read operation is required to replace the data from a failing
    disk with calculated data). On the assumption that disks don't fail very
    often, the performance degradation of having to read data from a second
    stripe to get the parity I would have thought would be acceptable - but
    there ...

    So RAID1 and 2 disks should compare with RAID 5 and 3 disks. Given that a
    controller for RAID 5 is generally more expensive than one for RAID 1 it
    would seem that RAID 1 represents good value unless you actually need 4 or 5
    disks for RAID 5 - even then performance might be improved at better value
    by choosing faster SAS disks rather than the slower SATA disks.

    Any comments, please?
     
    Graham J, Jul 3, 2009
    #1
    1. Advertisements

  2. Graham J

    Graham J Guest

    I'm reposting this because I see already I have a failure to understand the
    performance improvement offered by RAID1 ...

    I've read this up on the web and want to clarify my ideas - can I run this
    past you for comment, please?

    This is RAID for use in a server where a UPS guarantees power and an orderly
    shutdown. I assume a hardware RAID controller. Write performance I assume
    to be asynchronous, i.e. totally managed by the controller so in general it
    won't be affected by the number of disks or the RAID style.

    What I'm interested in is read performance - I imagine this will have the
    most effect on the performance as perceived by users.

    Data is stored in stripes across all the disks. I don't know whether a
    stripe contains a block from each disk, or a track, or a cylinder; but I
    presume a cylinder. It follows that performance is in any case only
    improved where the data volume read exceeds one stripe - for smaller volumes
    the disk rotational latency will be the overriding consideration.

    What I'm after is a general rule of thumb comparing RAID1 and RAID 5, as
    follows:

    RAID 1 improves disk performance slightly. A stripe is stored on one disk.
    The same stripe is stored on the second disk. The stripe data is available
    after the rotational latency of the first disk. The virtue of RAID 1 is
    reliability i.e. it will continue to work normally if one disk fails.

    RAID 5 improves disk performance roughly by a factor of the number of disks
    less 1. The stripe contains data from all the disks, and is again read
    after the rotational latency of the last disk. But the descriptions I've
    seen show that within the stripe, data from one disk is parity (so that no
    further read operation is required to replace the data from a failing disk
    with calculated data). On the assumption that disks don't fail very often,
    the performance degradation of having to read data from another stripe to
    get the parity I would have thought would be acceptable - but there ...

    Given that a controller for RAID 5 is generally more expensive than one for
    RAID 1 the question is what value to put on the increased performance of
    RAID 5 ??

    Any comments, please?
     
    Graham J, Jul 3, 2009
    #2
    1. Advertisements

Ask a Question

Want to reply to this thread or ask your own question?

You'll need to choose a username for the site, which only take a couple of moments (here). After that, you can post your question and our members will help you out.