Amazing RAID array performance

Discussion in 'DIY Computers' started by Jonathan Eales, Oct 20, 2004.

  1. I've just built my first striped RAID array and I can't believe how easy it
    was and the terrific performance I'm getting from it.

    I was ordering a couple of cheap 40GB hard drives (Maxtor, ATA133 7,200rpm)
    one to replace a crashed drive and one as a spare. I thought that I'd add a
    cheap Silicon Industries 0649 PCI RAID card for an extra tenner and play
    with it. Easy to install, setup the striped two disk array in the BIOS, add
    the driver in Windows XP and then format it as NTFS.

    I tested it with Sandra and it says a bandwidth of 93MB/s! My C: drive
    (80GB Samsung 7,200) only gets 29MB/s. OK so the RAID drive is completely
    empty but that shows an amazing improvement over a single drive. According
    to Sandra examples I should be getting 65MB/s.

    Is this typical for RAID arrays? How reliable are they in reality? Do many
    folks out there use them for desktops?
     
    Jonathan Eales, Oct 20, 2004
    #1
    1. Advertisements

  2. Jonathan Eales

    [ste parker] Guest

    I'm thinking about doing this myself with two SATA drives, but would I
    be correct in thinking that with RAID 0 you're doubling the chance of
    losing all your data, if one of the drives goes thats it, right?

    I must admit I've never had a HD fail (but then I've only ever had 4 or
    5!), nor do I have much important data that I don't have backed up
    either (any new space I'd probably only use to install games into and
    perhaps make a copy of all my CD's or something).
     
    [ste parker], Oct 20, 2004
    #2
    1. Advertisements

  3. Jonathan Eales

    Doug Ramage Guest

    I assume you are referring to RAID0?

    You may find that your actual experience will not live up to the
    benchmarks - depending on your system usage.

    Also, RAID0 (as opposed to RAID1) will double your chances of failure, so
    it's not advisable to hold mission critical data on the array.

    I tried RAID0, but found 2 single SATA drives to be about the same or
    slightly faster. My data is held on a separate hard drive.
     
    Doug Ramage, Oct 20, 2004
    #3
  4. Jonathan Eales

    logized Guest

    Raid 0 gives the performance increase at the cost of fault tolerance - if
    either disk fails then you lose everything.
    If you intend to store important files on there, then back-up regularly or
    buy two more similar drives and create a Raid 0+1 if the controller supports
    it. (Mirrors your raid 0 onto the two extra drives so get almost as high
    performance plus improved fault tolerance) - see
    http://www.acnc.com/04_01_0_1.html

    Dave
     
    logized, Oct 20, 2004
    #4
  5. Jonathan Eales

    Alan Guest

    Why not just buy one more drive and do RAID 5 ?
     
    Alan, Oct 20, 2004
    #5
  6. Jonathan Eales

    Alex Fraser Guest

    Pretty good, but throughput is only one measure of performance. General
    performance is more influenced by access time, IME. RAID 0 doesn't improve
    this, and may actually make it worse depending on a number of factors.
    Half as reliable as a single drive, assuming no driver bugs.
    Many who probably shouldn't ;).

    Alex
     
    Alex Fraser, Oct 20, 2004
    #6
  7. Jonathan Eales

    Philip Jones Guest

    The message <>
    RAID coverage always comes down to cost in the end.
    You have to decide if recovering from a disaster is worth the cost of
    purchasing extra HDD's and possibly a dedicated RAID controller.
    Using multiple HDD's for RAID-0 gives great performance increases, but
    for every drive you add to the RAID-0, you add extra possibility
    to loosing everything on your system.
    RAID1 or even better RAID0+1 gives you security but has the burden of
    the extra cost.
    In my mind RAID-5 is the most economic whilst giving you reasonable
    security from a single HDD failure.
    As always it's 'swings and roundabouts', and you get what you pay for.
    If you want performance "and" security, then RAID 0+1 with as many
    drives as you can afford is the way to go. Few people go down this route
    because of cost.
    Best of luck, Phil..
     
    Philip Jones, Oct 20, 2004
    #7
  8. Jonathan Eales

    logized Guest

    I think the controller card mentioned by the OP only supports raid 0, 1 and
    0+1.
    Raid 5 is a common solution used on servers which use a more complex card to
    handle parity and may have non-volatile memory backup should a fault occur
    before data is completely written across the drives.
    Also, Raid 5 is much slower than raid 0.

    Dave
     
    logized, Oct 20, 2004
    #8
  9. Actually it'll be worse. You have to wait for BOTH drives to reach the
    right sector (in rotation .. track-to-track head seek should be about
    the same) .. by definition this can be no better than the best one (is
    in fact as bad as the worst one). So your normal 'allow half a rotation
    on average' becomes rather worse than that (more like 3/4 of a rotation
    on average, iirc).
     
    GSV Three Minds in a Can, Oct 20, 2004
    #9
  10. Jonathan Eales

    Alex Fraser Guest

    According to my model, for reasons I don't understand, a stripeset of N
    drives gives an average of N/(N+1) of a revolution. I'd love to know why.

    For 7200rpm drives, 2/3 of a revolution as opposed to 1/2 equates to ~1.4ms
    extra, or around 10% on top of the random access time for a single drive.

    There are two factors that are harder to put figures to that I can think of.
    First, typical access patterns are likely to be "closer" than random access,
    which would increase that percentage. Second, some gain is possible when
    requests fit inside a stripe, decreasing the percentage (but only if the
    controller is clever enough; I think there's a good chance cheaper ones
    aren't).

    Alex
     
    Alex Fraser, Oct 20, 2004
    #10
  11. Jonathan Eales

    Trevor Best Guest

    Yes. But then if you had two drives separate (or just the one drive) and
    one failed, you'd lose all the data on the failed drive so it's not a
    lot of difference there except in the case of RAID-0 there's the
    frustration of knowing there's still one perfectly good disk left and
    the data is still gone.

    Backups are as important on any disk set up, including RAID-1
    (mirroring) as HDD failure is not the only cause of lost data. I'd
    hazard a guess that more data is lost through software errors than
    hardware and in the case of a software error, the data loss is
    duplicated immediately on a mirrored system so mirroring only protects
    against one disk failing, i.e. one possible failure out of many that can
    destroy data.
    Having the OS and the swap file (or partition depending on OS) on a
    RAID-0 array is OK, programs too, any hardware failures and these can be
    re-installed. This is probably good for Windows anyway as that tends to
    get tired with old age and slows right down.

    (the next bit is my RAID-0 disaster story)

    I did have a SQL Server at work that started life with RAID-0+1 but
    after bits going back to the shop for RMA and certain disks not being
    returned in time it ended up with 2 RAID-0 arrays, I used to ghost one
    to the other every couple of days or so until the disks on the spare
    array started getting pilfered and put into other boxes then bang! One
    disk failed and the whole caboodle was gone.

    There was a daily tape backup and the failure happened over a weekend so
    on Monday I went to Friday's tape to restore the databases onto another
    machine but that decided to chew the tape so it was back to Thursday's.
    The starTeam server decided it wouldn't play with Friday's vault data
    and Thursday's SQL data so development was being held up while we
    figured out what the starTeam server needed. As paying client's can't be
    held up with our silly little problems we ended up creating a new
    project in starTeam and losing source control history on our project,
    given a little more time we would have sorted it out.

    What were the chances of two disasters in one sitting? HDD failure
    followed by the backup tape being chewed. The odds must be phenomenal
    but it happened anyway.
     
    Trevor Best, Oct 21, 2004
    #11
  12. Jonathan Eales

    Trevor Best Guest

    I'm not aware of any software RAID-5, at least not on a controller. (I
    know Win NT could do it just by having 3 drives attached, but then the
    whole array was only available to that NT system, not sure you could
    transfer the array to another machine).
     
    Trevor Best, Oct 21, 2004
    #12
  13. Jonathan Eales

    Trevor Best Guest

    Dorothy Bradbury wrote:

    [on RAID-0]
    That's a bit harsh, zero redundancy yes but reliability is a little
    better than that.
    I like that :)
    I have a cheapo (£13) DabsValue RAID-1 on our mail server (also DC, DNS,
    DHCP), the impact of the rebuilds on that wasn't noticable. I would
    expect on a SQL Server or File Server it would be noticable. It (the
    Dabsvalue) also supports hot spares, which is good considering the price.
     
    Trevor Best, Oct 21, 2004
    #13
  14. Jonathan Eales

    Trevor Best Guest

    Yes I know, all the cheap controllers are like this, however I've not
    seen one that does RAID-5, only 0,1 & 0+1.
     
    Trevor Best, Oct 22, 2004
    #14
  15. Jonathan Eales

    Dave Guest

    In
    The hpt374 based controllers from highpoint (rocketraid 404/454) support
    RAID5 and they are only software based cards.

    Cheers

    Dave
     
    Dave, Oct 22, 2004
    #15
  16. Jonathan Eales

    Philip Jones Guest

    The cost of controllers. Having said that, modern CPUs can do software
    As I've said before, "You Get What You Pay For" .!!
    Yes you can do Software Raid at levels 0 and 1, but if you want
    "Serious" protection then you purchase a hardware RAID controller and
    get RAID-5 or RAID-10.
    It all comes down to the simple equatition of what it would cost you if
    you "TOTALLY" lost All your data. If the cost of re-creating All your
    Data is more than the cost of purchasing a Hardware RAID controller and
    an extra HDD or two, then the answer is simple - Purchase a decent
    Protection System ...
    Phil..
     
    Philip Jones, Oct 22, 2004
    #16
  17. Thanks for all your posts to my original message about your experiences with
    RAID, especially RAID0.

    I've decided not to implement it on my main system as the system disk, but
    to wait and build a second system which I can use as a data/media/games
    store that is less of a hassle if it fails with a disk crash. That will
    give me the confidence to incorporate and recommend it in systems that I
    build for other people.

    I'm still amazed at the 'streaming' performance of this I/O technology and
    for the low cost and hassle levels that it offers.

    Thanks again everyone,
    Jonathan
     
    Jonathan Eales, Oct 23, 2004
    #17
  18. Jonathan Eales

    Philip Jones Guest

    Jonathon, quite a coincidence - but my first call on Monday is to a
    Compaq ML370 server. Apparently this Server Running Novell and being the
    Main Database server fell over with one of three RAID Disks having
    failed.??? Yep, you've guessed it. It had been set up as RAID-0 because
    of the fantastic speed over their LAN.!!
    The system went down Midday Friday, and will not be up much before
    Tuseday morning.
    That's 1 and 1/2 days for a company to be down. In this case a lot more
    than the price of three additional drives to make RAID-1.
    Phil..
     
    Philip Jones, Oct 24, 2004
    #18
  19. Jonathan Eales

    Trevor Best Guest

    My SQL Server (RAID-0) was down longer. By contrast, I had a disk
    failure in our mail server too, this was RAID-1, downtime was as long as
    it took to change a disk as the downtime was planned.

    hmmm! if you have a failure you'll re-evaluate that last statement :)
     
    Trevor Best, Oct 26, 2004
    #19
  20. Jonathan Eales

    Philip Jones Guest

    Jonathon, quite a coincidence - but my first call on Monday is to a
    It's a sad fact that this is often the point where small companies
    realise the error of their ways. For the sake of an extra 3 or 4 hundred
    pounds, many small companies choose not to have redundancy on their
    servers. Unfortunately because of my job, I see all the sad cases where
    systems have failed....
    The most extreme example I ever saw was approx five years ago. This was
    a fairly large Kitchen Manufacturer in the NW of England. Their
    personnel and wages server had failed to boot one morning (at the end of
    the month to make things worse). The <Single> SCSI HDD was stuffed. I
    replaced this and asked how were they going to restore their system, as
    there was No Backup device on the machine........
    I received blank looks, turning into horror when I told them "No", I
    could not magic up a backup to restore their system. I heard later that
    it took them nearly two weeks to get back to anything like what they had
    before the HDD failure. Once again, <Many> times more than the cost of
    putting in some redundancy and backup.
    Phil..
     
    Philip Jones, Oct 26, 2004
    #20
    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.