Re: RAID 0+1 sofware or dedicated adapter for 7025 F50?

From: Steve Greatbanks (steve_greatbanks_at_hotmail.com)
Date: 10/04/03

  • Next message: jean-marc.pieau: "Re: hyperterminal connexion on RISC/6000 with Null Modem Cable" 
    Date: Sat, 4 Oct 2003 08:38:41 +1000

    "Bill" <bverzal@komatsuna.com> wrote in message
    news:51157c68.0310031250.422a0fdc@posting.google.com...
    > Collin McClendon <cbmcclendon3aat@netscapedot.net> wrote in message
    news:<blk2h3$1tj$1@murdoch.acc.Virginia.EDU>...
    > > Hello all,
    > > My company is currently looking at boosting our performance on our main
    > > data processing server, a 4 way 7025-F50 (just went from 1 way to 4 way
    > > last week). We use Unidata which from what I have read uses much random
    > > I/O. I am thinking of buying 6 more disks which are faster than the 7200
    > > RPM ones (10K) and either using LVM to do RAID 0+1 or get two RAID
    > > adapters for a duplex 0+1 RAID setup. with 4 processors will my overhead
    > > for RAID 0+1 be too much? Will I get a huge increase in performance with
    > > a dedicated RAID adapter with this config? (Basically two 3 disk
    > > stripes mirrored across two non raid LVD controllers vs the two 3 disk
    > > stripes created via hardware RAID and mirrored using LVM (we have to
    > > split mirrors every day so we can backup our data)
    > > Thanks for any advice,
    > > Collin
    >
    > Hardware RAID is far faster than LVM RAID. RAID 10 (or 0+1) is
    > fastest for writes. Raid 5 is fastest for reads.

    Misinformation. A read from a RAID10 device should be faster than a read
    from a RAID5 device for the obvious reason that the RAID5 has a disk worth
    of distributed parity information compared to the RAID10.
    In short, the only advantage RAID5 has over RAID10 is you get more usable
    disk space from a given set of disks. You lose out in both read and write
    speed.

    In response to the original poster's question, the LVM mirroring (and more
    particularly the mirror write consistency) will slow you down in either
    situation. I agree that the hardware option would probably be somewhat
    quicker (especially if the RAID controllers have a decent cache on them),
    but you will be constrained by the LVM waiting for both writes to complete.
    If you have the money, go with the hardware RAID.

  • Next message: jean-marc.pieau: "Re: hyperterminal connexion on RISC/6000 with Null Modem Cable"

    Relevant Pages

    • Re: SATA RAID 5 Performance
      ... RAID5 certainly isn't designed to perform write operations as fast as ... but then again writes aren't written transactionally to disk on ... but RAID5 often cuts the mustard for many small to mid range transaction ... Are there any RAID vendors I should look ...
      (microsoft.public.sqlserver.server)
    • Re: filesystem: 12h to delete 32GB of data
      ... RAID5 is good for sequential reads -- eg. ... blob of data up to the CPU. ... -- which works a great deal better on RAID10. ... For most H/W raid that equates to ...
      (freebsd-performance)
    • Re: filesystem: 12h to delete 32GB of data
      ... RAID5 is good for sequential reads -- eg. ... blob of data up to the CPU. ... -- which works a great deal better on RAID10. ... For most H/W raid that equates to ...
      (freebsd-questions)
    • Re: Software RAID partitions not detected at boot time with linux 2.6.25.1.
      ... md: Autodetecting RAID arrays. ... raid5: device sdc2 operational as raid disk 2 ... Subject: block: do_mounts - accept root=<non-existant partition> ...
      (Linux-Kernel)
    • Re: What is the most widely used RAID type for Oracle databases?
      ... Is RAID 10 pretty much now the rule of thumb? ... I think RAID5 is the most widely used these days. ... buffer saturation and degraded hardware situations (ie, bad disk). ... technical superiority or proper requirements analysis. ...
      (comp.databases.oracle.misc)
    Loading