Oracle FAQ Your Portal to the Oracle Knowledge Grid
HOME | ASK QUESTION | ADD INFO | SEARCH | E-MAIL US
 

Home -> Community -> Usenet -> c.d.o.server -> Re: Oracle and RAID-5

Re: Oracle and RAID-5

From: Michael Brown <mike_at_mlbrown.com>
Date: Wed, 16 May 2001 22:53:41 -0400
Message-ID: <mge6gtsbmgkhsf2spa35ohvp9fbmju32e1@4ax.com>

RAID-5 requires at least 4 I/O operations per small write. For each small write operation, the current data and parity blocks must be read, changes to the parity must be calculated based upon the data changes and the new data and parity blocks need to be written. With hardware RAID, cache memory may be used to try to group i/o's in order to reduce the total number. The "at least" comes in with the fact that some newer RAID 5 implementations (HP's 7100, XIOtech, etc.) are allowing more than one parity block, so you have to add a write op for each additional parity disk. With a mirrored RAID level (1, 1+0, 0+1), you just have the two writes. During read operations, RAID 5 can service only one access to a single spindle at a time, but a mirrored RAID level, can handle two (the two mirrored copies). If you know the I/O pattern against the data, you can make a decent calculation of the amount of degradation in performance you will see in your environment.

On Thu, 17 May 2001 01:36:32 GMT, gedau_at_isa.mapson.com.au (George Dau) wrote:

>Andrew Mobbs <andrewm_at_chiark.greenend.org.uk> wrote:
>
>]RAID-5 gives poor write performance
>
>For us, the performance was not "poor". We looked at the cost of Mirror, vs
>RAID-5, and the performance difference, and went RAID-5. Yes, RAID-5 writes are
>slower than Mirror writes, but I could not describe the result as "poor".
>
>We found ardware RAID-5 faster than software.
>
>This wasn't on NT though.
  Received on Wed May 16 2001 - 21:53:41 CDT

Original text of this message

HOME | ASK QUESTION | ADD INFO | SEARCH | E-MAIL US