Re: RAID and Oracle

From: Marty Himmelstein <marty.himmelstein_at_valley.net>
Date: 1996/03/18
Message-ID: <314D80A8.145C_at_valley.net>#1/1


George Dau wrote:
>
> steve.miles_at_ci.seattle.wa.us wrote:
>
> ]We are planning to set up several new disks for our Oracle database. I'm a little confused about the
> ]issue of RAID. Our database will be used for both data entry applications and read-only queries.
> ]Obviously the nice thing about RAID 5 is the potential for hot-swapping disks (and little down time) in
> ]case of media failure. However, I hear that there is a performance hit when implementing RAID

For data warehouse applications, I have determined that RAID is more trouble than it is worth.
My views are specifically applicable to Parallel Query Option (PQO), which is the direction that large warehouses are headed. My basic reasoning is:

  1. Oracle likes lots of spindles. RAID uses lots of spindles, but Oracle loses control on data placement.
  2. Standard RAID: Reliability is inversely proportional to the number of disks in a stripe. 6 disks, 6 times the chance of failure as a single disk. Disks are much larger, and more reliable than when the initial RAID concept was introduced. If your tables are very big, why is RAID better than explicitly striping the table across several datafiles? If you lose a single 4Gig drive, restoring it can be painful, but if you lose a single disk in a 6 disk stripe, even with hot swap, you will take an ugly performance hit, anyway.
  3. If reliability and availability is an issue, as it should be, RAID can not replace sound backup and recovery procedures. Again, backing up and restoring a single disk (if failure is not hw related) will be easier if the volume is smaller rather than larger. Disk failures are responsible for a minority of failures.
  4. Performance tests I've done have steered me away from using RAID (even RAID 0) along with Oracle in a data warehousing environment. There are several technical reasons for this, if you are interested, send me a note.

Marty Himmelstein
marty.himmelstein_at_valley.net Received on Mon Mar 18 1996 - 00:00:00 CET

Original text of this message