Re: RAID and Oracle

Bruce Pihlamae wrote:
>
> David E. Daniel wrote:
> > I have been attempting to perform some research regarding the use of
> > Oracle in a RAID environment. The university at which I am employed is
> > moving my Oracle from an Intel/NT platform to a Midrange Unix
> > evironment. In the process, they are setting me up with whatever disk
> > setup I desire.
> > In my research, I have found out a few things.
> > - Oracle states that you should not use RAID.
>
> You may be misquoting them. I've heard recommendations NOT to use RAID5
> because of its write overhead but RAID 0+1 is in common use.
>

I have also not seen any such advice from Oracle regarding RAID ??

> > - RAID 5 works best for a data warehouse but only for DBs that are more
> > read intensive.
>
> RAID5 works well and is suited to ANY 90% or better READ requirements.
> RAID5 is a cheap reliability configuration, NOT a performance
> configuration and should be selected carefully.
>
> RAID5 will protect against disk failure but NOT against bad block media
> failure where the disk block and not the disk as a whole fails.
>
> To operate "properly" (database recovery wise) an Oracle database needs
> to separate CONTROL files to different disk devices, REDO Logs to a
> separate device(s), Archived REDO Logs to separate device, and ideally
> separate tables from indexes. Most RAID5 devices give you all your
> storage in one huge virtual disk and you cannot separate these database
> components without buying multiple RAID5 boxes.
>

This is excellent advice. As you may have "whatever you desire", my recommendation
is RAID 0+1, that is striped and mirrored. Good performance - read&write, plus
redundancy - must be a good deal.

If that seems too expensive, RAID 5 can still be an option. I have experience of Oracle
running with RAID 5 perfectly well, but we threw in Prestoserve. This is a type of write cache
which reduced the impact of disk latency (prob with write on RAID 5).

My experience is that Prestoserve dramatically improves performance. The factor improvement
is of course dependent on the %write of the particular DB.

Good luck ! Received on Fri Nov 08 1996 - 00:00:00 CST