Re: Configuring Oracle on RAID 5

Maybe the bozos that were explaining RAID to you were just salesman.

But, RAID 5 can work, and is used on major databases.

The problem is the amount of writes that are required and associated.

The entire problem is not RAID, it is a transaction per second you need to sustain.

So, you need to solve for transactions per second that your DB needs to maintain. There is a formula that Compaq uses to determine the amount of drives needed in a RAID 5 array to meet your transactions per second rate due to the amount of write operations needed to perform a transaction with RAID 5 due to overhead.

It may take 80 hardrives in a RAID 5 array to meet your transaction rate, but it is possible, that is the problem with RAID 5, the amount of write operations needed to stripe the information across the array, it isn't necessary that they can't do it, it is just the formula has to be crunched to meet the DB requirements of sustained transaction rate. Then, it will tell you how many drives you need in a RAID 5 array to meet this rate.

Keith

"Kirt Thomas" <kremovethisspamthingthomas_at_gfsiinc.com> wrote in message news:6a8sato8jt0gt8b6940dd6iuf4rvtufkbf_at_4ax.com...
> This whole RAID thing chaps my hide :) We were in the process of
> adding disk to our AIX (SSA and FCAL) system, and so, we decided to
> look at the new 'SAN' type products - IBM has it's Shark, and there
> are several others around. All of them however use RAID 5. We
> initially decided to take a pass, but a large company that's name
> starts with H, let us have there system on 90 take it or leave it
> basis. This system had a substantial write cache (6gb), but in
> overall performance, it was 'half as fast' as our existing FCAL disk
> subsystem (at RAID 0+1). (It equalled the SSA, but these were tired
> old 6000rmp SSA drives, that were faux striped). We ended up punting
> the E7xxx from H after they could find no way to get performance up to
> snuff. I found it humorous that the sales staff refered to RAID 0+1
> as RAID 10 - (both IBM and H).
>
> I saw an interesting implementation of RAID 4 (where the parity is
> written to only one drive) at OpenWorld. They made a pretty good case
> for this, mirroring the parity drive, and caching all writes. But you
> still have a write penalty eventually, and even with a large cache, it
> must be written to disk sometime :)
>
>
> On Tue, 13 Mar 2001 23:07:19 +1100, "Howard J. Rogers"
> <howardjr_at_www.com> wrote:
>
> >
> >Nothing to do with the number of disks. If it were, RAID-0 would be
> >interesting, would it not??!
> >
> >I'm not real hot on my RAID numbers, but IIRC (and I think I do) 0 is
> >striping, 1 is mirroring (hence RAID 0+1 is striping AND mirroring), and
> >RAID-5 is striping with parity. Real RAID gurus will tell you all about
> >RAIDS 3, 4 and 6, but if you're like me, you will be asleep before
 they've
> >finished.
> >
> >The numbers don'r relate to anything very much, just the standard to
 which
> >that flavour of RAID adheres.
> >
> >Incidentally, RAID 5 works very well with 3 disks. Though, since I have
> >shares in Seagate, I think it only fair to say that more is always
 merrier.
> >
> >Regards
>
Received on Tue Mar 13 2001 - 18:51:06 CST