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: RAID 5 vs RAID 10 benchmark

Re: RAID 5 vs RAID 10 benchmark

From: Howard J. Rogers <hjr_at_dizwell.com>
Date: Mon, 29 Nov 2004 20:20:52 +1100
Message-ID: <41aae9e3$0$24380$afc38c87@news.optusnet.com.au>


Jan wrote:
> Thank you for your comments.
> I did read maybe all these papers about RAID (BAARF etc.) before
> posting so I had no intention to provoke another general debate about
> all the already said points again (costs, "what if" scenarios etc...).
>
> I`m sorry that I was not exact enough about my question.
> What I wanted to know is that:
>
> If a write intensive database is placed on RAID 5 and there are some
> performance problems due to IO, and I have this set of parameters:
>
> 1. I have one logical volume from the storage like mentioned in above
> benchmark
> (IBM ESS 800)
> 2. I have to put all the database files (redo, archive redo, dbf,
> controlfiles) on this volume only.
> 3. No application design to reduce logical IO is possible etc.
>
> Question: Will be helpful to change from RAID 5 to RAID 10 to reduce
> the performance problems significantly? Please note that I cannot test
> the impacts and it (change from RAID 5 to RAID 10) will not be without
> costs.
>
> What I have up now is the only conclussions comming:
>
> A) from BAARF, and saying YES, it will help,
> B) from the mentioned IBM benchmark, and saying NO, it will not help
>
> Regards, Jan
>

Well, if you cannot test, then neither can anyone here. So it's pray and   play time, and one opinion is going to be about as helpful as any other.

Personally, I would say it is not going to help, because you're dealing with the same number of hard disks. The only thing you would be eliminating, it seems to me, is RAID5's write penalty... and as the discussion here mentioned (so it wasn't pointless) the write penalty is possibly the least significant of RAID5's drawbacks.

Given it (the change) is a lot of work. And given that you cannot design better. And given you cannot test beforehand. Then I'd say, leave well alone.

If you really want to eliminate an I/O bottleneck, I would do whatever it takes to separate redo from data files, and both from archives.

Regards
HJR Received on Mon Nov 29 2004 - 03:20:52 CST

Original text of this message

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