Re: RAID Groups and ASM - I've got a problem, I think
- From: Niall Litchfield <niall.litchfield@xxxxxxxxx>
- To: david.barbour@xxxxxxxxxxxxx
- Date: Tue, 6 Dec 2005 10:34:44 +0000
Are they duplicating the redundancy of the RAID system by using ASM
redundancy as well? In my view if you are using ASM with LUNs on a SAN or
other volume manager, you are better off declaring the redundancy as
EXTERNAL for the disk groups. You also mention Dell 2850s in the plural, is
this a RAC database? If so you'll get some other IO overhead (though not
much IME) from that alone.
Finally we found issues with firmware and driver versions with fibre
attached storage (though we run Windows and different hardware so clearly it
isn't a straight comparison, just another thing to consider.
On 12/5/05, David Barbour <david.barbour@xxxxxxxxxxxxx> wrote:
>
> Looking into a performance problem on RHEL 3.0 AS QU4 running on Dell
> 2850s with 8gB RAM and dual 3.6gHz PIV processors and a fibre-attached EMC
> CX-500.
>
> A simple 100000 row insert on an identical database running on a Windoze
> XP desktop with 1gB RAM and internal IDE drives takes 137 seconds. On the
> Linux box it takes 443 seconds!
>
> They're using ASM, and there's one diskgroup defined for the DB, that has
> six LUNS from four different RAID Groups on the SAN. I think the intent of
> the folks who installed it was to try to spread the IO across as many
> spindles as possible (laudable goal), but I don't think using ASM to to that
> is the best solution.
>
> Might I not have an issue with the overhead of striping across four
> different RAID Groups? When I run the insert, the cpu on the box never goes
> above 5% or so, but IO Waits are in the 60 - 70% range. I't's the only
> activity happening on the box , period. To make it somewhat worse, the
> numbers of LUNS from each RAID Group aren't consistent ( 2 from RG0, 2 from
> RG1, 1 from RG4 and 1 from RG5) and the sizes, although not radically
> different, aren't identical. The LUNS from RG1 are 150M vs. 166M for
> RG2/4/5.
>
> I really don't think ASM was designed (at least at this point - 10.1.0.4)
> to stand in as a substitute for metaluns at the storage level.
>
> Any thoughts?
>
>
--
Niall Litchfield
Oracle DBA
http://www.niall.litchfield.dial.pipex.com
Other related posts:
