From: Michael Austin on
Now that we have sufficiently beat the RAID5 horse to death, let's add a
twist to the debate. What is your experience either in the lab or real
sites where you have the virtualization at the array level (pick your
RAID level) in addition to the virtualization provided by ASM in an
Oracle eBusiness environment.

Let's say you have Tier1 storage (high end whatever...) you have 50
spindles in a RAID0+1 providing ~12.5TB of raw storage that we carve out
100 50GB LUNS and present them to our ASM environment. How, in your
testing and research, does performance either improve or suffer with the
addition of ASM. The database is scattered across 6 separate disk
groups with logfiles etc in their own disk group.
From: AlexB on
On Mon, 15 Dec 2008 21:03:53 -0600, Michael Austin <maustin(a)firstdbasource.com>
wrote:

>Now that we have sufficiently beat the RAID5 horse to death, let's add a
>twist to the debate. What is your experience either in the lab or real
>sites where you have the virtualization at the array level (pick your
>RAID level) in addition to the virtualization provided by ASM in an
>Oracle eBusiness environment.
>
>Let's say you have Tier1 storage (high end whatever...) you have 50
>spindles in a RAID0+1 providing ~12.5TB of raw storage that we carve out
>100 50GB LUNS and present them to our ASM environment. How, in your
>testing and research, does performance either improve or suffer with the
>addition of ASM. The database is scattered across 6 separate disk
>groups with logfiles etc in their own disk group.

While not being a very large database shop, all I can say about ASM on top of
RAID is that it uses extra CPU, but makes DBA life a bit easier (no need to
specify datafile names and easier to create new instances, for example). We are
using IBM DS4000 series SAN with RAID10 LUNs presented (through VIO server) to
AIX/Linux LPARs on a pSeries 570. AIX LPARs are using direct access to raw
devices by Oracle 9iR2 processes, while Oracle 10gR2 on Linux uses ASM. If we
look at 'sar' output - we can hardly notice the overhead (our CPUs are 1.65GHz
POWER5), while looking at nmon/topas screens during data loads or queries we can
see ASM processes taking some 7-10% of CPU at peak times. Did not encounter any
reliability matters so far.
We are about to install a new p570-MMA and I will be definitely going with ASM
on AIX LPARs as well.

On the RAID5 matter - I do not see anything terminally wrong with RAID5 even
after reading the whole thread, but we only use RAID5 LUNs for disk backups.

Alex
From: Robert Klemme on
On Dec 17, 10:11 am, AlexB <b...(a)pisem.net> wrote:
> While not being a very large database shop, all I can say about ASM on top of
> RAID is that it uses extra CPU, but makes DBA life a bit easier (no need to
> specify datafile names and easier to create new instances, for example). We are
> using IBM DS4000 series SAN with RAID10 LUNs presented (through VIO server) to
> AIX/Linux LPARs on a pSeries 570. AIX LPARs are using direct access to raw
> devices by Oracle 9iR2 processes, while Oracle 10gR2 on Linux uses ASM. If we
> look at 'sar' output - we can hardly notice the overhead (our CPUs are 1.65GHz
> POWER5), while looking at nmon/topas screens during data loads or queries we can
> see ASM processes taking some 7-10% of CPU at peak times. Did not encounter any
> reliability matters so far.

This is in line with an article I recently stumbled across:

http://www.linuxjournal.com/article/8539#N0xa50890.0xb3d4b8

Cheers

robert
From: robert on
Michael Austin wrote:
> How, in your
> testing and research, does performance either improve or suffer with the
> addition of ASM. The database is scattered across 6 separate disk
> groups with logfiles etc in their own disk group.

I've no numbers to contribute (though I'm also very interested in the
collective experience of this group on this).
But I do have a question: how much would we reasonably *expect*
performance to be impacted one way or the other by ASM, when we are on a
high performance SAN?
From: DA Morgan on
robert wrote:
> Michael Austin wrote:
>> How, in your
>> testing and research, does performance either improve or suffer with
>> the addition of ASM. The database is scattered across 6 separate disk
>> groups with logfiles etc in their own disk group.
>
> I've no numbers to contribute (though I'm also very interested in the
> collective experience of this group on this).
> But I do have a question: how much would we reasonably *expect*
> performance to be impacted one way or the other by ASM, when we are on a
> high performance SAN?

On a high end SAN you are reading and writing to a cache not to disk so
the number of spindles becomes less important. This is part of what
makes NetApp's RAID4 different from RAID4 from other vendors.

ASM seems to prove either a neutral or slightly positive impact (a
few percent) on performance if compared to raw disk with no LVM. If
comparing with other LVMs what we see is a substantial drop in CPU
utilization. Most LVMs are cpu pigs: ASM is not.
--
Daniel A. Morgan
Oracle Ace Director & Instructor
University of Washington
damorgan(a)x.washington.edu (replace x with u to respond)
Puget Sound Oracle Users Group
www.psoug.org