Re: Physical layout of a DB in RAC environment

From: Paul Drake <drak0nian_at_yahoo.com>
Date: 29 Jan 2004 16:12:26 -0800
Message-ID: <1ac7c7b3.0401291612.40728c3d@posting.google.com>

Roman Klesel <rupa_at_firemail.de> wrote in message news:<bvb4c8$2ku$1_at_ork.noris.net>...
> Dear DBAs,
>
> from what I have learned one should keep things seperate in an
> produktive Oracle DB:
>
> online redologs:
> on 2 different dedicated physical disks (raid sets), groups swiching
> from one disk to another
>
> archive logs:
> on another dedicated disk (raid set)
>
> datafiles for undo/redo TSP on RAID 1
> datafiles for temp TSP on RAID 1

tempfiles

> other datafiles on RAID 5
>
> controlfiles 3 all on different physical disks.
>
> Now I'm confronted with a setup where all data is stored in a T1 storage
> system which internally does a RAID 5. Everything in this storage is
> software mirrored in another T1.

some might call this RAID 15

> Both RAC nodes use this storage system for all oracle relevant files
> redologs, archivelogs, controlfiles ...
>
> Plus the shadow DB uses this storage system for all its files too.
>
> Everything is connect through fiberchannel switches.

I hope its multipathed.

> Is this insane?

Check out James Morle's paper "Sane SAN" up on OakTable http://www.oaktable.net or BAARF http://www.baarf.com

There are papers up there also that discuss RAID, but it seems that the RAID config is already determined, out of your control, etc.

> Or does it just look mad to me? Or is this the modern
> way of doing things?
>
> Can you please comment on this!
>
> Greetings Roman

Roman,

Do you get to keep a backup set on something other than the T1 units? (and I don't mean tape).
use at least 4 redo log groups, with at least 2 members per group. size the logs so that the log switches co-incide with your recovery objectives.
If you want a log switch every 60 minutes and all of your transactions occur during an 8am to 6pm window, then you could roughly estimate your log size from:

select to_char(completion_time,'Day HH24') day_hour, count(1),

       trunc(sum(blocks*block_size)/1048576) mb,
       round(trunc(sum(blocks*block_size)/1048576)/count(1),2) avg_log

from v$archived_log
where substr(to_char(completion_time,'Day'),1,1)!='S' and to_char(completion_time,'HH24') between '08' and '18' group by to_char(completion_time,'Day HH24') /

and set the parameter archive_lag_target to the switch interval in seconds (e.g. 3600).

If you size the online redo logs too small, you'll introduce more log file switches, and therefore have more checkpoints occur, which includes overhead.

hth.

Pd Received on Thu Jan 29 2004 - 18:12:26 CST