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: Physical Layout of disk to use Oracle

Re: Physical Layout of disk to use Oracle

From: Howard J. Rogers <howardjr2000_at_yahoo.com.au>
Date: Thu, 18 Sep 2003 08:12:19 +1000
Message-ID: <3f68dd14$0$28117$afc38c87@news.optusnet.com.au>


Hari Om wrote:

> Hi Howard,
>
> What about this approach. I have two 18GB hdisk (mirrored with OS
> installed on it) and SIX 36GB hdisk.
>
> It seems that since SYSTEM and REDO Log files are "NOT GOOD" for
> RAID 5, I came up with following. Instead of RAIDing (level 5) on 6
> hdisk I would RAID only 3 disk and use them soley for DATA files.
> So here is my layout:
>
> hdisk0 --> AIX OS + Control File1 + Redo Log (Group 1)
> hdisk1 --> AIX OS + Control File2 + Redo Log (Group 2)

I take it these are the two 18GB hard disks, in a RAID0 (mirroring) setup.

Good, Redo Logs will do well on a RAID0 setup, but they won't appreciate sharing that with the operating system.

But your redo log configuration looks a bit dodgy anyway. Either that, or your terminology is wonky. Redo log *groups* can all share the same hard disk (leaving aside archiving issues at the moment). You need at least two *groups*. Each group should then have multiple *members*, to prevent software or user corruption. hdisk0 can therefore have 2 redo logs on it, and those will be mirrored onto hdisk1. But that still means, from Oracle's perspective, that you only have 1 member per group. LGWR writes only once, and your mirror ensures its write is replicated to hdisk1. Which is better than a kick in the teeth, but what happens if you accidently practice your 'rm *' skills one day? Er, you lose the log on both disk0 and disk1... total loss of redo logs, and hence lost data.

Somewhere in your scheme, you need to make provision both for redo log mirroring (good for protection against hardware failure), which is what it would seem to me your hdisk0/hdisk1 mirror would provide. And for redo log member multiplexing... which you don't appear to have in this proposal at all....

> hdisk2 --> Oracle SW + SYSTEM Tablespace files
> hdisk3 --> REDO LOGS (Group 3)

...although hdisk3 could well be a candidate for the second *member* of all your groups. (In other words, get the difference between groups and members sorted before you go much further).

> hdisk4 --> ARCHieve Log Destination + UNDO

Why have you separated out UNDO?

Can I make one suggestion? Put down the Loney book, and stop trying to match his suggestions (which are valid only for non-striped, non-RAIDED systems).

UNDO is a tablespace. It has datafiles. Datafiles, in your plan, belong on the RAID5 set. So stick the UNDO in there.

> RAID 5 Disk (hdisk5 + hdisk6 + hdisk7) ---> DATA Files + TEMP
>
> I cannot put my "hdisk3" under RAID 5 becos to Ohdisk0 --> AIX OS +
Control File1 + Redo Log (Group 1)
> hdisk1 --> AIX OS + Control File2 + Redo Log (Group 2)S RAID 5 is just one
> LOGICAL DISK and so I cannot multiplex them on different disk.

See above. You appear to me not to grasp the distinction between members and groups.

> Will there be any issues in above setup....? Does someone here nay
> better idea then this one....?
>
> THANKS!
This isn't ideal, because of the logs+O/S issue, but try this:

hdisk0 --> AIX OS + Control File multiplex A + Redo Log Group 1, 2 and 3, member A
hdisk1 --> AIX OS mirror + Control File Mirror + Redo Log Group 1,2,3 mirror of member A
hdisk2 --> Archive Redo Logs
hdisk3 --> ControlFile multiplex B + Redo Log Group 1,2,3 Member B RAID 5 Disk (hdisk4 + hdisk5 + hdisk6 + hdisk7) ---> DATA Files + TEMP + UNDO + SYSTEM But it all rather depends on the number of controllers you've got, as well as hard disks.

Regards
HJR Received on Wed Sep 17 2003 - 17:12:19 CDT

Original text of this message

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