Placing Oracle Files, OFA or ....

[Shahbaz_1]

Hi Friends,
Your input is required, for placing oracle files.
- We have four volumne group, so four mount points.
(/u01 /u02 /u03 /u04)
- os is in diff volume group.

- I have planed for 3 contrl files & two log members

Please ...

Thanks & Regards
Syed

[Scott Van Kalken]

we have a similar setup for oracle.

We split data file across the volue groups (usually mirrored sets of disks). We determine this at random initially and tune as necessary based on disk I/O.

Scott.

[Alexander M. Ermes]

Hi there.
Your plan is ok so far.
My two cent to this :
separate the index tablespace files from the database tablespace files.
And try to put the offline redo logs somewhere else ( continously zipping and backing them up twice a day ).
Rgds
Alexander M. Ermes

[Steven Sim Kok Leong]

Hi,

Try to spread the Oracle index and data filesystems over as many controllers and as many spindles (disks).

If possible, stripe your data filesystems across spindles to increase I/O throughput for random-accessed data. Remember that for performance reasons, the Oracle block size must be a multiple of the OS stripe size. The optimal (depends on your data) is usually has Oracle block size equals filesystem block size (eg. 8 KB)

For control files and archive logs, they are sequentially accessed, thus striping is never encouraged. Also, if you are already mirroring the control files at the oracle level, there is no need to mirror them at the OS level. As far as possible, try to have your control files use different spindles and controllers.

Hope this helps. Regards.

Steven Sim Kok Leong

[Volker Borowski]

Well,
4 units is not that much ....

3 controlfiles is OK
2 members for online logs is a good choice

no datafile on a volume/disk where the offline redologs are located (loss of data may occour if this disk goes down).

No problem to put datafile and online-redologs (if several members) together, beside performance and a bit recovery overhead.

Hope this helps
Volker

[Dennis J Robinson]

Sri,

Would love to help you.

Today Oracle is advocating SAME strategy.

SAME means "Stripe and Mirror Everything", most of us old school (I'm not that old) do not advocate this.

Actually some of us make a pretty good living of reversing SAME and providing a customer a nice boost in performance.

Here goes a strategy that will work for you:

Lets say you have 12 disks, we won't consider mirroring here:

Mount points would follow the "OFA on Steroids" strategy:



/u01 - Oracle install - single disk or chunk of vg00
/dba/u02 - Oracle odd redo primary - single disk
/dba/u03 - Oracle even redo primary - single disk
/dba/u04 - Oracle odd redo mirror - single disk
/dba/u05 - Oracle even redo mirror - single disk
/dba/u06 - Oracle SYSTEM, TEMP, TOOLS - single disk or 2-4 disk stripe
/dba/u07 - Oracle RBS - single disk or 2-4 disk stripe
/dba/u30 to u69 - Oracle datafiles - 2-4 disk stripes, 128kb-1M stripe width
/dba/u2*
/dba/u70-u100 - Oracle indexes - 2-4 disk stripes, 128kb-1M stripe width

This is something. If you only have 4 disks, then SAME them.

[Andreas D. Skjervold]

Hi

Dennis' input is perfect.. remember to add a disk for archiving redologs.

I do also prefer the OLD school of OFA.

and after introducing the Localy managed tablespaces, placement of the system tablespace has become less important in terms of performance...

Andreas