Re: Asynchronous I/O and AIX and jfs = perf ?

Believe it or not, both answers are right, and which one is better for you depends on how badly you want absolutely all the performance available for your DB.

Using raw logical volumes in AIX with Oracle yielded ~18% better performance in our testing (8 way R50, 4GB RAM, 100 GB database, 32 mirrored SSA drives=16 spindles for Oracle data, 6 mirrored SSA drives = 3 spindles for Oracle logging and other admin stuff, 3 SSA adapters) over JFS. Aynch I/O was used in RLV testing. We saw much more I/O wait with JFS and longer response times to our sample SQL queries as opposed to RLV, which had less I/O wait and faster response times.

The rub comes when you try to back up a RLV database. If it was JFS, you just shut down the database and backup the filesystems. To back up a database that uses RLV's, you have to use the cpio command and copy from the raw logical volume into file(s) in a filesystem, then back up the file(s) later on.

The DB admins I worked with were extremely leery of using the backup utility in Oracle (this was at version 7.3.2). They claimed it was unreliable, and that the cpio method was the only way they would implement production backups of RLV's.

So the option is really up to you: easier backups with a drop in performance, or more difficult backups with better performance.

  Hope this helps,

  Dan

Emmanuel Cortes wrote:
>
> To increase performance of oracle database 7.3.x.x, is it better to use
> Asynchronous I/O on AIX 4.3.x whith jfs file systems or not ?
>
> I read in a redbook (Database Performance on AIX in DB2 UDB ans Oracle
> Environnements "SG24-551-00") it's a good thing.
>
> But oracle tell if you don't use raw device is not good idea.
>
> Where is the truth ?
>
> Thanks
>
> Emmanuel Cortes

--

<<<<<<<<<<<<<<<<<<<< * >>>>>>>>>>>>>>>>>>>>

Dan Jennings
Speaker to Machines

When the only tool you have is a hammer,

   all problems tend to look like nails.

• Masured

<<<<<<<<<<<<<<<<<<<< * >>>>>>>>>>>>>>>>>>>>