there are a lot of good whitepapers and how-to's on the IBM pages - maybe have a look here :
http://www-03.ibm.com/servers/enable.../9a46/9a46.pdf or
http://www-03.ibm.com/support/techdo...Index/WP100883 - but all this can be only a guide since system tuning happens on the system, not on some whitepapers ... and it depends widely as well on what exact AIX TL/SP you are running - since a lot changed with TL2 and TL5 for AIX - so tunables for earlier versions might not be that optimal any more - and it depends as well on your hardware / infrastructure and the kind of workload and the oracle settings themselves in the system. You might want to have a look here
Jaqui's AIX Performance and Security Blog - Jaquie might be THE IBM goddess of system tuning - we're running most of these values with a bit finetuning on our 160 oracle lpars - only a few servers have very particular requirements where we had to amend widely.
I would go as well with all recommendations made previously (though I would indeed set the min/max values to 1-3 and 90 if its an oracle-only system). Have a look on your buffer sizes, make sure that oracle is set to use asynch IO since this is reducing the waits immensely, you might want to set oracle to manage if its using cio or not instead of AIX, make sure that your redo-logs (and controlfiles) are in a filesystem with smallest possible blocksize and that you use different filesystems for archivelogs, redologs, datafiles and dumps each, put as much load as you can onto your system and start tuning.
That means - monitor - amend - monitor - amend further or revert - have users testing and sharing their feelings using the database ... until your system fulfills your and your user's expectations. Make sure your SGA sizes are not too small but also not too big as this is contraproductive as well - make sure your AVM remains around 70% - yes oracle needs filecaching even when you're running cio on filesystem base ...
Kind regards
zxmaus