OK, we have now some of the information my colleagues and i asked from you. We still do not know:

- what the system is built of: is it a LPAR or WPAR? If the former, which configuration do you use (capped/uncapped CPU assignments, shared/dedicated networks, which types of disks are attached and how, etc., etc.)

- what the system exactly does: which applications (and versions/releases/patch levels thereof) are running, what the expected performance should be (versus what the performance actually is - which part of the systems operation can you name that is doing slower or less than expected), how is the application used (for instance: is the database used for user interaction and therefore expected to have its maximum load during day? Or is it batch oriented and the heavy load can be expected to start every day at the exact same time? Or ...)

- how the users connect to the system: Not at all (because only application servers connect directly and the users connect to the app-server?), via a shell session (say: telnet, rlogin, ssh, etc.), via a graphical interface (CDE, X-session, ... ?), etc.


As much as i am willing to help there are two general constraints: i will help you to help yourself but i will definitely not do the work you cannot do - its simply not my job and if you want to have me working for you please hire me, i make a living from that. Right now i ask ten questions, you answer the two which require the least effort and i try to make as much as possible from incomplete data. Have a look at my first post and ask yourself which of the questions i posed there (and repeated here) you have in fact answered. Which of the theories i have developed did you comment on and tried to approve/disprove? How much have you written about your system and how much did i or, generally said: how much effort are you putting into this and how much do i? Sorry to be that blunt, but: this is not showing the right attitude.

The second thing is: what exactly is the problem? We still do not know what your SLA (service level agreement) is. The thing is: you (the metaphorical "you") agree to deliver a system which performs function X at the rate of Y. If the actual performance is less than Y you have a performance problem, otherwise not. "Tuning" is always a process, where some function X operating at the rate of some Y' is raised (or lessened) to this Y. If you reach this goal you again have no problem while if you don't you have one. The expected "rate of Y" is what is called SLA.

Example: your boss tells you to have the system do 100k transactions per second (your SLA). It does 80k/s right now, so you have to do some tuning. You change the system and it does 90k/s - goal not reached and you are not done. You change the system further and it does 105k/s - mission accomplished, go back to surfing the web.

I hope this helps.

bakunin