I'm doing performance testing for one application which works on AIX.

But I don't know which performance parameters of memory need to be collected. Now, I just know very few:
1. page in
2. page out
3. fre

They are all collected by "vmstat" command.

I want to know, except for above parameters, what else do I need to collected?

Thanks in advance!

Try nmon and capture the system stats using it. It provides very detailed and Excel Ready log files. Get a baseline of your first.

NMON will have Processor, memory, disk, IO and almost everything you can think of. Pretty good. It's not part of standard AIX, so you may need to download (free) and install it.

It is a widespread misconsception that "performance" is some value you can simply measure, like "length" or "weight". "Performance" means something like "fitness for a given purpose" and without defining this purpose it means nothing at all.

If you want to "test performance" first agree with your customer what the goal of the system is. Something like "the system has to respond within 2 seconds to every message" or "the application has to finish every user action within 0.5 seconds" or "the system has to do 5000 transactions per hour" or whatever. But this parameter(s) have to be defined first, this is your "given purpose".

Only then you can start measuring (by counting transactions, seconds, ....) and find out how far off your requirement you are, if at all.

Everything else is aggregation of meaningless data*) to come up with meaningless curves (or other graphs) displaying them to meaningless people. Refuse such time-killing nonsense and concentrate on your work.

I hope this helps.

bakunin

PS: if, after defining your requirements, you need help in collecting or interpreting data for this defined purpose please write again. I will gladly help you further in this case.
_________________
*) To further expand on that: you can measure the number of transactions of an OLTP system and measure its weight, then come up with some "weight-to -transactions-quotient". You can now cut away some parts of the systems case and "measure" again and you will find that you have "optimized" the systems because afterwards it weighs, say, 3 pounds less. Would you think that you have made the system any better doing this? Would a nice curve generated from the data before and afterwards change your opinion?