Hello.
I have couple of applications installed on my AIX server. Since yesterday the admins have installed one more software installed on it. I am asked to monitor the performance and to conclude if the new application is really affecting the performance for all the applications users.
Could you please tell me the commands in AIX for performance monitoring, which could help me preparing the statistics of noting figures in morning , noon and evening times. I can arrange for application uninstall to take down the performance and the reinstall and again to take down the performance stats.
I believe there are some thousend freeware tools that monitor some sort of performance. Unfortunately most of those monitors are not aware of AIX and not aware of pSeries hardware. Hence they will fail to monitor more than the basics. Especially when it comes to RSCT and Kernel those tools often are not up to the job. Things might get worse when you have to compile the monitor yourself. That all is unfortunate but at the same time unneccessary. As an AIX admin you'd probably use one of the monitors from the IBM wiki: IBM developerWorks: Wikis - AIX - Performance Other Tools
A very fast and easy montoring can be set up with nmon. It can collect perf data into a file that can be analysed with the nmon analyser. So to set up monitoring on a standalone server in two steps you'd
You are done. This will collect a 24 hour period of performance data with a five minute intervall into a file every day.
Two more hints for using nmon
Nmon uses public API hence no need to run it as root.
If nmon stops writing into the file there might be an error with the device that is being queried. In that case either remove the error or prevent nmon from querying that device (by commandline option).
@OP: You are right: to find out if/how the new app affects performance of your system you need to know what the load was before the new app was installed. So for comparison you need to uninstall the app, monitor, then install the app and monitor again. However if monitoring with the app reveals that performance is still ok you might simply leave it.
resource usage is NOT equal to performance ... you can have a system with loads of unused memory/cpu/io and its still performing badly - because the code used is bad. And you can have a system using a very high amount of resources, that is perfectly tuned and performing just fine.
When I add an application to my system, I expect the box to use more resources (what would be shown by nmon, topas and others) but I do not necessarily expect the performance to go down.
I would suggest if you really want to know something about performance, that you run a benchmark before and after applying the application - this can be a commercial tool or just a batch you're running anyway on that box frequently. We're using a stress test batch in our company that just produces a predefined amount of work onto the box and performes certain tasks. We run it a couple of times before and after and compare the average time taken - if its similar, you don't have a performance impact, if it takes recognizable longer, you do.
Nmon is a brilliant tool to identify bottlenecks as soon as they occur but since performance is something pretty much subjective, you need something that comes close to your normal workload but needs to be measurable. Oracle and Sybase eg have such functionalities per default ...