I am doing a performance analysis on one of our psystem. Most of time I am using Nmon analyser to do my trend graph. But I can't find any help with it. We are interesting in the time spend by tasks in Aix run queue.
After looking the Aix documentation, I am pessimist to find any straight metric in order to measure that time.
This would depend on what you are trying to collect.
One, classic approach, might be to use sar perhaps.
A different approach would be to use AIX Advanced Accounting.
If you are looking for a deep approach, and are looking at a specific process (rather than all processes) of a know PID and/or TID you could use various trace based tools (e.g., tprof, curt, etc) and/or trace and trcrpt . Rather than PID this could also be command name - there are too many options to name them all.
And, another option could be using probevue mechanism.
Thank you for your answers. I should have specify my working environment :
Business app : SAP R3 on Oracle Cluster-ware
DB : Oracle 11g + RAC
FS : GPFS
OS : AIX 6.1
Servers : IBM psystem (Power7 CPU)
We are investigating divergence between real CPU consumption versus 'sap + rac oracle + gpfs' benchmarks in order to challenge our advanced support from SAP and Oracle.
By the way, I am not a AIX sysadmin neither I am an AIX expert. And unfortunately, I do not have hands over the system. To do something on the system, I have to ask our contractor in charge of. However, I understand thing well ... I think
About your propositions :
First thing first, I want just graph this waiting time over the time no matter the process. I should read more carefully the SAR doc. I didn't know about the AIX advanced accouting system. Hence, I printed the doc and now I have to read it
If we find something particuliar, I may go for a deep analysis but not with a real time trace I that is possible. We ran some CURT (3 time 1 minutes) and heavily impacted our 6000 users :x
Thank you again for your time. I will be certainly back with others questions and I hope results.
AIX 6.1 - which TL, how recently patched?
"Applications" - level and patch levels
How are you defining real CPU consumption?
Also,
Is the a shared, or dedicated processor partition?
If shared: how much entitlement; how many VP; how many active threads (normally); what is overal system utilization; how many activated processors. Partition utilization using the command
# lparstat 6 10 # simple command to review processor utilization for 60 seconds
please only outputs during "stress". "Idle" may also be presented, but only as a comparision.
And, what also helps - a ballpark figure of what you expected to see compared to what you are seeing.
Ah yes, almost forgot: Power7 Model and Firmware level.