Cpu consumption reduced to 100% during capture

[duanetiemann]

Using update 4 on Linux, I notice that our cpu consumption which is otherwise 300%-400% is reduced to 100% during capture. This occurred all both times I tried it.

Is this normal? It seems odd.

Capture interval was 50ms for 2 minutes, if that matters.

The cpu consumption was captured via a ps every 10 seconds.

Duane

[Kirill_R_Intel]

Hi Duane,

This looks strange. Could you please provide some additional info about your setup:
- Amplifier XE command line
- Do you runyour app from Amplifier or attach to it? If attaching - does the CPU time comes back to 300%-400% when analysis is complete?
- ps command line you use
- Platform info: OS and CPU type.

I'll try to see if this behavior is reproducible in my lab.

[duanetiemann]

Thanks.

Our app is an IONIX server. It does network diagnostics. It was doing a discovery of a simulated network. There were probably around a dozen threads active. But about 100in all. The active stacks were probably 40-80 frames deep, depending.
VTune is installed in my home directory on nfs.

This was done through the GUI. The command line was just "./amplxe-gui".
I attached to the process. CPU consumption does go back up to 300%-400% after the capture.

"ps -eo pid,ppid,stime,pri,time,rss,vsz,nlwp,comm,args | grep sm_server >>outputFile"
This is part of a Perl script. It does it every 10 seconds.
I also ran a script that captured cpu by thread. "ps -L -p >>outputFile". I didn't see any one thread stand out during the measurement period.

Red Hat Enterprise Linux Server release 5.2 (Tikanga)
Kernel 2.6.18-92.1.13.el5 on an x86_64
Intel Xeon CPU X7350 @ 2.93GHz
4 quad cores.

Duane