Amplifier XE on Linux locks up system

[Michael_J_6]

I am running Amplifier XE 2013 (update 12) on RHEL 6.4 (32-bit). As soon as I try to start data collection, the entire system freezes and has to be powered off hard and rebooted. It works fine on other development systems we have, but this is our "production" system where the performance measurement actually matters. I'm sure it's something system-specific. Looking for any clues on how to start diagnosing this, please.

I did RTFM about the CPU C-states and certain models of Core i7 processors. The system in question has dual Xeon E5440 CPUs which, if I read the notes correctly, should not be affected by that. I looked in the BIOS anyway and it does not have any options to mess with things like that. For the record, it's a SuperMicro X7DCA-3 motherboard.

Thanks for any advice/hints that anyone can offer!

[robert-reed]

Some extra information would be useful to help diagnose: were you trying to run a "basic hotspots" collection or "advanced"?  Or perhaps one of the other collectors?  There should be nothing particular in Intel Core architecture that would differentiate it from a collection standpoint.  How big is the application from which you're trying to collect data?  Does your application stress system resources?  Do your production systems have restricted resources (less memory or something like that) that might explain the difference in VTune Amplifier behavior?

[Bernard]

In Windows such a behavior could be related to crucial components deadlock or to high Irql or even interrupt storm.There also probability of infinite loop while executing in kernel mode or even hardware error.

As you are on Linux unfortunately I cannot help you more.

[Michael_J_6]

I was going for the gold and trying to collect Advanced with everything turned on (call stacks, context switches, etc). When I tried just doing the Basic analysis, it works fine. And that really has what I wanted to see, so that's good enough for now. I don't believe I'm exceeding any resource limitations like memory and it's not a very stressful application anyway.

[David_A_Intel1]

Hi Michael:

Thanks for the update!  Regarding "Looking for any clues on how to start diagnosing this", I would start with the Advanced Hotspots and none of the extra options (i.e., stacks, context switches).  Does that lock up the system?  If so, it sound like a hardware problem, since the Advanced Hotspots uses the PMU of the processor and interrupts to collect samples.  If not, then try adding just stacks.  Etc.  Iterate and see where it fails.

[Bernard]

Maybe try to do as Mr Anderson advised.There could be some probability of hardware error when you enaled the broadest spectrum of profiling options.