When I run the PEBS on the intel Syklake "Intel(R) Xeon(R) Silver 4114 CPU @ 2.20GHz" and then run the Perf concurrently to measure the PEBS overheads, I have found that number of micro operations delivered from microcode sequencer (MS) is increasing a a lot. Particularly by lowering the sampling rate of PEBS it grows exponentially. Can we say that these micro operations delivered from the MS are related to running the PEBS assist microcode on the cores ?
Thank you for posting in Intel Communities. Could you please answer our below questions so that we can debug your issue further?
- Which Analyzing tool are you using?
- What kind of analysis are you trying to run using the above tool?
- Could you please share the steps to reproduce your issue?
- System details
Actually I am using Perf, and try to collect some metrics to do analysis regarding the overhead of running PEBS. So, while using the Perf I was investigating this metric IDQ.MS_UOPS. Based on my observation, by lowering the sampling rate of PEBS this metric experiences a significant increase.
We apologize for the repeated query but we would want to get a clear confirmation from your side. You mentioned that you're using Perf. So, are you not using any of the Intel Analyzers i.e. Intel VTune Profiler/Intel Advisor/Intel Inspector?
We apologies for the delay. Thanks for the confirmation again. This is an Intel Analyzers forum. As you mentioned that you're using Perf tool which is not one of the Intel Analyzers i.e. Intel VTune Profiler/Intel Advisor/Intel Inspector, we cannot help you further with the Perf tool error that you've got.
However, Using Intel VTune profiler, you could actually investigate this metric IDQ.MS_UOPS. Please refer the below documentation for more info:
If this resolves your issue, make sure to accept this as solution. This helps others with similar queries. Thank you!