Raw Time-series Data for hardware performance counters

Raw Time-series Data for hardware performance counters

Hi,

  1. I understand that we can get hardware performance event samples by attaching to a process. Are we able to use VTune to get the raw time-series counts of the hardware performance counters? For example, if I would like to see how the L1 cache misses change over time as I use say Internet Explorer to view a specific webpage, can we retrieve that kind of data?
  2. Does VTune allow us to add to the list of hardware performance events that can be monitored?
  3. I am trying to understand how accurate the per-process measurements are in VTune. I understand that VTune uses hardware performance counters in the CPU. Such measurements collected apply to the whole system. How does VTune differentiate the measurements from process to process?

Thanks!

7 posts / 0 nouveau(x)
Dernière contribution
Reportez-vous à notre Notice d'optimisation pour plus d'informations sur les choix et l'optimisation des performances dans les produits logiciels Intel.
Portrait de iliyapolak

>>>I understand that we can get hardware performance event samples by attaching to a process. Are we able to use VTune to get the raw time-series counts of the hardware performance counters? For example, if I would like to see how the L1 cache misses change over time as>.>

Do you mean "plot  of counters data as a function of the elapsed time"

Citation :

iliyapolak a écrit :

>>>I understand that we can get hardware performance event samples by attaching to a process. Are we able to use VTune to get the raw time-series counts of the hardware performance counters? For example, if I would like to see how the L1 cache misses change over time as>.>

Do you mean "plot  of counters data as a function of the elapsed time"

Yes. Say a plot of the L1 cache misses over time. Can we export such raw data to perform analysis?

>>...as I use say Internet Explorer to view a specific webpage...

Internet Explorer does internal caching of web pages ( look, for example, at IE option 'Do not save encrypted pages to disk' ) and I don't think VTune is a right tool to verify that a web-page is cached or not.

>>>>...as I use say Internet Explorer to view a specific webpage...
>>
>>Internet Explorer does internal caching of web pages ( look, for example, at IE option 'Do not save encrypted pages to disk' ) and
>>don't think VTune is a right tool to verify that a web-page is cached or not.

Also, since Internet Explorer supports OLE Automation and has lots of COM interfaces it is possible that there is a method in some COM interface that could provide such details for a web page. Like, is a web page cached or not?

Portrait de iliyapolak

>>>Such measurements collected apply to the whole system. How does VTune differentiate the measurements from process to process?>>>

One of the method is resolve the address of the function(s) and exporting dll by looking at the current state of the instruction pointer.Also PDB files can be useful in obtaining information about the exported functions.Enumerating of import and export tables of the PE format (.exe) exactly after the file is loaded and correct addresses are resolved by NtLoader functions can be also used.

Portrait de iliyapolak

>>>For example, if I would like to see how the L1 cache misses change over time as I use say Internet Explorer to view a specific webpage, can we retrieve that kind of data?>>>

I do not think that you can obtain such a measurement granularity.Mostly you will get some kind of plot for example processor load over the time but this is measured per process and per thread.

Connectez-vous pour laisser un commentaire.