Experimental Physics and
Industrial Control System

On 7/18/20 4:56 PM, Mark Rivers wrote:
> Hi Michael,
> 
>  
> 
> Thanks for the tip, that is very nice.  This is perf looking at the ADSimDetector.
> 
>  
> 
>  
> 
> 49.4% is in unknown, which is not too helpful.

This TUI is interactive.  Press '?' for a list of hot keys.
Especially note 'f' to pause/resume sampling so things don't
change while you're moving around.

The leading '+' indicates that you can dig down (press '+').
Note the different between the Children and Self column shows that
all of the time is being spent further down the call tree.

I also forgot that you will want to to pass '--call-graph dwarf'.  eg.

$ perf top -g --call-graph dwarf -p <pid>

>  But it shows where the statistics plugin is spending its time.  It shows a lot of time in the tcp routines, which is because pvAccess  is transmitting 200 images per second, and each image is 1024x1024 Float64, which is 8 MB per image,  for a total of 1600MB/s.

This tracks with what I recall from past investigations.
The places where pixel by pixel iteration are the hot spots.
I'm a little surprised that convertTypeSwitch() isn't at least
equal to the others.


>  
> 
> Mark
> 
>  
> 
> -----Original Message-----
> From: Tech-talk <tech-talk-bounces at aps.anl.gov> On Behalf Of Michael Davidsaver via Tech-talk
> Sent: Saturday, July 18, 2020 6:18 PM
> To: Siddons, David <siddons at bnl.gov>
> Cc: tech-talk at aps.anl.gov
> Subject: Re: profiling an IOC
> 
>  
> 
> On 7/17/20 1:19 PM, Siddons, David via Tech-talk wrote:
> 
>> I'd like to understand where my IOC code is taking its time, and gprof seems like the right tool for this, is that right? Where would I add the compile flags (-pg)?
> 
>  
> 
> You might also investigate the Linux 'perf' utility.
> 
> It doesn't need any special compiler flags, though it's hard to find much sense without debug symbols.
> 
>  
> 
> I especially like the 'top' mode, which can inspect an already running process.  eg.
> 
>  
> 
> $ perf top -g -p <pid>
> 
>  
> 
>  
> 
> https://perf.wiki.kernel.org/index.php/Main_Page
> 

References:

profiling an IOC Siddons, David via Tech-talk

Re: profiling an IOC Michael Davidsaver via Tech-talk

RE: profiling an IOC Mark Rivers via Tech-talk

Navigate by Date:

Prev: RE: profiling an IOC Mark Rivers via Tech-talk

Next: two IOC run as a time Vishnu Patel via Tech-talk

Index: 1994  1995  1996  1997  1998  1999  2000  2001  2002  2003  2004  2005  2006  2007  2008  2009  2010  2011  2012  2013  2014  2015  2016  2017  2018  2019  <2020>  2021  2022  2023  2024 

Navigate by Thread:

Prev: RE: profiling an IOC Mark Rivers via Tech-talk

Next: archiver viewer plug in with archiver appliance in Linux astrid morreale via Tech-talk

Index: 1994  1995  1996  1997  1998  1999  2000  2001  2002  2003  2004  2005  2006  2007  2008  2009  2010  2011  2012  2013  2014  2015  2016  2017  2018  2019  <2020>  2021  2022  2023  2024