Argonne National Laboratory

Experimental Physics and
Industrial Control System

1994  1995  1996  1997  1998  1999  2000  2001  2002  2003  2004  2005  2006  2007  2008  2009  2010  2011  2012  2013  2014  2015  2016  <2017 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
<== Date ==> <== Thread ==>

Subject: Re: ADSimDetector
From: Mark Engbretson <markatwork@att.net>
To: Mark Rivers <rivers@cars.uchicago.edu>
Cc: "tech-talk@aps.anl.gov" <tech-talk@aps.anl.gov>
Date: Fri, 26 May 2017 17:26:32 -0500
Fantastic! Just the info I was looking for.

On May 26, 2017, at 17:24, Mark Rivers <rivers@cars.uchicago.edu> wrote:

Hi Mark,

 

Most plugins that support multiple threads in ADCore R3-0 do have linear increase in performance as the number of threads is increased, at least up to 5 threads.  However, I have not proven this for NDPluginStdArrays because it is so fast that it is hard to send it frames fast enough.

 

Here is I benchmark I just ran on a 20 core RHEL 7 machine:

 

- simDetector 1024x1024, UInt8 data.

- AcquireTime and AcquirePeriod both 0.0001 seconds

- Peaks mode, PeakNumX=1, PeakNumY=1, PeakWidthX=2, PeakWidthY=2

- NDPluginStdArrays BlockingCallbacks=No

 

- All plugins disabled

simDetector ArrayRate_RBV = ~2200 frames/s

simDetector task using 66% of 1 core.

 

- NDPluginStdArrays plugin enabled, 1 thread

simDetector ArrayRate_RBV = ~2200 frames/s (no change)

simDetector task using 66% of 1 core.

NDPluginStdArrays  ArrayRate_RBV = ~1550

NDPluginStdArrays task using 99.5% of 1 core.

 

So with 1 thread it cannot quite keep up with the simDetector

 

- NDPluginStdArrays plugin enabled, 2 threads

simDetector ArrayRate_RBV = ~2200 frames/s (no change)

simDetector task using 66% of 1 core.

NDPluginStdArrays  ArrayRate_RBV = ~2200

NDPluginStdArrays tasks (2) each using  89% of 1 core.

 

So by using 2 threads for NDPluginStdArrays it can keep up with the simDetector at 2200 frames/s.

 

The NDPluginStdArrays plugin is so fast that it is hard to test how many cores it can use because the simDetector cannot really generate frames fast enough to saturate it.

 

Mark

 

 

From: Mark S. Engbretson [mailto:Engbretson@anl.gov]
Sent: Friday, May 26, 2017 11:56 AM
To: Mark Rivers
Cc: tech-talk@aps.anl.gov
Subject: ADSimDetector

 

Mark –

 

Hypothetical question.

 

You have posted at various time ADSimDetector beanchmarks. Historical values from a PDF that I am looking at now being about 485 FPS for a 1024 x4 1024 image. If I increase this image dimensions  to 3078 by 4096, basically 12 times the data, the frame rate  drops by about the same factor of 12 down to 40 FPS or so (This is confirmed using one of your old AD1.9 windows prebuilts and playing with image sizes).  This FPS  value is pulled from at the very top level NDStdArray plugin, blocking, so frames are not dropped.

 

In AD3-0, This plugin is multi-thread enabled.  I’m assuming that if someone did not specify a value, 1 thread is allocated for this plugin and that more *might* make it handle bigger data or faster data.  I.e., 5 threads might make that original 1024 x 1024 image collect at . . . 2400+ FPS . . . and that larger image perhaps 200 FPS without dropping frames.

 

Or am I mistaken on how the AD3.0 threads work?

 

I have a newer model Xeon computer with 12 real cores, most of which are idle at any given moment, and was wondering which plugins may be worth committing cores to.

 

Me

 

 

 

 


References:
ADSimDetector Mark S. Engbretson
RE: ADSimDetector Mark Rivers

Navigate by Date:
Prev: RE: ADSimDetector Mark Rivers
Next: Re: ADSimDetector Michael Davidsaver
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
Navigate by Thread:
Prev: RE: ADSimDetector Mark Rivers
Next: Re: ADSimDetector Michael Davidsaver
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
ANJ, 30 May 2017 Valid HTML 4.01! · Home · News · About · Base · Modules · Extensions · Distributions · Download ·
· EPICS V4 · IRMIS · Talk · Bugs · Documents · Links · Licensing ·