Experimental Physics and Industrial Control System

To add a data point with Debian Linux 8 and 9 and Base 3.16.1.  (I've
also see significant slowness)

Building the simplest IOC for 6 targets.  So dbExpand.pl is run once and
registerRecordDeviceDriver.pl is run 6 times.

> $ ../bin/linux-x86_64/makeBaseApp.pl -t ioc foo
> $ time make -s
> 
> real    0m44.828s
> user    0m43.800s
> sys     0m0.436s
> $ make distclean
> $ time make -sj4
> 
> real    0m27.115s
> user    1m5.956s
> sys     0m0.744s

For a single target the "real" times are 8.7 and 7.6 sec.  Basically no
parallelization.

For 6 targets, looking at the CPU load graph (attached) shows a
bottleneck at ~25% (1 of 4 cores) early in the build.

The dbdExpand.pl and registerRecordDeviceDriver.pl steps specifically

> (cd /home/mdavidsaver/work/epics/base-git/tmp/fooApp/src/O.linux-x86_64 && time perl -CSD /home/mdavidsaver/work/epics/base-git/bin/linux-x86_64/dbdExpand.pl   -I. -I.. -I../O.Common -I../../../dbd -I../../../../dbd -o foo.dbd base.dbd)
> 
> real    0m1.241s
> user    0m1.228s
> sys     0m0.012s

> (cd /home/mdavidsaver/work/epics/base-git/tmp/fooApp/src/O.linux-x86_64 && time perl -CSD /home/mdavidsaver/work/epics/base-git/bin/linux-x86_64/registerRecordDeviceDriver.pl   -I. -I.. -I../O.Common -I../../../dbd -I../../../../dbd    -o foo_registerRecordDeviceDriver.cpp ../O.Common/foo.dbd foo_registerRecordDeviceDriver /home/mdavidsaver/work/epics/base-git/tmp)
> real    0m5.836s
> user    0m5.828s
> sys     0m0.004s

I hadn't noticed this difference before.  It seems like these scripts
should be doing almost the same work.


I don't think this is perl interpreter startup time.  (I've never
noticed makeMakefile.pl taking any time)

> 
> time perl -e 'print "Hello\n"'
> Hello
> 
> real    0m0.007s
> user    0m0.008s
> sys     0m0.000s
> 
> time perl -CSDL -e 'print "Hello\n"'
> Hello
> 
> real    0m0.008s
> user    0m0.004s
> sys     0m0.000s
> 
> time perl -C -e 'print "Hello\n"'
> Hello
> 
> real    0m0.004s
> user    0m0.000s
> sys     0m0.000s



On 09/27/2017 10:32 AM, Johnson, Andrew N. wrote:
> Hi Hinko,
> 
> On Sep 27, 2017, at 5:15 AM, Hinko Kocevar <[email protected]> wrote:
>>
>> Under 3.15.4 perl script execution takes a lot of time on Linux Mint, while it executes in a jiffy on CentOS.
>> I added a time in front of the perl call in RULES.db:
>>
>>
>> time perl -CSD /devel/root/R3.15.4/base/bin/linux-x86_64/registerRecordDeviceDriver.pl   -I. -I.. -I../O.Common -I../../../dbd -I/devel/root/R3.15.4/modules/adcore-master/dbd -I/devel/root/R3.15.4/modules/adsimdetector-master/dbd -I/devel/root/R3.15.4/modules/asyn-master/dbd -I/devel/root/R3.15.4/modules/autosave-master/dbd -I/devel/root/R3.15.4/modules/busy-master/dbd -I/devel/root/R3.15.4/modules/calc-master/dbd -I/devel/root/R3.15.4/modules/sscan-master/dbd -I/devel/root/R3.15.4/base/dbd    -o simDetectorApp_registerRecordDeviceDriver.cpp ../O.Common/simDetectorApp.dbd simDetectorApp_registerRecordDeviceDriver /devel/stage/R3.15.4/iocs/simdetectorioc-master
>>
>>
>> 29.95user 0.02system 0:29.98elapsed 99%CPU (0avgtext+0avgdata 19256maxresident)k
>>
>>
>> I think other calls to perl are awfully slow, too..
>> Any idea why it might be so slow?
> 
> This sounds more like a problem with your Linux Mint system configuration than with the EPICS scripts, I run Linux Mint on my personal laptop at home and don’t see these kinds of issues.
> 
> I wonder if this may be related to different Locale settings. Is this slow:
>     time perl -CSD -e 'print "Hello\n"'
> 
> If yes, please try this
>     time perl -e ‘print "Hello\n"'
> and then this
>     time perl -CSDL -e 'print "Hello\n"'
> and this
>     time perl -C -e 'print "Hello\n"'
> to see if the flags make any difference. What is your locale set to (echo $LANG) and maybe try changing it to see if that has any effect.
> 
> Please report any interesting findings back here...
> 
> HTH,
> 
> - Andrew
> 

Attachment: cpuload.png
Description: PNG image

References:

registerRecordDeviceDriver.pl takes a long time to finish Hinko Kocevar

Re: registerRecordDeviceDriver.pl takes a long time to finish Johnson, Andrew N.

Navigate by Date:

Prev: Re: registerRecordDeviceDriver.pl takes a long time to finish Henrique Almeida

Next: Re: issue with epics in python Matt Newville

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: registerRecordDeviceDriver.pl takes a long time to finish Johnson, Andrew N.

Next: Re: registerRecordDeviceDriver.pl takes a long time to finish Luca Cavalli

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