issue tracking with PELEGANT

[Philippe Piot]

After installing PELEGANT on a new cluster I am unable to track particles while the same ele file works fine in ELEGANT. Specifically, PELEGANT seems to do everything fine (e.g. fitting quadrupole magnets) but then does not track the distribution. A simplified example that reproduces the issue is attached along with the outputs from PELEGANT and ELEGANT. You will see that the output is essentially the same except for the "tracking xxx particles" (ELEGANT says 10000 while PELEGANT says 0). I suspect this is a silly mistake on my side. -- Philippe.

simpleTest.tar

(20 KiB) Downloaded 22 times

[michael_borland]

Philippe,

That is very weird. I don't see a problem with either the serial or the parallel version. I tried it on my linux workstation and Improv, with the latest source and version 2023.4.

Are you using Improv? If so, check that you have the right modules loaded:

Code: Select all

$ module list

Currently Loaded Modules:
  1) .binutils/2.41        (H)   3) tcltk/8.6.13        5) aocc/4.1.0                 7) amdfftw/4.1-openmpi-5.0.1-aocc-4.1.0
  2) aocl/4.1.0-gcc-13.1.0       4) subversion/1.14.2   6) openmpi/5.0.1-aocc-4.1.0

Also, you can try ~borland/improv/beta/Pelegant and Pelegant_2023.4 (last release).

--Michael

[Philippe Piot]

Michael,
This gets weirder... I did try to run your exec of pelegant and elegant (on IMPROV) on my .ele file have the same observations: Pelegant does not track while elegant does. The same problem occurs on our new linux computer at NIU. I suspect something is wrong with my ele or lte file, but I cannot just figure it out...
On improv you can see the output and file run in the scratch /lcrc/globalscratch/philippe_testPelegant the stdout have the .output extension.
Thank you! -- Philippe.

[Philippe Piot]

and my loaded module and running commands are:
Currently Loaded Modules:
1) .binutils/2.41 (H) 2) aocc/4.1.0 3) openmpi/5.0.1-aocc-4.1.0 4) amdfftw/4.1-openmpi-5.0.1-aocc-4.1.0 5) a
ocl/4.1.0-gcc-13.1.0

~borland/improv/beta/Pelegant_2023.4 tracktest.ele > pElegantMichael.output
~borland/improv/beta/elegant tracktest.ele > ElegantMichael.output

[michael_borland]

Philippe,

This command

Code: Select all

~borland/improv/beta/Pelegant_2023.4 tracktest.ele > pElegantMichael.output

looks wrong. Are you using mpiexec to invoke Pelegant? E.g., to run with 8 cores you'd use

Code: Select all

mpiexec -np 8 Pelegant ... 

It shouldn't be possible to run it any other way, but it is behaving as if it is using only 1 core, since particles are generated and tracked only on the slave cores.

Some other suggestions:
* Check your ELEGANT_CONFIGURATION environment variable, if defined. Anything odd about the file it names?
* Try setting bunch=%s.bun in &bunched_beam. Is a file created? If so, does it look odd?
* Try using &sdds_beam instead of &bunched_beam

--Michael

[Philippe Piot]

Michael,
Thank you. I checked and the .bun file was not be written. I later realize than pelegant has to be run with a np>1 -- for testing purpose in interactive mode I was using mpiexec -np 1 . Changing to -np 4 worked, i.e. particle got tracked. The other codes I used are usually OK with -np 1 so this potential issue did not cross my mind... Anyway as expected it was a user silly mistake. -- Philippe.

[michael_borland]

Philippe,

The reason that happens in Pelegant is that the master processor is "non-working." It handles coordination among the slave processors, but (usually), doesn't do anything related to particle generation or tracking.

I'll add a check to the code for this problem.

--Michael