Pelegant freezes up sometimes during touschek_scatter

Moderators: michael_borland, soliday

Post Reply
GPenn
Posts: 38
Joined: 28 Aug 2018, 09:46

Pelegant freezes up sometimes during touschek_scatter

Post by GPenn » 02 Oct 2018, 17:40

I am having occasional trouble running simulations of particle losses due to touschek scattering. Pelegant will stop doing new calculations at some point, but there will not be any error message, it just hangs. It seems to happen whenever the number of "particles selected for tracking" is smaller than the number of processes. Since the number of selected particles is random, I encounter this issue sometimes even when running with 1M particles if I am using a lot of processes.

I include files for a very artificial run with n_simulated=1000, including output from a run using 4 processes. It hangs indefinitely at the 31st TSCATTER element, where there is only 1 particle to be tracked. In previous attempts I observed Pelegant to freeze up even when the number of particles is several times bigger than the number of processes, but that was with a different input file that was using WATCH elements and other features.

I would appreciate any ideas for a workaround. I can reduce the number of processes, but there is no guarantee that the program won't hang anyway. I am also going to try to continue the simulation if it hangs by setting 'overwrite_files=0' and running on a single process, but that will probably be too slow to do on a regular basis.

Cheers,
Gregg
Attachments
touschek_scatter_4proc.zip
(19.27 KiB) Downloaded 210 times

michael_borland
Posts: 1797
Joined: 19 May 2008, 09:33
Location: Argonne National Laboratory
Contact:

Re: Pelegant freezes up sometimes during touschek_scatter

Post by michael_borland » 03 Oct 2018, 09:35

Gregg,

If you set CHANGE_T=0 on your RFCA elements, it should resolve the problem. For your particular simulation this setting makes no difference. We'll try to fix the bug in the next release.

Thanks for reporting it.

--Michael

GPenn
Posts: 38
Joined: 28 Aug 2018, 09:46

Re: Pelegant freezes up sometimes during touschek_scatter

Post by GPenn » 08 Oct 2018, 16:51

Thank you Michael, using CHANGE_T=0 fixes that problem.

Even after making that change, I do see one other occasional problem where Pelegant hangs as soon as the calculations of the losses starts. It only happens if I set overwrite_files=0 in touschek_scatter, and it's kind of erratic. When I use many processes, it seems to always hang, but if I only use 4 (for example) then sometimes it hangs and sometimes it works fine, even if I don't change anything in the input files. The systems I am using all have openmpi installed instead of mpich. Also, I am not 100% sure I compiled Pelegant correctly in terms of linking to the MPI routines. Anyway, I don't really need to use the overwrite_files option any more because the calculations are reliably going to completion, but maybe there is a bug related to overwrite_files as well.

Files are attached, they are basically the same as before except for the one line where overwrite_files is set.

-Gregg
Attachments
overwrite.zip
(10.27 KiB) Downloaded 204 times

Post Reply