Dear all,
I am currently not able to run my simulation with MPI. While it works just fine for the "FlowStepKe"-Tutorial,
for my project it crashes. As it involves radiation, ViewingFactors need to be calculated, and it seems as this causes the code to crash:
From "umf4sym" I'd guess that this happens in the linear solver. The simulation runs fine without the parallel solver.
Calling a serial instance of ViewingFactors also works.
I am not sure where to start for fixing the error, could anyone give me a hint?
I'm running this on a windows 10 machine with the recent mpi-version of elmer
Many thanks!
Error for ViewingFactors with parallel computation
Error for ViewingFactors with parallel computation
Last edited by kdarte on 29 Nov 2022, 10:42, edited 1 time in total.
-
- Posts: 2313
- Joined: 25 Jan 2019, 01:28
- Antispam: Yes
Re: Error for ViewingFactors with parallel computation
This usually means running out of memory, RAM, mpi generally takes more RAM because each processor needs a certain amount of RAM for each process. Try monitoring RAM usage during the runs.
Re: Error for ViewingFactors with parallel computation
Thanks for the suggestion.
I am pretty sure that RAM is not the issue here - it hasn't even started to build up when the simulation crashes.
I can reproduce this with the Gui-Tutorial "radiation" - attached is my slightly adjusted version (essentially changing the parallel-solver settings)
I am pretty sure that RAM is not the issue here - it hasn't even started to build up when the simulation crashes.
I can reproduce this with the Gui-Tutorial "radiation" - attached is my slightly adjusted version (essentially changing the parallel-solver settings)
- Attachments
-
- radiation_parallel.zip
- (73.1 KiB) Downloaded 56 times
Last edited by kdarte on 29 Nov 2022, 10:43, edited 1 time in total.
-
- Posts: 2313
- Joined: 25 Jan 2019, 01:28
- Antispam: Yes
Re: Error for ViewingFactors with parallel computation
I have the same issue on Linux Mint. I faintly remember that calculation of view factors in parallel required precise control of the partitions so the the radiation boundary and the target boundary were in the same partition. This would not happen by default.
-
- Posts: 2313
- Joined: 25 Jan 2019, 01:28
- Antispam: Yes
Re: Error for ViewingFactors with parallel computation
Thanks a lot, I must have missed this thread... using
Code: Select all
ElmerGrid 2 2 %msh -connect 1 2 -partdual -metiskway %n
For my setup with 6 radiation regions this results in the following decomposition: The results seems ok (heating up the parts within the holes by radiation)