Parallel Settings for ElmerGUI

[modelman]

I have successfully, and happily, run numerous models with ElmerGUI (8.4) using a single processor on a Windows 10 machine. Now I would like to use parallel processing, but have run into difficulties with the Parallel Settings in ElmerGUI. I would prefer to continue running my problems using the GUI instead of the command line.

An excerpt from my output, shown below, indicates that partitioning of my single domain does not happen.

LoadInputFile: Number of BCs: 2
LoadInputFile: Number of Body Forces: 0
LoadInputFile: Number of Initial Conditions: 0
LoadInputFile: Number of Materials: 1
LoadInputFile: Number of Equations: 1
LoadInputFile: Number of Solvers: 2
LoadInputFile: Number of Bodies: 1
Loading user function library: [StressSolve]...[StressSolver_Init0]
Loading user function library: [ResultOutputSolve]...[ResultOutputSolver_Init0]
WARNING:: LoadMesh: Requested mesh > ./. < in partition 2 does not exist!
ElmerSolver: *** Elmer Solver: ALL DONE ***

WARNING:: LoadMesh: Requested mesh > ./. < in partition 2 does not exist!
ElmerSolver: The end
SOLVER TOTAL TIME(CPU,REAL): 0.08 0.08
ELMER SOLVER FINISHED AT: 2020/11/06 15:24:39

Please see the parallel settings that I have used in the ElmerGUI, shown below: (Because the board attachment quota has been reached, I cannot attach a jpg)

Executable: C:/Program Files/Microsoft MPI/Bin/mpiexec.exe
Arguments: -n %n -genvlist PATH,ELMER_HOME ElmerSolver_mpi.exe

Divide: ElmerGrid 2 2 %msh -metis %n
Merge: ElmerGrid 15 3 %ep -partjoin %n

It appears that my "system commands for domain decomposition" are incorrect.
I would greatly appreciate if someone could steer me in the right direction. BTW, I haven't found any postings in the forum that address this issue.

Thank you.

[kevinarden]

You can test ElmerGrid in the command line to see if it runs

ElmerGrid 2 2 meshname -metis 4 (assuming you want 4 partitions)

if ElmerGrid doesn't run then it is not in your path and you could change it to

C:\whereiselmer\bin\ElmerGrid 2 2 meshname -metis 4

if that works you can put the same thing in the GUI window.

I believe windows uses backslashes not forward slashes.

[modelman]

Kevin, Thank you for your quick reply. The problem was that I was using forward slashes (/). When I changed the path to use backward slashes (as I should with Win 10) the parallel solver worked.

Another problem surfaced when using the parallel solver. Using the parallel solver with a linear elastic simulation no data from the output (vtu) could be displayed, although paraview indicated that stress and displacement data was present. Also, when opening a pvtu file paraview produced the error message "Can not set attribute Vectors. Incorrect number of components."

Please tell me your thoughts and/or suggestions. Thank you.

[kevinarden]

not run across that one before. can you post the sif and geometry?

[modelman]

Hi Kevin,

I tried to attach files to this post, but receive the following message:
Sorry, the board attachment quota has been reached.

Will PM you with file attachments.

[modelman]

Kevin,

Sorry for all the confusion regarding attachments. I deleted an old attachment that I sent many years ago; now my file attachments are accepted.

Please see attached files. In the meantime, I will experiment with different meshing options. I wonder if there are some unattached nodes.

Thank you very much for your help.

[kevinarden]

I downloaded these files and ran it on my system using 2 processors, there were no issues.

[modelman]

Kevin, that is interesting. The model also runs also runs fine from my Win10 ElmerGUI (8.4), but the vtu and pvtu files can't display any results. Were you able to show any results with paraview?

I have since passed the mesh through autoclean and everything, including the output data, is OK. I might have neglected to do autoclean on the problematic model.

Once again, thank you for looking into this.

[kevinarden]

Yes paraview open the results files, with no issue, It even opened the ones you posted, although the values in the ones you posted were nunerically zero.

[modelman]

Yes, the problem is that the values are numerically zero with parallel processing. The single processor solution produces reasonable non-zero values. Now, with the mesh passed through autoclean the parallel values are identical to those from the single processor solution.

Bottom line is that I most likely forgot to use autoclean and posted this issue prematurely. Sorry for wasting your time, but I really appreciate your replies.