Save Scalars: mixed up result file in parallel run - SOLVED

Numerical methods and mathematical models of Elmer
Post Reply
thomasatelmer
Posts: 47
Joined: 18 Jan 2019, 18:18
Antispam: Yes

Save Scalars: mixed up result file in parallel run - SOLVED

Post by thomasatelmer »

Hello alltogether,

currently I am trying to use the SaveScalars solver to output some temperature data on a very simple transient model that I made for a test case. Once I know the different things to watch out for, I'll enlarge it to the real task I am after :-)

However, right now I am stuck with the funtionality, that savescalars solver shall output temperature data on two defined nodes into a data file. It works fine when ElmerSolver runs only in one instance, but when I use MPI, the data file is screwed up, no matter what I try. Sometimes it stores just the y and z coordinates of the nodes, skipping the x coordinate, but in all cases, it does not store the two temperatures. Sometimes, the parallel run crashes with segmentation fault, but runs flawlessly when started once more without changes... :o

I have attached the case.sif, the result data files and the result.names files for both cases, with and without MPI. Also, I attach a ZIP with the complete project directory for anyone wanting to try this small case out. Thank you for your kind attention,
Thomas.

p.s. here only the file without MPI are added, the next posts have with MPI and the project ZIP
Attachments
result_without_mpi.csv
(1.73 KiB) Downloaded 99 times
result.csv.names.txt
(500 Bytes) Downloaded 102 times
case_without_mpi.sif
(2.89 KiB) Downloaded 97 times
Last edited by thomasatelmer on 29 Nov 2021, 21:52, edited 1 time in total.
thomasatelmer
Posts: 47
Joined: 18 Jan 2019, 18:18
Antispam: Yes

Re: Save Scalars: mixed up result file in parallel run

Post by thomasatelmer »

here with MPI...
Attachments
result_with_MPI.csv
(1.3 KiB) Downloaded 101 times
result.csv.names_with_MPI.txt
(418 Bytes) Downloaded 105 times
case_with_MPI.sif
(2.89 KiB) Downloaded 99 times
thomasatelmer
Posts: 47
Joined: 18 Jan 2019, 18:18
Antispam: Yes

Re: Save Scalars: mixed up result file in parallel run

Post by thomasatelmer »

and lastly, the project file
Attachments
SaveScalars_mixup.zip
(36.79 KiB) Downloaded 99 times
Rich_B
Posts: 421
Joined: 24 Aug 2009, 20:18

Re: Save Scalars: mixed up result file in parallel run

Post by Rich_B »

Hello,

It might be interesting if you would try using SaveLine, with a polyline from one node to the other node, and see if that solver works any better for parallel runs?

Best regards, Rich.
Rich_B
Posts: 421
Joined: 24 Aug 2009, 20:18

Re: Save Scalars: mixed up result file in parallel run

Post by Rich_B »

Hello,

I loaded the elmergui project and noticed the menu for SaveScalars is crossed with StressSolve, as shown in this screen shot.
edf-menu.png
(43.46 KiB) Not downloaded yet
You may want to try moving the savescalars.xml file from edf-extra into the edf folder. Recreate the project and that may help fix the crossed up menu. I worked for me once, it might help you too.

When the additional xml menu files have been moved from edf-extra to edf, then all of the solver tabs will be in alphabetical order and they shouldn't be crossed.

Rich.
raback
Site Admin
Posts: 4812
Joined: 22 Aug 2009, 11:57
Antispam: Yes
Location: Espoo, Finland
Contact:

Re: Save Scalars: mixed up result file in parallel run

Post by raback »

Hi

Try instead

Code: Select all

Save Coordinates(3,2) = x1 y1 z1 x2 y2 z2
The code seems to look for local indexes which does not master well parallel cases where each partition has just its local nodes. If you give coordinates then the closest one will be found in parallel.

-Peter
thomasatelmer
Posts: 47
Joined: 18 Jan 2019, 18:18
Antispam: Yes

Re: Save Scalars: mixed up result file in parallel run

Post by thomasatelmer »

Hello, Peter, Rich,

thank you for your quick reply. For the phenomenon of the solver orders in the menu, the savescalars.xml is actually in the edf folder. I normally move the files when I need the solver the first time for some simulation, before it is just memory and speed invest... But for some reason, the graphical menu is mixed up in order. This reminds me of another effect I have noticed: the entries in the case.sif are sometimes quite shaken up; then the instruction for setting coordinate_3 is given first, then e.g. coordinate_1 and after another settings line there comes the last coordinate, or the like. Up to now, I have not noticed any issues with this, it is just weird to check such a SIF file, as you have to read back and forth :-).

Peter, this hint actually solved the problem. I just tried it out, and it works fine with local and distributed calculation. Honestly, I used the nodes route initially, as I understood this option better in the Elmer Models Manual. But I have the coordinates as well, so now I can use the identification by X|Y|Z.

Thank you both,
Thomas.
Rich_B
Posts: 421
Joined: 24 Aug 2009, 20:18

Re: Save Scalars: mixed up result file in parallel run - SOLVED

Post by Rich_B »

Hello Thomas,

Looking at your ElmerGui project and the generated sif file, one can see a 'Run Control' section, as shown in this screen shot:
run control.png
(66.32 KiB) Not downloaded yet
What is interesting is the clever use of the free text input field in the Header section. This trick allows storing the Run Control section in the ElmerGui project file, without having to manually edit the sif file.

Rich.
tomatoma
Posts: 12
Joined: 12 Feb 2016, 23:04
Antispam: Yes

Re: Save Scalars: mixed up result file in parallel run - SOLVED

Post by tomatoma »

Hello,

from ElmerModelsManual:

Code: Select all

Save Coordinates(n,DIM)
So it should be:

Code: Select all

Save Coordinates(2,3) = x1 y1 z1 x2 y2 z2
(notice that 2 and 3 are swapped)

Matteo
Post Reply