[Zane]

Hi Foamers,


I'm using dynamicTopoFvMesh with mesquite(motion solver) in [foam-extend 4.1] to realize dynamic mesh with topo change.

I choose the two tutorials named "circCylinder3d" and "Balltranslation" to test. They both run well in serial, however, in parallel they both crash at the first time step.

From serial to parallel, All I did was the following:
1. Getting a decomposeParDict(using method scotch).
2. Running decomposePar.
3. Running moveDynamicMesh with mpirun.

My cases and logs are attached.

I don't know if I made a mistake or it is a bug.


Does anyone have the same problem with me?
Any suggestions would be greatly appreciated!

Best regard,
Zane

----------------------------------------------------------------------------------

Part of ERROR:
[DESKTOP-IT4S0I8:03194] *** Process received signal ***
[DESKTOP-IT4S0I8:03194] Signal: Segmentation fault (11)
[DESKTOP-IT4S0I8:03194] Signal code: (-6)
[DESKTOP-IT4S0I8:03194] Failing at address: 0x3e800000c7a
...
[DESKTOP-IT4S0I8:03196] [ 0] /lib/x86_64-linux-gnu/libc.so.6(+0x3ef10)[0x7fcf35d8ef10]
[DESKTOP-IT4S0I8:03196] [ 1] /lib/x86_64-linux-gnu/libc.so.6(gsignal+0xc7)[0x7fcf35d8ee87]
[DESKTOP-IT4S0I8:03196] [ 2] [DESKTOP-IT4S0I8:03194] [ 0] /lib/x86_64-linux-gnu/libc.so.6(+0x3ef10)[0x7f57a2263f10]
...
[DESKTOP-IT4S0I8:03194] [ 3] /home/zane/foam/foam-extend-4.1/lib/linux64GccDPInt32Opt/libfoam.so(_ZN4Foam8polyMesh15resetPrimitivesERKNS _4XferINS_5FieldINS_6VectorIdEEEEEERKNS1_INS_4List INS_4faceEEEEERKNS1_INS9_IiEEEESI_RKSF_SK_b+0xfb)[0x7f9e9188052b]
[DESKTOP-IT4S0I8:03193] [ 4] /home/zane/foam/foam-extend-4.1/lib/linux64GccDPInt32Opt/libdynamicTopoFvMesh.so(_ZN4Foam17dynamicTopoFvMes h20moveCoupledSubMeshesEv+0x5bc)[0x7f9e6ecf09fc]

[Zane]

Can anyone help me, please?

Thanks in adavance.

[Bloerb]

Topology changes aren't easily parallelized. If you have topology changes on the boundary between processors the solver won't know what it should couple to. Or if you delete all cells on a processor it can't solve anything anymore. I am uncertain about foam extend, but the topology changing solvers inside openfoam in general do not like that. Tools like load balancing and so on exist, but in all likely-hood you have to integrate those yourself and the solvers won't use those out of the box. The easiest way is to make sure that all your topology changes are happening inside the processor domain and that you do not delete to much of the processors cells.

[Zane]

Hi Bloerb,

First, thanks to your kind reply!

You're right. Changing topology in parallel is complicated, but I think dynamicTopoFvMesh in foam-extend is capable of doing that. Because when I checked the code of dynamicTopoFvMesh, I saw some words like "in parallel", "multiThread" and "loadBalancing". Therefore, I just copied a tutorials using dynamicTopoFvMesh and ran it in parallel, but somehow I failed (it crashed in the very beginning).

Originally Posted by Bloerb

Tools like load balancing and so on exist, but in all likely-hood you have to integrate those yourself and the solvers won't use those out of the box. The easiest way is to make sure that all your topology changes are happening inside the processor domain and that you do not delete to much of the processors cells.

Do you mean decomposing manually to make sure processor boundaries do not intersect with the topoChanging domain? That is a good idea. However, as far as I know, dynamicTopoFvMesh is applied to global mesh. In other words, topoChanging could happen anywhere. It is very difficult to realize what you mean, so I am stuck again. Thank you all the same.

Best regard.

[deepsterblue]

I haven't maintained this code for years now, so it's quite likely that recent changes in the foam-extend code have caused things to go awry. If you're willing to revert to OpenFOAM-2.3.x (the last maintained version that I know worked in parallel), you're welcome to try the Port-2.3.x branch on github.com/smenon. Unfortunately, I don't have the time to keep up with recent API changes an maintain it for newer versions.

[Zane]

Hi Sandeep,

Thank you very much for your reply and your code implementation.

I will try dynamicTopoFvMesh in OpenFOAM-2.3.x in the near future. If I succeed, I will reply in this thread later.

[treem22]

Zane,

I take it you were never able to successfully use dynamicTopoFvMesh in parallel with OpenFOAM-2.3.x? I'm currently using foam-extend 4.0 on a heart valve simulation that requires mesh deformation and am wondering if it's worth switching versions around and trying again.

Any update?

[Zane]

I’m sorry I didn't try to use dynamicTopoFvMesh in parallel in OpenFOAM-2.3.x, and I have given up, because it's time-consuming for me now. So I can't answer your question, you'll have to try it yourself.

[cdz_CFD]

Originally Posted by treem22

Zane,

I take it you were never able to successfully use dynamicTopoFvMesh in parallel with OpenFOAM-2.3.x? I'm currently using foam-extend 4.0 on a heart valve simulation that requires mesh deformation and am wondering if it's worth switching versions around and trying again.

Any update?

Did you try dynamicTopoFvMesh in parallel with OpenFOAM-2.3.x ? Did it run successfully in parallel ？ I am looking forward to your reply, thanks.