[Gmsh] meshing partitions

[Christophe Geuzaine]

> On 11 May 2020, at 16:32, Hansjoerg Seybold <hansjoerg at fisica.ufc.br> wrote:
> 
> Hello,
> Thanks a lot for the reply. Using the visibility filter worked. I
> created a simple test case (attached)
> which creates the two mesh pieces. So now I have two meshes
> multizone_1.msh and multizone_2.msh
> which I can load in a geo file
> --------------------------------------
> Merge "multizone_1.msh";
> Merge "multizone_2.msh";
> Coherence Mesh;
> Save "multizone_merged.msh";
> Save "multizone_merged.bdf";
> Exit;
> --------------------------------------
> I understood that the coherence mesh removes the duplicate nodes on
> the common edge and the above
> script displays the mesh in the gui correctly.
> However the export is distorted as the node ordering is not updated.
> I would get the development snapshot to try to use the mesh tag
> suggestion, but how would I do the reordering
> starting from the script above?
> 

You would do:

Mesh.FirstNodeTag=1;
Mesh 2;
Save "multizone_1.msh";

...

Mesh.FirstNodeTag=Mesh.NbNodes + 1;
Mesh 2;
Save "multizone_2.msh";


> regarding openmp:
> 
> We (the IT cluster staff at ETH Zurich and myself) tried to get the
> openmp version of gmsh running on the cluster (CentOS 6), but without
> success.
> I managed to compile and run it on my ubuntu PC, but with very little
> speedup for the selected meshing algorithm.
> I tested openmp with two physical groups active and -nt 2, on a finer
> LC as in the attached geo for speed measurements and Mesh.Algorithm =
> 9; gmsh-4.4.1,
> a significant speedup could only be observed for del2d, but quality
> tests showed that the 'pack' algorithm gave the best meshing result

Ok I see. The "pack" algorithm has not been worked on for a while and is indeed currently sequential. We are working on updating this algorithm, so hopefully it will get much faster (and parallel) soon.

> BTW, is it possible to apply different meshing algorithms to different
> physical groups?
> 

Yes. In .geo files: MeshAlgorithm Surface {...} = ...;

Christophe


> Regarding the 5 days meshing: I am using quad meshing (algo 9) with a
> not too sophisticated sizing function but with a very large domain,
> del2d is orders
> of magnitudes faster.
> 
> Thanks a lot for the help.
> 
> Hansjoerg
> 
> 
> 
> 
> 
> 
> 
> On Sun, May 10, 2020 at 9:05 AM Christophe Geuzaine <cgeuzaine at uliege.be> wrote:
>> 
>> 
>> 
>>> On 9 May 2020, at 18:09, Hansjoerg Seybold <hansjoerg at fisica.ufc.br> wrote:
>>> 
>>> Hello,
>>> I am trying to mesh a model consisting of several physical groups each
>>> group at a time and then merge the meshes for the different physical
>>> groups afterwards. The reason why I am trying to perform this "domain
>>> decomposition" is that the meshing of the full model takes over 5 days
>>> and exceeds the runtime limit of the queuing system.
>>> 
>>> However I could not find a simple way to apply the  "Mesh 2" to a
>>> specific physical group. Gmsh always tries to mesh the whole model.
>>> 
>> 
>> You could either delete the parts you don't want to mesh; or hide the parts you don't want to mesh (cf. the `Show` and `Hide` commands in .geo script, or `setVisibility()` in the api) and use the `Mesh.MeshOnlyVisible` option.
>> 
>>> My question would be if anybody has a hint how to perform this meshing
>>> in parts and how to combine the resulting meshes into a single final
>>> model.
>>> 
>> 
>> That's trickier as each mesh will be independent. The development snapshot allows you to set the starting node/element tag (Mesh.FirstNodeTag, Mesh.FirstElementTag), which will help. Removing duplicate nodes when you merge things together can be done with Coherence Mesh (in .geo files) or removeDuplicateNodes() in the api.
>> 
>> PS: 5 days to perform a 2D mesh ? Anything special in the geometry/size field? If you don't do this already at least recompile Gmsh with OpenMP enabled, and mesh in parallel?
>> 
>> Christophe
>> 
>>> Thank you very much.
>>> 
>>> Best,
>>> hansjoerg
>>> 
>>> _______________________________________________
>>> gmsh mailing list
>>> gmsh at onelab.info
>>> http://onelab.info/mailman/listinfo/gmsh
>> 
>> —
>> Prof. Christophe Geuzaine
>> University of Liege, Electrical Engineering and Computer Science
>> http://www.montefiore.ulg.ac.be/~geuzaine
>> 
>> 
>> 
> <multizone_1.geo>_______________________________________________
> gmsh mailing list
> gmsh at onelab.info
> http://onelab.info/mailman/listinfo/gmsh

— 
Prof. Christophe Geuzaine
University of Liege, Electrical Engineering and Computer Science 
http://www.montefiore.ulg.ac.be/~geuzaine