[AnnaF]

Hi all

I am trying to decompose my case with the scotch (and simple) method with baffle/patch preservation, but my baffle (set up as cyclic) gets split between two processors.

It is a 2D mesh with a vertical surface which is defined as a baffle/faceZone. The baffle gets split between two processors when I decompose the domain into 14 subdomains, which leads to problems later on. I don't get any error messages, but I get a message that the baffle is specified:

  “…
  Selecting decompositionMethod scotch
  Selecting decompositionConstraint preserveBaffles
  preserveBaffles : setting constraints to preserve baffles
  Selecting decompositionConstraint preservePatches
  preservePatches : adding constraints to keep owner of faces in patches 1(plateBaffle) on same processor. This only makes sense for cyclics.
  Selecting decompositionConstraint preserveFaceZones
  preserveFaceZones : adding constraints to keep owner and neighbour of faces in zones 1(plateBaffle) on same processor
  …

Please find pictures attached that show the baffle in red and the mesh parts of the two processors containing the baffle in blue and gray.

Things do work (baffles/patches exist and give correct results) when I decompose into e.g. 4 subdomains, but this could just be lucky as it doesn't have to split the baffles.

One question in general: Shouldn't baffles (or patches) be preserved with the scotch method automatically?


As it didn't preserve my baffles automatically, I added constraints for preservation to my decomposeParDict as follows:

numberOfSubdomains 14;

method          scotch;

constraints
{
    // Keep owner and neighbour of baffles on same processor
    // (ie, keep it detectable as a baffle).
    // Baffles are two boundary face sharing the same points
    preserveBaffles //baffles
    {
        type    preserveBaffles;
        enabled true;
    }
    preservePatches //patches
    {
        type    preservePatches;
        patches (plateBaffle); //(plateFront plateBack plateBaffle); //NO COMMAS!
        //patches   (plateFrontPatch_half0 plateBackPatch_half0 plateBafflePatch_half0);
        enabled true;
    }
    preserveFaceZones //faces
    {
        type    preserveFaceZones;
        zones   (plateBaffle); //(plateFront plateBack plateBaffle);
        //zones   (plateFrontPatch_half0 plateBackPatch_half0 plateBafflePatch_half0);
        enabled true;
    }
}

I have tried a few options as you can see, but all give me the same result: a decomposition across the baffle, no error messages. Of course, I could find a different number of subdomains which works out as a workaround, but I would like to have 14 for comparative reasons and it might be a bug (?)....

Please find my case attached.

Maybe someone has an idea?

Best wishes
Anna

[AnnaF]

I still do have this problem... does anyone have an idea?

[syavash]

Originally Posted by AnnaF

I still do have this problem... does anyone have an idea?

Hi Anna,

I am working with an older version of OpenFOAM (2.3.1), but I've got the same problem. I need to preserve a certain planar inlet patch on one process.
I have used similar options in decomposePar, but they apparently not working.

Did you find a solution for this issue?

Kind regards,
syavash

[AnnaF]

Dear syavash

Unfortunately, I couldn't solve the problem with the scotch decomposition method. Instead I used manual decomposition to avoid this.

Good luck

Anna

[syavash]

Originally Posted by AnnaF

Dear syavash

Unfortunately, I couldn't solve the problem with the scotch decomposition method. Instead I used manual decomposition to avoid this.

Good luck

Anna

Anna,

Happy to hear that it worked for you.
How many processes you decomposed into?
I have seen how manual method works, but I wonder if there is a straightforward way to use it for large number of processes (over 100)?

Kind regards
Syavash

[AnnaF]

Hi again

I have not looked into it for a larger number of processes. I only decomposed into 14 subdomains which was obviously not much effort to do it manually. I will have another look at my case and if I find out more I would give you an update.

Best wishes
Anna

[simply-alex]

Are you still facing the same problem?

Depending on which type of internal boundary you introduce, OpenFOAM wants different constraints (as you know).

For baffles

    baffles
    {
        type    preserveBaffles;
        enabled true;
    }

For cyclics

    patches
    {
        type    preservePatches; //set (".*") for all patches, & DONT USE COMMAS!
        patches (cyclicPatchA0 cyclicPatchA1 cyclicPatchB0 cyclicPatchB1);
        enabled false;
    }

for AMIcyclics

    processors
    {
        type    singleProcessorFaceSets;
        sets
        (       (cyclicAMI_A    -1) //these names are specefied in topoSetDict
                (cyclicAMI_B    -1) //these names are specefied in topoSetDict
        );
        enabled true;
    }

could it be that you are using AMI cyclics in your setup?

[l_fornasari]

Hey Ana,

I was having a similar problem with baffles aparently being split between processors, but it turns out the problem had to do with using GAMG to solve the matrices.

[8] --> FOAM FATAL ERROR: (openfoam-2306)
[8] Attempt to cast type cyclic to type processorLduInterface
[8] 
[8]     From Type& Foam::refCast(U&) [with Type = const Foam::processorLduInterface; U = const Foam::lduInterface]
[8]     in file lnInclude/typeInfo.H at line 154.
[8] 
FOAM parallel run aborting

I still dont quite understand why, but changing my solvers to conjugate gradients solved my problem.


Hope it helps..