[Sean95]

I'm just wondering how I could equally decompose a cylindrical fluid domain so that each processor has the same number of cells. I can quarter the domain using simple but I want to divide it further. The simulation crashes unless each processor has the same number of cells, the picture below is of the YZ plane.

[simrego]

Hi!


The simplest if you use scotch. It'll try to reduce the common faces of the proc. regions (smaller processor patches -> less communication between processors)
It is easy to use, you only have to specify the number of processors.

[Sean95]

I tried using scotch but the simulation crashes. The only time the mesh motion solver hasn't crashed is when each processor has the same amount of cells.

simple (1 2 2)

Processor 0
    Number of cells = 320640
    Number of faces shared with processor 1 = 10736
    Number of faces shared with processor 2 = 10736
    Number of processor patches = 2
    Number of processor faces = 21472
    Number of boundary faces = 12320

Processor 1
    Number of cells = 320640
    Number of faces shared with processor 0 = 10736
    Number of faces shared with processor 3 = 10736
    Number of processor patches = 2
    Number of processor faces = 21472
    Number of boundary faces = 12320

Processor 2
    Number of cells = 320640
    Number of faces shared with processor 0 = 10736
    Number of faces shared with processor 3 = 10736
    Number of processor patches = 2
    Number of processor faces = 21472
    Number of boundary faces = 12320

Processor 3
    Number of cells = 320640
    Number of faces shared with processor 1 = 10736
    Number of faces shared with processor 2 = 10736
    Number of processor patches = 2
    Number of processor faces = 21472
    Number of boundary faces = 12320

Number of processor faces = 42944
Max number of cells = 320640 (0% above average 320640)
Max number of processor patches = 2 (0% above average 2)
Max number of faces between processors = 21472 (0% above average 21472)

Time = 0

Processor 0: field transfer
Processor 1: field transfer
Processor 2: field transfer
Processor 3: field transfer

End

scotch (4 equally weighted processors)

Processor 0
    Number of cells = 321064
    Number of faces shared with processor 1 = 5353
    Number of faces shared with processor 2 = 5336
    Number of processor patches = 2
    Number of processor faces = 10689
    Number of boundary faces = 10773

Processor 1
    Number of cells = 319067
    Number of faces shared with processor 0 = 5353
    Number of faces shared with processor 3 = 5082
    Number of processor patches = 2
    Number of processor faces = 10435
    Number of boundary faces = 10655

Processor 2
    Number of cells = 321884
    Number of faces shared with processor 0 = 5336
    Number of processor patches = 1
    Number of processor faces = 5336
    Number of boundary faces = 15494

Processor 3
    Number of cells = 320545
    Number of faces shared with processor 1 = 5082
    Number of processor patches = 1
    Number of processor faces = 5082
    Number of boundary faces = 12358

Number of processor faces = 15771
Max number of cells = 321884 (0.3879741% above average 320640)
Max number of processor patches = 2 (33.33333% above average 1.5)
Max number of faces between processors = 10689 (35.5526% above average 7885.5)

Time = 0

Processor 0: field transfer
Processor 1: field transfer
Processor 2: field transfer
Processor 3: field transfer

End

[yambanshee]

Do you have a cyclic domain boundary? If so, it's important to have those two boundaries on the same processor. This can be done with scotch by using the 'singleProcessorFaceSets' option. For example:

FoamFile
{
    version         2;
    format          ascii;
    class           dictionary;
    object          decomposeParDict;
}

numberOfSubdomains 24;

method          scotch;

singleProcessorFaceSets ( ( cyclic -1 ) );

where cyclic is a faceSet I defined in a topoSetDict:

FoamFile
{
    version     2.0;
    format      ascii;
    class       dictionary;
    object      topoSetDict;
}
// * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * //

actions
(
    {
        name    cyclic;
        type    faceSet;
        action  new;
        source  patchToFace;
        sourceInfo
        {
          name side1;
        }
    }

    {
        name    cyclic;
        type    faceSet;
        action  add;
        source  patchToFace;
        sourceInfo
        {
          name side2;
        }
    }  

}