[hconel]

Hello all,
I have used to generate my mesh in this order:

METHOD 1

runApplication blockMesh
runApplication decomposePar -copyZero
runParallel snappyHexMesh -overwrite

i.e. decomposing the domain before refining certain regions with snappyHexMesh (by the way, I am using scotch method in decomposition). But then, I thought that running decomposePar before snappyHexMesh would cause inefficient decomposition of the domain (which is just generated by blockMesh and is coarse) and reversing the order would provide more efficient decomposition, such that:

METHOD 2

runApplication blockMesh
runApplication snappyHexMesh -overwrite
runApplication decomposePar -copyZero

Now, the second method obviously requires a significantly longer computation time. The thing is, both methods grants similar cells/processor ratio, in contrary to my intuition! Does OpenFOAM redistributes cells to processors after running snappyHexMesh in METHOD 1?

Thanks in advance.

PS: My final mesh includes unevenly distributed refined regions.

[Antimony]

Hi,

I did some similar testing a while back and had put it up on cfd-online. Here is the link to that thread: Parallelize snappyHexMesh optimally?

Hope this helps.

Cheers,
Antimony

[hconel]

Originally Posted by Antimony

Hi,

I did some similar testing a while back and had put it up on cfd-online. Here is the link to that thread: Parallelize snappyHexMesh optimally?

Hope this helps.

Cheers,
Antimony

This is exactly what I am looking for, thanks! So the execution order does not matter, snappyHexMesh will redistribute the processor loading.

[Antimony]

Hi,

Yes, if your decompositionMethod is scotch, then snappyHexMesh seems to do load balancing at every step.

Cheers,
Antimony