[PeterShi]

Dear Foamers,

I have encountered a weird problem as follows. I am running LES in parallel and my mesh has like 15 million cells. Therefore, I used decomposePar to decompose my domain into sub-domains. However, some processors do not contain any cells, i.e. they are empty.

Processor 17
Number of cells = 0
Number of processor patches = 0
Number of processor faces = 0
Number of boundary faces = 0


--> FOAM Warning :
From function Foam:olyMesh:olyMesh(const Foam::IOobject&)
in file meshes/polyMesh/polyMesh.C at line 330
no points in mesh

Because of that, my simulation is unable to run.

I appreciate any help, comments, suggestions, etc. Thank you.

Best regards,
Peter

[ssa_cfd]

Which decomposition method are you using,? Try using scotch method in decomposePardict.

[PeterShi]

Originally Posted by ssa_cfd

Which decomposition method are you using,? Try using scotch method in decomposePardict.

Hi,

Thank you for your reply.

Originally, I used simple for domain decomposition. As you suggested, I tried other methods, both scotch and hierarchical.

scotch is bit difficult to use and I still got empty processors (probably I did not use it correctly). However, hierarchical gave me pretty good results.

Does the the domain decomposition method affect the simulation time? In addition, does the domain decomposition method affect the OpenFOAM scalability?

Thank you.

Best,
Peter

[ssa_cfd]

Hi,

scotch is easy to use. Just specify the method and the number of processors. According to me, scotch is good for doing simulation in high performance computers with many processors. I have used this method in a supercomputer and it works pretty good.

I don't know about the simulation time.!

[PeterShi]

Originally Posted by ssa_cfd

Hi,

scotch is easy to use. Just specify the method and the number of processors. According to me, scotch is good for doing simulation in high performance computers with many processors. I have used this method in a supercomputer and it works pretty good.

I don't know about the simulation time.!

Hi,

Thanks for the input. So you mean the decomposition method does affect the scalability of OpenFOAM?

Here is an example of my scotch:

numberOfSubdomains 128;

method scotch;

scotchCoeffs
{
processorWeights
(
1
1
1
1
1... (128 ones in total, means I want my mesh to be decomposed to processors evenly)
);
strategy "b";
}

distributed no;

roots ();



Perhaps something is wrong, but I still got "empty" processors (processors do not have any mesh points) with above scotch. Could you please point out what is wrong in my way of using scotch? Specifically, I am not sure what is "strategy" I should use (here I used "b", same as an example in a tutorial case)?

If possible, could you please post your scotch as well?

Thank you so much.

Best regards,
Peter