[nicolarre]

Hi,

I've been having some problems when trying to run simpleFoam from a decomposed mesh.

I want to run simpleFoam on a very dense mesh made with snappy from an .Stl file with several objects in it. The mesh is so big that i cannot mesh it on a single PC without running out of memory, nor can i reconstruct if after meshing it in parallel for the same reason.

For this i decomposed the case in 20 parts and snappyHexMesh'd it. Prior to decomposing, i set up the boundary and initial conditions on my U, p, K etc.. files at time 0, then proceeded to make the blockMesh and snappy. In doing so i ran into several problems.

Problem #1: For some strange reason, all my boundary conditions for the patches in my stl model do not appear in the /processor*/U , p , K files and im left only with the boundary conditions the six faces of the blockMesh and the processorXtoY patches.

Problem #2: My geometry tends to have skewfaces and my solutions usually diverge because of them. I made a bash macro that removes cells wiith Skewfaces using cellset and subsetMesh. The problem is that even if i make it so that the macro removes skewfaces one processor at a time, if some processor has 0 skewfaces in it, it will no longuer generate the oldInternalFaces patch on that processor. idk if this might cause problems or not since this patch will only apear in some processors but not in all of them.

i have more problems but they all can be fixed if i can solve problem #1

am i doing something wrong? or was it always complicated to simulate from a decomposed mesh?

[gwierink]

Hi Nicolarre,

Have you used "simple" or e.g. "metis" to decompose the case? Did you first decompose and then run snappy? Perhaps it's an idea to do a fairly coarse snappy mesh, decompose the case (metis method), and then use refineMesh in parallel to refine the mesh to the level you want. I'm not sure how well this will wok, but it may get you further ...

[nicolarre]

Originally Posted by gwierink

Hi Nicolarre,

Have you used "simple" or e.g. "metis" to decompose the case? Did you first decompose and then run snappy? Perhaps it's an idea to do a fairly coarse snappy mesh, decompose the case (metis method), and then use refineMesh in parallel to refine the mesh to the level you want. I'm not sure how well this will wok, but it may get you further ...

Hi qwierink,

Thats exactly what i did, but now i have a diferent problem

Using the same decomposition method for both parallel runs, i first made a coarse mesh in a single pc, then decomposed to get the correct /0/ boundary fields, then i made a fine mesh in parallel.

i use the /0/ time from the first run and the mesh from the second run. In theory it should work but for sime strange reason, when i run snappy i get a message that goes something like:

"processor16to5 is not defined in processorXX/0/p::boundaryfields"

even though i decomposed both the coarse and fine mesh with the same decomp options (and in heriarchical), the finer mesh has more processorXtoY patches than the coarse one.

Does it make sense? i though heriarchical would take care of this.