OpenFOAM in parallel: Repartitioning filtered-out mesh

Official adapters and tutorials
,
1

I am trying to run parallel simulation of flow-over-heated-plate tutorial in a HPC.

The tutorial runs well in my local PC (using single processor and also in parallel using multiple processors).

The tutorial runs well in HPC when using single processor.

The tutorial fails when trying to run in parallel using 1 node or multiple nodes of HPC.

When using multiple nodes, the two openfoam sovlers are unable to communicate with each other.

When using a single node, I am getting the following error

error message
error message1657×211 14.8 KB

First, I would like to deal with running the simulation in parallel in a single node.

Based on the error message, I have tried the following

[1] Coupled participants do not match geometry wise - I am sure there might not be any issue with the geometry because it runs well (in parallel) in my local PC.
[2] I have tried to increase the safety-factor = 1.0. It did not work.
[3] I have set the geometrifilter=no-filter. It did not work.

I have attached my case files, SLURM job file and config file, herewith.
flow-over-heated-plate.zip (16.6 KB)

Thanks in advance for any help or direction.

2

@Deepak you probably found the reason already, but I am trying to not leave any topic without an answer.

I can reproduce this with 1 fluid rank and 17 solid ranks on my laptop.

The reason is very simple: The solid interface only has 32 cells per x-direction. In the system/decomposeParDict of the solid-openfoam, you are defining a partitioning in 26 subdomains in the x-direction. This leads to some parts of the interface having only one cell.

While one could argue whether not supporting one-cell interfaces is a bug or not, I would argue that in practice one should keep in mind that such dividing the case in so small subdomains typically increases communication so much that kills any kind of performance benefits that parallelism brings.