COMM_LEVEL setting affecting flow solution and residual evolution when using mpirun on SU2_CFD

[MARCUS-021]

Changing between COMM_LEVEL = FULL and MINIMAL affects the output of the solution when using mpirun. This setting does not seemingly affecty SU2_CFD run without MPI.
This was tested on the standard rae2822 test case with mpirun -n 16 and the following files taken directly from the repository:

Config file: turb_SST_RAE2822.cfg
Mesh: mesh_RAE2822_turb.su2

The only modification made is adding COMM_LEVEL at the bottom of the cfg.

When using FULL the simulation converges in 2618 iterations and satisfies the rho residual at -8.
When using MINIMAL the simulation converges in 1787 iterations, although the final rho residual is -6.66147.

Attached are the config and mesh files, as well as a screenshot of the rho residuals vs number of iterations as seen from the history output (in green the one where COMM_LEVEL = FULL).

My assumption is that COMM_LEVEL is affecting more than just the warning output, and is instead changing the way parallel computation is performed.

The simulation is run on

• OS: DockerDesktop container using Ubuntu 24.04 (see below)
• C++ compiler and version: GCC v13.3.0
• MPI implementation and version: Open MPI v4.1.6
• SU2 version: v.8.1.0

For the sake of completeness I am also attaching the Dockerfile used to generate and build the docker image used, which includes some custom compiler options.

Files.zip

[pcarruscag]

Comm level is an option for testing performance ignoring communications to help assess the communication overhead. It should not be used outside of that context. The results will be wrong.

[MARCUS-021]

I understand, I was told to post this on the issues page by somebody else on the CFD online forum. Good to know it isn't an issue and just a misunderstanding of the option on my part. Thank you very much.