OpenFOAM
Quick-start: Example use of OpenFOAM on BASE cluster
For the example we will use one of the tutorial cases.
module load green-spack
module load openfoam
First time users need to create their $WM_PROJECT_USER_DIR:
export WM_PROJECT_USER_DIR=$HOME/OpenFOAM/$USER-$WM_PROJECT_VERSION
mkdir $WM_PROJECT_USER_DIR --parent
copy the damBreak tutorial case into the $WM_PROJECT_USER_DIR:
cp -r /gpfs/mariana/software/CentOS-spack/0.17.1/opt/spack/linux-centos7-skylake_avx512/gcc-10.3.0/openfoam-1912-iic3o75w73mduhfhqxb3gqra73txale7/tutorials/multiphase/interFoam/laminar/damBreak/damBreak $WM_PROJECT_USER_DIR/
cd $WM_PROJECT_USER_DIR/damBreak
pwd
Now we can run the OpenFOAM case step-by-step or as a batch job.
srun --partition=common -t 2:10:00 -−pty bash
blockMesh
setFields
interFoam
NB: Do not use the Allrun script(s) of the tutorials, as these may try to launch parallel jobs without requesting resources.
Visualization of the results (create case.foam file to load in ParaView):
touch case.foam
paraview
Open case.foam in ParaView.
Interactive single process
For a non-parallel run of the tutorial case, the decomposeParDict needs to be removed:
mv system/decomposeParDict system/decomposeParDict-save
Running the damBreak case step-by-step interactively:
module load green-spack
module load openfoam
srun --partition=common -t 2:10:00 -−pty bash
blockMesh
setFields
interFoam
Batch-job (non-interactive) parallel job
Alternatively, we can run the job in parallel as a batch job:
(If you removed/renamed the decomposeParDict before, copy it back: cp system/decomposeParDict-save system/decomposeParDict)
The openfoam.slurm script:
#!/bin/bash -l
#SBATCH -n 4
#SBATCH -t 00:10:00
#SBATCH -J openfoam-damBreak
# #SBATCH --partition=green-ib
#the following 3 lines are only needed if not done manually in command-line
#before submitting the job
source /usr/share/lmod/lmod/init/bash
module load green-spack
module load openfoam
blockMesh
decomposePar
setFields
srun interFoam -parallel
reconstructPar
and then run in the command-line (module commands are only needed if not in sbatch script):
module load green-spack
module load openfoam
sbatch openfoam.slurm
Which module and which node to use?
Here are timings for the simpleFoam solver with the motorBike case on empty nodes:
Empty nodes:
| module\ node | green (empty) | gray (empty) | green (full) |
|---|---|---|---|
| openfoam-v1912 | 16m0.781s | 20m30.122s | 40m15.321s |
| spack-green | 15m18.835s | - | 37m17.363s |
| spack-gray | 15m18.537s | 19m8.511s | 37m53.517s |
Surprisingly, the timing between the different modules is not much different. However, on full nodes we experienced a significant difference to the empty nodes. This is probably due to two reasons, full nodes cannot run on boost-clock-frequency and there may be congestion of the memory lanes. A Xeon Skylake CPU has only 6 memory lanes to be shared by 20 cores. If a memory intensive application is on the other cores, this may cause a slow-down.
Pre-processing (geometry and mesh generation)
The geometry and mesh can be either hand-coded using blockMesh or with Gmsh, FreeCAD or Salome. When using Gmsh, be sure to save the mesh in v2 ASCII format (see separate page on CAD-mesh. This creates a volume mesh.
To convert a Gmsh volume .msh file for OpenFOAM, use
gmshToFoam meshfile.msh
Another possibility is to use CAD for a surface mesh and use the snappyHexMesh utility to adapt a blockMesh volume mesh to the surface (see OpenFOAM motorcycle tutorial).
Visualizing the results (post-processing)
Login to viz, change to the case directory, create an empty .foam file for the case
touch damBreak.foam
and then use the regular ParaView
paraview
and open the .foam file from the menu
Comparison of the execution time
It is educational to check the runtime of the code using the time command, e.g. for the single-thread
time interFoam
and for the parallel run (in the openfoam.slurm script)
time mpirun -n $SLURM_NTASKS interFoam -parallel
As the damBreak case is quite small, it is likely that the parallel run is not faster than the sequential, due to the communication overhead.
In a testrun, the resuls have been as follows:
| time type | sequential | parallel |
|---|---|---|
| real | 0m8.319s | 0m39.463s |
| user | 0m6.927s | 1m1.755s |
| sys | 0m0.432s | 0m2.922s |
Lesson to be learned: Parallel computation is only useful for sufficiently large jobs.
NOTE: Parallel does not (necessarily) mean faster!!! Parallel execution introduces overhead (starting threads, communication)! For optimal execution time and optimal use of resources one needs to test and find the sweet spot.
The division into the areas is a combined decision taking into account “real” (wall clock) and “user” (summed time of all threads) time (from the time command). “Wall clock” (real) time is the time one needs to wait till the job is finished, “Summed thread time” (user) is the sum of the times that all individual threads needed, it should be roughly user = numtreads x real. For parallel programs, one can expect that “user” time of the parallel run is larger than for the sequential, due to communication overhead, if it is smaller, that probably means the individual threads could make better use of cache.
| area | why | explanation |
|---|---|---|
| sweet spot | minimal "user" time | = minimal heat production, optimal use of resources |
| good range | linear speedup for "real", with constant or slightly increasing "user" | |
| OK range | slightly less than linear speedup for "real", and slightly increasing "user" | |
| avoid | ascending slope in the diagram for "real" and "user" | one actually needs to wait longer compared to the case with fewer cores |
Recommended in this case would be to request 8 threads -n 8 --ntasks-per-node 8 but use mpirun -n 4. OpenFOAM does not seem to benefit from hyperthreading.