Learning Forum

[memonz]

I’m running ANSYS Fluent v2025 R2 (3ddp solver) on a Slurm-based HPC cluster. I have access, modules load correctly, but when I submit a batch job distributed over multiple nodes, Fluent/OpenMPI ends up placing all 64 MPI ranks on one node and then stalls (application times out).

#!/bin/bash

#SBATCH --job-name=fluent_geometry7

#SBATCH --account=mime

#SBATCH --partition=share

#SBATCH --ntasks=64

#SBATCH --cpus-per-task=1

#SBATCH --time=06:00:00

#SBATCH --output=fluent_geometry7_%j.out

#SBATCH --error=fluent_geometry7_%j.err

#SBATCH --constraint=avx2

#SBATCH --hint=nomultithread

 

module load ansys/v2025r2

module load fluent

export OMP_NUM_THREADS=1

 

cd /nfs/hpc/share/memonz/fluent_porous_liquid

 

fluent 3ddp -g -t${SLURM_NTASKS} -mpi=openmpi -i run_geometry7.jou

Journal file (run_geometry7.jou):

/file/start-transcript fluent_geometry7_run.trn
/file/read-case-data geometry_rev6-2-0.510000.cas.h5
/file/auto-save/data-frequency 200
/file/auto-save/case-frequency 200
/file/auto-save/name-format "autosave-%t"
/solve/monitors/residual/print? yes
/solve/monitors/residual/convergence-criteria? yes
/solve/set/time-step 0.001
/solve/dual-time-iterate 500 5
/file/write-case-data geometry7_completed.cas.h5
/file/write-data geometry7_completed.dat.h5
/file/stop-transcript
/exit yes


Observed output (fluent_geometry7_.out, head):

Connected License Server List: 1055@linlic8.engr.oregonstate.edu
--------------------------------------------------------------
This is an academic version of ANSYS FLUENT. …
Host spawning Node 0 on machine "cn-c20.hpc.engr.oregonstate.edu" (unix).
… -mpi=openmpi -ssh -mport … 
mpirun … --np 64 --host cn-c20.hpc.engr.oregonstate.edu:64 … 
Application is timed out, aborting …