Fluent is used to simulate matter and/or energy flow in 2-3 dimensions.
First, here's the correct path to fluent:
To ensure that this path preceeds any other fluent executable that might exist in your path, include the following line in your .bashrc:
If you don't know what .bashrc is, or cannot edit it by some reason, then simply enter the following commands to your terminal:
touch ~/.bashrc echo "PATH=/usr/local/packages/ansys-13.0/v130/fluent/fluent13.0.0/bin/:$PATH" >> ~/.bashrc . ~/.bashrc
Now, you have two options: batch and interactive (and also with or without GUI). Either way, you must first ask for a compute node from the scheduling system (PBS).
(1) Batch Mode: You need to prepare a PBS script, similar to:
# ====================================== #PBS -N fluent_test #PBS -l nodes=1:ppn=4 #PBS -l walltime=12:00:00 #PBS -l mem=6gb #PBS -q force #PBS -k oe #PBS -m abe /usr/local/packages/ansys-13.0/v130/fluent/bin/fluent -t4 -g < inputfile > outputfile #======================================
You could also include the following flags, where relevant:
2d : runs the two-dimensional, single-precision solver,
3d : runs the three-dimensional, single-precision solver,
2ddp : runs the two-dimensional, double-precision solver, and
3ddp : runs the three-dimensional, double-precision solver.
Note the -t parameter, which is the number of processors (or cores). In this case, 't' is equal to nodes*ppn=4 (make sure it is!!). The -g parameter tells fluent to disable GUI, since we are not using it at the moment.
(2) Interactive mode: You need to first SSH to Force head node with X-window forwarding option (ssh -Y email@example.com). If using windows, X-win would also allow you to do that. See this FAQ entry for more details.
Then, you need to ask for an interactive compute node from the PBS server:
qsub -I -X -q force -l nodes=1:ppn=4:mem=6gb,walltime=12:00:00
Note the similarity between these flags and the parameters in the PBS script file. The system will log you into an available compute node (could take a few minutes, depending on the availability). Now you can run fluent, with GUI if you like, interactively for 12 hours (or any other value for wallclock time) simply by invoking:
Don't forget to set the number of parallel nodes from the GUI!
Hints for Improved Performance:
Fluent options that make a difference in speed
If running an MPI job, you MUST specify the number of processes "-t
The MPI distribution seems to make a difference.
-mpi=pcmpi - works best on a small number of nodes. Uses shared-memory message passing where possible. Lowest latency on intra-node communication. Lowest latency on Reduce and Scatter operations. (mpitest shows this)
-mpi=intel - works best on a large number of nodes. Higher bandwidth than pcmpi. Higher intra-node latency, lower inter-node latency. Lowest latency on Broadcast. (Again, mpitest)
Use INFINIBAND if available on the node!
Specify "-pib". Otherwise, things get slower.
Testing fluent speed:
# Run MPI bandwidth test using IB:
fluent 3ddp -alnamd64 -r14.0.0 -t32 -mpi=pcmpi -pib -g -mpitest -cnf=$PBS_NODEFILE
fluent 3ddp -alnamd64 -r14.0.0 -t32 -mpi=intel -pib -g -mpitest -cnf=$PBS_NODEFILE