FLUENT Speed Issues on Cluster
 

Hi,

I have several FLUENT sessions running concurrently at a supercomputing facility. The scripts are prepared and the job is launched from MATLAB. The geometries / mesh cell size are the same, with only minor differences between the sessions (eg velocity and angle of attacks).

I have noticed there are drastic speed issues between the sessions. Submitting a direct job through qsub, I get a converged solution (eg maximum iterations of 3000) within minutes, yet the jobs prepared and launched from MATLAB are taking hours.

Therefore, I'm not getting the accelerated computation that would be expected by parallelising the geometry (in this case I use 4 nodes per geometry) on the cluster. However, when I directly submit the same case by using qsub we get an accelerated solution; but not from MATLAB.

Can you please advise what is causing these speed issues? Attached below
is the code:
================================================== ========================
#!/bin/bash

. /etc/profile.d/modules.sh

export LD_LIBRARY_PATH=/usr/lib64
cd shape1
module load fluent/v120
fluent 2ddp -g -t4 -i
/home/ /matlab/xfoil_optz/shape1/Fluent_Airfoil_1.jou >&
/home/ /matlab/xfoil_optz/shape1/fluent_1.out
================================================== ========================
All flow conditions are in Fluent_Airfoil_1.jou.

Thanks

Maybe Matlab doesn not let you to fork four processes when you launch Fluent from it. Do you have a parallel Matlab license ?

If you want to be safe better use a simple bash script to launch Fluent !!!

Do

No, MATLAB is indeed running in parallel. I have alot of operations prior to calling FLUENT (eg. mesh generation) which all run as expected in parallel. Also in the case reported, I had 8 concurrent FLUENT sessions running; 5 converged in ~17 mins, while the remaining 3 took over 5 hours!

I do use a bash script for FLUENT (attached in original msg.), so that works fine (sometimes). Again the mesh size and type are the same, but it seems in some cases there are some major issues which affect the speed for some random cases.

Any ideas?