Executing a FEniCS script in parallel

Hi,

I have a FEniCS script which has not been written in parallel. According to the website of FEnics program, I found that by calling "mpirun -np 8 python script.py", I will be able to execute it in parallel. However; it takes longer than it took before. Does anyone know what is the reason?
Is this command enough or should I add some commands in the script?
Any helps will be appreciated.

Thanks.
Dan

asked Jun 16, 2017 by Danial FEniCS Novice (120 points)

The command is correct. My Assumtion:

Your test problem is too small. In this case, distributing takes longer than solving on a single core. Just try a bigger mesh and test different numbers of procs.

my experience in 3D with MUMPS solver and p1 polynomials:

< 5k nodes = 1 proc
< 50 k nodes = max. 10 procs
< 200 k nodes = max. 20 procs

commented Jun 16, 2017 by RR FEniCS User (3,330 points)

Agree. Also depends where it is slowing down. You can add some flags (print statements) to see where the problem is. A lot depends on what exactly you are doing.

commented Jun 16, 2017 by mwelland FEniCS User (8,410 points)

Thanks for sharing your experience. I tried with a bigger mesh and now it solves faster. Also, by using "mpirun -np 8 python script.py" I got strange results (piecewise motion) compared to the results obtained python script.py (Continues motion). Do you think what can be the reason? Should I add some other commands? Thanks.

Best,
Danial

commented Jun 17, 2017 by Danial FEniCS Novice (120 points)

What do you mean by 'piece-wise motion' mean?

commented Jun 19, 2017 by mwelland FEniCS User (8,410 points)

"What do you mean by 'piece-wise motion' ?"

That's my question, too. Maybe you can post a minimal example?

commented Jun 19, 2017 by RR FEniCS User (3,330 points)

...