parallel calculation on multiple computers
 

Hi,

I recenty figured out how to run snappy parallel on multiple computer. The method described in the user guide (Chapter 3) didn't work for me and I stumbled upon the solution by sheer accident. Instead of typing cpu=x after the name of the PCs, I used slots=x.

I tried various methods to execute the command such as:

mpirun --hostfile machines -np 6 snappyHexMesh -parallel

OR

mpirun --hostfile machines -np 6 foamJob -p -s snappyHexMesh

But I only received "orted: command not found. A daemon (pid x) died unexpectantly with status x while attempting ..."

Afterwards I just executed the snappy command for one PC.

foamJob -p -s snappyHexMesh

That actually worked.

But now I am unable to run simpleFoam on multiple PCs. If I use the command with "mpirun --hostfile ... simpleFoam -parallel", then I receive the same error as above.

Does anyone have any experience with this?

Thanks in advanced.

Patrick

Problem solved.

Can you please elaborate?

I have similar problems, but with a high perfomance cluster. (bsub on multiple machines).

I can run mpirun.lsf interFoam, but mpirun.lsf foamJob -p interFoam gives the error mentioned above

Hi,

to be honest, I wasn't the one who fixed the problem, my supervisor did and I'm not exactly sure what he did. Everytime I used OpenFOAM, I had to type in setfoam to get all the environment settings for OF adjusted. He just changed a few settings so I didn't have to type in setfoam everytime I opened a new terminal. Afterwards he told me to try run the calculation. I used the command in the user guide and it worked. He is often not at the office, so I can' t really asked him for the specifics.

Hope this helped a bit.

Patrick

Not really, but I've found a work-around for my problem already. My main problem, is that I have to use bsub for a run on a high performance cluster and there are some problems combining that with openmpi. Not really OpenFoam's fault though.
The work-around is to use only 1 processor on the HPC, with several nodes instead of spreading a job on multiple processors.