[Schag]

Hi,

I'm looking for a solution to define a cutting plane for parallel running.
My domain is a U pipe, with an inlet lenght different from the outlet one.
Looks like that:

inlet...... xxxxxx
.................... x
.................... x
outlet xxxxxxxxx


The problem is that methods simple/metis/herarchical do not cut it at the middle of the U turn, so the interface is not optimal. I could eventually use a manual method, but don't have an example of implementation. Is there another solution?

Regards

Julien

[henrik]

Julien,

how many processors?

It's important to have an equal amount of cells on each processor. The size of the processorProcessor patch is secondary.

Henrik

[Schag]

Only 2 processors for the moment, about 160k cells each...and 1200 faces for interface.
I have no temporal gain compared to monocore simulation...

[henrik]

Dear Julien,

did you start your simulation in parallel?

mpirun -np 2 julienFoam -parallel

Henrik

[Schag]

yes, I did that correctly I think, I used that command line, and my 2 processors are busy.

I have a gain in executionTime, but not in clockTime (and the computer isn't doing anything else at the same time).

I though, maybe my computer is not adapted to that kind of parallel using. (AMD Athlon(tm) Dual Core Processor, Debian 64bit)

[henrik]

Julien,

your hardware is fine.

I have a gain in executionTime, but not in clockTime (and the computer isn't doing anything else at the same time).

I am confused about this. Please post the log of last few iterations.

Please check in top that the two processes together consume about the same memory than the single processor process.

Which parallel comms library are you using?

echo $WM_MPLIB.

Try a different one. LAM rather than OPENMPI or vice versa. You probably need to compile Pstream.

Incidentely, is this a binary installation? If yes, it might be worth a try recompiling everything.

Henrik