[Typ]

Hello,
i wanted to run the Case "damBreak" in Parallel on 6 Processors. But when i excecute my Allrun file the calculation just finish really quick and i have no time directory or processor0, processor1... directorys in my case directory.

First I changed the decomposeParDict like this:

/*--------------------------------*- C++ -*----------------------------------*\
| =========                 |                                                 |
| \\      /  F ield         | OpenFOAM: The Open Source CFD Toolbox           |
|  \\    /   O peration     | Version:  v1912                                 |
|   \\  /    A nd           | Website:  www.openfoam.com                      |
|    \\/     M anipulation  |                                                 |
\*---------------------------------------------------------------------------*/
FoamFile
{
    version     2.0;
    format      ascii;
    class       dictionary;
    object      decomposeParDict;
}
// * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * //

numberOfSubdomains 6;

method          simple;

coeffs
{
    n           (3 2 1);
}

distributed     no;

roots           ( );


// ************************************************************************* //

Then i changed the Allrun file in this way:

#!/bin/sh
cd "${0%/*}" || exit                                # Run from this directory
. ${WM_PROJECT_DIR:?}/bin/tools/RunFunctions        # Tutorial run functions
#------------------------------------------------------------------------------
runApplication decomposePar -force
runApplication blockMesh
runParallel setFields
runParallel $(getApplication)
runApplication reconstructPar -newTimes

#------------------------------------------------------------------------------

That what the Bash says:

/mnt/c/users/typus/tutorials/multiphase/interfoam/laminar/damBreak/damBreak$ ./Allrun
Running decomposePar on /mnt/c/users/typus/tutorials/multiphase/interfoam/laminar/damBreak/damBreak
Running blockMesh on /mnt/c/users/typus/tutorials/multiphase/interfoam/laminar/damBreak/damBreak
Running setFields (6 processes) on /mnt/c/users/typus/tutorials/multiphase/interfoam/laminar/damBreak/damBreak
Running interFoam (6 processes) on /mnt/c/users/typus/tutorials/multiphase/interfoam/laminar/damBreak/damBreak
Running reconstructPar on /mnt/c/users/typus/tutorials/multiphase/interfoam/laminar/damBreak/damBreak

I think that i missed to change something but i couldnt find anything in the OpenFoam Manual.

[LuckyTran]

Some output from OpenFOAM would be helpful.

You need to run blockMesh before decomposePar

You should have processor dirs if decomposePar ran. And if that was successful, you should have 0 dirs in each of the processor dirs if setFields ran.


You don't need to use setFields in parallel. Can't you just stick with the regular tutorial?

runApplication blockMesh
runApplication setFields
runApplication decomposePar
runParallel $(getApplication)
runApplication reconstructPar

[Typ]

Originally Posted by LuckyTran

Some output from OpenFOAM would be helpful.

You need to run blockMesh before decomposePar

This was the problem now it works. Thank you.

[Typ]

I have one more question.
Is there a way to omptimize the decomposePar if you for example have 40 Processors then you can split them up in different ways in:
n (10 1 4)
n (5 4 2)
...and so on

is there a way to find the best allocation? With the shortest simulation time?

[LuckyTran]

You are doing a tutorial, why worry about these things...

The answer is no, there isn't. At least, there isn't a magic wand like you want.

The time it takes you to find the best mesh with fastest possible simulation time I can almost certainly guarantee will be longer than the time it takes you run the case even with a bad mesh partitioning.

When you move on to doing productive simulations with more complex geometry, you'll likely stop using simple and move on to things like scotch and metis.


Even still, the time it takes to solve a given partition depends also on what is being solved in the partition. You could have multiple physics for example and then one partition ends up being solved much faster than another. Even with a single set of physics, the time it takes to solve the same partition is dependent on the flow itself. To complicate things even further, the time it takes to solve each partition is also dependent on your hardware... There's an endless list of things you need to optimize for to find the perfect mesh.

[Typ]

Originally Posted by LuckyTran

You are doing a tutorial, why worry about these things...

Because i want to learn and practice besides this was an generally question.
Ok so i have to test out whats the best allocation in the decomposeParDict.
Thank you.