[shimakasaei]

Hi all,
I am running a simulation with olafoam solver, I want to run in parallel but I can't although it works when I run it with one processor.
I did put decomposeParDict in system folder:
FoamFile
{
version 2.0;
format ascii;
class dictionary;
location "system";
object decomposeParDict;
}
// * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * //

numberOfSubdomains 32;

method hierarchical;

hierarchicalCoeffs
{
n ( 4 2 4 );
delta 0.001;
order xyz;
}

I want to use 2 nodes with 16 processors each,
I want to check the order of the commands that I am using to see whether I make mistake while writing commands.
These are the commands that I am using:

blockMesh
decomposePar
mpirun -np 32 setFields -parallel
mpirun -np 32 olaFlow -parallel


I will appreciate for any help.

Thanks,
Shima.

[Andrew.Coughtrie]

Hi Shima,

What are the errors when you run your commands, are you running on a system which uses a job scheduler.

Andrew

[shimakasaei]

Originally Posted by Andrew.Coughtrie

Hi Shima,

What are the errors when you run your commands, are you running on a system which uses a job scheduler.

Andrew

yes, I assign a job for running my simulation. Now, when I try it with this order, it seems to work:

decomposePar
mpirun --bind-to-core blockMesh -parallel > err0
mpirun --bind-to-core setFields -parallel > err1
mpirun --bind-to-core olaFlow -parallel > err2

Do you think that the problem was for running blockMesh without parallel running.?

[Andrew.Coughtrie]

To start with blockMesh can't be run in parallel so that isn't the problem. Doing decomposePar before blockMesh won't help, it will either give an error or just decompose the mesh that is already there if you haven't cleaned the case.

If I were you I'd run the following before you submit the job to only run the olaflow command:

blockMesh
setFields
decomposePar

Then in your job submission script:

mpirun -np 32 olaFlow -parallel

See how that goes and post any errors you get.

Andrew

[shimakasaei]

Originally Posted by Andrew.Coughtrie

To start with blockMesh can't be run in parallel so that isn't the problem. Doing decomposePar before blockMesh won't help, it will either give an error or just decompose the mesh that is already there if you haven't cleaned the case.

If I were you I'd run the following before you submit the job to only run the olaflow command:

blockMesh
setFields
decomposePar

Then in your job submission script:

mpirun -np 32 olaFlow -parallel

See how that goes and post any errors you get.

Andrew

Thanks , now it seems working too, further if I get error I will post it here,

[shimakasaei]

Originally Posted by Andrew.Coughtrie

To start with blockMesh can't be run in parallel so that isn't the problem. Doing decomposePar before blockMesh won't help, it will either give an error or just decompose the mesh that is already there if you haven't cleaned the case.

If I were you I'd run the following before you submit the job to only run the olaflow command:

blockMesh
setFields
decomposePar

Then in your job submission script:

mpirun -np 32 olaFlow -parallel

See how that goes and post any errors you get.

Andrew

Hi
I used the same method that you have told me and it did work, but now, when I want to draw the water time series, it is wrong. I did reconstruct and writeCellcenter before extracting data, do you know if there is any other thing that I need to do?
I believe that the problem is something related to parallel run because when I run with one processor it doesn't looks like that.

Best,
shima

[Andrew.Coughtrie]

Sorry but I'm not sure what you mean by the water time series. Why the data is wrong could be due to any number of things.

Possibly try reconstructParMesh before reconstructPar and then the rest.

I don't have any other suggestions.

Andrew

[shimakasaei]

Originally Posted by Andrew.Coughtrie

Sorry but I'm not sure what you mean by the water time series. Why the data is wrong could be due to any number of things.

Possibly try reconstructParMesh before reconstructPar and then the rest.

I don't have any other suggestions.

Andrew

I mean water surface elevation var time , Thank you for your hint, I will do that, just one question, should I do reconstructPar -latest time?

[Andrew.Coughtrie]

If you're only interested in the latest time and it would otherwise take a long time to reconstruct all the time steps then yes. If you need to analyse the change in data over time you'll need more than just the latest time.

Andrew