potentialFoam results not written in parallel run

ishan_ae · September 29, 2020, 09:27

Hello.

I am trying to figure out what's going wrong in my potentialFoam run.

I am using this command to run the solver on 30 processors:

Code:

mpirun -np 30 potentialFoam -writep | tee potentialFoam.log

It runs all right, but the fields of U,phi and p are never written to the 30 processor directories. The 0 files within the processor directory always remain as they were originally.

But the main 0 directory is updated with the calculated fields.

The domain was decomposed using Scotch.

I have attached here the data from the solver run using MPI and also the controlDict.

I ran the pitzDaily tutorial under decomposed domain and there the results are written to the processor directories.

I am using OF6.

I am not sure what I am doing wrong.
controlDict.txt

potentialFoam.txt

ishan_ae · September 29, 2020, 10:00

I found the reason why it was happening.

I was decomposing the domain into parts with a number greater than the available physical cores.

I have 18 but I was decomposing the domain into 30.

Yann · September 30, 2020, 03:36

Hello Ishan,

I think your error is not related to the number of physical cores.
When running in parallel with mpirun, you have to use the "-parallel" option with your command:

Code:

mpirun -np 30 potentialFoam -writep -parallel | tee potentialFoam.log

With the command you wrote in your initial post, you are basically launching 30 times a single process of potentialFoam instead of running it in parallel mode. So the results are probably written at the root of your case instead of the processor directories.

I hope this helps.
Yann

ishan_ae · October 1, 2020, 03:14

Hello.

I think that's correct. I got confused with it since this is my very first time using OpenFOAM to work on a big simulation.

Carlo_P · October 1, 2020, 04:03

I can confirm Yann hypotesis.

It is not a problem to run on more processors than you have available. It has nosense, but it is possible.

but, you have to use -parallel.

Otherwise, you are running 30 potentialFoam on 1 processor each.

You can check also on your log file.

In the first lines, you should have 30 different headers and each one tells you that potentialFoam is running on 1 processor.

September 29, 2020, 09:27	potentialFoam results not written in parallel run	#1
ishan_ae Member ishan Join Date: Oct 2017 Posts: 77 Rep Power: 8	Hello. I am trying to figure out what's going wrong in my potentialFoam run. I am using this command to run the solver on 30 processors: Code: mpirun -np 30 potentialFoam -writep \| tee potentialFoam.log It runs all right, but the fields of U,phi and p are never written to the 30 processor directories. The 0 files within the processor directory always remain as they were originally. But the main 0 directory is updated with the calculated fields. The domain was decomposed using Scotch. I have attached here the data from the solver run using MPI and also the controlDict. I ran the pitzDaily tutorial under decomposed domain and there the results are written to the processor directories. I am using OF6. I am not sure what I am doing wrong. controlDict.txt potentialFoam.txt

September 30, 2020, 03:36		#3
Yann Senior Member Yann Join Date: Apr 2012 Location: France Posts: 1,074 Rep Power: 26	Hello Ishan, I think your error is not related to the number of physical cores. When running in parallel with mpirun, you have to use the "-parallel" option with your command: Code: mpirun -np 30 potentialFoam -writep -parallel \| tee potentialFoam.log With the command you wrote in your initial post, you are basically launching 30 times a single process of potentialFoam instead of running it in parallel mode. So the results are probably written at the root of your case instead of the processor directories. I hope this helps. Yann Carlo_P likes this.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
[Salome] Salome import "Cannot find file "points" in directory..."	mismichael	OpenFOAM Meshing & Mesh Conversion	5	June 22, 2019 02:56
[mesh manipulation] Cannot get refineMesh to run in parallel	smschnob	OpenFOAM Meshing & Mesh Conversion	2	June 3, 2014 11:20
The results difference between parallel and serial run.	Hkp	OpenFOAM Running, Solving & CFD	2	April 17, 2014 02:26
Transient Run - Output "Time" in partial results?	evcelica	CFX	2	May 16, 2012 21:36
SnappyHexMesh OF-1.6-ext crashes on a parallel run	norman1981	OpenFOAM Bugs	5	December 7, 2011 12:48

September 29, 2020, 10:00		#2
ishan_ae Member ishan Join Date: Oct 2017 Posts: 77 Rep Power: 8	I found the reason why it was happening. I was decomposing the domain into parts with a number greater than the available physical cores. I have 18 but I was decomposing the domain into 30.

October 1, 2020, 03:14		#4
ishan_ae Member ishan Join Date: Oct 2017 Posts: 77 Rep Power: 8	Hello. I think that's correct. I got confused with it since this is my very first time using OpenFOAM to work on a big simulation.

October 1, 2020, 04:03		#5
Carlo_P Senior Member Carlo_P Join Date: May 2019 Location: Italy Posts: 176 Rep Power: 7	I can confirm Yann hypotesis. It is not a problem to run on more processors than you have available. It has nosense, but it is possible. but, you have to use -parallel. Otherwise, you are running 30 potentialFoam on 1 processor each. You can check also on your log file. In the first lines, you should have 30 different headers and each one tells you that potentialFoam is running on 1 processor.