Verify icoFoam result

qnib · November 16, 2014, 08:57

Hey there,

I am playing around with docker and thought about conducting a little study about OpenFOAM in a containerized environment.
I conducted a study on MPI workloads with docker in general using HPCG:
http://qnib.org/qnibterminal/2014/11...MPI-workloads/

The goal is to show how OpenFOAM performs using different operating systems and (maybe more importantly) how to rerun an input deck which provides the same outcome.

Now I wonder how this 'equal outcome' thing can be defined. Would it be accurate to simply check the md5sum of some
of the result files?

Code:

job_results/openfoam/openfoam/13/omp-sys/2014-11-16_13-29-00/slurm-708.out:U_MD5=22d7cb33d17f8a604c83a0411232d16e  ./0.005/U
job_results/openfoam/openfoam/13/omp-sys/2014-11-16_13-29-03/slurm-709.out:U_MD5=22d7cb33d17f8a604c83a0411232d16e  ./0.005/U
job_results/openfoam/openfoam/13/omp-sys/2014-11-16_13-29-05/slurm-710.out:U_MD5=22d7cb33d17f8a604c83a0411232d16e  ./0.005/U
job_results/openfoam/openfoam/13/omp-sys/2014-11-16_13-29-08/slurm-711.out:U_MD5=22d7cb33d17f8a604c83a0411232d16e  ./0.005/U

Furthermore, what should I expect in terms of reproducibility of results?

As far as I understood so far I can not expect to get the same result if I change the decomposition?
'n = ( 16 1 1)' is going to be different from 'n = ( 8 2 1)' no matter what?

I use the cavityFine test case found here:
http://www.cfd-online.com/Forums/ope...processor.html

If someone suggest a better testcase that scales up to 64 cores, I am all open.

Kind regards
Christian

qnib · November 16, 2014, 09:16

The previously shown results were run on 2 nodes, each with 2 sockets with a QUAD-core XEON.

Running on four nodes I got the same checksum (only using 4 tasks/node):

Code:

U_MD5=22d7cb33d17f8a604c83a0411232d16e  ./0.005/U

qnib · December 7, 2014, 14:45

No one has an opinion about how to verify if two results are identical?

November 16, 2014, 08:57	Verify icoFoam result	#1
qnib New Member Christian Join Date: Nov 2014 Location: Berlin Posts: 3 Rep Power: 12	Hey there, I am playing around with docker and thought about conducting a little study about OpenFOAM in a containerized environment. I conducted a study on MPI workloads with docker in general using HPCG: http://qnib.org/qnibterminal/2014/11...MPI-workloads/ The goal is to show how OpenFOAM performs using different operating systems and (maybe more importantly) how to rerun an input deck which provides the same outcome. Now I wonder how this 'equal outcome' thing can be defined. Would it be accurate to simply check the md5sum of some of the result files? Code: job_results/openfoam/openfoam/13/omp-sys/2014-11-16_13-29-00/slurm-708.out:U_MD5=22d7cb33d17f8a604c83a0411232d16e ./0.005/U job_results/openfoam/openfoam/13/omp-sys/2014-11-16_13-29-03/slurm-709.out:U_MD5=22d7cb33d17f8a604c83a0411232d16e ./0.005/U job_results/openfoam/openfoam/13/omp-sys/2014-11-16_13-29-05/slurm-710.out:U_MD5=22d7cb33d17f8a604c83a0411232d16e ./0.005/U job_results/openfoam/openfoam/13/omp-sys/2014-11-16_13-29-08/slurm-711.out:U_MD5=22d7cb33d17f8a604c83a0411232d16e ./0.005/U Furthermore, what should I expect in terms of reproducibility of results? As far as I understood so far I can not expect to get the same result if I change the decomposition? 'n = ( 16 1 1)' is going to be different from 'n = ( 8 2 1)' no matter what? I use the cavityFine test case found here: http://www.cfd-online.com/Forums/ope...processor.html If someone suggest a better testcase that scales up to 64 cores, I am all open. Kind regards Christian

November 16, 2014, 09:16		#2
qnib New Member Christian Join Date: Nov 2014 Location: Berlin Posts: 3 Rep Power: 12	The previously shown results were run on 2 nodes, each with 2 sockets with a QUAD-core XEON. Running on four nodes I got the same checksum (only using 4 tasks/node): Code: U_MD5=22d7cb33d17f8a604c83a0411232d16e ./0.005/U

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
strange result for non-orthogonality mesh using icoFoam	xiaof	OpenFOAM	1	November 2, 2013 08:48
How to display the whole result of a periodic model via one period result?	dixylo	FLUENT	2	February 23, 2012 05:11
Read Result Files or User Defined Result	Aquilaris	ANSYS	0	February 9, 2011 06:25
Density in icoFoam Densidad en icoFoam	manuel	OpenFOAM Running, Solving & CFD	8	September 22, 2010 05:10
Kubuntu uses dash breaks All scripts in tutorials	platopus	OpenFOAM Bugs	8	April 15, 2008 08:52

December 7, 2014, 14:45		#3
qnib New Member Christian Join Date: Nov 2014 Location: Berlin Posts: 3 Rep Power: 12	No one has an opinion about how to verify if two results are identical?