chtMultiRegionFoam on multiple cores, run-time increases?

vince_cfd · August 28, 2019, 07:24

Hello fellow foamers,

I am currently working on some heat transfer problems using the chtMultiRegionFoam solver. I want to speed up the solving process by running a decomposed case. To find out how much I would benefit from using multiple cores I tried to run a similar case in single core and multi-core mode. The results had the same quality (in terms of U, T and p), but the run time wasn't as I expected. Here are my results:

~20 000 cells
1 core -> 875s
2 cores -> 1022s

~225 000 cells
1 core -> 57046s
4 cores -> 58539s

Now I have some questions:
- is it possible, that I didnt use enough cells, per core? I read at about 50 000/core the time will reduce.
- is it important to decompose the case in a special way? I used the simple method.

Thank you already,
Vince.

simrego · August 28, 2019, 08:08

Hi!

Try with scotch method. It will tries to decompose the case on a way that the communication between the processors is minimal.
Also a renumberMesh can help.

Based on my experiences OF is scaling pretty well.

vince_cfd · August 28, 2019, 08:35

Thank you for your reply simrego.

I decomposed with scotch and am currently running the case. It might take some time till I can update, as I only work with OF on one day per week.

Greetings

August 28, 2019, 07:24	chtMultiRegionFoam on multiple cores, run-time increases?	#1
vince_cfd New Member Vincent Join Date: Aug 2019 Location: Germany Posts: 14 Rep Power: 6	Hello fellow foamers, I am currently working on some heat transfer problems using the chtMultiRegionFoam solver. I want to speed up the solving process by running a decomposed case. To find out how much I would benefit from using multiple cores I tried to run a similar case in single core and multi-core mode. The results had the same quality (in terms of U, T and p), but the run time wasn't as I expected. Here are my results: ~20 000 cells 1 core -> 875s 2 cores -> 1022s ~225 000 cells 1 core -> 57046s 4 cores -> 58539s Now I have some questions: - is it possible, that I didnt use enough cells, per core? I read at about 50 000/core the time will reduce. - is it important to decompose the case in a special way? I used the simple method. Thank you already, Vince.

August 28, 2019, 08:08		#2
simrego Senior Member anonymous Join Date: Jan 2016 Posts: 416 Rep Power: 14	Hi! Try with scotch method. It will tries to decompose the case on a way that the communication between the processors is minimal. Also a renumberMesh can help. Based on my experiences OF is scaling pretty well. vince_cfd likes this.

August 28, 2019, 08:35	scotch	#3
vince_cfd New Member Vincent Join Date: Aug 2019 Location: Germany Posts: 14 Rep Power: 6	Thank you for your reply simrego. I decomposed with scotch and am currently running the case. It might take some time till I can update, as I only work with OF on one day per week. Greetings

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
pimpleDyMFoam computation randomly stops	babapeti	OpenFOAM Running, Solving & CFD	5	January 24, 2018 05:28
How to run OpenFOAM using Shared Memory method on local machine with multiple cores?	ronithstanly	OpenFOAM Running, Solving & CFD	1	August 4, 2017 00:31
pressure in incompressible solvers e.g. simpleFoam	chrizzl	OpenFOAM Running, Solving & CFD	13	March 28, 2017 05:49
Transient run continues from last time (when startover is desired)	bongbang	CFX	2	March 22, 2015 23:05
calculation diverge after continue to run	zhajingjing	OpenFOAM	0	April 28, 2010 04:35