Speeding up 2D simulations

bobuhito · December 19, 2021, 16:19

EDIT: I would just delete this post, but it's not allowed. Nevermind my starting thought below (since I realize now that the phi is actually flux through the faces, so that causes the doubling and there is no waste here for 2D meshes because the faces to the unused dimension are not included). I assume that the code is already optimized for 2D structures, and no further easy speed up is known.
----

Considering the damBreak example, in which interFoam is run on a 2D structure, I noticed that "phi" and "alphaPhi0.water" output files are about double the size of the "p" file.
I am guessing this is because they store data for each point in the mesh (not each cell in the mesh like "p" does), right?

In these 2D simulations (since the mesh for 2D is just points in 2D repeated in the unused dimension), this seems a little wasteful, doubling some space and time requirements (because CPU is probably using the doubled phi representation). Then again, I might be misunderstanding this, so please correct me if I am wrong.

Anyway, I'm wondering how much speed up would be possible if the code were optimized for 2D structures?

bobuhito · January 22, 2022, 23:33

As I learn more and more, I do see some things that could be cut out for speed for 2D simulations. At least, the U solving/corrections could drop the z component. So, maybe there is potential for a faster "2D interFoam" which could finish simulations in 33% less time.

December 19, 2021, 16:19	Speeding up 2D simulations	#1
bobuhito New Member BH Join Date: Dec 2021 Posts: 9 Rep Power: 4	EDIT: I would just delete this post, but it's not allowed. Nevermind my starting thought below (since I realize now that the phi is actually flux through the faces, so that causes the doubling and there is no waste here for 2D meshes because the faces to the unused dimension are not included). I assume that the code is already optimized for 2D structures, and no further easy speed up is known. ---- Considering the damBreak example, in which interFoam is run on a 2D structure, I noticed that "phi" and "alphaPhi0.water" output files are about double the size of the "p" file. I am guessing this is because they store data for each point in the mesh (not each cell in the mesh like "p" does), right? In these 2D simulations (since the mesh for 2D is just points in 2D repeated in the unused dimension), this seems a little wasteful, doubling some space and time requirements (because CPU is probably using the doubled phi representation). Then again, I might be misunderstanding this, so please correct me if I am wrong. Anyway, I'm wondering how much speed up would be possible if the code were optimized for 2D structures?

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Water-Air simulations	Miguel Hernandez	OpenFOAM Running, Solving & CFD	26	September 18, 2021 05:38
Confusion choosing between static / transient simulations	TharunSK	CFX	1	March 15, 2019 17:53
3D simulations OpenFoam running too low(more 1 000 000 elements) Suggestion speed up?	silvai	OpenFOAM Running, Solving & CFD	4	April 6, 2017 11:44
Hardware selection for steady/unsteady incompressible, turbulent and cht simulations	maddalena	OpenFOAM	2	July 13, 2011 08:55
URANS and Transient Simulations	bob	Main CFD Forum	0	October 1, 2003 03:54

January 22, 2022, 23:33		#2
bobuhito New Member BH Join Date: Dec 2021 Posts: 9 Rep Power: 4	As I learn more and more, I do see some things that could be cut out for speed for 2D simulations. At least, the U solving/corrections could drop the z component. So, maybe there is potential for a faster "2D interFoam" which could finish simulations in 33% less time.