CFD Online Discussion Forums

CFD Online Discussion Forums (http://www.cfd-online.com/Forums/)
-   OpenFOAM (http://www.cfd-online.com/Forums/openfoam/)
-   -   Openfoam optimization question (http://www.cfd-online.com/Forums/openfoam/88980-openfoam-optimization-question.html)

atareen64 May 31, 2011 14:38

Openfoam optimization question
 
Hello folks,

I've been doing some simulations in openfoam using the rhoSimpleFoam solver. The geometry is a convergent-divergent nozzle with approximate dimensions of 20 mm by 4 mm by 2 mm. Air is blown in through the inlet on one side and there is an outlet on the other side and I observe the results.

The issue I am having is that with a mesh of a million cells, it takes four days for my simulation to give good results. That's far too long. I know that the same simulation takes 8 hours in solidWorks using 8 processors. I am also using 8 processors using the 'simple' decomposition scheme. Can anybody think of any optimizations I can make? Or Is there a general consensus that OF is slower than solidWorks or other CFD packages?

Any help of comments would be appreciated. Please let me know if you would like me to post additional information.

Thank you!
-Ammar.

niklas May 31, 2011 15:17

Four days!!!
You are doing something seriously wrong.

First off, I would decompose it with scotch.

Even 8 hours sounds a bit long to me, but then again I dont know your hardware.

atareen64 May 31, 2011 16:18

Thank you!
 
Dear Niklas,

Thanks for saying that, It gives me a lot more hope. I'm using the mac port of openfoam (OF 1.7.x) so I'm not quite sure how I'll get the scotch decomposition scheme running here, but I think I should be able to figure it out.

What I just did that was a little eye opening was that I increased my relaxation factors by a lot and it gave me an improvement by a factor of ~ 300!!!

Now I just have to figure out how to chose optimal relaxation factors for different mesh sizes. Any suggestions?

Thanks again! :)

-Ammar.

niklas June 1, 2011 08:36

depends on scheme, flow and mesh quality how aggressive you can be.

I prefer slow convergence before the oscillating behavior you get with too high under-relaxation.
I usally use
p 0.1
and the rest 0.3

check the convergence history with
foamLog logFile
gnuplot
set logscale y
plot "logs/p_0" using 1:2 with lines

N


All times are GMT -4. The time now is 08:58.