CFD Online Discussion Forums

CFD Online Discussion Forums (http://www.cfd-online.com/Forums/)
-   OpenFOAM (http://www.cfd-online.com/Forums/openfoam/)
-   -   Very Very SLOW interFoam on Cluster (http://www.cfd-online.com/Forums/openfoam/76597-very-very-slow-interfoam-cluster.html)

ovie May 30, 2010 14:26

Very Very SLOW interFoam on Cluster
 
Hi..

I am running some modified version of interFoam on a cluster. My parallel computation uses 40 processors and the problem has about 700k grid cells. The modified solver includes an energy eqution and uses the MULES Implicit solver for the alpha1 transport equation. The computation has been shockingly slow even though I am not writing out results at any time steps. The computation has been running for almost 24 hours with an average time step of 1e-4 and it is yet to compute 2 seconds of real time. Has anyone encoutered similar problems? Any ideas to fix this?

Thanks

chiven May 31, 2010 01:31

Quote:

Originally Posted by ovie (Post 260909)
Hi..

I am running some modified version of interFoam on a cluster. My parallel computation uses 40 processors and the problem has about 700k grid cells. The modified solver includes an energy eqution and uses the MULES Implicit solver for the alpha1 transport equation. The computation has been shockingly slow even though I am not writing out results at any time steps. The computation has been running for almost 24 hours with an average time step of 1e-4 and it is yet to compute 2 seconds of real time. Has anyone encoutered similar problems? Any ideas to fix this?

Thanks

You can use your parallel computer to do a simple case to check whether it is the computer's problem or use another parallel computer to calculate your case to check whether it is the case or solver's problem.

Best regards,
Chiven

kumar May 31, 2010 03:50

Hello ,
i also had the same experience while using a modified version of interFoam with MULES implicit, but i was not solving the energy equation in my case. Since you are using MULES implicit, i assume that you are using CrankNicholson as dt scheme for all the equations. Why dont you just try to run one case with backward or Euler timestep and see if the selection of time integration schemes is causing the problem .

You can also try reducing the CrankNicholson coefficient to 0.9 and see the difference.

regards
K.Suresh kumar

herbert May 31, 2010 04:06

Quote:

Originally Posted by ovie (Post 260909)
My parallel computation uses 40 processors and the problem has about 700k grid cells.

Hi,

I think your problem is much to small to use 40 processors. You are spending more time with communication than with solving. I'd suggest you start your case with only 4-8 processors and look, if you still don't get some speed-up.

Regards,
Stefan


All times are GMT -4. The time now is 09:17.