CFD Online Discussion Forums

CFD Online Discussion Forums (https://www.cfd-online.com/Forums/)
-   Hardware (https://www.cfd-online.com/Forums/hardware/)
-   -   looking for a cluster to run a case (https://www.cfd-online.com/Forums/hardware/93311-looking-cluster-run-case.html)

miles_davis October 11, 2011 10:56

looking for a cluster to run a case
 
HI


I have a multiphase case on Openfoam 2.0.1.
Actually I am running it on a 12 cores PC under ubunutu-linux.
The total human time for the simulation on such a machine should be 15 days (360h).
I need to speed up the resolution.
Case feature
initial file size : 200Mo
Results : 200Mo*30 saves => approx. 6Go


Does anyone know where and how I can find some solution to run it on a supercluster?


Regards,


Sydney TEKAM

abdul099 October 11, 2011 17:38

How many cells does your case contain?

miles_davis October 12, 2011 04:39

not that much: one million.
the problem is my simulation are unsteady.

puma October 12, 2011 06:14

Hi,

cloudnumbers.com offers computing resources also for Openfoam. But I didn't try it myself.

Cheers

abdul099 October 20, 2011 22:57

That's what I thought, the cell count is too low.
Just imagine, you've got a transportation task to solve. You've got 12 vans to do this, every van has to drive the route only once to finish the job. Now you find out, it takes to long. Do you think you will get a significant speedup when using 100 vans? No one of this vans can drive faster due to the speed limit, so it only might take longer due to a self-caused traffic jam...

Translated to your problem: With that low cell count you will not get a significant speedup when using a cluster. You've got only 80 000 cells per core, which is about the lower limit to use per core. When further reducing the cells per core, the solution will even take LONGER, as communication effort increases.
Maybe you could go down to 60 000 cells per core as it is a multiphase flow, but you will never obtain a speedup of factor 2 or 3 just by using more cores.

Unsteady simulations always need some time. As long as it is no trivial testcase, it takes enough time for the user to get impatient. This doesn't make happy, but is close to a law of nature.

miles_davis October 24, 2011 14:06

Thanks a lot for this clear explanation.
You saved me a lot of effort!

Regards,

Miles

federvo.mala March 3, 2012 18:08

Is there one for Ansys Fluent too?

thanks


All times are GMT -4. The time now is 19:02.