CFD Online Discussion Forums

CFD Online Discussion Forums (https://www.cfd-online.com/Forums/)
-   OpenFOAM Meshing & Mesh Conversion (https://www.cfd-online.com/Forums/openfoam-meshing/)
-   -   [Other] How to create uneven load for the processors using decomposePar utility. (https://www.cfd-online.com/Forums/openfoam-meshing/135481-how-create-uneven-load-processors-using-decomposepar-utility.html)

shinde.gopal May 14, 2014 06:09

How to create uneven load for the processors using decomposePar utility.
 
Hi Foamers,

I am interested in doing OpenFoam simulations on heterogeneous architecture wherein I want to exploit the system architecture to the fullest extent. For that I need uneven load for the processors.

Has anybody attempted this kind of task ?

I need your help on how to create uneven load for the processors using decomposePar utility. I have explored the option of scotch decomposition using processorWeights but it does not suffice my requirement. It creates almost same load for all the processors.

Also the communication between the processors is increased if I compare the decomposition stats using scotch method (with processorWeights) with the usual hierarchical or simple decomposition stats.

Please help.

-Gopal

derekm May 24, 2014 09:49

Quote:

Originally Posted by shinde.gopal (Post 491697)
Hi Foamers,

I am interested in doing OpenFoam simulations on heterogeneous architecture wherein I want to exploit the system architecture to the fullest extent. For that I need uneven load for the processors.

Has anybody attempted this kind of task ?

I need your help on how to create uneven load for the processors using decomposePar utility. I have explored the option of scotch decomposition using processorWeights but it does not suffice my requirement. It creates almost same load for all the processors.

Also the communication between the processors is increased if I compare the decomposition stats using scotch method (with processorWeights) with the usual hierarchical or simple decomposition stats.

Please help.

-Gopal

have you used very large differences in processor weights e.g 1000:1


All times are GMT -4. The time now is 12:19.