scotch decompose for dsmc
 

Hi All,
i'm doing some simulations with the dsmcFoam solver and read this paper:
http://eucass2011.conferencecenter.r...papers/382.pdf

There's a discription of a "new" scotch decompose-method witch deals with the number of particles per CPU.(Section 5)
This looks quite promising, due to the particle-based calculations. :)

Does anyone knows if its implemented in OF2.3.0 ? I couldn't figure out... :confused: (If, how to use it, strategy-keyword or something like this?)

Tüdelü
qwert

Hi,

thanks for pointing on that dsmc paper

http://www.openfoam.org/download/source.php depending on your Linux distribution installing Scotch can be sudo apt-get install libscotch-dev
http://www.openfoam.org/docs/user/ru...s-parallel.php
http://www.openfoam.org/mantisbt/view.php?id=1048

Hi Elvis,
thanks for your reply. Maybe there's a misunderstanding: I know the 'regular' scotch decomposition.
In the paper i've pointed in my last post, there is a discription of an extention of these method:

Quote:

In order to increase the computational efficiency of dsmcFoam, we have developed and tested an extension to the Scotch decomposer. This new decomposer splits the domain so that each processor has the same number of particles to work with, meaning that processors are concentrated in regions of high number of particles, for example, in shock waves. To assess the new Scotch technique, which is open-source and available from the OpenFOAM website (version 1.7.x), all three decomposers were tested ...

Ahmad et al.

At their cases they could reduce time consumption by ~30% in contrast to the cell-based scotch method. :)

Is this extension (mentioned in the paper) includet in version 2.3.0?
I didn't found out (searched in ~/../OpenFoam-2.3.0/scr
Or I misunderstood and there's a hidden hint somewhere in your links and I didn't get it :cool: (blind).
Did anyone tried out this method. As I mentioned, it seems to be very promising for larger problems.

best regards,

qwert