Limiting the number of integration of fluxes in boxes


Adding Boxes can significantly increase the total computation time. This can be reduced significantly if the integration of fluxes between boxes are computed every N-iterations only.
The proposal consists in adding a STEPS_TO_BOXCALL keyword (with value 1 by default), that determines the number of iterations to wait until the next integration of fluxes between boxes.
This would help tremendously to control the total computation time and keep it close to the total computation time without boxes.
------- UPDATE ---------
To keep in line with the philosophy of the code, the proposal was changed, and instead of a STEPS_TO_BOXCALL a DT keyword was implemented in the boxes configuration file.
Closed Jun 4, 2011 at 12:07 PM by guillaume7
The feature is implemented and tested.


guillaume7 wrote Mar 29, 2011 at 10:11 AM

Just implemented. The results: http://goo.gl/8z51o. Now with a boxes DT equal to 20 times a model DT the impact of boxes in performance is neglectable (w/ openmp).

wrote Mar 29, 2011 at 10:14 AM

wrote Jun 4, 2011 at 12:07 PM

wrote Feb 14, 2013 at 1:13 AM

wrote May 16, 2013 at 7:02 AM