open TELEMAC-MASCARET The mathematically superior suite of solvers

Hello,
does anybody experience the performance of the coupled models TEL2d+TOM on linux cluster machine?
I run tests on a cluster machine, performing 1 hour of simulation and varying the number of cores.
My mesh has 180000 nodes; in TEL2d, I used TPXO tidal model, and imposed velocities varying in time and space at the offshore boundary (I modified subroutines); in TOM, I imposed varying wave conditions on the offshore boundary as well (modified limwac subroutine).
I first used tel2d and tom in standalone mode.
Then I coupled them.
The results in terms of elapsed time versus numb of cores are reported in the attached image.
Time in the coupled mode are 10 times grater than the (maximum) elapsed time in TOM.
Any comments on this? Is it what you experienced too?

Hi

What is the coupling period?
Maybe you should post the steering files of the three cases

Regards

The coupling period is 1.
Here they are the steering files.





Thanks.

G

Hello,

What you get, i.e. a much larger computing time when coupling, is quite normal and well known, though people think it is a bug: Tomawac has to compute the characteristics only once when it is not coupled, and at every time step when it is coupled, because the currents and depth are updated and it changes the trajectories. Most of the computing time is spent in computing the characteristics. The only solution to speed-up is then to couple with Tomawac with a larger period, because generally the wave climate does not change as rapidly as the hydrodynamic time step.

With best regards,

Jean-Michel Hervouet

Perfect! Thanks JM.
I was thinking about the exact reason you clearly explained, but I asked to be sure..
Hope to speed-up better the models, increasing the coupling period..
Best regards,

G