3D Interpolation performance about diva HOT 3 CLOSED

Hi, thanks for the benchmark.

The parameter optimisation is a usually a step in the analysis that takes a lot of time so your last comment is totally correct: turning off the optimisation will speed up the operations.

A diva 2D analysis is made up of two parts:

1. the mesh generation, which can be slow if the length is set to a small value,
2. the calculation, which is supposed to not depend too much on the number of data.

So if you work on 2 levels, two meshes have to be generated, and this may explain the difference.

from diva.

@ctroupin Thanks for responding.

The parameter optimisation is a usually a step in the analysis that takes a lot of time so your last comment is totally correct: turning off the optimisation will speed up the operations.

Thanks for that clarification.

A diva 2D analysis is made up of two parts:
1. the mesh generation, which can be slow if the length is set to a small value,
2. the calculation, which is supposed to not depend too much on the number of data.
So if you work on 2 levels, two meshes have to be generated, and this may explain the difference.

1. In the 3D interpolation case where I have turned parameter optimization off, I set the correlation length to the same value that I use for the 2D variational analysis.
2. Even assuming that the mesh generation is happening twice for two levels of 3D interpolation, I would expect the 3D performance (with parameter optimization turned off) to be twice that in 2D. However, it seems the 3D analysis (for 2 levels) takes ~8x amount of time (24s) as compared to the 2D interpolation (3s).
3. Note that in both the 2D case (using divadress) as well as the 3D case (using diva3Ddress), cleaning of the mesh happens each time between runs. (Since both divadress and diva3Ddress both do mesh cleaning.)

from diva.

from diva.