Hi,

I am currently working on car external aero simulations and I am trying to figure out wether one can save significant computation time by re-partitioning a large model (several millions of cells) after each pressure gradient-based adaption. It looks like there's a compromise between model reading/writing time and computation time savings... Does anyone have experience with this issue ?

Thanks for your interest.

Alexis Scotto

Probably the threshold for adaption is quite high and the number of cells was increased so as to balance the effort to reduce computational time by repartitioning.

It depends on what you are looking for...

What I want to achieve is pretty simple: now that I have satisfying results for aero coefficients computations, I would like to optimize the CPU time required to complete an analysis.

Typically, my models are fairly large (million sof cells). They run on a 4 processor machine and, after convergence, I usually perform a few adaptions based on pressure gradient.

It seems like re-partitioning after each adaption really speeds up the solver. However, this approach is penalized by all the model read/write operations (models get very large). I am trying to optimize this procedure and thought that someone probably had experience with it...

Alexis Scotto