pisoFoam faster in parallel after recomposition
 

Hi all,

I'm calculating a 3.6 millions hexahedrals case with SpalartAllmarasDDES (pimpleFoam in piso mode) in parallel on a sixteen cores cluster. I use GAMG for everything except the pressure correctors, for which I use GAMGPCG. I noticed that reconstructing and re-decomposing the domain reduces the calculation time per iteration by a 20% or more, and not only for a pair of time steps, it seems to remain stable for a while. Does anyone knows why, and if this practice affects the solution in any (bad?) way?