non reproducible results with respect to domain decomposition
 

Hi guys,
I'm doing simple tests using the application BuoyantBoussinesqPimpleFoam. It's basically the same example than the hot room given, where I added an inflow on every sides of the box.
However, I did the exact same simulation on 24 processors, but specifying in system/decomposeParDict a different domain decomposition. I used the following:
(24 1 1)
(6 4 1)
(4 6 1)

I joined to this message the results of the simulations at an equal time. The plots are showing a cut in the z direction. On the left hand side, the Uz component of the velocity is given, and on the right hand side, mag(U) is given.

We can see that Uz changes from a simulation to another.
I also joined a decomposeParDict file I use as text file.

Did anybody experienced the same problem? Is there a way to fix that?


Attachment 36605

Attachment 36606

Attachment 36607

Attachment 36608

Any solution?
 

Hi Vinci,

Did found the reason for the not able to reproducibility results with different processors.

Could any of you provide a reproducible example for this?