strange behaviour of GGI in parallel on axis symmetrical case
i'm running an axis symmetrical case with a moving element. This case runs perfectly with a simple decomposition on 4 CPUs (2 in x and 2 in y), but remains unfortunately quite slow, even using the last svn of 1.5-dev, and even if my domain is fairly small (1.4 Mcells).
This same case has a strange behaviour when i want to raise the number of CPUs to 8:
- cutting x y z axis in 2 2 2 brings my mesh to move where it shouldn't, quite far away from my GGI zone, even if i "preserve" my GGI interfaces when i decompose.
- i thought my problem came from the decomposition in z, which is the axis of symmetry, so i decomposed the case in an "azimuthal" (pie-like) way on 8 CPUs with a manual decomposition dictionary, which didn't help: my mesh was still moving at some other places where i didn't expect it.
Does anyone have an idea or observe the same behaviour? May my dynamicMeshDict or decomposeParDict be useful, i could of course post them here.
|All times are GMT -4. The time now is 10:46.|