-
Notifications
You must be signed in to change notification settings - Fork 69
Description
There are indications that partitioning the mesh can change elvls.out and nlvls.out (I hope I am interpreting the content of these files correctly). Is this behaviour expected?
In particular, before partitioning dist_384 in /work/ab0246/a270092/input/fesom2/for_alex/core2 on Levante, nodes 125226 and 125228 had depths of 18 and 19, respectively. After partitioning it got deeper (22 resp. 20).
This can lead to unexpected blowups of simulations which used to work fine before, if in the mean time an additional mesh partitioning is generated. Changing those values back to the original ones bitidentically reproduces output from before generating the partitioning.
Below is the diff output. It is an interesting coincidence that the change in depth occurs at element number 1.
Diff on nlvls.out:
125226c125226
< 22
---
> 18
125228c125228
< 20
---
> 19
Diff on elvls.out:
1c1
< 22
---
> 18
187115c187115
< 20
---
> 18
187171c187171
< 22
---
> 18
241598c241598
< 20
---
> 19