Framework for adaptive dt / independent dt specs in each solver input #107
Conversation
|
will be incorporating this into a reg test |
|
depends on: |
|
This approach is working for the reg test, but it is not working at scale for an actual run. Nalu-Wind and AMR-Wind end up using different time step sizes. I'm not sure how this is happening. I would suspect that it is a parallelism issue, but every processor should have a valid value of dt, so it doesn't make sense to have it be due to parallelism. |
|
Even though each processor has a valid dt, not every processor has both solvers, and usually they're separate. Need an MPI All Reduce min. |
|
Reg test still works as a valid test case (revealing the problems at scale) provided that the processors are divided between the two solvers. |
3 participants
No description provided.