Running regression tests with seed dispersal on and the fates_seed_dispersal_max_dist set to 2.5E6 resulted in tests with grid resolutions of f19_g17 and smaller to fail with either of the the following errors:
1659 MPT ERROR: STDOUT or STDERR buffer exhausted.
1660 Set MPI_UNBUFFERED_STDIO=true or output no more than about 8192
1661 characters between newlines.
1662 (HPE MPT 2.25 08/14/21 03:06:24)
My suspicion after conducting some research into the failure mechanism is that this is possibly due to there being too many neighbors for the MPI passing routines to handle. Running with a maximum distance that was approximately an order of magnitude smaller, results in the tests running successfully. As such, there should be some method of determining what a maximum limit on the dispersal maximum based on the grid resolution. Using this we could then gracefully fail and let the user know that the distance is too high.
Running regression tests with seed dispersal on and the
fates_seed_dispersal_max_distset to2.5E6resulted in tests with grid resolutions off19_g17and smaller to fail with either of the the following errors:or
My suspicion after conducting some research into the failure mechanism is that this is possibly due to there being too many neighbors for the MPI passing routines to handle. Running with a maximum distance that was approximately an order of magnitude smaller, results in the tests running successfully. As such, there should be some method of determining what a maximum limit on the dispersal maximum based on the grid resolution. Using this we could then gracefully fail and let the user know that the distance is too high.