Opened 16 years ago
Closed 16 years ago
#178 closed bug (Fixed)
tile bug in ini_fields
| Reported by: | kate | Owned by: | arango |
|---|---|---|---|
| Priority: | major | Milestone: | Release ROMS/TOMS 3.2 |
| Component: | Nonlinear | Version: | 3.2 |
| Keywords: | Cc: |
Description (last modified by )
When running with multiple tiles in serial mode, Zt_avg1 gets set on one tile, set_depth_tile gets called and Hz is recomputed internal to that tile. Then, Hz is used (including one ghost cell) to compute the vertical sums of u and v for resetting ubar, vbar. The ghost cells aren't up to date and that one edge of ubar, vbar is not quite the same as it should be.
Change History (1)
comment:1 by , 16 years ago
| Description: | modified (diff) |
|---|---|
| Resolution: | → Fixed |
| Status: | new → closed |
Note:
See TracTickets
for help on using tickets.
Oh yes, good catch!!! This is a shared-memory bug and a serial with partition bug. The distributed-memory algorithm is correct. I missed this one because we usually initialize the model with zero free-surface. The problem arises when the free-surface is not zero like in restarts or forecasting.
The solution was to split ini_fields into two routines (ini_zeta and ini_fields) so we can have the parallel synchronization. The TLM, RPM, and ADM were also corrected for consistency. Recall that we can only run the ADM in distributed memory.