Agenda

Review status of action items for mid-term review Tues 09/29

Attending

Tom, Thomas N., Don, Xin, Beau, Hongli, Walls

Minutes

Round table of work related to WRF+ in the past 2 weeks.

Walls
  • Working on WRF ADJ code optimization: a careful analysis of the code showed two potential areas of improvement:
  • Memory "I/O": 80% of the variables are stored without being used. Potential for 20% speed-up
  • Improve cash usage by breaking large "do loops". Estimated improvement in performance of 10%
  • estimated time for the work: 1 or 2 months
Beau
  • Namelist entry to disable solver in WRF+
  • Namelist entry stop_on_error to force fatal stop on tladj-related error
  • Working on documentation of Namelist entries. Many namelist entries seem not to be used. Discussed strategy: clean the obvious unnecessary Namelist entries (e.g. physics options that are not implemented in WRF+) and document as much as possible
  • Working on expanding TL and ADJ tests in test_solve_em: from one time step to multiple time steps
Xin
  • J_LBC: report is finished
  • clean-up for 3DVar and 4DVar (e.g. adjoint variables disabled in 3DVar)
  • minimize I/O between Var&NL and also Var&Plus
  • Penn State student uses Real.exe routines to balance "ndown" and "nup" + other bugfixes for multi-incremental 4DVar
Don
  • Working on user document about single executable
  • Working on task related to ESMF report
  • will start at RAL on Oct 5
ThomasN
  • working on testing MPAS TL/ADJ generated with OpenAD
  • next step: will contact Argonne (OpenAD) team to provide feedback and figure their strategy to solve problems encountered
  • No labels