pyPDAF.PDAF3.init_parallel

pyPDAF.PDAF3.init_parallel(screen: int, type_parallel: int, online_coupling: int, dim_ens: int, n_modeltasks: int, COMM_model: int) → Tuple[int, int, int, int, int, int, int, int]

Initialize PDAF3’s flexible parallelization layout.

This helper delegates communicator setup to PDAF3. It is intended for applications that use the PDAF3 interface and want PDAF to derive the model and assimilation communicators from the selected parallelization strategy. Call it during PDAF3 setup before routines that rely on PDAF’s internal parallel configuration.

The configuration of the parallelization is what we found to be ‘typically working’. If one has a particular case, e.g., the ensemble parallelization is readily prepared by the model, one can do the communicator configuration in the user code and then call pyPDAF.set_parallel() to provide PDAF with the configuration.

Parameters:

  • screen (int) – Verbosity flag for PDAF screen output.

  • type_parallel (int) –

    PDAF3 parallelization strategy selector. (currently not used).

    Type of parallelization:

    • 0: common setup using task 1 for assimilation

    • 1: setup using separate task 0 for assimilation

  • online_coupling (int) – Coupling mode flag. Use the value expected by PDAF for online or offline coupling. 1: online DA coupling, 0: offline DA coupling

  • dim_ens (int) – Ensemble size.

  • n_modeltasks (int) – Number of parallel model tasks supplied to PDAF3. If online coupling is used, this value must match the number of model tasks that are running the model code. If offline coupling is used, this value is always 1 regardless of the input.

  • COMM_model (int) – MPI communicator handle. This should be the communicator that contains all processors used to run the entire ensemble. This can be MPI_COMM_WORLD. In some cases, e.g., NEMO, the processors are split into model and XIOS communicators. In that case, this should be model communicator.

Returns:

  • n_modeltasks (int) – Number of parallel model tasks supplied to PDAF3.

  • COMM_model (int) – This will be the model communicator after splitting by the number of model tasks. This communicator contains processors that run a single model task instead of the entire ensemble. This is different from the input argument COMM_model.

  • mype_model (int) – Rank of the current process in COMM_model.

  • npes_model (int) – Number of processes in COMM_model.

  • COMM_assim (int) – MPI communicator contains all processors that run actual DA filters.

  • mype_assim (int) – Rank of the current process in COMM_assim.

  • npes_assim (int) – Number of processes in COMM_assim.

  • task_id (int) – Ensemble task identifier of the current process.