demo/SFSW2_project_settings.R

In Burke-Lauenroth-Lab/rSFSW2: Simulation Framework for SOILWAT2

#------------------------------------------------------------------------------#
# rSFSW2: FRAMEWORK FOR SOILWAT2 SIMULATIONS: CREATING SIMULATION RUNS,
#         EXECUTING SIMULATIONS, AND AGGREGATING OUTPUTS
#
# See demo/SFSW2_project_code.R for details
#------------------------------------------------------------------------------#


################################################################################
#----------------------- SETTINGS FOR A RUN OF A SIMULATION PROJECT ------------

# NOTE: The values may be changed/adjusted from run to run a rSFSW2 simulation
#  project. The values of the description of a project
#  (file `demo/SFSW2_project_description.R``) cannot be changed once a
#  rSFSW2 simulation project is set up.
stopifnot(exists("SFSW2_prj_meta"), !is.null(SFSW2_prj_meta[["opt_platform"]]))


opt_behave <- list(
  # Resumes/continues with unfinished part of simulation after abort if TRUE,
  # i.e.,
  #  - It doesn't delete an existing weather database, if a new one is requested
  #  - It doesn't re-extract external information (soils, elevation, climate
  #    normals, climate scenarios) if already extracted
  #  - It doesn't lookup values from tables if already available in input
  #    datafiles, i.e., 'LookupEvapCoefs',
  #    'LookupTranspRegions', and 'LookupSnowDensity'
  #  - It doesn't repeat calls to 'do_OneSite' that are listed in 'runIDs_done'
  resume = TRUE,
  # Use preprocessed input data if available
  use_preprocin = TRUE,
  # Update dbWork for each output element if TRUE (can cause traffic jams in
  # highly parallel runs); if FALSE, dbWork is not updated continuously
  keep_dbWork_updated = TRUE,
  # If action == "check_dbOut" detects missing Pids, then workDB is updated
  # (so that a new run of the script can be started to add missing runs)
  check_updates_dbWork = TRUE,
  # Check linked BLAS library before simulation runs
  check_blas = FALSE
)



#------ Options for parallel framework
opt_parallel <- list(
  # Should job be run in parallel
  parallel_runs = !interactive() &&
    !SFSW2_prj_meta[["opt_platform"]][["no_parallel"]],

  # Number of cores/workers/slaves if job is run in parallel
  num_cores = if (identical(SFSW2_prj_meta[["opt_platform"]][["host"]],
    "local")) {
      2
    } else if (identical(SFSW2_prj_meta[["opt_platform"]][["host"]], "hpc")) {
      39
    },
  # Parallel_backend:
  #   - "socket" = "cluster" (via package 'parallel') or
  #   - "mpi" (via 'Rmpi')
  parallel_backend = if (identical(SFSW2_prj_meta[["opt_platform"]][["host"]],
    "local")) {
      "socket"
    } else if (identical(SFSW2_prj_meta[["opt_platform"]][["host"]], "hpc")) {
      "mpi"
    },

  # Computation time requests: time limits are only enforced if
  #   parallel_backend is "mpi"
  opt_job_time = list(
    wall_time_s = 12 * 3600, # requested wall time
    one_sim_s = 60, # time needed to complete one call to do_OneSite()
    one_concat_s = 60 # time needed to process one temporary SQL file
  )
)



#------ Options for printing progress and debugging information
opt_verbosity <- list(
  # Prints status of progress to standard output
  verbose = TRUE,
  # Prints details of progress to standard output
  print.debug = interactive(),
  # Calculates and prints estimated time of job completion at end of each
  #   call of 'do_OneSite' (a somewhat expensive operation)
  print.eta = interactive(),

  # Sets global option 'warn' for the duration of a simulation project
  # Possible values: -1, 0, 1, 2; for details: ?options -> Value: warn
  debug.warn.level = max(1, 2 * interactive()),
  # Should R objects be dumped to disk on error (including for each call to
  #   the function `do_OneSite`)
  debug.dump.objects = interactive()
)


#------ Output options
opt_out_run <- list(
  # Write rSOILWAT2 input and output to disk for each SOILWAT2 simulation
  saveRsoilwatInput = FALSE,
  saveRsoilwatOutput = FALSE,

  # Enforce that rSOILWAT2 objects meet the current version requirement
  enforce_rSW2_version = TRUE,

  # Write data to big input files for experimental design x treatment design
  makeInputForExperimentalDesign = FALSE,

  # Delete any previous dbOutput; be careful to not wipe your data!
  wipe_dbOutput = FALSE,
  # Delete temporary text files produced by workers (temporary storage of
  #   simulation output which is copied to dbOutput byu action 'concatenate')
  deleteTmpSQLFiles = TRUE
)


#----- Define chunk size for operations on massive data
opt_chunks <- list(
  # chunk_size == 1e4 && n_extract 6e4 will use about 30 GB of memory
  ExtractSkyDataFromNOAAClimateAtlas_USA = 10000,
  # Extracting data from NCEP/CFSR is also OS-limited by the number of
  #   concurrently open files (on 'unix' platforms, check with 'ulimit -a')
  ExtractSkyDataFromNCEPCFSR_Global = 100,
  DailyWeatherFromNCEPCFSR_Global = 100,
  # This value is the chunk size for reads of 'runID' from the database, i.e.,
  #   chunk size = ensembleCollectSize * scenario_No.
  #   Find a balance between available memory, cores, read/write times, etc.
  #   For instance, on Yellowstone, 500 seems to work well.
  ensembleCollectSize = 500
)



################################################################################