Suggest a safe tile size for the tiled seasonal analysis engine

Computes the largest square tile (in pixels) that fits within target_ram_mb of working memory, given the number of raster layers, variables and parallel workers. wapor_plan_processing() chooses a tile size automatically; use this helper to pick one by hand for wapor_run_seasonal_analysis_tiled().

Usage

wapor_suggest_tile_size(
  n_layers,
  n_workers = 1L,
  bytes_per_val = 8L,
  target_ram_mb = 4096,
  min_tile = 64L,
  max_tile = 4096L,
  n_vars = 1L,
  overhead = 2.5
)

Arguments

n_layers

Integer. Number of raster layers to hold in memory at once (e.g. total dekadal layers in the season).

n_workers

Integer. Number of parallel tile workers. Each worker holds one tile in RAM simultaneously. Default 1L.

bytes_per_val

Integer. Bytes per cell in memory. terra holds values as 8-byte doubles, so the default is 8L.

target_ram_mb

Numeric. RAM budget in megabytes. Default 4096 (4 GB).

min_tile

Integer. Minimum tile size to return. Default 64L.

max_tile

Integer. Maximum tile size to return. Default 4096L.

n_vars

Integer. Number of variables read per tile. Default 1L.

overhead

Numeric. Multiplier for temporaries. Default 2.5.

Value

A single integer: recommended tile side length in pixels.

Details

Formula: bytes_per_tile = tile_size^2 * n_layers * n_vars * bytes_per_val * overhead total_bytes = bytes_per_tile * n_workers

The result is clamped to [min_tile, max_tile] so extreme inputs produce a usable value rather than an error.

Examples

# 36 dekadal layers, 2 workers, 8 GB RAM budget
wapor_suggest_tile_size(n_layers = 36L, n_workers = 2L, target_ram_mb = 8192)
#> wapor_suggest_tile_size: 2442 px (n_layers=36, n_vars=1, n_workers=2, budget=8192 MB, 4094.7 MB/tile)
#> [1] 2442