r.watershed crashing on large rasters

[dnewcomb]

Attempting to run r.watershed on a 13.7 billion pixel raster elevation layer

r.watershed --overwrite elevation=cfsdem05 accumulation=cfs_accum drainage=cfs_drain

SECTION 1a (of 4): Initiating Memory. WARNING: Subprocess failed with exit code 11 WARNING: category information for [cfs_accum] in [PERMANENT] missing or

invalid

WARNING: category information for [cfs_drain] in [PERMANENT] missing or

invalid

I have tried this with Memory=100000 MB

Does r.watershed need more than 100GB of memory to run on a 14 billion pixel raster?

The raster is linked via r.external, doe it need to be imported ?

Looking backwards, the enhancement to run large raster was done for GRASS 7.0 , is this a regression, or was it not tested to larger than 9 billion cells?

[mmetz]

Replying to dnewcomb:

Attempting to run r.watershed on a 13.7 billion pixel raster elevation layer

r.watershed --overwrite elevation=cfsdem05 accumulation=cfs_accum drainage=cfs_drain

SECTION 1a (of 4): Initiating Memory. WARNING: Subprocess failed with exit code 11 WARNING: category information for [cfs_accum] in [PERMANENT] missing or

invalid

WARNING: category information for [cfs_drain] in [PERMANENT] missing or

invalid

I have tried this with Memory=100000 MB

The memory option has an effect only when using the -m flag. I don't think r.watershed in all-in-memory mode (no -m flag) is supporting such large input maps.

Does r.watershed need more than 100GB of memory to run on a 14 billion pixel raster?

The raster is linked via r.external, doe it need to be imported ?

Looking backwards, the enhancement to run large raster was done for GRASS 7.0 , is this a regression, or was it not tested to larger than 9 billion cells?

[martinl]

Ticket retargeted after milestone closed