Difference between revisions of "Replacement raster format"

From GRASS-Wiki
Jump to: navigation, search
(Recent discussions: + link to ML)
(Disadvantages of tiled storage)
Line 24: Line 24:
 
** Reasoning:  Glynn said on the mailing list: "In most cases, single-level tiled storage will give you close to the same performance with a lot less complexity."
 
** Reasoning:  Glynn said on the mailing list: "In most cases, single-level tiled storage will give you close to the same performance with a lot less complexity."
 
** Function needed to check whether tiles are all null, or all the same value.
 
** Function needed to check whether tiles are all null, or all the same value.
** What tile size should be used? Could be user/program specified, or standard value of something like 64x64. If there isn't a fixed value then there should be utility program that can convert tiled rasters to different tile sizes. - [[User:Ferrouswheel|Ferrouswheel]]
+
** What tile size should be used? Could be user/program specified, or standard value of something like 64x64. If there isn't a fixed value then there should be utility program that can convert tiled rasters to different tile sizes. -  
 +
*** However, row-oriented storage has the advantage that you can easily skip entire rows when downsampling. Using e.g. 64x1 "tiles" would provide both optimisations, but at the expense of reduced compression, as you need a pointer (8 bytes) for each tile, and the compression has to be restarted for each tile.
 +
[[User:Ferrouswheel|Ferrouswheel]]
 
* Merge NULL file into main data array.
 
* Merge NULL file into main data array.
  

Revision as of 00:43, 4 October 2008

 GRASS's long-standing raster format is overdue for a major overhaul.
Below you will find some ideas and roadmaps for future work.
The idea of this page is to collect ideas and flesh out a specification so that when the change occurs, all the components will be in place, pitfalls expected, and the implimentation, when it comes, quick and painless. Most importanly it can serve to keep interested parties informed and working together instead of in parallel forks.
Any changes to the data format will necessitate a bump in major version number (i.e. from GRASS 6 to GRASS 7) so if possible changes should happen in the same development cycle, and relatively minor changes should be held back in experimental status until a major change is committed.

see also GRASS 7 ideas collection#Raster

Core raster format

Lead developer: Glynn Clements

  • Storage in tiles instead of by row.
    • Reasoning: Glynn said on the mailing list: "In most cases, single-level tiled storage will give you close to the same performance with a lot less complexity."
    • Function needed to check whether tiles are all null, or all the same value.
    • What tile size should be used? Could be user/program specified, or standard value of something like 64x64. If there isn't a fixed value then there should be utility program that can convert tiled rasters to different tile sizes. -
      • However, row-oriented storage has the advantage that you can easily skip entire rows when downsampling. Using e.g. 64x1 "tiles" would provide both optimisations, but at the expense of reduced compression, as you need a pointer (8 bytes) for each tile, and the compression has to be restarted for each tile.

Ferrouswheel

  • Merge NULL file into main data array.

Directory structure

  • Centralize map components in "$MAPSET/raster/$MAPNAME/*" instead of many "$MAPSET/cell/$MAPNAME", etc. directories.

Many library functions and modules will need to be updated. The GRASS 6 vector format has already been ported to this structure.

Meta-data support

The existing raster meta-data handling is rather weak. (currently stored in $MAPSET/hist/$MAPNAME) Total replacement will be the best option.

Brad Douglas suggests:

 It would be very advantageous to at least support metadata as specified
 in FGDC-STD-012-2002.
 XML is an ideal file format.

Nick Lawrence suggests: The GIMP project is examining the concept of a general mult-layer bitmap format.

OpenRaster - Create Wiki OpenRaster Sandbox

Recent discussions

Other suggestions: