[OSM-dev] OpenStreetMap

[Jon Burgess]

On Mon, 2007-11-19 at 18:19 +0000, Robert (Jamie) Munro wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
> 
> Steve Coast wrote:
> > Derek
> > 
> > Thanks much for your email. I'm CC'ing Jon Burgess who now manages our  
> > main tile server and may be able to push this in to our tiles.
> > 
> > I'm also pulling in Oliver and Chris from the tiles at home effort as  
> > this applies to them too.
> 
> Tiles at Home passes all tiles through PNGcrush which AFAIK does similar
> optimisations.
> 
> If Mapnik were to use PNGcrush, it would probably make it infeasibly
> slow at it's rendering on demand process, however we can probably find
> more optimal settings for whatever mapnik uses to encode PNGs, and this
> would certainly be worth doing.

We already perform colour reduction to a 256 colour palette. The
original images are rendered in 32bit RGBA. Last time I tried pngcrush
the results were not worth the extra processing overhead. The reduction
was always < 10% from what I recall.

Most of the benefit claimed on the blog page seems to be due to reducing
the palette from 256 to 64 colours. This is fine for simple tiles but I
think is likely to cause problems on tiles with many colours. 

The tool which Mapnik uses to perform the colour reduction is in SVN at 
http://trac.openstreetmap.org/browser/applications/utils/color255
If anyone comes up with any enhancements or alternatives to that tool
with some evidence to show an improvement over the current code then I'd
be happy to consider adopting something better.

Approximately half the time spent to rendering a single tile in the
Mapnik layer is spent in the colour conversion (color255) so the speed
of this code is very important to the overall tile processing rate.

	Jon