I stumbled across optipng (for Linux) recently, and wondered whether it would be worth running this on the OSM tiles (or rather, incorporating the algorithm into newly generated ones).
It seems to get maybe 25% size reduction on "empty" tiles (ocean, fields) but only 2-3% on "busy" urban tiles.
討論
由 andrewpmk 於 2009年08月11日 00時18分 發表的評論
It would probably take far too much CPU power to be worth running on the tile server.
由 Falcorian 於 2009年08月11日 00時51分 發表的評論
And of course optipng isn't the only compression program, pngout for example generally gets smaller files at the cost of more processing time.
由 adaviel 於 2009年08月11日 01時03分 發表的評論
I remember I used to get significant reduction in GIFs by reducing the palette size, before everyone had broadband. With people getting images over cell data networks (and maybe paying per kb) there may be some point to it again.
On a related topic, I forget whether I saw it on OSM or Maemo Mapper - is there any scheme to save downloads by redirecting all requests for "empty blue tile" to the same URL ?
由 Firefishy 於 2009年08月14日 08時56分 發表的評論
The OSM Mapnik tiles are 256 colour PNGs to improve their size.
The tile server is CPU bound and currently bandwidth usage is not a problem.
由 adaviel 於 2009年08月15日 03時18分 發表的評論
I tried pngout. Much slower than optipng, at least on the maybe 1.5Mpixel image I used.
I was thinking about bandwidth on mobile clients, not the server.