[Tilesathome] Colour resolution

[Kai Krueger]

Frederik Ramm wrote:
> Accidentally sent to Kai only:
> 
> Hi,
> 
>> I have had the impression, that the osmarender/tah layer feels slower to browse 
>> than e.g. the mapnik layer, so I tried to find out what may be the cause for 
>> that. One of the things I have noticed is that the osmarender tiles are much 
>> bigger than the mapnik tiles, often on the order of 3 to 4 times bigger. Given 
>> that the difference between a single 100kb tile and a 27kb tile (e.g. z12 of 
>> Karlsruhe) can be noticed even over a reasonably fast dsl connection and 
>> definitely over slower links, it might be a good idea to try and reduce the tile 
>> size. This would also reduce up and download bandwidth needed by the server.
> 
> Are you sure that you have not accidentally used a non-pngcrush tile for
> your comparison? Normally, tiles at home puts tiles through pngcrush before
> they're uploaded. I believe it may be possible to run t at h without this
> program so if you used tiles by someone who does not have pngcrush
> enabled then they'd look very big...

Well, I had simply pulled random tiles from the server, so I don't know 
if they were put through pngcrush or not. But given it was a selection 
of tiles, it is what is on the server.

I have now tried to put the files through pngcrush and optipng myself 
and in each case it said that the files were in fact already optimal. So 
that is not really the explanation.

Here are the numbers for some random tiles:

High complexity tiles:

15/16373/10895 mapnik: 19kb tah: 49kb tah-8bit: 22kb  2.5x bigger
14/8414/5384   mapnik: 22kb tah: 79kb tah-8bit: 33kb 3.5x bigger
12/2143/1406   mapnik: 27kb tah: 100kb tah-8bit: 41kb 3.7x bigger



Low complexity tiles:

16/34282/22518  mapnik: 3.87kb  tah: 9.3k  tah-8bit : 5.4k 2.3x bigger
15/16220/10821 mapnik: 2.8kb tah: 3.0kb tah-8bit: 2.5kb 1.07x bigger

for tah-8 I took the tiles I downloaded from the tah server and manually 
converted the tiles to 8bit color pallet.

Ok, so 4 times bigger I guess was a bit exaggerated and the Karlsruhe 
tile is an exrtreme case, but  still, apart from the most trivial tiles, 
the size of the tah tiles do seem a lot bigger and neither optipng nor 
pngcrush help in anyway in improving the situation. In fact it helped 
the mapnik tiles more which it could actually compress by a few bytes.

So it seems like a thing worth looking into a bit more of how to reduce 
the tile sizes.

Kai


> 
> If it turned out that even though we use pngcrush we STILL have tiles 3
> to 4 times larger than Mapnik's, that would be a reason to bin pngcrush ;-)
> 
> Mapnik's use of 8bit images is not without problems; sometimes colours
> look wrong and people have complained about this from time to time.
> 
> Bye
> Frederik
>