[Tilesathome] Saving a few cycles

[Jiri Klement]

On Tue, Sep 9, 2008 at 9:53 PM, Christian Ehrlicher <Ch.Ehrlicher at gmx.de> wrote:
> Jiri Klement schrieb:
>> Running pnqnq on the whole tileset doesn't make much sense. What pngnq
>> do is find most commonly used colors and build palette of them. If you
>> run pngnq on the whole tileset, then there will be more colors and
>> selected colors will be less optimal. It would be much faster then to
>> simply make some good palette and apply it every tile, no matter what
>> colors the tile use.
>>
> Afaik there's a ongoing work to create a common palette. But for now we
> don't have it.
> I don't see your point - imho it's better to have more tiles with the
> same color palette than every tile with it's on one.

The problem is that every tile has lots for colors (mainly due to
antialiasing). If every tile has it's own optimalized palette, then
there is bigger change that colors in the pallete will be closer to
real colors. What pngnq takes so long is looking for the optimal
pallete. If you have one pallete for all tiles, then you wouldn't need
pngnq and you will save lots of cpu cycles.

>> Btw. If you are worried about I/O and have enough memory (3GB is more
>> than enough when you use batik), then you can set t at h working
>> directory to be on ramdisc. I have /tmp on ramdisc and it's
>> considerably faster.
>>
> I've no problems with I/O - I just render now and then (and normally on
> windows where batik is sloooooooooow) :-)

Try batik in agent mode. It's slightly faster then inkscape on linux
because it runs long enough to allow Java to compile into native code.

--
Jiri