[OSM-dev] Improvement of tile rendering

[Stefan de Konink]

Udo Giacomozzi wrote:
> Hello Stefan,
> 
> Friday, March 20, 2009, 12:55:27 AM, you wrote:
>>> Instead I'm thinking about writing some multithreaded FreePascal or
>>> C++ FastCGI server that does it much more efficiently (by using
>>> libmapnick internally) but I still need to have a look at mod_tile as
>>> it might already provide a highly performant solution.
> 
> SdK> Investigate where the time is; it is not in the wrapper layer.
> 
> Most of the time will of course be spent in the rendering process. But
> very often there are optimizations possible (reusing instances or data
> structures, multithreading strategies, avoiding unnecessary rendering,
> trying to predict following requests - to name a few quick ideas).
> 
> As said, I have to investigate the current solutions in depth before I
> can think of concrete alternatives...

What you are attempting with a predictive strategy would probably 
require a render queue. Or if you are really living on the edge 
rendering big tiles (4096x4096) that are cutted in smaller tiles, as one 
request so that every tile in the surrounding is made available by one 
single database select.

Now I think this process could potentially be useful especially when 
updating large amounts of data. For the rest I put my money on the webcache.


Stefan