[OSM-dev] Slippy map's future

[Nick Hill]

You could set up 100 machines with the current set up. You will end up 
with 100 slow, crippled slippy servers.

The bottleneck is no longer the database. It hasn't been for months.

The bottleneck is the lack of rendering scheduling system. The 
bottleneck is lack of intelligent tile invalidation system.

Currently good tiles are thrown away - tiles which take ages to render. 
Whilst bad tiles are stored - tiles where the rendering process times 
out. A renderer starts for every concurrent HTTP request bungung the CPU 
scheduler and memory with mostly moribund processes. If the tile 
rendering system were re-architected as discussed, current hardware will 
probably be adequate for a year's growth, and be fairly scalable for 
years to come.




Nick Whitelegg wrote:

> That is a good idea, particularly if combined with the idea that someone 
> else mentioned about different servers focusing on one particular 
> geographical location. Indeed, the whole of the UK is proving a bit much 
> for the Freemap server (Bytemark VM) but if it only offered southern 
> England between say 0.5W and 2W, and below 51.5N that would be much more 
> manageable.