[MapProxy] Benchmarking Mapnik seeding performance

[Paul Norman]

I'm doing some investigations of metatile settings with MapProxy, and 
wanted to run my method by the list in case anyone can spot problems 
with it.

I have mapnik source pointing at Mapnik XML, and a file cache based on 
it, saving in TMS directory layout. The metatile size is adjusted 
depending on what I'm testing. For a seed file, I'm seeding that cache 
to z8 with mapproxy-seed, adjusting the parallelism with -c.

A couple of potential problems I already know about are

- Mapnik has problems with images over 16384px in any dimension, so the 
biggest metatiles that are power of two are 32x32
- inode limitations can be a problem for higher zooms, but aren't an 
issue at z8, or even z12.

Are there any others that I might hit, particularly any unique to 
MapProxy? I recall hearing that MapProxy needs to reload part of Mapnik 
for each tile it renders, but found nothing documenting this and have no 
idea if it's true.