[Tile-serving] openstreetmap-carto benchmark results

[Paul Norman]

> From: Frederik Ramm [mailto:frederik at remote.org]
> Sent: Thursday, May 16, 2013 4:16 PM
> To: tile-serving at openstreetmap.org
> Subject: Re: [Tile-serving] openstreetmap-carto benchmark results
> 
> Possibly stupid question since I haven't followed too closely, but have
> we already established that the performance loss is something one has to
> live with?

Yes and no. Because carto compiles to mapnik XML, carto will never be
faster. Because mapnik XML is a lower level language, there is the
possibility of doing stuff more directly than in carto, leading to faster
rendering. On the other hand, the switch to carto has never been about
speed, it's about having a sane way to edit the stylesheet.

> If the carto style is slower when generating the exact same images then
> it must be making (even more ;)) unnecessary queries than when the XML
> style is used; maybe it is possible to fix either the style or the
> toolchain to avoid that? Does the benchmark perhaps offer some kind of
> starting point here - can one somehow compare the database queries made
> by both solutions during the test run, find out which of them differ
> vastly, and then work on making them differ less?

There is some work at fixing issues related to attachments,
https://github.com/gravitystorm/openstreetmap-carto/issues/15 is tracking it

I'm working at it from a different angle - I'm trying to identify
performance improvements, most of which would likely apply to osm.xml as
well. I'm looking at postgresql/postgis stuff, mainly with limiting the rows
that reach mapnik or having simpler geometries. There might be optimizations
that apply to both that are more related to the mapnik and styling side, but
that's not my area so I'm not looking for them.