[Tile-serving] openstreetmap-carto benchmark results

Andy Allan gravitystorm at gmail.com
Sat May 18 10:40:26 UTC 2013

On 18 May 2013 10:00, Paul Norman <penorman at mac.com> wrote:

> I won't be able to easily figure this out, since it would require manipulating osm.xml. I'm a bit reluctant to do this, because I really dislike mapnik XML.

Mapnik has a rendering stats option, which provides per-layer
breakdowns. It's human-viewable, but I'd love to see some work (either
in mapnik or as a post-processor) to parse these out and give
per-layer statistics.


It's great as-is for single images, but it would be better to parse
the output and aggregate them over the 9k queries that you're running.

> Three slow layers are roads, buildings and admin. I'm sure there's more. Removing them speeds up z13 by 46%, 34% and 22% respectively. I doubt the buildings can be optimized - there's just a lot of objects, few of which can be omitted and few of which intersect the bbox and aren't within it.

That leads me to suspect that it's the endcaps on the roads, or the
dasharrays on paths, which are both hard to avoid. But it shows that
given these percentages, there's not much point in investigating any
other layers!

> I also want to re-format all the SQL - right now it's formatted to produce pretty mapnik XML with the \n and tabs, I'd much rather an easier to edit .mml.

No, it's formatted that way to produce readable SQL within the
Tilemill GUI. Unfortunately the lack of multi-line query support in
carto is a known issue, and I don't see a way around it without moving
away from json, which is unlikely.


More information about the Tile-serving mailing list