[Tile-serving] [openstreetmap/osm2pgsql] Use libosmium for geometry building (#684)

[Sarah Hoffmann]

> That is an impressive reduction of CPU from 30k to 12k seconds. Can that be explained by the simpler geometry processing?

Yes. Using geos functions to build OSM geometries was not a straightforward process while libosmiums algorithm is tailored for the peculiarities of OSM data. There is also a lot less memory allocation going on when preparing the data as more data is processed directly in place in the libosmium buffers.

> Given that the overall time is 20k seconds, the bottleneck must be somewhere other than osm2pgsql.

Our biggest bottleneck is still postgres. Given that all queries (including inserts) are done synchronous, osm2pgsql probably spends a lot of time waiting for results.

Looking at the numbers it might be worth testing if reducing the number of parallel threads won't actually be faster.

-- 
You are receiving this because you are subscribed to this thread.
Reply to this email directly or view it on GitHub:
https://github.com/openstreetmap/osm2pgsql/pull/684#issuecomment-279943923
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstreetmap.org/pipermail/tile-serving/attachments/20170215/1ba960d2/attachment-0001.html>