Bad Rendering Performance (Tirex)

Question

We have set up a rendering server with a Intel(R) Xeon(R) CPU E5-1620 v2 @ 3.70GHz, 32 GB RAM and SSD discs for the database storage.

Import was done using osm2pgsql -C14G --number-processes 8 -v --slim planet-latest.osm.pbf and postgresql 9.3 and the settings as suggested by Frederik Ramm.

But the performance on low zoom levels is worse than what Frederik gets in his talk. Maybe because of additional map data in the past 2 years?

Some extreme examples:

Map                           X          Y   Z Prio   Age
example                      32         16   6    1   486

Currently rendering: (num=4)
Map                           X          Y   Z Prio   Age
example                     136         80   8    1   295
example                     128         80   8    1   295
example                     264        168   9    1   285
example                     528        352  10    1   260

I am using the standard osm.xml (mapnik-osm-carto).

generate_image.py with the standard bounding box (GB) takes 6m10.450s.

Is this normal?

Accepted Answer

You've already partially answered your own question: two years ago we had roughly half the data we currently have in the database see for example http://osmstats.altogetherlost.com/index.php?item=nodes (I'm actually surprised that you are not still importing given that the cache value you used is far too small for a current planet).

However the queries for low zoom tile data tend to already substantially reduce the ammout of data, so that is likely not to be a major factor. Frederiks numbers seem to be very good, I've not actually ever seen low zoom tiles render as fast as his numbers show. IIRC for the standard style we pre-render down to zoom level 14 to work around that.

Bad Rendering Performance (Tirex)

Follow this question

Related questions