[Tile-serving] [osm2pgsql] Disable ORDER BY way "clustering" (#102)

Paul Norman notifications at github.com
Wed Nov 13 20:39:37 UTC 2013


Testing indicates that a modified pinning number is no better with
ORDER BY way than by clustered/ordered on osm_id. As the table at
this point are approximately ordered by ID, skip ordering completely.

The modified pinning number is the pinning number, but instead of a
point, an area equivalent to a rendering meta-tile.

For high zooms and a true pinning number ordering by ID is effective,
probably because typically the ways right around a point were created
at about the same time.

For low zooms, a clustering on gist (way) may be desirable, but this
is something the user will have to consider, particularly as it takes
a decent amount of time to do this clustering, and significantly
increases the disk space requirements during the import process.

Benchmarks show the pinning number is within 10% for both a full planet and an extract and the ID-ordered z12 modified pinning number is 50% of the `ORDER BY way` z12 modified pinning number for the line table and 90% for the polygon table.

@apmon, I know there's some conflicts between this and the threading branch, but this also completely eliminates a section you had to change, making the merge simpler.
You can merge this Pull Request by running:

  git pull https://github.com/pnorman/osm2pgsql no_ordering

Or you can view, comment on it, or merge it online at:

  https://github.com/openstreetmap/osm2pgsql/pull/102

-- Commit Summary --

  * Disable ORDER BY way "clustering"

-- File Changes --

    M output-pgsql.c (10)

-- Patch Links --

https://github.com/openstreetmap/osm2pgsql/pull/102.patch
https://github.com/openstreetmap/osm2pgsql/pull/102.diff
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstreetmap.org/pipermail/tile-serving/attachments/20131113/34715d00/attachment.html>


More information about the Tile-serving mailing list