[postgis-users] PostGIS 1.3.3 on Windows XP Unstable

Thu Aug 14 16:00:37 PDT 2008

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA256

> It's your bad luck to have geometries that make PostGIS fail...   8^)
> 
> If there's geometries which cause problems with certain operations,
the
> most helpful thing to do is to post them along with a description of
the
> problem.  That gives a fighting chance to fix them when further
> development is done.

The problem of ST_SimplifyPreserveTopology() producing geometries that
ST_Union() chokes on is already known. When I tried to scrub the
polygons I ran into performance problems, particularly with
ST_Union(geometry). I ran into similar performance problems with
Oracle's sdo_aggro_union().

> So Java2D worked on your problem?  I have to admit, I'm impressed how
> robust Java2D is.  If I could understand their algorithm I'd be keen
on
> trying to follow their approach to get the same level of robustness in
> JTS/GEOS...

Yes, Java2D was able to perform a union on the geometries produced by
ST_SimplifyPreserveTopology() and it does the union relatively fast. 

The only thing I have found that beats Java2D is ArcGIS. In one test I
did I performed a union of the USGS state boundaries
(http://www.census.gov/geo/cob/bdy/st/st00shp/st99_d00_shp.zip). Here
are the results:

Oracle 11.1:	25 min
PostGIS 1.3.3:	42 sec
Java2D 1.5:		4 sec
ArcMap 9.2:		1 sec

Any thoughts why PostGIS is about 40x slower than ArcMap at this
operation?

Any thoughts why Oracle takes 25 min? lol

The bottleneck in my application now looks to be
ST_SimplifyPreserveTopology(). I'm finding that it's often best to do
computation outside the database. However, one thing that Java2D is
missing is a polygon simplification function. I might look to see if
Geotools has one.