[postgis-users] Buffer bug ?

[strk at refractions.net]

Ok, I downloaded your test case and tried comparing JTS,
GEOS-2.1.4 and GEOS-1.0.0.

I used PostGIS version 1.1.0CVS as test interface.
Nothing changed in collect() and npoints() since PostGIS-1.0.3,
so results should be about the same in the two versions.

The query is:

	-- ### is Number of features
	SELECT npoints(buffer(collect(geom), 1000))
	FROM ( select lga_geom as geom from lg_annee limit ### ) as foo;

Number of features: 200 (1106 input points in collected MultiPolygon)

	[JTS-1.7]    output_points 1754
		real    0m8.705s

	[GEOS-2.1.4] output_points 1816
		real    0m5.594s

	[GEOS-1.0.0] output_points 1735
		real    0m16.450s

Number of features: 400 (2320 input points in collected MultiPolygon)

	[JTS-1.7]    output_points 1897
		real    1m3.761s

	[GEOS-2.1.4] output_points 1959
		real    0m43.020s

	[GEOS-1.0.0] output_points 1912
		real    6m50.783s

Number of features: 1000 (5992 input points in collected MultiPolygon)

	[JTS-1.7]    output_points 2598
		real    9m20.873s

	[GEOS-2.1.4] output_points 2618
		real    10m54.414s

	[GEOS-1.0.0] 
		KILLED AFTER:
		real    94m14.421s


I don't really understand why number of output points change, anyway
the test shows that GEOS-2.1.4 is *much* faster then GEOS-1.0.0
(and sometimes faster then JTS-1.7). For 1000 features GEOS-1.0.0
is completely stuck after 94 Minutes !!

So, to conclude, I could not verify the reported bug.

I suggest you update GEOS to version 2.1.4 and PostGIS
to the upcoming 1.0.4 release. Then, if you can reproduce the
OLD vs. NEW performance loss problem yourself, please provide a
*simple* test case.

As for killing GEOS during operation this is an old and known
problem we might approach soon or late, but it is not first
priority at the moment.

--strk;


On Mon, Aug 29, 2005 at 11:28:13AM +0200, Daniel wrote:
> Yes: it's harder for the whole community when you're away ! ;-)
> 
> My dataset is about 5 Mo, so I would not post it on a public list ... 
> Can I upload it on a server and give you a private access ?
> 
> I tried tests on reduced datasets, too: it works, but very slowly. Far 
> much slowly than with earlier versions. The execution time is 
> exponential. For example:
> 100 features: 303 ms
> 200               : 2478 ms
> 500               : 34690 ms
> 
> With our first implementation (firsts postgis versions with geos), we 
> got about 25 seconds for 7500 features.
> 
> Sincerely yours,
> 
> dF
> 
> 
> 
> strk at refractions.net wrote:
> 
> >On Fri, Aug 26, 2005 at 04:33:27PM +0100, Mark Cave-Ayland wrote:
> >...
> > 
> >
> >>Unfortunately I'm now away from work for at least a week, but I hope that
> >>someone else on the list will be able to give you some help in the 
> >>meantime.
> >>   
> >>
> >
> >Hello guys, it looks like I'm missing something here...
> >Daniel, would you post a test case for me to check out ?
> >A reduced case working on your old setup and hanging
> >on the new one ?
> >
> >--strk;
> >_______________________________________________
> >postgis-users mailing list
> >postgis-users at postgis.refractions.net
> >http://postgis.refractions.net/mailman/listinfo/postgis-users
> > 
> >
> 


> _______________________________________________
> postgis-users mailing list
> postgis-users at postgis.refractions.net
> http://postgis.refractions.net/mailman/listinfo/postgis-users