I've been getting some slow responses from STIntersection() using NTS,
and have tried performing the same intersection with Postgis and
Microsoft.SqlServer.Types.dll to see how they compare. The differences
are remarkable:

Postgis: 300 - 400ms
SqlServer.Types.dll: 50 - 100ms
Nts: 2 minutes, 21 seconds!

I'm using version 1.7.3 on an XP VM. (I also tried running the NTS
intersection test using mono on a mac with similar results). I've
tried to add the WKT strings for the two shapes to this message, but
the post was rejected, probably because one of the strings was too
large - will see if I can attach.

Any suggestions would be most welcome!

Many thanks.

WKT strings attached.

wkt.txt 175K Download

Diego Guidi

On Apr 26, 7:25 am, Diego Guidi <diegogu...@gmail.com> wrote:

the same test runs in about ten seconds in JTS 1.8... now it's time to find
how NTS loses all the time

On Apr 26, 12:05 pm, Diego Guidi <diegogu...@gmail.com> wrote:

Could you let me know what the svn branch is for v2.11? I'll try a
build at this end and test performance..

On Apr 26, 3:59 pm, Felix Obermaier <o...@IVV-AACHEN.DE> wrote:

OverlayOp.CopyPoints for the multipolygon geometry is executed in a fraction
of a second by JTS, and in 10 seconds or more by NTS (trunk)
I think that here is the main problem, now the hard part: discover why this
happens :)

Hi Diego, perhaps try with r538 or earlier to make sure I didn't break anything aliasing ArrayList = List<Object> etc when doing the Silverlight stuff...

From: nettopologysuite@googlegroups.com [mailto:nettopologysuite@googlegroups.com] On Behalf Of Diego Guidi
Sent: 26 April 2011 16:27
To: nettopologysuite@googlegroups.com
Subject: Re: R: Re: STIntersection performance

OverlayOp.CopyPoints for the multipolygon geometry is executed in a fraction of a second by JTS, and in 10 seconds or more by NTS (trunk)
I think that here is the main problem, now the hard part: discover why this happens :)

--
You received this message because you are subscribed to the Google Groups "NetTopologySuite" group.
To post to this group, send email to nettopologysuite@googlegroups.com.
To unsubscribe from this group, send email to nettopologysuite+unsubscribe@googlegroups.com.
For more options, visit this group at http://groups.google.com/group/nettopologysuite?hl=en.

sorry, I'm wrong. Maybe a temporary result due to some problem in my pc... I
check better and see

Thinking about it the alias only applies within Silverlight - in .Net it stays as ArrayList...

From: nettopologysuite@googlegroups.com [mailto:nettopologysuite@googlegroups.com] On Behalf Of John Diss
Sent: 26 April 2011 16:32
To: nettopologysuite@googlegroups.com
Subject: RE: R: Re: STIntersection performance

Hi Diego, perhaps try with r538 or earlier to make sure I didn't break anything aliasing ArrayList = List<Object> etc when doing the Silverlight stuff...

From: nettopologysuite@googlegroups.com [mailto:nettopologysuite@googlegroups.com] On Behalf Of Diego Guidi
Sent: 26 April 2011 16:27
To: nettopologysuite@googlegroups.com
Subject: Re: R: Re: STIntersection performance

OverlayOp.CopyPoints for the multipolygon geometry is executed in a fraction of a second by JTS, and in 10 seconds or more by NTS (trunk)
I think that here is the main problem, now the hard part: discover why this happens :)

--
You received this message because you are subscribed to the Google Groups "NetTopologySuite" group.
To post to this group, send email to nettopologysuite@googlegroups.com.
To unsubscribe from this group, send email to nettopologysuite+unsubscribe@googlegroups.com.
For more options, visit this group at http://groups.google.com/group/nettopologysuite?hl=en.
--
You received this message because you are subscribed to the Google Groups "NetTopologySuite" group.
To post to this group, send email to nettopologysuite@googlegroups.com.
To unsubscribe from this group, send email to nettopologysuite+unsubscribe@googlegroups.com.
For more options, visit this group at http://groups.google.com/group/nettopologysuite?hl=en.

If I disable NodingValidator (OverlayOp.NodingValidatorDisabled = true;)  
the operation is executed immediately.
Of course this is not a good thing to do, and JTS validates noding, so I  
think that is the nodingvalidator that actually is slow.

Il giorno , Diego Guidi <diegogu...@gmail.com> ha scritto:

--
You received this message because you are subscribed to the Google Groups "NetTopologySuite" group.
To post to this group, send email to nettopologysuite@googlegroups.com.
To unsubscribe from this group, send email to nettopologysuite+unsubscribe@googlegroups.com.
For more options, visit this group at http://groups.google.com/group/nettopologysuite?hl=en.

CheckEndPtVertexIntersections 00:00:00.1641412
*CheckInteriorIntersections 00:02:11.3173618*
CheckCollapses 00:00:00.0017678

On Apr 26, 5:48 pm, Diego Guidi <diegogu...@gmail.com> wrote:

FastNodingValidator is used by 2.x branch, and looks that contains "only"
check for interior intersections, actually the same heck that is so slow in
trunk. So maybe here's why trunk and 2.x branches are so slow.

I've added a test that validates a single segmentstring (the slower of all
the checks, alone it executes in 5 seconds).
In NTS the check is executed in 5 seconds (as expected), in JTS the
execution is immediate.
I hope that analyzing the single test I can find how is the problem

why god simply using a one-dimensional array can improve performances to 2x
or more?
the test on segmentstring now executes in 2 seconds instead of 5, and all
I've changed is that now in LineIntersector
protected ICoordinate[] inputLines = new ICoordinate[4];
and not
protected ICoordinate[,] inputLines = new ICoordinate[2,2];

On May 2, 11:02 am, Diego Guidi <diegogu...@gmail.com> wrote: