[postgis-devel] GeomUnion Performance Info

Bill Binko bill at binko.net
Wed Jun 22 18:52:17 PDT 2005


strk,

I'm going to have to rerun all of the performance graphs if it's going to 
be a reasonable comparision.  The reason is that the database has simply 
not stayed static since I ran those.  I have added columns, indexes, etc. 
and it simply wouldn't be meaningful.

So, I will get your short-circuit code in there, and run the following 
tests (with the same data and load):

1) Random Sort (clustered)
2) GiST Clustered
3) BTree clustered (a bit hard, but I can do it)

And I'll run them both with and without the short-circuit code.  Is there 
a runtime switch?  If not, I'll recompile gdal with and without (I 
shouldn't have to rebuild postgis right?).

It will take at least tomorrow to do that.  However, I'll aim for being 
complete before the weekend in case there's development to be done then.

Bill

 On Wed, 22 Jun 2005, Bill Binko wrote:

> On Thu, 23 Jun 2005 strk at refractions.net wrote:
> 
> > Similar in terms of performances also or just visual effect ?
> > Did you try the short-circuit tests ? Would you add curves
> > for them as well if you do ?
> > 
> 
> The performance in in the graph as "Spatial Index".  I thought that's what 
> ORDER BY was using when I did the graphs.  It is similar to, but slower 
> than, the clustered GiST.
> 
> I suppose I can cluster on the attribute I just added which would give a 
> clustered btree number.  I can add that to the list if you think it's 
> worth it.
> 
> I have not doing the short-circuit yet, as I will need to upgrade PostGIS 
> from 1.1rc6 to CVS HEAD which may be painful.  (I just don't want daytime 
> downtime on it.)
> 
> Bill
> _______________________________________________
> postgis-devel mailing list
> postgis-devel at postgis.refractions.net
> http://postgis.refractions.net/mailman/listinfo/postgis-devel
> 
> 



More information about the postgis-devel mailing list