[postgis-devel] Calling Index Genii

Nicklas Avén nicklas.aven at jordogskog.no
Fri Dec 31 02:36:47 PST 2010


yes, it seemed like a great theory. are you sure it is not building a 3d index anyway. 
can you confirm that the "DIMENSIONALITY
 DETERMINED BY THE DATA" thing is working. Is there no mechanism somewhere that adds an empty z coordinate like get_3d_point (or whatever  the name is) is doing.


2010-12-31 skrev Paul Ramsey :

Well, it was a great theory, but my test did not bear it out. Even
>with 2D inputs the new index is still taking about 5x longer to return
>a small result set as the old one. :(
>On Thu, Dec 30, 2010 at 3:32 PM, Paul Ramsey wrote:
>> You edit the postgis/gserialized.h file and define or undef GSERIALIZED_ON
>> BTW, I have a theory now which I just need to test... the data are 3D!
>> And in fact the new GiST index will build an index WITH DIMENSIONALITY
>> DETERMINED BY THE DATA. Not by the opclass. So the tree built with the
>> new index is going to be more expensive to search in 2D than an actual
>> 2D tree would have been. I just need to change my input data to 2D and
>> see if the performances fall into line to confirm my guess.
>> P.
>> On Thu, Dec 30, 2010 at 12:50 AM, Andrea Peri wrote:
>>>>On my lap top the loading and index building take about the same
>>>>amount of time, but the query runs in about 1ms for the old index and
>>>>about 5ms for the new one. My assumption (thus unconfirmed) is that
>>>>the problem is not in the consistent checks (doing things like
>>>>reducing the size of the slice detoasted to the minimum necessary for
>>>>2d queries only improved the times by about 10%) but rather in the
>>>>tree itself: that the new index is just building a crappier tree, so
>>>>the query has to traverse more of it to get the same answer.
>>> There even an increase of memory resource needed ?
>>> To test this version there is a new key setting to apply in the .configure,
>>> or need to change manually the definitions ?
>>> --
>>> -----------------
>>> Andrea Peri
>>> . . . . . . . . .
>>> qwerty àèìòù
>>> -----------------
>>> _______________________________________________
>>> postgis-devel mailing list
>>> postgis-devel at postgis.refractions.net
>>> http://postgis.refractions.net/mailman/listinfo/postgis-devel
>postgis-devel mailing list
>postgis-devel at postgis.refractions.net
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.osgeo.org/pipermail/postgis-devel/attachments/20101231/fcb5c5a2/attachment.html>

More information about the postgis-devel mailing list