[postgis-users] Small speed comparison

[strk at refractions.net]

I'm happy to hear about this, speed was my primary concern.

I'd exclude your first guess, second and third could be,
also consider the reduced levels of indirection now that 
points scan is done using real objects vs. pointers.

Would people having problems with qgis and mapserver try
their tests again ? 

--strk;

On Fri, Feb 11, 2005 at 11:39:30AM +0100, Markus Schaber wrote:
> Hi,
> 
> I re-ran my benchmarks with a 1.0.0 CVS checkout from today morning, to
> see the impact of the alignment issues. I simply added the new timings
> after the quoting of the old ones. The hardware setup is the same, and
> the table was not changed, so the test environment should be comparable.
> 
> Please note that all Geometries are polygonal, so this does not say
> anything about other geometries.
> 
> > lwgeom=# select geom from adminbndy1;
> > Zeit: 8629,822 ms
> > Zeit: 6401,693 ms
> 
> Zeit: 10556,032 ms
> Zeit: 6301,014 ms
> Zeit: 5809,858 ms
> 
> > lwgeom=# select asBinary(geom) from adminbndy1;
> > Zeit: 11648,607 ms
> > Zeit: 7845,229 ms
> 
> Zeit: 7604,296 ms
> Zeit: 7465,358 ms
> Zeit: 8062,700 ms
> 
> > lwgeom=# select asEWKB(geom) from adminbndy1;
> > Zeit: 7824,009 ms
> > Zeit: 7308,423 ms
> 
> Zeit: 7191,243 ms
> Zeit: 7342,762 ms
> Zeit: 7214,146 ms
> 
> > lwgeom=# select asText(geom) from adminbndy1;
> > Zeit: 33655,132 ms
> > Zeit: 33154,569 ms
> 
> Zeit: 32816,896 ms
> Zeit: 32792,680 ms
> Zeit: 32645,417 ms
> 
> > lwgeom=# select asEWKT(geom) from adminbndy1;
> > Zeit: 34402,441 ms
> 
> Zeit: 37156,399 ms
> Zeit: 35210,384 ms
> Zeit: 33757,256 ms
> 
> > lwgeom=# select asEWKB(geom)::bytea from adminbndy1;
> > Zeit: 7444,995 ms
> 
> Zeit: 8681,531 ms
> Zeit: 7225,146 ms
> Zeit: 7230,526 ms
> 
> I woult not try to read too much out of those results (as the test
> environment was not 100% clean), but it seems go be a slight speedup. I
> could think of three possible reasons:
> 
> - Possibly, some optimizations at other places between my first test CVS
>   checkout (start of january) and now overrule the effect of your
>   alignment work.
> 
> - The new code somehow allows better use of processor caches and/or
>   pipelines
> 
> - My processor (Mobile Intel Pentium 4 @ 2Ghz) profits from the better
>   alignment speed-wise.
> 
> Personally, I would think that the third variant is the most likely one.
> 
> Can anyone do some similar tests on another hardware?
> 
> Markus
> 
> 
> --
> markus schaber | dipl. informatiker
> logi-track ag | rennweg 14-16 | ch 8001 zürich
> phone +41-43-888 62 52 | fax +41-43-888 62 53
> mailto:schabios at logi-track.com | www.logi-track.com



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