[Benchmarking] GS 2.0.1 results shed some light on the GS-MS similarities

[Andrea Aime]

Hi,
lately I managed to run the FOSS4G 2009 vector benchmarks
against GS 2.0.1 (as well as against the older GeoServer 1.7.0
used in the FOSS4G 2008 benchmarks) and the results
are interesting:
http://geoserver.org/display/GEOS/GeoServer+benchmarks

The interesting part is in a number of vector tests where
GS 2.0.1 and GS 1.7.0 are substantially faster than the
pack of results we got during the FOSS4G 2009 results,
in particular the areawater and gnis 2009 ones.

GeoServer 2.0.0 was benchmarked by mistake with a very
high PNG compression ratio, that actually slowed GS
down quite a bit. Once I restored a compression level
similar to the one used in MapServer I got the speedup
you can see.

Long story short, in the areawater and gnis tests there
was no strangely hidden bottleneck in the network,
the speedup clearly shows those tests are dominated by
the image compression, and thus by the CPU.

The results also show that for GeoServer labelling is
a bottleneck in the labelled tests, as we did not get
much of a speedup there.

In the light of these new results I don't really see
the network playing any significant role, and I'm
guessing the very similar results we got were just
due to chance (bad chance for us in GeoServer land ;-) )

Opinions?

Cheers
Andrea

-- 
Andrea Aime
OpenGeo - http://opengeo.org
Expert service straight from the developers.