[Benchmarking] Vector tests: first oddity

[Andrea Aime]

Hi,
so I managed to make a couple of runs of the vector only tests
and here pops up the first oddity: the reprojected tests are
faster than the non reprojected ones.

Sure, I only have GS results, but it's not possible to do more
work and take less time.
Looking at the outputs the reason is evident thought: the
reprojected requests are more zoomed in, visually, they do
use less data.

Now, I tried to make the two csv files similar, but the thing is,
the whole calculation is based on the OGC suggested scale
computation formulas, which are completely different beasts for
the geographic and projected case. At least, this is the
explanation I have in mind at the moment, I'll be pleased to
hear your thoughts on this one.

Anyways, I think it would be nice to have comparable results.
How do we get that? I'd say, by taking the 4326 bounds and
reproject them to 3857, and then adjusting the image form
factor to follow (e.g., keep the width, adjust the height,
or be even smarter and make it so the number of pixels
in the two images is roughly the same so that png compression
and memory usage are comparable as well).

I think I can do the above writing a GeoTools based script.
But if you want it lighter and written in python, I'm not
the right person to do it.
Opinions?

Cheers
Andrea

PS: I also noticed another thing. The first run is fully
disk bound (as expected), the second is sort of disk bound
the third run still reads from disk a bit but appears to edge
on the CPU bound instead (as a consequence the throughput
increases significantly between one run and then next)

PS2: once we've fixed the raster bounds to avoid the empty
area we should probably do the same reprojection script
against the raster data as well

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