[Benchmarking] Testing setup

[Adrian Custer]

On Mon, 2010-08-09 at 11:41 +0200, Andrea Aime wrote:
> Adrian Custer ha scritto:
> > 
> > Projections:
> >         (vector only?) EPSG:22662, Platte-Carré,    (no reproj)
> >         EPSG:25831, UTM       (original, no reproj for raster)
> >         EPSG:3857,  google    (reproj, no datum change)
> >         EPSG:23031, UTM/ED50  (reproj, with datum change)
> >         Lambert Conformal Conic, 60N,30N; 20East    (not mercator)
> 
> Adrian,
> thanks for the summary. It all makes sense to me, I just have a couple
> of concerns about the projections.
> 
> One is that they might too many and we should just pick 2, or at
> most 3, to reduce the overall number of tests we're running.

Let us define first what we would like to do overall, then we can cut
back for the sake of practicality when we have the full picture. 

Clearly we can't do combinatorial tests of every parameter but if we
have two series testing projections (or just one if we combine raster
and vector) we might as well make them cover a minimum of the complexity
of that world.

[...snip...]

> 
> The other is about the usage of EPSG:22662 as the "non reprojecting"
> one. For the non reprojecting case I'd go for EPSG:4326 directly.
> While I know it's not classically meant to be used for representation,
> it has become common practice (e.g., NASA WMS-es do publish data almost
> exclusively in EPSG:4326) 

Given that this EPSG code as used in WMS 1.1.1 returns a false answer,
is the major error in that version of the standard, and has been flipped
in WMS 1.3.0, it seems like a really bad idea to use this version of
unprojected data as a lazy way of saying "Platte-carre". 

If any participant cannot produce EPSG:22662 then perhaps we will have
to live on with this broken beast but I would like to avoid it if
possible.


> and it guarantees the server is not
> exercising the referencing subsystem at all, making it easier
> to show what the real effect of the referencing subsystem is.

No, I don't think the use of such a referencing code guarantees any
internal pathway on the servers: that is a detail of each implementation
about which nothing can be presumed externally.

--adrian

> 
> Cheers
> Andrea
> 
>