[Benchmarking] ECW data set, a proposal

[Andrea Aime]

Frank Warmerdam ha scritto:
> Andrea,
> 
> I think this would be a nice sort of medium sized image to work with.
> I think we *might* still want a big image.  /me checks ... If I read the
> "df" output right there is 380GB available on the disk so I guess I had
> better not go too crazy about big images.

Wow, that's big indeed.

>> I guess it could be interesting for the ECW benchmarking. As for turning
>> it into a mosaic of TIFFs for the mosaic part heh, that will be a little
>> painful, especially since the tiles need to be less than 2GB (last
>> time I tried I think I had to limit myself to North America to be able
>> and extract a file of less than 2GB, uncompressed).
> 
> There should be no problem translating the image to bigtiff.

Yep, but GeoServer does not support bigtiff yet, so we would
not be able to make a apples to apples comparison

>> If people think it's good I can upload it to the server.
> 
> Please do.  I can work on converting it to some other formats
> once it is there.

Ok

>> Oh, btw, what's the MapServer status vs jpeg compression? Shall we
>> use JPEG compressed or just go for the uncompressed TIFFs?
> 
> I am hoping to deploy this image in a few raster formats to
> investigate the performance effects in MapServer.  One of those
> will likely be tiled, jpeg compressed TIFF which is popular
> with at least one prominent MapServer user.

Yeah, tiled, jpeg compressed TIFF with overviews is quite common
in GS land too.

> I have even vaguely contemplated trying to deploy an image using
> the new postgres WKTRaster extension.  I have a GSoC student working
> on this in GDAL and I have yet to even try his code!   But that might
> be a bit much to bite off for now.

GeoServer wise we do have a generic in-database pyramid which first
uses gdal_retile to generate a on disk pyramid and then slurps 
everything up into a db (as blobs).
However, since the two servers would end up doing quite different
things we'd end up comparing apples to oranges.
Afaik Paul was not so keen on the idea, thought there is the
argument that comparing mosaic on the fs vs stuffing images in
a database is a hot topic today, so it would raise interest on
the presentation

Cheers
Andrea


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