[gdal-dev] Reprojecting a tiled dataset

[Mike Leahy]

On Tuesday, November 23, 2010 23:54:34 Frank Warmerdam wrote:
> Mike Leahy wrote:
> > On another note...if the output image is internally tiled, is it still
> > worthwhile to run gdaladdo on it?  Is it generally about the same
> > efficiency to use the single internally-tiled file if I'm going to use
> > it as a layer in MapServer, or should I re-tile it again into separate
> > images?
> 
> Mike,
> 
> Absolutely you should build overviews for it.  I don't see any value in
> turning it into a bunch of files - I much prefer serving from one, or only
> a few large GeoTIFF files instead of all this tiles-as-files business.
> 
> Best regards,

To carry on the externally-tiled vs. internally-tiled discussion - another 
question I have is what should I do about the large areas of empty space 
around the outside of a non-rectangular dataset?  In the dataset I'm working 
with at the moment, about 30% of the one-file image area is black/nodata.  This 
can be excluded if I tile the dataset an remove any tiles with all zero 
values.  Is there an equivalent way to exclude this within an internally-tiled 
image in order to conserve disk space?  Would it have made a difference if I 
had set the -srcnodata flag when I used gdalwarp?

I must say, however, the performance of working with the internally tiled 
image noticably better.  I gind I get significant loading/lag time when I view 
the data using the *.vrt for the original tiled data within QuantumGIS, while 
it's virtually instantaneous with with the single internally-tiled image.  So 
if the cost for this is the extra disk space for the nodata regions, it's 
worth it anyway.

Best regards,
Mike

P.S.: I forgot to copy the list on my previous reply - sorry about that.