[gdal-dev] LZW Compression on geotiffs

[Jukka Rahkonen]

Andrea Aime <andrea.aime <at> geo-solutions.it> writes:


> Interesting... wondering how much the predictor affects decoding speed
though (it is
> a common case to have to compress once, and read many times out of the
> compressed file)

A very quick test by decompressing deflate -> uncompressed from disk to disk
shows no difference in speed. Most time is used for disk read/write with
this test and actually with predictor the speed is perhaps a bit better
because there are less bytes to read. Would need more repeats to get
reliable results, though.

I now remember that I have mean measuring the WMS output from Mapserver with
different formatotions for png compression
http://www.mapserver.org/mapfile/outputformat.html. Conclusion was that it
was good to put some more load for the processor and compress more than the
default and make less bytes for the output. But png compression is perhaps
something different than the use of predictor and I guess that QUANTIZE
probably shrinks file size more effectively. But for a top quality 24-bit
png output it might be good to make WMS server to compress with predictor=2.
I wonder if Mapservers and Geoservers do it already. I can't see such
formatoption in Mapserver documents.

-Jukka Rahkonen-