[Paraview] Bug 10274

[burlen]

I noticed this bug report up on Mantis.
http://www.paraview.org/Bug/view.php?id=10274
I feel kind of responsible for this since I suggested geometry 
compression should be investigated.

Just to clarify: The investigation I made 
(http://www.paraview.org/Bug/file_download.php?file_id=2866&type=bug) 
showed at most a 2x speed up, and only for very large data and slower 
networks. If you are going to implement geometry compression then there 
are a couple of things to keep in mind:

1) geometry compression hurts performance on fast networks
2) only makes much of a difference (and at most 2x speedup) on slower 
networks for very large data
3) for binary floating point data, zlib compression level affects run 
time drastically, but higher levels don't increase the compression ratio.

I noticed that in vtkMPIMoveData the compression ratio is set by the 
compile time variable Z_DEFAULT_COMPRESSION, which is typically 6, that 
will kill any hopes of a speedup. You should always use zlib compression 
level of 1 on the binary floating point data.

Also, if you're going to compress geometry or other binary floating 
point data during communication you might look at LZO
http://www.oberhumer.com/opensource/lzo/
I didn't do any benchmarks with LZO but they claim it's faster than zlib.