Lists Home |
Date Index |
On Nov 20, 2003, at 2:36 PM, Jeff Lowery wrote:
> Real-time stream compression algorithms certainly exist, although I
> understand their compression rates are around 50% or less. I suppose
> applications would require better than that, though.
One thing that sticks with me from the binary infoset serialization
workshop is what really matters to users is latency; compression is a
way people TRY to improve latency, but it seldom works well except for
the hardware-based streaming compression built into modems, etc. In
experiments of which I'm aware, conventional (gzip I believe)
compression only produced a net decrease in latency in situations where
there was lots of spare processor power on the compressing side and the
network itself was fairly slow. If the network is fast or the
processor slow, conventional compression showed no improvement in
overall latency, and could easily slow things down.
The holy grail would be a compression/decompression scheme that added
very little to the processor load (which directly relates to battery
life in wireless apps) but significantly reduced the number of bits to
be sent over the wire. I guess the holy spear :-) would be a format
that is fast to compress, decompress, and parse back to an Infoset.
I don't know this subject very deeply, but I suspect that the Holy
Spear is unobtainable, and there will be a tradeoff between compression
efficiency and the amount of work it takes to reconstruct an Infoset.