Re: [hybi] deflate-stream and masking

["Bob Gezelter" <[e-mail]>]

As I have noted before, Greg's observation about the interaction of
masking and compression is undoubtedly correct.

To be effective, compression must be done when the steam of data is most
strongly correlated. In short, this should be BEFORE multiplexing,
encryption, and masking. Data on different multiplexed streams cannot be
assumed to be correlated, and correlation is needed for effective
compression. 

Encrypting or masking data to create randomness similarly negates the
effectiveness of compression. 

When dealing with correlated data (e.g., the contents of a printable
file, or even a PostScript file), I have seen compression factors of
100:1. If the data had first been encrypted, or randomly masked, the
ratio would have been likely 2:1 (what is often achieved by disk or tape
compression facilities), or even less than 1:1 (worst case).

Compression should be done within each sub-stream (assuming
multiplexing, which is not in the current specification), and in any
event, before encryption and masking.

- Bob Gezelter, http://www.rlgsc.com