Re: LPF Statement on the GIF controversy (LZW patent)

[Phil Karn <karn@qualcomm.com>]

>Compression is a link layer issue. Over a fast hop compression is often
>counter productive in the CPU time used. Thus the compression ought to be
>part of the AX.25 layer.

Valid arguments can be made for placing compression at any of several
layers.

End-to-end compression generally yields the best performance since it
doesn't have to deal with the sudden changes in statistics that appear
in multiplexed data streams. A single end-to-end compression operation
can save bandwidth over an entire network path, not just a single
link. And it often gets to run on faster host CPUs with lots of memory
for table space rather than on slow modem micros with limited memory,
making better compression possible.

Note also that low-level compression breaks completely when a higher
layer encrypts. This is a significant issue for some non-amateur
networks. If you're going to compress, you must do it before you
encrypt.

On the other hand, end-to-end compression is often impractical in
existing systems, especially when you only control one end. That's why
it's so popular in dialup modems.

Phil