madler@nntp-server.caltech.edu (Mark Adler) (04/03/91)
In article <1991Apr02.192952.8405@looking.on.ca> brad@looking.on.ca (Brad Templeton) writes: >No, I believe you. Well, no wonder it's easy to make a compressor that >does better than ZIP then. Still, most of the compression is in the sliding dictionary and not in the S-F trees. Also, the representations of the trees take up 25 bytes for binary files, 23 bytes for short text files, and 121 bytes for large text files, so leaving off the trees don't make much difference. As a result, PKZIP really doesn't do so bad, getting 2.98 bits/byte in Peter's tests compared to LHARC's 2.86 and LZA's 2.82. The design decision was to not quibble about a few percent compression here or there when you can pump 32K/s through the thing. Not a bad choice in my opinion. Of course, it would be nice to have an option for better compression at the cost of speed. Mark Adler madler@pooh.caltech.edu