[ current-users removed from the CC; this discussion _really_ doesn't
  belong there... ]

> [ compressed VM foo. ]
>
> When I was thinking about this I came to the conclusion that the (possible)
> Huffman tree would be static or there would be a few versions of it and
> as an old one was not used it was reused as a new one.

Every time i've looked at "in kernel" compression for various things,
the algorithms which have always appealed to me most are the LZRW
family.  They provide high speed, reasonably good compression ratios,
and require limited memory.  "all of these things are good."  8-)


As i've mentioned in private mail, there exist various papers on this
sort of thing, and i'd encourage people to look at them.  As i recall,
the following is one of them:

%A Fred Douglis
%T The Compression Cache: Using On-line Compression to Extend Physical Memory
%P 519-529
%I USENIX
%B USENIX Technical Conference Proceedings
%D Winter 1993
%C San Diego, CA
%W Matsushita Information Technology Laboratory

and (if i remember) it did a fair amount of characterization of
performance curves, etc.  I don't have any more citations, but it can
probably point to some more...


cgd