Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Path: utzoo!utgpu!water!watmath!clyde!rutgers!husc6!cmcl2!beta!hc!ames!amdcad!rpw3
From: rpw3@amdcad.UUCP
Newsgroups: comp.sources.d
Subject: Re: Problems with Compress
Message-ID: <18513@amdcad.AMD.COM>
Date: Fri, 2-Oct-87 22:52:11 EDT
Article-I.D.: amdcad.18513
Posted: Fri Oct  2 22:52:11 1987
Date-Received: Sun, 4-Oct-87 01:43:31 EDT
References: <13006@bu-cs.BU.EDU> <780@sugar.UUCP> <1493@cognos.UUCP> <1488@geac.UUCP> <4786@ncoast.UUCP>
Reply-To: rpw3@amdcad.UUCP ()
Organization: [Consultant] San Mateo, CA
Lines: 47

My Fortune System has only about 700K usable user memory, and running 16-bit
"compress" (triggered by unbatching news) made it swap so bad that UUCP would
time out! So I did some comparisons of what the various "-b" settings did to
the compression ratio, versus the memory size of "compress" compiled for various
maximum supported "-b" values. I'm sorry I don't have the table at hand, but
the following table of pseudo-values is roughly what I remember for compress
version 3.6 (I have *not* tried compress 4.0 yet, although I've heard it's
considerably smaller -- folded some tables or something) when run on several
large batches of news (*not* with any uuencoded binaries in them):

	WARNING: From vague recollection!

	MaxBits	MemSize	Compressed file (normalized to -b16 == "1")

	-b 16	640K	1
	-b 15	360K	1.02
	-b 14	220K	1.05
	-b 13	150K	1.15
	-b 12	105K	1.25

Note that the memory size goes up as a power of 2 of the MaxBits, plus some
base size for the program and misc variables. Note also that (as best I recall)
the real "knee" in compression was between "-b 14" and "-b 13", as was the
knee of the memory size. Since my problem was at a different hard limit than
a 64k+64k machine, "-b15" worked just fine for me, and given the figures, none
of my news feeds minded cranking down the compression by 1 (even if it meant,
as it did in one case, cranking *all* their outgoing compressed batch down to
"-b 15").

Given compress 4.0, somebody might want to repeat my experiment again, and see
if there is a MaxBits value which is above the compression "knee", but within
the memory size limits for these small machines. (Remember to include mainly
text groups in your test.) For example, if it turns out that "-b14" will fit,
and the penalty is small, your neighbors (indeed, the net) may be willing to
switch to "-b14" as the "standard".

By the way, you big guys would win from this, as well. "-b16" makes your systems
thrash around a lot more than they really need to...


Rob Warnock
Systems Architecture Consultant

UUCP:	  {amdcad,fortune,sun,attmail}!redwood!rpw3
ATTmail:  !rpw3
DDD:	  (415)572-2607
USPS:	  627 26th Ave, San Mateo, CA  94403