Path: utzoo!utgpu!jarvis.csri.toronto.edu!cs.utexas.edu!uunet!mcsun!unido!mikros!mwtech!martin
From: martin@mwtech.UUCP (Martin Weitzel)
Newsgroups: comp.dcom.modems
Subject: Re: poor uucp performance - help! (LONG)
Message-ID: <670@mwtech.UUCP>
Date: 24 Feb 90 02:01:56 GMT
References: <125@cdss.UUCP> <1990Feb17.063412.18455@rancho.uucp> <959@codonics.COM> <1318@ispi.UUCP>
Reply-To: martin@mwtech.UUCP (Martin Weitzel)
Organization: MIKROS Systemware, Darmstadt/W-Germany
Lines: 41

In article <1318@ispi.UUCP> jbayer@ispi.UUCP (Jonathan Bayer) writes:
}bret@codonics.COM (Bret Orsburn) writes:
}
}>In article <1990Feb17.063412.18455@rancho.uucp> rock@rancho.uucp (Rock Kent) writes:
}>>
}>>4.  Make sure that you have compression turned off.  You don't want to
}>>    be compressing already compressed news batches.
}>>
}
}>OK, I'll bite: Why not?
}
}>Surely, compression will be less effective the second time, but is there
}>any good reason to disable it? Do you have any data to demonstrate that
}>throughput is *decreased* by enabling compression for a compressed newsfeed?
}
}
}Yes it is.  Remember, compression takes time.  The idea behind
}compression on the fly is to save characters at the cost of time,
}hopefully there will be a net gain.  However, compressed files usually
}are not compressable.  Therefore if the modem is trying to compress the
}data and it can't it will slow down because it is trying to do a
}compress, and may actually send MORE characters due to a
}non-compression.

Quite easy to explain: Suppose, you had an algorithm which allways
compresses some input-data to *less* output-data. Why not feed the
output again and again, until its size is reduced to zero. Such an
algorithm cannot exist for arbitrary input-data.

Or view it vice versa: If *uncompressing* yields N data-bits in total,
there are 2^N possible bit patterns. No two of these can have the
same representation when compressed, because uncompressing could
not decide then between the two. So, if some bit patterns need less
bits when compressed, there must be other bit patterns that need more.

A little more mathematics shows, that algorithms, which tend to give
high compression rates with certain kind of data, are much prone to
fail on arbitrary data - esp. on the output of their own or other
compression algorithms!
-- 
Martin Weitzel, email: martin@mwtech.UUCP, voice: 49-(0)6151-6 56 83