From: Jim Gillogly <jim@rand.org>
Date: Fri, 16 Sep 94 12:24:04 PDT
To: perry@imsi.com
Subject: Re: thoughts on RC4
In-Reply-To: <9409151452.AA03618@webster.imsi.com>
Message-ID: <9409161923.AA06121@mycroft.rand.org>
MIME-Version: 1.0
Content-Type: text/plain



> perry@imsi.com (Perry E. Metzger) writes:
>         Can anyone see any reason why one could not change RC4 to
> being a word oriented stream cipher, call it "ERC4"?

> The reason I ask is because this would speed things up by a factor of
> four on 32 bit machines, which would mean modest hardware could
> possibly break 100mbps speeds. The 64 bit extension on 64 bit RISC
> processors could go far, far, faster still.

Is mbps megabits per second?  If so, I'm within a factor of 3 of confirming
your numbers.  If it's megabytes, I'm more than an order of magnitude away
from understanding what "modest hardware" means.

The original code plods along on my 50 Mhz '486 laptop (Borland C++ Pro)
at a paltry 1.43mbits/s.  Turning the inner loop into obfuscated C picks up
a little to 3.84mbits/s, and doing it with 8086-compatible assembler
yields only 8.40mbits/s.  The compiler could certainly be a lot smarter,
but the assembler probably couldn't be improved by a factor of 2 without
modifying the algorithm as you suggested -- the current incarnation is at
15 instructions per encrypted byte.

Anybody else have timing numbers?

	Jim Gillogly
	25 Halimath S.R. 1994, 19:18