Re: 3-Way Described on web site

From: jsavard@ecn.ab.ca
Date: 03/10/03 

From: jsavard@ecn.ab.ca ()
Date: Mon, 10 Mar 2003 06:13:39 GMT

Paul Rubin (http://phr.cx@NOSPAM.invalid) wrote:

: I remember thinking 3-way was interesting because it can be
: implemented using very little memory. However, there are some other
: ciphers with similar low requirements and more traditional design, so
: it's not that big a deal.

Why I found 3-Way to be worthy of inclusion - in fact, long overdue for
inclusion - on my page is:

- its historical status as an ancestor of both SERPENT and Rijndael, and

- its ingenious approach to the issue of using the same code, or nearly
the same code, for both encryption and decryption

- its uniqueness as a cipher with a 96-bit block size

and so it wasn't necessary for me to view it as highly secure.

A slight modification of the round, from:

key addition
theta
pi-1
gamma
pi-2

to, say

key addition
S-box
key addition
pi-1
gamma
theta
gamma
pi-2

where an S-box with 8-bit inputs and outputs, designed so that it becomes
its own inverse when the bit order of inputs and outputs is reversed is
added would, of course, make the cipher slower and increase its memory
requirements, but it would eliminate the problem with the Courtois and
Pieprzyk attack.

John Savard

Relevant Pages

  • Re: Security for embedded device
    ... >will need at least 8 bytes of memory, ... I once did a very insecure "cipher" by simply filling up a block of ROM ... Do you have an "impossible" engineering project that only someone like Doc Brown can solve? ...
    (sci.crypt)
  • Re: sci.crypt sandbox?
    ... Perhaps the scoring system should be changed: ... less memory it uses the higher the score. ... Also, when a cipher uses ... It shouldn't be that hard to include the speed into the rating system, ...
    (sci.crypt)
  • Re: Security for embedded device
    ... >Does anyone know a simple cipher for a 8-bit processor with only a few bytes ... AES? ... Your performance goals sound quite aggressive. ... will need at least 8 bytes of memory, ...
    (sci.crypt)
Quantcast