From: Eric Young <eay@mincom.oz.au>
Date: Thu, 21 Mar 1996 23:29:16 +0800
To: cypherpunks@toad.com
Subject: RC2 speed
Message-ID: <Pine.SOL.3.91.960321104204.26664B-100000@orb>
MIME-Version: 1.0
Content-Type: text/plain



Just a quick RC2 update, since I have not seen anything about it's speed 
on this list.

I have implemented RC2 from the posting from 
Message-ID: <4fk39f$f70@net.auckland.ac.nz> in sci.crypt
(This is the 'text' description of the algorithm).

The following times are from C code using gcc 2.7.0 on a sparc 20 and
cc on 'some old slow alpha box'

	     sparc 20	Alpha
rc4           4521k/s	3835k/s
des cbc        993k/s	 833k/s
des ede3 cbc   370k/s	 476k/s
idea cbc       862k/s	 726k/s
rc2 cbc        975k/s	1083k/s

Documentation I had seen previously about RC2 made the claim it was about
3 times faster than DES is software.  From my times it appears to be of a
similar speed (depending on the box).  It as has been speculated that RC2
was origionally written for effiecent implementation on 16bit hardware,
this could well be true, but for modern 32bit processors, it appears to
have no real speed advantage over DES or IDEA.

I'm mostly interested in RC2 because it is in the SSLv2 and S-MIME
specifications :-).

Any comments?

eric

PS If RC2 was implemented in assember, it would obviously be faster
   due to direct use of machine rotate instructions.  I don't quite
   know how much this would speed things up but obviously the other
   algorithms would also benifit from hand coding.
--
Eric Young                  | Signature removed since it was generating
AARNet: eay@mincom.oz.au    | more followups than the message contents :-)