From: stewarts@ix.netcom.com
Date: Sat, 31 Aug 1996 05:00:24 +0800
To: cypherpunks@toad.com
Subject: Re: Encryption
Message-ID: <199608301753.KAA21913@toad.com>
MIME-Version: 1.0
Content-Type: text/plain


At 05:40 PM 8/30/96 +0200, Germano Caronni <caronni@tik.ee.ethz.ch> wrote:
>Dale Thorn wrote:
>> Algorithm: Select bit-groups of random length from the file until the file is
>>            completely processed.  Shuffle the bits in each group randomly and
>>            save each group back to the file. Repeat if needed using different
>>            key-strings for each successive encryption, for increased
security.
>Very nice. There are just two little issues:
>a) How do you generate the random bytes?
>b) How do you transmit them to the other side, without having a secure channel?

I agree.   This method can be looked at as two parts:
1) Generate a stream of random numbers
2) Use the stream to transform the plaintext into cyphertext by shuffling.

Generating a stream of random numbers can be done by cryptographically
strong methods, or by physical methods such as throwing dice or counting 
gamma rays, or by cryptographically weak methods such as using the RANDOM
function provided by your BASIC compiler.  If you're using strong crypto,
the method is strong if the transform is strong.   If you're using physical
randomness, that's also true, and you'll definitely need to use 
agents with briefcases handcuffed to their arms to haul the randoms around,
which doesn't gain you much operational security, since the random stream
needs to be longer than the plaintext.  If you're using weak crypto,
we can use those fast computers you're so impressed with to try
all 32767 or 2**32 random streams on your input to see what works -
brute force isn't that hard when you've got a small keyspace,
even if you don't take advantage of the special forms (e.g. there are
good methods for reversing linear congruential multiplicative generators.)

How long a stream of random numbers do you need to do the transform?
And how secure _is_ the transform?  For a simple XOR, you know that
each bit in the stream will diddle one bit in the plaintext,
so you're fine (as long as you only use the random stream once.)
But for your shuffling method, you'd have to shuffle much longer
to make sure that the "random" selection of bitgroups hits all the bits,
and hits them often enough that adjacent bits will be separated.
For instance, if your message is a 3000-word set of instructions
to your fellow plotters, you need to be sure that the phrases
"gunpowder", "Parliament", and "November 5th" get shredded, not merely
moved to different places in the document.  This takes much longer
than deterministic methods, and risks not being as good.

There are some advantages with the shuffling method - assuming you've 
cranked the system long enough to really shred everything,
it's hard to take cyphertext and reconstruct plaintext, and 
brute force is a bit more work than with a deterministic method
that uses one pass.  But you can get that effect by using
stronger methods, such as the
        des | tran | des | tran | des
where "tran" is a simple key-or-input-driven shuffle that
doesn't need to be totally strong.  (Is Carl Ellison the person
who proposed this?  I've forgotten.)  


#			Thanks;  Bill
# Bill Stewart, +1-415-442-2215 stewarts@ix.netcom.com
# <A HREF="http://idiom.com/~wcs"> 	Reassign Authority!