From: hfinney@shell.portal.com
Date: Thu, 4 Nov 93 22:37:51 PST
To: cypherpunks@toad.com
Subject: message depots, packet routing?
Message-ID: <9311050634.AA00949@jobe.shell.portal.com>
MIME-Version: 1.0
Content-Type: text/plain


-----BEGIN PGP SIGNED MESSAGE-----

A few discussions of DC-Nets here show a small misconception.  The
network topology does not have to be the same as the DC-Net topology.

What I mean is this: the communication in the network may be a ring
or any other topology.  But that does not mean that you only share
random numbers with your neighbors on the ring.  You can share random
numbers (on a pairwise basis) with every other participant in the network,
in the most extreme case.

Look at this diagram, showing four people communicating in a DC-Net.
The lines represent shared random number one-time pads.

		A---------B
		| \     / |
		|   \ /   |
		|   / \   |
		| /     \ |
		C---------D

Each person shares random numbers individually with every other person
in the network, in this example.  A and B share their own random numbers,
A and C share a different set, A and D have their own, B and C do,
and so on.  But that does not mean that the network communication
topology has to be all to all.  Instead, a ring topology could be used,
with packets passing around the network A-B-D-C-A-B-D-C-....  At each
step, A would xor in the next random number from each of the three pads
that he is using, then xor in his message bits if he has anything.  Then
he would pass it on to B.  After the packet has gone all the way around,
the message (if any) would be revealed.

And in this case it doesn't matter who your "neighbors" are in the
communication network.  B and C colluding can't distinguish whether messages
come from A and D despite the fact that they separate them in the comm
network.

So this concern about "knowing your neighbors" in the DC-Net is not as
serious as it sounds.  If truly paranoid people want to participate in
a DC-Net (and who else would?) then they can use a DC-Net topology
which does not allow partitioning.  This adds overhead and inconvenience
of distributing shared random numbers, but it does not require the
communication pattern to change.

BTW, I like the name someone proposed for a DC-Net: "Ouija Net".  The idea
is that messages appear in a DC-Net somewhat like messages appearing on
a Ouija board.  The true source of most Ouija board messages, IMO, is people
pushing the indicator around.  But because everyone is touching it, each
person has plausible denial.  You know that SOMEONE is moving it, but
there is no way to tell who.  This is similar to DC-Net messaging.

Hal
hfinney@shell.portal.com

-----BEGIN PGP SIGNATURE-----
Version: 2.3a

iQCVAgUBLNnG0qgTA69YIUw3AQFigQP+LDMO6S7HkS4YkLdctLus4GamIvb/BxSX
uG8VcZ/0eujQt8ZEIlNEzwNZvBR3Sio8gKko2jjvmWDGyobibSpctfqcr5Qf42xz
42TWVzYCjg+tka6FttosZ0phwGP1m7Dy+sC/zE0YmEQagS6jDSn/RiqR2PDMSJZn
FEo2lfh8tmQ=
=ZUfW
-----END PGP SIGNATURE-----