On Thu, Mar 14, 2002 at 12:17:02AM +0000, jft 628 wrote:
<snip>
> Above are the design goals of layer 2.  I am not an expert in the area, so 
> I am hesitant to share the rough idea I have of how
> they could be achieved.  But from what I read of anonymous remailers, it 
> seems like you could use the same basic notions
> here.  That is, a node on the network communicates via anonymous 
> remailer-type things.  Certain configuration information
> would have to be set for a networked node, such as what anonymous remailers 
> are trustworthy for what types of data, and
> reliableness.  So then the two participants in a message exchange each have 
> a "line of defense": their list of anonymous
> remailer-type things through which the message passes.  (The configuration 
> information itself could be shared between users at
> layer 4.)
<snip>

be careful. anonymous re-mailers do not deal w/ real-time communication. you
can describe both in terms of packet-based communication, but the devil is
in the details. a good remailer, a mixer, queues some amount of messages,
then chooses message out of the queue randomly to send so that you cannot
correlate a message going in w/ the one coming out of the remailer. the best
an attacker could do is say, the message i was tracking that came in could
be one of 5 leaving. if you follow those five to the next hop, you have the
same dilemma, you track more and more, and your attack become intractable,
if not impossible.

unless you erase the correlation, the _quality_ of anonymity is spurious at
best. the problem in real-time is how long do you wait to queue enough
packets and still claim real-time. real-time SMTP is different than
real-time HTTP, which is different than real-time telnet, from than
real-time voip, etc. the only concrete solution i know of is for each node
to continually send a steady of strem of packets between nodes, and to
inject real packets into the stream, aka padding.

but this is resource intensive, and most applications try to ignore the
issue. if the public TCP/IP network topology were 100% even, you might get
away w/ it. but as it is, its probably much easier to leverage these
correlation attacks than most people give credit, because all kinds of
traffic converge at specific points (MAE-EAST, etc) which becomes a birds
eye view of a huge swath of the internet.

but, your idea of seperating out the anonymity layer from the rest is a good
start. and don't forget to differentiate asymmetric anonymity (i.e., i know
you, you don't know me) and symmetric (where we are both blind to each
other). those should also be seperated, in terms of solution building.

_______________________________________________
freenet-tech mailing list
[EMAIL PROTECTED]
http://lists.freenetproject.org/mailman/listinfo/tech

Reply via email to