mix questions

Roger Dingledine arma at mit.edu
Sat Jan 3 08:05:01 UTC 2004


On Wed, Dec 31, 2003 at 12:52:24AM +0100, Some Guy wrote:
> > You should recognize that Tor does _no mixing_. If you want
> 
> Ok sorry, bad term.  I mean circuit based annoymity, probably without
>too much reordering/batching
> of messages.  It depends on what kind of app is using the DHT.
>For file sharing you can do this
> because downloads take a while anyway.  Is there a better term for
>"circuit based annoymity"?  I
> used to call it onion routing.

I use the term 'circuit based anonymity' to mean any of the low-latency
designs like Tor (aka second-generation onion routing), Jap/Web mixes,
Cebolla, MorphMix, Freedom, Crowds, etc.

Remember that "messages" in a circuit-based system are sent as "cells",
and are generally 1KB or less. So one file might use thousands of cells.
If you want to collect 1MB messages and batch/reorder them at each hop,
then you're talking about a different design.

> Ok wait what attacks are possible, because of not checking consistancy
>on every hop?  It seems
> like they can all be solved.

The big attack Mixminion aims to address is correlating one hop on the
path to another hop farther down the path. Circuit-based anonymity,
because of its low latency, is vulnerable to this attack and for
the most part accepts it (but see the cascade-based web mix design,
http://freehaven.net/anonbib/#web-mix:pet2000).

Mixminion aim to block this attack. See Section 4.2 of
http://mixminion.net/minion-design.pdf for more details.

>  So the
> cell stays the same size going down the circuit?

Correct. Though really, based on our threat model we don't need this to be
the case (neglecting efficiency issues). Many of the design decisions in
Tor are made with the intent of making it easy to move to a mid-latency
design where we can resist some end-to-end attacks while still providing
acceptable performance for some activities.

(It's likely that we don't want to do this, because "usability is
a security parameter" -- we'll shed most users and thus be *less*
anonymous. But that's a different story too.)

--Roger



More information about the tor-dev mailing list