Re: RfC: Taxonomy of mix batching approaches (draft)

[Roger Dingledine <arma@mit.edu>]

----- Forwarded message from owner-mixminion-dev@freehaven.net -----

To: Len Sassaman <rabbi@abditum.com>
Cc: mixminion-dev@freehaven.net, freehaven-dev@freehaven.net,
	syverson@itd.nrl.navy.mil
Subject: Re: RfC: Taxonomy of mix batching approaches (draft) 
Date: Wed, 04 Sep 2002 10:46:02 +0100
From: Andrei Serjantov <Andrei.Serjantov@cl.cam.ac.uk>

> On Mon, 2 Sep 2002, Roger Dingledine wrote:
> 
> > http://freehaven.net/doc/batching-taxonomy/taxonomy.pdf
> > (or http://freehaven.net/doc/batching-taxonomy/taxonomy.ps)
> 
> About Mixmaster dummy messages:
> 
> Dummy messages are not fixed at 5 hops. They use the value of the CHAIN
> setting for picking their chains (which is by default *,*,*,*), or four
> random remailers. The user/remop can change this.
> 
> Also, Mixmaster currently does not generate dummy messages when sending
> from the pool. Dummy message generation is entirely up to the user (and
> many remops have a cron setting for dummy message generation.)

Do you mean that dummies get generated at times specified by the cron job, 
placed in the pool and then flushed together with the next batch of real 
messages?
 
> We're thinking of changing that and having dummy messages generated each
> time a message is placed in the pool.

Don't get this either -- do you mean that for every message arrival you put a 
dummy into the batch? So, each batch is always 1/2 full of dummies?
 
> The attack described in the paper, where the anonymity set is limited to
> 2, is not as easy if we generate multiple dummy messages upon message
> arrival. Then, if two good messages exit the pool, the attacker won't know
> for certain that one of them is good. You could have a pool_send where
> multiple dummy messges, but no real message, are sent.

Indeed. It turns out that generating between, say, 0 and 10 (chosen uniformly 
at random) dummies per batch works even better against the n-1 attack. I am 
writing a paper with Claudia Diaz about this at the moment. It'll appear on 
this list in due course.

A.

----- End forwarded message -----