[freehaven-dev] Re: comments on proposed mixmaster v3 protocol

[William Rowden <rowdenw@eskimo.com>]

Thanks for your comments, Adam.  Your ideas intrigue me; one in
particular catches my attention at the moment.

Yesterday, Adam Back wrote:
> http://www.eskimo.com/~rowdenw/crypt/Mix/draft-moeller-v3-01.txt
> 
> Some comments on the proposed mixmaster 3 protocol (see above).
[snip]
> I suspect that's a little long, and it might be better to make
> the thing more efficient by having a 18k payload and 2k header
> and have a max of 30 or so hops (depending on remailer address
> lengths).

Regarding chain lengths, we could learn from the practical
experience with the current anonymous remailer network:  Statistics
are available in alt.privacy.anon-server for chains of length 2, and
in some cases of length 3.  This is partly to avoid placing a large
load on the network for gathering statistics.  Posts in a.p.a-s
don't suggest that there is regular use of longer chains, however.

I haven't seen a sample size large enough to draw a statistically
valid conclusion, but anecdotal evidence and personal experience
suggest that reliability decreases dramatically for longer chains.  
Does anybody have some contrary evidence?

Consequently, until the network is more reliable, even the current
20 hop maximum is more than sufficient.  Like you suggest, reducing
the header and increasing the payload would be better than
increasing potential chain lengths, if the headers can be more
efficient.

This reasoning begs the question as to *why* longer chains are so
much less reliable.  Depending on the threat model, hypotheses range
from the analysis by RProcess on one hand to the statement that "the
network is run by volunteers, after all" on the other.  I'd like to
see the version 3 protocol incorporate strategies for increasing
reliability so that we could test some of these hypotheses.
-- 
    -William
PGP key: http://www.eskimo.com/~rowdenw/pgp/rowdenw.asc until 2000-08-01
Fingerprint: FB4B E2CD 25AF 95E5 ADBB  DA28 379D 47DB 599E 0B1AA
When they find out who *you* are they'll pad the cell.