Re: [RePol] Re: Remailer Logs (and Accountability)

17 Dec 2003

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

On 3 Jan 1998, Charlie Comsec wrote:
...
...
The "reordering pool" is
always a minimum of 5 messages so people can opt for how long their
message wil be 'stashed' at replay. (use 'Latency: +00:00' for zero latecy).
Is that default reordering pool size the same for Type I and Type II
messages?
Type-II messages don't support latency yet (or not until fairly recently,
I can't remember). Type-I remailers don't, by default, do any reordering,
but reordering is not as useful for type-I messages (unless you remix). 
Cracker uses Mixmaster to reorder type-I messages. In addition to the pool
size of 5, it also will remail a maximum of half the messages present, so
in reality, the pool size floats up and down (but not lower than 5).
...
Perhaps someone can double-check my math on this.  Assuming
equally-sized messages that are otherwise indistinguishable, and a
reordering pool size of five, then the odds of matching up an
encrypted incoming message with an encrypted outgoing message are
one in five.
The odds are somewhat worse (for the attacker). In the case of the
scenario above on cracker, the odds of any given message being cycled out
of the pool (the pool is processed at six minute intervals) is generally
50% per cycle. Thus, there is a 75% chance that it has been sent by the
end of the second cycle, and therefore a 25% (.5*.5) chance that it is
still in the queue. The current RMS latency (from the remailer's own
stats) is 19.5 minutes. You want to about double that to be sure that the
message has really come out (and then, you still can't be sure), so that's
about six cycles. If we are doing 5 per cycle, then the odds are 1 in 30. 
A very rough estimate.  However, by my estimates, it's more like 12
messages per cycle (typically; it's variable for the reasons above), so
that runs it up to 1 in 72 or so.

(And of course, the remixing ensures that all messages are
indistinquishable, whenever possible.)
...
What I'm unclear on is how setting a Latency:  flag affects the
diffusion of the output.  Is that lattency IN ADDITION to the pool
size of five, or does Latency:  +00:00 bypass the reordering process
altogether?
Latency occurs before reordering. Latency: +00:00 does nothing, BTW, and
it's the default. Latency: +12:00r adds a 0-12 hour random delay before
reordering.
...
My main concern is the security of chained reply blocks which are
more vulnerable to attack than normal anonymous messages.  A single
anonymous message can only be traced BACKWARDS after it's been
received.
Which is probably not so hard when you record all inter-remailer traffic,
which is probably what happens somewhere.
...
An anonymous reply, OTOH, could, theoretically, be
"walked" through each remailer in the chain until the identity of
the recipient was discovered.  While that process would require
convincing each remailer operator in the chain to cooperate, it's a
lot more feasible than tracing a message backwards to its source.
(Yes, I know about posting to a public message pool, such as a.a.m,
but NG posting seems to be rather unreliable lately.)
Yep, that was part of the motivation for remixing: To keep eavesdroppers
from intercepting partially-decrypted reply blocks. It also prevents
traffic analysis based on the message section after the reply block. Reply
blocks, of course, tend to get smaller after each hop, while the message
getting delivered tends to get larger. Automatically encapsuling type-I
messages in type-II format whenever the recipient supports it prevents
this type of traffic analysis.

Andy Dustman / Computational Center for Molecular Structure and Design
For a great anti-spam procmail recipe, send me mail with subject "spam".
Append "+spamsucks" to my username to ensure delivery.  KeyID=0xC72F3F1D
Encryption is too important to leave to the government. -- Bruce Schneier
http://www.athens.net/~dustman mailto:andy@neptune.chem.uga.edu   <}+++<