Remailers vs Nyms - conflicting assumptions?

[E. ALLEN SMITH]

From:	IN%"raph at cs.berkeley.edu"  "Raph Levien" 19-MAY-1996 05:46:40.03

>Bruce Baugh wrote:
 
>> What would it take to create a nym server that could route around the death
>> or disability of any given mailer?

>Well, that would be a serious problem. The big question is: who decides
>the routing? With the existing nym setup, the client decides the entire
>route. The nymserver knows only the first hop. For the nymserver to be
>able to route around damage, it would have to know that there is damage,
>and that implies knowing the route.

	Could a reply block be restructured so as to have a series of
binary/trinary/whatever decisions embedded in it, in encrypted form? In other
words, the remailer doing the remailing knows (transiently, hopefully) to
whom it's sending. It would have a choice of two remailers to use for this,
and would wipe the other reply block (replacing it with random gibberish). If
two were down, it could do a random choice; if one was down, it would send it
via that one; if both were down, you get the same situation as currently.
	Now, one problem that I can see is increasing the length of the reply
block-perhaps greatly if it was done enough. Given the need for fixed lengths
(e.g., Mixmaster) to really get around traffic analysis, this could be a
problem.
	There's also the question (to be left up to the user) of whether you
should try any remailers again in the chain if the choice had earlier said no.
If the answer was yes, then you'd have a greater chance of it hitting a
remailer that didn't realize another remailer was down. If the answer was no,
traffic analysis would be helped by knowing - if a remailer was subverted -
that it was less likely that a particular remailer was going to be used.
Indeed, with subversion you could choose the easier-to-trace remailer to send
messages through when you had a choice.
	-Allen