Re: Mixminion to become type 3 remailer protocol?

[Nick Mathewson <nickm@alum.mit.edu>]

On Sun, 2002-04-21 at 15:07, Roger Dingledine wrote:
 [...]
>      He wants to have the mix nodes support both type 2 and type 3 for a
> while, at least until we can phase over to 3. This may make implementation
> significantly yuckier; perhaps we can take something that supports type
> 2 and just graft it on somewhere until he's willing to drop it.

Probably the best solution is to have them run as completely separate
services, implemented by separate servers, on separate ports.  Node
operators could run both servers, or just the t3 server, as they saw
fit.

[I suggest this because of the desirability of writing a standalone t3
server, and the undesirability of making t2 an official 'transition'
feature of t3.]

To ease transition, backward-compatibility clients for the t3 protocol
should exist.

 [...]
>                                            then we can get
> comments on it and start figuring out how we want to go about building
> a spec and eventual implementation.

My personal take on what we need from the spec before we can implement
was in the release criteria I sent out earlier today. 

>                                      Len wants the protocol spec to
> be standalone so he can build one in C. More power to him. :)

Yeah, that one's his puppy. :)  Multiple independent implementations is
cool, though, so it's good that _somebody_ wants to do it in C.

[Personally, I'm betting that all the CPU will go to PKI anyway, so that
you won't see much of a speed improvement so long as everybody does RSA
in C.  But I'd be fascinated to find out I'm wrong.]

-- 
Nick