Re: Feedback for mixminion specs

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

[Hi, all.  I'm working through my Minion backlog.]

On Fri, 2003-02-21 at 14:14, George Danezis wrote:
> Dear All,
> 
> I have been using miximinion for the last few weeks and supervising an 
> implementation in C. I thought that I should share with you my thought 
> about some parts of the spec. 
> 
> 1) SURB length
> 

Personally, I'd say, "don't worry."  Yes, it's a lot of information to
cut and paste, but the length issue is hardly the most inconvenient
aspect of SURBs.  IMO, SURBs should be managed by nymserver software,
and the user shouldn't need to spend time cutting and pasting them.

> 2) SSL/TLS
> 
> Early in the design we have decided to go for SSL/TLS instead of designing 
> our own forward secure channel. Was this wise? 

IMO, yes.  I'm not smart enough to analyze the proposed protocol, and
nobody else has stepped forward to do so.  It looks (superficially)
okay, but what do I know?  

The "everything else is simple" issue may not be so: optimizing DH is a
bit of a black art, and doesn't seem (yet?) to be cleanly supported in
any OpenSSL alternative.

Code bloat: In the Python implementation, 1808/21382 lines are spent
implementing MMTP and wrapping TLS.  That's only 8%.  If/when we switch
to Twisted, this number will/would decrease.  

Compilation and licensing: Once OpenSSL 0.9.7 is available everywhere,
we can link it dynamically, and not worry about compilation.  The bugs
seem to be minor (and well understood); the licensing issues can be
solved if NSS ever gets server-side DHE.

I also hereby place my support behind Lucky's argument: TLS is tricky to
do, and we need most of it.

-- 
Nick Mathewson <nickm@alum.mit.edu>