[tor-dev] tor2web proposal: comments appreciated before I commit this

[Roger Dingledine]

On Fri, Feb 28, 2014 at 12:03:10PM -0800, Virgil Griffith wrote:
> 2. Tor2web optimizations
>  2.1) self-rendezvous
>    In the current tor2web mode, the client establishes a 0-hop circuit
> (direct connection) to a chosen rendezvous point.

We actually call that a 1-hop circuit, because it has one relay in it.

A 0-hop circuit would be what you say below:

>    We propose that, alternatively, the client set *itself* as the
> rendezvous point. This coincides with ticket #9685 [2].

As Nick points out in that ticket, the tor2web node will need to
have an ORPort open. For example, it could run as a bridge and set
"PublishServerDescriptor 0" so the only incoming connections are
rendezvous circuits.

It's actually an interesting research question whether this would indeed
be faster -- in the proposed case some relay would have to do a new v3
tls handshake with the tor2web node (since you need an established OR
connection before you can extend a circuit on it), whereas in the old
case maybe there is already an established OR connection from the final
hop in the hidden service's circuit to the rendezvous point (relay) that
tor2web picked.

>  2.2) direct-introduction
>    Identical to the non-tor2web mode, in the current tor2web mode, the
> client establishes a 3-hop circuit to the introduction point.
>    We propose that, alternatively, the client build a 1-hop (or
> 0-hop?) circuit to the introduction point.

1-hop. 0-hop would be if the tor2web node is a relay and fortuitously
the hidden service picked it as one of its intro points.

I'm surprised we don't already do this part -- it looks like a clear
win for tor2web.

--Roger