Re: getting more exit nodes

[Roger Dingledine <arma@xxxxxxx>]

On Sun, Apr 20, 2008 at 05:41:57PM +0200, Dominik Schaefer wrote:
> I think, any routers with uptimes less then 4-6 hours will be more or
> less useless for the network and rather tend to produce more traffic
> for distributing the router descriptor than relay for clients.

Well, in theory the "client-exit" nodes don't need to send their
descriptors to any users. They just need to have an association with
a Tor relay so the relay learns quickly when they're available. Then
the relay can advertise itself with the exit policy of the client-exit
node, and so long as it has some client-exits available whenever an exit
request shows up, it can pass it on.

Here are a few other things that would be useful to consider before we
can evaluate the idea:

- Bug 98. If you run too many connections through a Tor process on
  Windows, the OS will crash. We (meaning Nick) are slowly working on
  that, but it is not yet solved.
- Robert Hogan's questions in this thread are good to look at. In
  particular, does the relay actually advertise the client-exit's exit
  policy as its own?
- Need to consider load balancing. Right now users will choose the
  relay proportional to the bandwidth it advertises. But if it's a fast
  relay and the client-exit node is slow, the client-exit node will get
  overloaded and things will be even slower than they are now.
- Related to load balancing: how much additional latency are we talking
  about, from adding a fourth hop to the circuit? Because it would seem
  that you need four hops, since the "relay to client-exit" hop isn't
  adding much additional anonymity. (Or is it?)
- How is the crypto going to work between the relay node and the
  client-exit node? Are you planning to put a Tor process on the
  client-exit node, or were you just hoping to put a tiny proxy on it?

Probably there are more issues to explore after these.

Hope that helps,
--Roger