Proposal: Optimistic Data for Tor: Server Side

[Ian Goldberg]

On Tue, Aug 03, 2010 at 02:27:35PM -0400, Nick Mathewson wrote:
> On Mon, Aug 2, 2010 at 10:31 AM, Ian Goldberg <iang at cs.uwaterloo.ca> wrote:
> > (My first Tor proposal; hopefully it's in a sensible form...)
> >
> > Here's the server side of the proposal I promised in my rump session
> > talk at PETS.  The server side seems harmless, and getting it deployed
> > to a bunch of nodes before the client side gets out there seems like a
> > good idea in any event.
> >
> > The client side yields both the performance improvements, as well as
> > potential client fingerprinting issues.  That side will have to be
> > carefully considered.
> >
> > Discuss.  ;-)
> 
> Thanks, Ian!  I've added it as proposal 174.
> 
> The proposal looks good to me.  I'll try to answer some of the points
> that it was confused on:
> 
> > What do version numbers for hypothetical future protocol-compatible implementations look like, though?
> 
> They look like Tor version numbers, for whatever Tor version merges
> the patch that implements this, and later.

No, I mean that if someone writes a totally different implementation of
the Tor server (say in some other language like Java or Haskell), how
will the client be able to tell whether that implementation supports
this feature?

> > It is not clear exactly what an "unrecognized" stream is
> 
> An "unrecognized" stream is one for which we haven't yet received a
> BEGIN cell.  We'll also need to modify the spec to say that older
> versions of Tor didn't handle RELAY_DATA cells the same way that newer
> ones did.
> 
> We should consider the patch independently from the proposal.  The
> proposal itself looks fine.  Generally, we try discussing patches on
> the tracker at trac.torproject.org.  We're in a loose feature-freeze
> right now in a hurried attempt to release an 0.2.2.x rc, but a later
> version of the patch in this proposal has a good shot IMO for 0.2.3.x.
>  Do you want to make the tracker entry for the patch, or should I? :)

Please go ahead.

Thanks,

   - Ian