[tor-bugs] #2286 [Tor Directory Authority]: We still use self-published relay bandwidth sometimes

[Tor Bug Tracker & Wiki]

#2286: We still use self-published relay bandwidth sometimes
-------------------------------------+--------------------------------------
 Reporter:  arma                     |          Owner:                    
     Type:  defect                   |         Status:  needs_review      
 Priority:  major                    |      Milestone:  Tor: 0.2.3.x-final
Component:  Tor Directory Authority  |        Version:                    
 Keywords:  arma-cares               |         Parent:                    
   Points:                           |   Actualpoints:                    
-------------------------------------+--------------------------------------

Comment(by nickm):

 Replying to [comment:31 arma]:
 > Replying to [comment:29 nickm]:
 > > Roger, I don't get at this point why you prefer your alternative
 approach (vote differently) vs the one I started to implement (consense
 differently).  I kind of prefer the latter, since it doesn't change the
 meaning of any vote, and it looks like we'll probably need a new consensus
 method anyway.
 >
 > I'm open to either approach. We'll want to make sure we don't cap all
 nodes to 50 in the case where there aren't enough bwauths putting in
 Measured votes.
 >
 > We may also find that we want to only cap weights when a certain
 threshold of the relays have a Measured vote. I'm not sure if we'll want
 to do that threshold, or what we'll want it to be, so I had figured
 leaving the flexibility to each authority might make it easier. For
 example, I worry about a case where all the bwauths restart afresh and
 have just a few votes each -- currently it makes sense for them to include
 the votes, but now they'd better think about not putting in any votes
 until they have quite a few.

 I think that approach  (Don't say "measured" until you have measured a
 bunch of routers) is smarter and way easier to implement than the approach
 where we don't cap until a threshold of routers has measured values.

 To avoid the first problem you talk about, I'd suggest a simple rule of
 "implement the cap approach when at least 3 authorities are publishing
 measurements."  It's not perfect, and has corner cases, but it should be
 basically ok.

 > That said, we also want to avoid an attack where the adversary floods
 the network with new nodes, and suddenly the bwauths take away their caps
 because they haven't covered the required threshold of the nodes yet.
 >
 > *That* said, now that it looks like we're going to be making a new
 consensus method anyway, there's not as much reason left to avoid doing it
 in the consensus.


 I've updated the branches "bug2286_part_a" in my public tor repo and in my
 public torspec repo to try to implement the behavior we've been
 discussing.  Sensible now?  Correctly implemented?

-- 
Ticket URL: <https://trac.torproject.org/projects/tor/ticket/2286#comment:32>
Tor Bug Tracker & Wiki <https://trac.torproject.org/>
The Tor Project: anonymity online