[tor-bugs] #29853 [Core Tor/sbws]: sbws should show relays for diagnostics, even when MIN_REPORT has not been reached

[Tor Bug Tracker & Wiki]

#29853: sbws should show relays for diagnostics, even when MIN_REPORT has not been
reached
-------------------------------------------------+-------------------------
 Reporter:  teor                                 |          Owner:  juga
     Type:  defect                               |         Status:
                                                 |  needs_revision
 Priority:  Medium                               |      Milestone:  sbws:
                                                 |  1.1.0
Component:  Core Tor/sbws                        |        Version:
 Severity:  Normal                               |     Resolution:
 Keywords:  tor-bwauth, sbws-1.0-must-           |  Actual Points:
  moved-20181128, sbws-11x-final-                |
  removed-20190312, sbws-110-proposed, changes-  |
  version-minor                                  |
Parent ID:  #28563                               |         Points:
 Reviewer:  nickm                                |        Sponsor:
-------------------------------------------------+-------------------------
Changes (by teor):

 * status:  merge_ready => needs_revision


Comment:

 Replying to [comment:6 nickm]:
 > LGTM. Before merging, a question: -- the last time that we marked
 anything as vote=0, we also marked it as unmeasued=1, bw=1.  Do we need to
 do that in this case?

 But the relays are measured. There just aren't enough of them to have a
 reasonable set of weighted bandwidths across the network.

 So the spec says `vote=0 under_min_report=1 bw=1 real_bw=N`:
 https://github.com/torproject/torspec/pull/68/commits/bce5edc7382bfe229f9b56124574b97df1939c1b
 Which means that authorities without #29806 will vote bandwidth 1 for
 these relays.

 But we might want legacy authorities to vote bandwidth=(the measured
 bandwidth) or bandwidth=(the reported bandwidth) for these relays.

 If 3 or more bandwidth authorities are already voting measured bandwidths
 for most relays, it doesn't matter what we choose. Even if we choose bw=1,
 the unmeasured relays just have a slightly lower median in their vote.

 But if we're just re-starting the bandwidth authorities after many of them
 have failed, here's what happens when we vote bw=
 * 1: measured relays get a consensus weight of 1 once 3 authorities know
 about them
   * then there is a big change in weight once 2 authorities have measured
 60% of relays, and start voting measured bandwidths
 * the reported bandwidth: measured relays get a consensus weight of their
 reported bandwidth once 3 authorities have measured them
   * then there is a smaller change to the measured weight once 2
 authorities have measured 60% of relays, and start voting measured
 bandwidths
   * relays can lie about their bandwidth more easily
 * the measured bandwidth:measured relays get a consensus weight of their
 measured bandwidth once 3 authorities have measured them
   * there is no change once 2 authorities have measured 60% of relays, and
 start voting measured bandwidths

 So I think we should put measured bandwidths in the bandwidth file in this
 case.
 (And we don't need real_bw.)
 I'll do a draft spec update,
 What do you think, juga, nickm?

--
Ticket URL: <https://trac.torproject.org/projects/tor/ticket/29853#comment:7>
Tor Bug Tracker & Wiki <https://trac.torproject.org/>
The Tor Project: anonymity online