[tor-bugs] #21394 [Core Tor/Tor]: connection timeouts are affecting Tor Browser usability

[Tor Bug Tracker & Wiki]

#21394: connection timeouts are affecting Tor Browser usability
-------------------------------------------------+-------------------------
 Reporter:  arthuredelstein                      |          Owner:  (none)
     Type:  defect                               |         Status:  new
 Priority:  Very High                            |      Milestone:  Tor:
                                                 |  0.3.3.x-final
Component:  Core Tor/Tor                         |        Version:
 Severity:  Normal                               |     Resolution:
 Keywords:  tbb-performance, tbb-usability,      |  Actual Points:
  performance, tbb-needs                         |
Parent ID:                                       |         Points:
 Reviewer:                                       |        Sponsor:
-------------------------------------------------+-------------------------

Comment (by arthuredelstein):

 I wrote a script using atagar's very nice stem library to try to test DNS
 behavior across all exits:
 https://github.com/arthuredelstein/tor_dns_survey/tree/cae14fc37758afb0963d9b67f5d92c0dfda9a507

 The script creates 2-hop circuits. Each exit was tested 5 times for
 example.com and 5 times for '93.184.216.34', the bare IPv4 for
 example.com.

 Here's the raw-ish data:
 https://arthuredelstein.net/tor/21394/exit_results_20171029.json
 The format is
 {{{
 { address : { exit : [ [ attempt1_result, attempt1_date_time,
 attempt1_duration ], [ attempt2... ] ... ] } }
 }}}
 with about 5 attempts made per address/exit combination.

 I did a quick analysis of the data and found the following.
 * For the raw IP address, the overall timeout frequency was 1.39%. 3 exits
 always timed out, 29 exits sometimes did, and 752 always suceeded (the
 rest were assorted fast failures).
 * For example.com, the overall timeout frequency was 9.03%, 41 exits
 always timed out, 74 sometimes did, and 653 always succeed (again the
 remainder were other failures).

 It's interesting that the observed timeout frequency here was only half
 that observed in comment:20. Perhaps that's because here I tested all
 exits equally, instead of using consensus weights to choose exits as Tor
 Browser does. But I would have expected more timeouts, not less, if I use
 the low bandwidth exits more, so I'm surprised by that.

 In any case, if the timeout probability were uniformly 9% for all exits, I
 would expect essentially no cases of an exit failing 5 out of 5 times. So
 it's clear that the probability of some exits failing  is much higher than
 others. Therefore we have "bad" and "flaky" exits.

--
Ticket URL: <https://trac.torproject.org/projects/tor/ticket/21394#comment:27>
Tor Bug Tracker & Wiki <https://trac.torproject.org/>
The Tor Project: anonymity online