[tor-bugs] #24785 [Core Tor/Fallback Scripts]: Reduce the fallback stability and flag requirements due to extra network load

[Tor Bug Tracker & Wiki]

#24785: Reduce the fallback stability and flag requirements due to extra network
load
---------------------------------------+-----------------------------------
 Reporter:  teor                       |          Owner:  teor
     Type:  enhancement                |         Status:  needs_review
 Priority:  Medium                     |      Milestone:  Tor:
                                       |  0.3.3.x-final
Component:  Core Tor/Fallback Scripts  |        Version:
 Severity:  Normal                     |     Resolution:
 Keywords:  fallback                   |  Actual Points:  0.5
Parent ID:  #22271                     |         Points:  0.2
 Reviewer:                             |        Sponsor:
---------------------------------------+-----------------------------------
Changes (by teor):

 * status:  assigned => needs_review
 * type:  defect => enhancement
 * actualpoints:   => 0.5


Comment:

 We get 101 fallbacks with the most recent list and current settings.
 Our goal is about 150, or more if we don't want to rebuild the list too
 often.
 (We want to rebuild when 25% are down. Ideally, we don't want to go much
 below 100 fallbacks, either, which implies that we want at least 134 to
 start with.)

 ADDRESS_AND_PORT_STABLE_DAYS is currently 30. But address or port changes
 are the major reason we lose fallbacks. So let's increase it to 90 days.
 (I think it's ok to assume that a relay that's had the same address for 3
 months will keep it for 12-24 months.)

 The cutoffs are currently:
 CUTOFF_RUNNING = .90
 CUTOFF_V2DIR = .90
 CUTOFF_GUARD = .80

 But there's no reason to require the guard flag when we have a minimum
 bandwidth requirement. What we really need is for fallbacks to be Running,
 and V2Dir:
 CUTOFF_RUNNING = .50
 CUTOFF_V2DIR = .50
 CUTOFF_GUARD = .00

 MIN_BANDWIDTH is currently 1.0 MByte/s, or 100x the expected load.
 I think it's fine to drop it to 50x the expected load.
 (Also, the bandwidth system isn't that accurate, anyway, so some relays
 with low measured bandwidth actually have lots of spare capacity to serve
 directory documents.)

 I also removed some old fallback list restrictions that we used to
 manually maintain, but haven't checked since 0.2.8 or 0.2.9. We don't try
 to work out if fallbacks are on the same machine, and we don't
 automatically blacklist fallbacks if their details change. (That's what
 ADDRESS_AND_PORT_STABLE_DAYS is for.)

 This gives us 152 fallbacks, and relay operators are still opting-in.
 So I think we'll go with that, and consider changing to opt-out for the
 next release (#24789).

 Please see my branch ticket24785 on https://github.com/teor2345/tor.git

--
Ticket URL: <https://trac.torproject.org/projects/tor/ticket/24785#comment:1>
Tor Bug Tracker & Wiki <https://trac.torproject.org/>
The Tor Project: anonymity online