[metrics-bugs] #31521 [Metrics/Analysis]: Investigate 10-second delay in TTFB

[Tor Bug Tracker & Wiki]

#31521: Investigate 10-second delay in TTFB
-------------------------------------+------------------------------
 Reporter:  karsten                  |          Owner:  metrics-team
     Type:  defect                   |         Status:  new
 Priority:  Medium                   |      Milestone:
Component:  Metrics/Analysis         |        Version:
 Severity:  Normal                   |     Resolution:
 Keywords:  scalability performance  |  Actual Points:
Parent ID:                           |         Points:
 Reviewer:                           |        Sponsor:
-------------------------------------+------------------------------
Changes (by mikeperry):

 * cc: arthuredelstein (added)


Comment:

 This sounds about the rate of bad exit connectivity that arthuredelstein
 noticed during his scans a year or so ago. IIRC he noticed it for DNS
 timeouts though. In this case DNS appears not to be involved/at fault.

 There are three ways to deal with this:

 0. Continue exit scanning constantly and hassle Exits to help diagnose/fix
 this.
 1. Test that pre-built Exit circuits can resolve a test domain like
 example.com (may not handle this case).
 2. Implement an Adaptive Stream Timeout (similar to/based on Circuit Build
 Timeout) rather than the fixed 10s value.

 Option 3 is the only sure shot. There is a trac ticket where we
 investigated doing adaptive stream timeouts in addition/instead of CBT as
 part of a Google Summer of Code + Research project. It did not help the
 general case, but it would help this long tail. I can't find the ticket at
 the moment, but it is in the performance R&D kanban under the "Develop"
 column.

--
Ticket URL: <https://trac.torproject.org/projects/tor/ticket/31521#comment:4>
Tor Bug Tracker & Wiki <https://trac.torproject.org/>
The Tor Project: anonymity online