[metrics-bugs] #34257 [Metrics/Onionperf]: Analyze unusual distribution of time to extend to first hop in circuit

[Tor Bug Tracker & Wiki]

#34257: Analyze unusual distribution of time to extend to first hop in circuit
-------------------------------+--------------------------------
 Reporter:  karsten            |          Owner:  metrics-team
     Type:  defect             |         Status:  new
 Priority:  Medium             |      Milestone:
Component:  Metrics/Onionperf  |        Version:
 Severity:  Normal             |     Resolution:
 Keywords:                     |  Actual Points:
Parent ID:                     |         Points:
 Reviewer:                     |        Sponsor:  Sponsor59-must
-------------------------------+--------------------------------

Comment (by dennis.jackson):

 Okay, this extra data is pretty helpful!

 Looking at the Histograms for `req2fb` they are look consistent:
 [https://github.com/galadran/onionperf-guard-
 analysis/blob/master/images/histograms/req2fb_ip-172-31-5-27.png 1],
 [https://github.com/galadran/onionperf-guard-
 analysis/blob/master/images/histograms/req2fb_op-hk2.png 2],
 [https://github.com/galadran/onionperf-guard-
 analysis/blob/master/images/histograms/req2fb_op-nl2.png 3],
 [https://github.com/galadran/onionperf-guard-
 analysis/blob/master/images/histograms/req2fb_op-us2.png 4]. So I would
 guess the actual circuits are performing reasonably similarly once built.

 Looking at the Histograms for `start2req` shows a very different pattern
 for each region. The distribution is offset roughly in proportion to the
 offset observed in the `bt1` distribution, but scaled up:
 [[Image(https://raw.githubusercontent.com/galadran/onionperf-guard-
 analysis/master/images/histograms/start2req_op-nl2.png, width=70%)]]
 [[Image(https://raw.githubusercontent.com/galadran/onionperf-guard-
 analysis/master/images/histograms/start2req_op-us2.png, width=70%)]]
 [[Image(https://raw.githubusercontent.com/galadran/onionperf-guard-
 analysis/master/images/histograms/start2req_op-hk2.png, width=70%)]]
 [[Image(https://raw.githubusercontent.com/galadran/onionperf-guard-
 analysis/master/images/histograms/start2req_ip-172-31-5-27.png,
 width=70%)]].
 I think the spikes in the last graph are just an artefact of the lower
 sample count, but it is pretty interesting that this is the only graph
 without a peak around 0.

 Why does the new Hong Kong node not see very fast circuit build times? Why
 does the time between initiation and success vary on geographic region in
 the same pattern as the first extension, but scaled up? It can't be caused
 purely by the `bt1` offset, because that is only on the order of ~1 second
 rather than ~8 seconds.

 I wonder if there is a bug in Onionperf? Some kind of race condition /
 incorrect mapping between requests and successful circuit builds? I'll try
 to find some time to sit down with the raw files.

--
Ticket URL: <https://trac.torproject.org/projects/tor/ticket/34257#comment:7>
Tor Bug Tracker & Wiki <https://trac.torproject.org/>
The Tor Project: anonymity online