[tor-bugs] #33434 [Metrics/Onionperf]: Allow users to select Onion Service version to measure

[Tor Bug Tracker & Wiki]

#33434: Allow users to select Onion Service version to measure
---------------------------------------+--------------------------------
 Reporter:  acute                      |          Owner:  karsten
     Type:  enhancement                |         Status:  needs_review
 Priority:  Medium                     |      Milestone:
Component:  Metrics/Onionperf          |        Version:
 Severity:  Normal                     |     Resolution:
 Keywords:  metrics-team-roadmap-2020  |  Actual Points:
Parent ID:                             |         Points:  0.5
 Reviewer:                             |        Sponsor:  Sponsor59-must
---------------------------------------+--------------------------------
Changes (by karsten):

 * status:  accepted => needs_review
 * cc: dgoulet, asn (added)


Comment:

 While working on #34023 (for deciding whether we can use partial
 completion times instead of full completion times) I realized that I could
 use the same analysis data and code to compare version 2 to version 3
 measurement results for this ticket.

 And while doing that analysis I realized that we're currently doing 1/3 of
 all OnionPerf measurements to version 3 onion servers---which we're not
 even graphing on Tor Metrics. But even if we did graph them, we'd be
 making 2/3 of measurements to onion services and only 1/3 via exits.

 Can we take out v2 onion service support and only do v3 onion service
 measurements in the future? Please take a look at the following graph. It
 shows time to download 50 KiB, 1 MiB, and 5 MiB over Tor to its own web
 server (public server) or to its own v2 or v3 onion server.

 [[Image(onionperf-v2vsv3-2020-05-17.png, 700px)]]

 I didn't follow all the v3 onion server changes in detail. Is there any
 conceptual difference that might show up in these performance graphs? If
 it's mainly different crypto, shared randomness, and a differently
 distributed directory, we could expect the same performance results for
 downloading files using either version.

 What do you think? Should we just kill the version 2 measurements? And
 should we include all onion server measurements on Tor Metrics, regardless
 of version?

 Copying onion service people to also get their input here.

--
Ticket URL: <https://trac.torproject.org/projects/tor/ticket/33434#comment:4>
Tor Bug Tracker & Wiki <https://trac.torproject.org/>
The Tor Project: anonymity online