[tor-bugs] #6341 [Tor Relay]: connection_or_flush_from_first_active_circuit() does wrong thing when ewma_enabled

[Tor Bug Tracker & Wiki]

#6341: connection_or_flush_from_first_active_circuit() does wrong thing when
ewma_enabled
-----------------------+----------------------------------------------------
 Reporter:  arma       |          Owner:                    
     Type:  defect     |         Status:  needs_review      
 Priority:  normal     |      Milestone:  Tor: 0.2.3.x-final
Component:  Tor Relay  |        Version:                    
 Keywords:             |         Parent:                    
   Points:             |   Actualpoints:                    
-----------------------+----------------------------------------------------

Comment(by arma):

 Replying to [comment:55 robgjansen]:
 > > Does that mean it's time to try the actual #6341 experiment again?
 >
 > Not sure, but
 [https://trac.torproject.org/projects/tor/attachment/ticket/6341/20120909-ec2
 -brokenewma-combined.pdf I already did anyway]. Does this provide any more
 answers than
 [https://trac.torproject.org/projects/tor/attachment/ticket/6341/20120813-ec2
 -brokenewma-combined.pdf last set of graphs]?

 It does provide more answers, insofar as the 0813 graphs look fantastic
 but it's because there's less load on the system overall. The 0909 graphs
 look like impressive but the load seems more realistic.

 The p2p-client time-to-first-byte (and time-to-last-byte) is worse with
 the patched ewma case. That's fine by me. Actually, it looks like time-to-
 first-byte is crummy for p2p clients in all three cases. Perhaps that's
 because they're blurting out many connection requests in parallel.

 The legend in 'network data read/written over time' is particularly poorly
 placed: what happens to the blue line underneath it? :)

 I guess my main question from considering these graphs is: what is the
 variance on these results? If you run with the same network but a
 different random seed, do you get the same general results or are they
 different by a bit? It looks like for the topic at hand, 'different by a
 bit' could easily swing things from 'helps to hurts' or vice versa. Does
 that mean we should close #6341 (now that we fixed the major Shadow bug)
 and move on to simulating something that we think ought to make a huge
 difference?

-- 
Ticket URL: <https://trac.torproject.org/projects/tor/ticket/6341#comment:58>
Tor Bug Tracker & Wiki <https://trac.torproject.org/>
The Tor Project: anonymity online