[tor-bugs] Re: #1335 [Tor - Tor client]: Low value for Xm gives high build timeout

["Tor Bug Tracker & Wiki" <torproject-admin@xxxxxxxxxxxxxx>]

#1335: Low value for Xm gives high build timeout
-------------------------------+--------------------------------------------
  Reporter:  mikeperry         |       Owner:  mikeperry         
      Type:  defect            |      Status:  closed            
  Priority:  minor             |   Milestone:  Tor: 0.2.2.x-final
 Component:  Tor - Tor client  |     Version:                    
Resolution:  fixed             |    Keywords:                    
    Parent:                    |  
-------------------------------+--------------------------------------------
Changes (by mikeperry):

  * status:  new => closed
  * resolution:  None => fixed


Old description:

> My tor client just learned a timeout of 42, despite having the majority
> of my timeouts below 2s. I suspect
> this is because the Xm value was 275, but I still had quite a bit of
> timeouts after that. This caused the tor client
> to calculate a really small alpha, resulting in a shallow curve.
>
> I am not sure what to do about this. I suspect what I'm actually looking
> at is 2 or 3 curves: one for each guard.
> We could try learning timeouts independently per guard (fragile, bulky),
> or we could try a different way to calculate
> Xm - maybe mean or median of 3 highest modes?
>
> It could also be that the synthetic timeouts I created were very high,
> and are also causing alpha to be low. I'm
> attaching the state file to experiment with in the future.
>
> [Automatically added by flyspray2trac: Operating System: All]

New description:

 My tor client just learned a timeout of 42, despite having the majority of
 my timeouts below 2s. I suspect
 this is because the Xm value was 275, but I still had quite a bit of
 timeouts after that. This caused the tor client
 to calculate a really small alpha, resulting in a shallow curve.

 I am not sure what to do about this. I suspect what I'm actually looking
 at is 2 or 3 curves: one for each guard.
 We could try learning timeouts independently per guard (fragile, bulky),
 or we could try a different way to calculate
 Xm - maybe mean or median of 3 highest modes?

 It could also be that the synthetic timeouts I created were very high, and
 are also causing alpha to be low. I'm
 attaching the state file to experiment with in the future.

 [Automatically added by flyspray2trac: Operating System: All]

--

Comment:

 Fix should appear in 0.2.2.14.

-- 
Ticket URL: <https://trac.torproject.org/projects/tor/ticket/1335#comment:5>
Tor Bug Tracker & Wiki <https://trac.torproject.org/>
The Tor Project: anonymity online