[tor-bugs] #4482 [Analysis]: Research: what's the relationship between capacity, bytes used, and torperf?

[Tor Bug Tracker & Wiki]

#4482: Research: what's the relationship between capacity, bytes used, and
torperf?
---------------------------------+------------------------------------------
 Reporter:  arma                 |          Owner:     
     Type:  task                 |         Status:  new
 Priority:  normal               |      Milestone:     
Component:  Analysis             |        Version:     
 Keywords:  performance scaling  |         Parent:     
   Points:                       |   Actualpoints:     
---------------------------------+------------------------------------------
Description changed by arma:

Old description:

> It seems that when our network capacity goes down, the torperf latency
> goes up. And vice versa. Except it's not a perfect correlation, and
> sometimes the response is damped more than other times.
>
> Seems to me that the less capacity there is (relative to actual load),
> the more losing further capacity will influence the torperf results. And
> the more capacity there i  (relative to actual load), the less gaining
> further capacity will influence the torperf results.
>
> Is there a way to predict where the 'healthy' or 'unhealthy' parts of
> this relationship graph are?
>
> Step one is to try to figure out what data points we have so far, and get
> an intuition for what we think happened, and how healthy we think the
> network was, in each case.
>
> Step two might be to design some experiments where we intentionally gain
> or lose capacity in a controlled way, and observe the effect on torperf
> results, either to confirm a hypothesis or to get more data points.

New description:

 It seems that when our network capacity goes down, the torperf latency
 goes up. And vice versa. Except it's not a perfect correlation, and
 sometimes the response is damped more than other times.

 Seems to me that the less capacity there is (relative to actual load), the
 more losing further capacity will influence the torperf results. And the
 more capacity there is (relative to actual load), the less gaining further
 capacity will influence the torperf results.

 Is there a way to predict where the 'healthy' or 'unhealthy' parts of this
 relationship graph are?

 Step one is to try to figure out what data points we have so far, and get
 an intuition for what we think happened, and how healthy we think the
 network was, in each case.

 Step two might be to design some experiments where we intentionally gain
 or lose capacity in a controlled way, and observe the effect on torperf
 results, either to confirm a hypothesis or to get more data points.

--

-- 
Ticket URL: <https://trac.torproject.org/projects/tor/ticket/4482#comment:2>
Tor Bug Tracker & Wiki <https://trac.torproject.org/>
The Tor Project: anonymity online