[tor-talk] mildly updated statistical analysis of Tor metrics data

[Virgil Griffith]

URL: https://docs.google.com/document/d/1SaBK664SchhZOP9XBsB8KK63k4xlmMTlkhfF28f2204/pub


New highlights:
* After filtering out the bot-net window, there's a stronger
relationship between increasing advertised bandwidth and faster
Torperf.

* Corroborated the above finding by showing that low "read bandwidth /
advertised bandwidth" reliably (r^2 > 60%) correlates with faster
Torperf.

* I am disheartened that Torperf speeds (especially for small files)
normalized by the average speed of client countries has barely budged
since 2011.

* I think I am largely done with this analysis.  I am ready to begin
the process of massaging this to be put on Tor Blog.

-V