[Bloat] Comparing bufferbloat tests (was: We built a new bufferbloat test and keen for feedback)

[Dave Taht]

I just wanted to note on this enormous thread that my primary beef
with most speedtests is the too short duration.

cable's "Powerboost" was optimized for speedtests' 20 sec duration.
The internet as a whole ended up optimized for a ~20sec download as a
result.... Run a test for 21 seconds, and boom.

flent defaults to 1minute tests by default and you can clearly see in
many results we've published, the long, slow, miserable, ramp.

A scientific way of compensating for the balance of user attention and
an accurate test would be to always run some
subset of the tests initiated by users for this longer duration and
cross check those results.