[tor-scaling] Analyzing the Predictive Capability of Tor Metrics

[Nick Mathewson]

This approach seems plausible to me, and I am excited to see what
results it produces.

One note: We should keep ourselves open to the null hypothesis that we
_haven't_ found the explanation for our performance. After we've found
whatever we think is the likeliest set of explanatory factors, we
should look for evidence to _disconfirm_ our conclusion, in case we've
found something that only applies under certain circumstances.

[My own suspicion is that the pre-2015 history of Tor's performance is
dominated by bottlenecks that don't actually tell us much about real
nature of the performance problem.  It's like we started driving down
the highway on a flaming bus with four flat tires, and we ran out of
gas once or twice.  With data like that we would correctly conclude
that the speed of our bus is inversely correlated with its
temperature, correlated with its height off the ground, and correlated
with the amount of fuel it's carrying.  But we can't conclude that in
order to go faster, we should supercool our bus, jack it up 3m off the
ground, give it a extra 500L fuel tank.]

But anyway, this is neither here nor there: I think that the approach
you outline will give us an excellent starting point if do it
carefully.

peace,
-- 
Nick