[tor-relays] BWauth kookiness

[Roger Dingledine]

On Wed, May 20, 2015 at 06:16:36PM -0400, starlight.2015q2 at binnacle.cx wrote:
> but in the last few days the BWAuths'
> opinion went from
> 
>  bw1-w Bandwidth=7382 Measured=7100
> *bw2-w Bandwidth=7382 Measured=9330
>  bw3-w Bandwidth=7382 Measured=13700 GuardFraction=69
>  bw4-w Bandwidth=7382 Measured=23700
> 
> to
> 
>  bw1-w Bandwidth=9375 Measured=17100
>  bw3-w Bandwidth=9375 Measured=77900 GuardFraction=75
> *bw4-w Bandwidth=9375 Measured=23000
> 
> While it's flattering to be promoted
> to double the actual capacity, with the
> potential to go 8x

Keep in mind that the bwauth weights are all relative, unit-less numbers.
They're not bandwidth values; they're fractions of total weights. So
at least in theory, it can be totally fine that your relay has a very
large weight, because it has to do with the relative chances that
clients should pick your relay compared to picking other relays, not
"how much bandwidth" they think your relay has.

That's not to say that we're computing these weights in the optimal
fashion at this point. But hopefully it will help you and others have
the right intuition. :)

--Roger