[tor-relays] amount of unmeasured relays continuously rising since 2 weeks

[Roger Dingledine]

On Thu, May 14, 2015 at 08:01:19PM +0000, nusenu wrote:
> Dear Tor BW Authories,
> 
> we'd like to bring the following issue to your attention.
> Since the last two weeks the amount of unmeasured relays is steadily
> rising [1].

Hello nusenu,

Thanks for the kick.

For other statistics, each hour I copy over the votes that moria1 sees:
http://freehaven.net/~arma/moria1-v3-status-votes

And there's also
https://consensus-health.torproject.org/#bwauthstatus

It is clear that the bwauth scripts are buggy and need a rewrite. And
not just buggy -- I think their design is probably flawed too.

I know of several research groups who have been working on better
bandwidth estimation schemes, with the main goal of making them more
robust to cheating, but with the secondary goal of still being adequately
accurate.

I had been hoping that one of those would be ready by now, but the
research group I've been paying the most attention to has quite a bit
of work left to do before they have a cheating-resistance estimation
scheme that doesn't open up too many other problems.

Another hope we have is Aaron Gibson's upcoming OTF fellowship, which
focuses on improving the bwauth system. I think we'll have to find the
right balance there between a redesign, which would take a while but
could be better for us in the long run, and bugfixes on the current
system, which would help things in the short term but take time away
from the redesign.

I'll call Aaron's attention to this thread too and see if he can provide
some insights.

--Roger