[tor-talk] SKKU43 Tor relay (was Re: Metrics fast exit page)

[Roger Dingledine]

On Tue, Feb 04, 2014 at 02:24:39AM +0100, Karsten Loesing wrote:
> There are two new graphs available as replacement:
> 
> https://metrics.torproject.org/network.html#advbwdist-perc
> 
> https://metrics.torproject.org/network.html#advbwdist-relay

Wow -- check out those URLs today.

And then see
https://compass.torproject.org/#?exit_filter=all_relays&links&sort=cw&sort_reverse&country=&top=10

which currently lists SKKU43 as 11% of the advertised bandwidth, and
7.3% of exit weights:
https://atlas.torproject.org/#details/9E4C3E2597AD49CED745BE024ED620B88BBBF002

This relay advertises a self-measured bandwidth of 1GByte/s:
https://exonerator.torproject.org/serverdesc?desc-id=2c414fc08f7cbb3b4b09a5cd5014cf03d12d20ca
which is fine because self-measured bandwidth isn't used (at least
not directly) by clients. But then the torflow bwauths measure it at
w Bandwidth=10000 Measured=1230000 (moria1's opinion)
w Bandwidth=10000 Measured=67000   (gabelmoo's opinion)
w Bandwidth=10000 Measured=1450000 (tor26's opinion)
w Bandwidth=10000 Measured=617000  (maatuska's opinion)

Leading to a median of 617000 weight, which is more than twice the
second-most-weighted relay:
https://atlas.torproject.org/#details/E0113C18726D3AD74D6081C43D2539CA9A0A8745

I think IPredator is actually carrying the traffic it claims. But I find
SKKU43's bandwidth line fishy-looking:
bandwidth 1200000000 1200000000 1000000000

So I guess the first question is, does it look like SKKU43 is actually
as good at carrying traffic as it appears to be?

And the second question is, should we lower the caps in torflow on how
much weight it will give a single relay? Or are the caps already low
and there's a bug?

Fun times,
--Roger