These relays all appear at least somewhat overrated by BWauths and are not rate-limiting with BandwidthRate.
DirPort performance is bad due to a saturated physical link or 'tc' bandwidth limit, so TCP congestion back-off kicks in.
If BandwidthRate were the limiting factor DirPort would return a 503 (busy) error.
Reason for the overrating is a serious bug in Torflow resulting in relays going unmeasured for extended periods, possibly more than 30 days at which time a BWauth will cease listing an affected relay.
I'm planning to work on fixing this, but have not had the time and presently don't see getting to it before early summer.
Dear Relay Operators,
Also while working on #17158, I found some relays whose DirPort responses made my python script hang. It wouldn't even respond to ^C, but that might just be an OS X thing, or a stem thing.
I'm not sure what to do about these relays. I'd like to do a tcpdump / Wireshark check, but I don't have time at the moment. Would anyone like to follow this up?
This is an incomplete list of broken relay IPs and DirPorts, starting with those with the highest consensus weight: 217.23.14.190:1194 151.80.164.147:80 148.251.255.92:80 78.142.19.59:80
Thanks
Tim
Tim Wilson-Brown (teor)