-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA512
Jesse Victors:
On 10/18/2013 11:46 PM, David Carlson wrote:
On October 8 something caused my non-exit relay speed to drop from around 50 KB to less than 10 KB according to Atlas graphs. I have checked with my ISP and run speed tests that verify my upload speed to be .96 Mb and download to be over 3 Mb, as they have been for years. I am running Tor 0.2.4.17-rc on Windows 7 and my consensus weight has dropped to 26. What could have caused this?
Same here. See the below and do the math on how much data I was and am now pushing per 24 hours, as well as open circuits.
Oct 07 09:04:41.242 uptime 15 days 22:00 hours, 791 circuits open. I've sent 55.41 GB and received 57.06 GB. Oct 08 09:04:41.237 uptime 16 days 22:00 hours, 1255 circuits open. I've sent 63.95 GB and received 65.80 GB.
Oct 17 23:04:41.236 uptime 26 days 12:00 hours, 87 circuits open. I've sent 90.08 GB and received 92.16 GB. Oct 18 23:04:41.259 uptime 27 days 12:00 hours, 87 circuits open. I've sent 91.85 GB and received 93.96 GB.
I've seen very similar logs (with occasional fluctuations upward, pushing much more traffic for a short time) on my Pi relay, which is struggling to get its Stable flag back. :( I just figured it was the lack of the Stable flag.
I've been seeing the same. My bandwidth limit is around 3 MB/sec, but it's been consistently at 400-600 KB/sec. I don't think it started that early for me though, not sure though. I was going to blame it on psad, but I can't see any evidence of that. The fact that all three of us experience this is interesting.
I suspect another user's assessment that Tor middle-node bandwidth is now abundant, and thus nodes below a certain consensus fraction are left out in the cold, may be correct. Just my hunch though.
I also wonder if this behavior is non-optimal. I'd prefer a better spread of traffic to nodes above some absolute floor (mainly to avoid well-meaning people on DSL connections with 128Kbps outbound, heh). I can push a couple hundred KB/sec very reliably - that ought to be worth something.
If nothing else, cutting traffic to near zero on slower-but-not-glacial nodes is problematic for a couple reasons:
1. If the operators are also using them as their entry point to Tor, it might make them more vulnerable to traffic analysis, since there's less "cover" traffic (just my hunch)
2. It may discourage node operators with very usable bandwidth, and cause them to shut down their relays. In the future, those relays may become very necessary.
Best, - -Gordon M.