I recently started my first relay, sTORMUR, and I have noticed that its behaviour seems backwards to the expected. Namely:
Relay server started 1800Z 18 Oct, quiet first 12 hours, as expected.
After 12 hours at 0600Z 19 Oct traffic suddenly jumped to configured max (200KiB/s). Download varied 50-400 KiB/s and upload was constant at 200KiB/s. More upload overall than down by a noticeable amount (~15%)
Consensus weight started at 1, over next couple days it slowly increased to 14. But as consensus weight increased, traffic decreased until after day 2 traffic was barely a trickle.
Consensus has now stagnated at 14.
Bandwidth data shows for the first three days, but no bandwidth data points for last two days.
My first thought was that the initial surge was a throughput test from a bwauth. But Is it normal for a bwauth test to run for a day and a half and for in and out traffic to be unbalanced during one?
I was using BandwidthRate to limit speed to the amount of data I wanted to donate per month, and then BandwidthRateBurst was set to the full connection speed. This was a poor choice, and I’m now using accounting and letting the connection run full speed.
I don’t think the initial burst was an attack. Each bwauth sends a fixed amount of traffic through for its bandwidth test, and at 200KiB/s that made each auth’s bandwidth test take much more time. On top of that, the bwauths aren’t smart enough to know not to step on each other’s test - I think they just assume if they are each randomly testing at some point over the first 24 hours they won’t step on each other. But with the slower speed, I think 4 or 5 of them stacked on each other at the same time which just snowballed. This killed each one’s bandwidth assessment and pegged the connection for a day.
The bandwidth assessment was hopelessly poisoned by this, so I’ve fixed the configuration and re-keyed the relay to reset it.
I recently started my first relay, sTORMUR, and I have noticed that its behaviour seems backwards to the expected. Namely:
Relay server started 1800Z 18 Oct, quiet first 12 hours, as expected.
After 12 hours at 0600Z 19 Oct traffic suddenly jumped to configured max (200KiB/s). Download varied 50-400 KiB/s and upload was constant at 200KiB/s. More upload overall than down by a noticeable amount (~15%)
Consensus weight started at 1, over next couple days it slowly increased to 14. But as consensus weight increased, traffic decreased until after day 2 traffic was barely a trickle.
Consensus has now stagnated at 14.
Bandwidth data shows for the first three days, but no bandwidth data points for last two days.
Actually, if you look at the history section on relay-search you see
bandwidth data points for you relay, although, granted, it’s not much.
But then a consensus weight of 14 is not much either…
My first thought was that the initial surge was a throughput test from a bwauth. But Is it normal for a bwauth test to run for a day and a half and for in and out traffic to be unbalanced during one?
No. It’s basically just a couple of second downloading a file.
I think what is meant here is the minimum requirement for running a useful relay. If you really, really want to you can run relays
configured to less than 100KiB/s (and there are such relays in the network).
···
As for traffic spikes, it is probably because of various attacks targeting Tor network.
Oh, and you can read up on bandwidth measurements a bit on our blog post from a couple of years back. As a bonus point you get a link to a blog post about research questions related to Tor, which talks among other things about challenges related to bandwidth measurement.