Dear all,

I have now been running an exit node for approximately two weeks, it has a fast connection (100Mbit) but I have a traffic limit of 1000GB per month. I have been closely monitoring it via [1] and Munin.

I believe to have read that using accounting/hibernation is preferrable over rate limiting with fast connections, but I can't seem to find the exact page at the moment.

As you can see (well, guess) from [1], the node is up for about 6-8 hours until its daily quota (currently 10GB, may become a little more) is exhausted, with peaks of around 900K/s but usually around 400-500K/s.

In my quest to maximize the usefulness of my node, I would very much appreciate if someone could explain or make educated guesses about the following questions:

1. Note that there are two significant dips in the exit probability chart [1]. The first one I can explain, because I had to reconfigure and restart the server. Nothing special happened after that. What may have caused the second dip? What tools could I use to figure it out/monitor it in the future?

2. My server is bored in terms of memory/CPU usage. I set NumCPUs to 2 and BandwidthRate 100 MB, BandwidthBurst 200 MB as suggested in [2]. Besides the fact that I wouldn't want to exhaust my daily quota in a few hours, shouldn't the traffic become more than 900K/s peaks?

3. What is a good trade-off in terms of speed and uptime with such a daily quota? Should I go for speed? If so, are there any pretty graphs that show at what time of the day the Tor network needs the most bandwidth? If I should go for uptime, what is a good bandwidth limit to keep this a fast node?

4. I enabled directory mirroring, but apparently this does not work with hibernation (i.e. it is not advertised). If I understand correctly, directory mirrors are very useful, so how does that weigh in on the decision to maybe limit the bandwidth to keep the node up 24/7?

5. Related to 4., why is DirPort not advertised when hibernation is configured? References to papers are sufficient if this requires a complicated answer :)

[1] https://atlas.torproject.org/#details/5A91910C1B3F3FCC15EA6C6538FB9A5FAF3607... [2] http://archives.seul.org/or/relays/Aug-2010/msg00034.html

Thank you very much for any pointer on any of these questions!

Cheers, Conrad