On Mon, Jan 21, 2013 at 04:31:01PM -0700, Nicolas Bock wrote:
every time I run a relay on a Comcast cable connection, after a while (a few hours to days) other network operations such as DNS lookups slow down to such an extent that the network becomes unusable. Unfortunately I don't know how to exactly pinpoint what the problem is, but found that the only reliable solution at such a point seems to be to reset the cable modem.
Is that a known issue? Could you suggest how I can diagnose this more accurately?
Two likely possible causes: either your relay is using a lot of bandwidth and causing congestion, or your cable modem is doing NAT and the many active TCP connections from the relay overload the modem's feeble mind. Try reducing the maximum bandwidth in your torrc and see if that resolves the problem.
To get closer to a root cause, I'd use mtr, http://en.wikipedia.org/wiki/MTR_%28software%29 , to compare performance during a "speedy" period and during a "slow" period. Specifically, I'd "mtr -u some-nearby-IP" and switch to "last" mode with the 'd' command, twice. This lets me see where in my pipeline the slowdown is likely occurring. If the location of the bottlenck changes over time as the visible performance gets worse, then there is a clue where the performance change is coming from.
-andy