Network throughput on tor relay stalls
Im am running the torland relay family. On one of my servers I experience sudden network outages. The relays normally pump at a rate of 450-550 MBit/s. When the outage happens, within minutes the rate goes down below about 5Mit/s. In this case it is difficult to connect to the server via ssh or to access the webserver with the traffic stats. I checked the logs but found nothing in there. During the outage the CPU runs at a pretty low level, so CPU is not the root cause of this issue. Doing a traceroute to the server showed that the server indeed was the point where packet loss happened. I also contacted the ISP and asked them to check their network but they found nothing. For me it looks as if the network stack on the machine runs at some point on some kind of lock which slows done throughout. Because the processes doing real work are tor processes (all are consuming nearly 100% CPU) I suppose tor is causing the outage. Are there any log options for tor I can use to investigate the cause of this issue?
Fingerprints of the relays are: D223399907113A1F216AAA64997BC1D4CFA8E1AC 945CBBA599808018749DDC4EBB592168F2858C1B 09C0C5800177BF3A11A78A98A1CAFD8E7AD2EA02
Just to visualize I attach a screen shot of network usage when the issue happens: http://goo.gl/vSe32
Any hint is appreciated.
Trac:
Username: torland