[Rack] UPS dropped off, sonic connection works better now
j at hug.gs
Sat Sep 18 17:45:56 PDT 2010
Miloh dropped off a Tripp-lite UPS (200 lbs, be careful) which is
charging underneath the EE bench closest to the server closet. I
plugged my computer into its USB port and was able to read the battery
state. I didn't do a load test.
The APC UPSes all say they want 24A peak and have L5-30 connectors.
Even if we change the plugs, we can't plug them into the 20A circuits.
If the Tripp-lite works out and no one objects, I'm going to move the
APCs to the FSF GCC compile farm since they need it for some of their
weirder embedded hardware that doesn't come up right after a reset.
I also enabled state tracking in pf on both routers to make it easy to
see who's torrenting and hosing the link:
# pfctl -s Sources
172.30.1.3 ( states 1, connections 1, rate 0.0/0s )
172.30.1.8 ( states 3, connections 3, rate 0.0/0s )
172.30.1.10 ( states 1, connections 1, rate 0.0/0s )
172.30.1.30 ( states 1, connections 0, rate 0.0/0s )
The torrenting nodes are the ones with high connection rates in the
Because pf doesn't have a DLM to enable updating distributed counters,
source tracking is per router and not shared across the pfsync link.
It's still pretty easy to look at both routers at the same time, just
add this to your .ssh/config:
Then do a ssh -fN root at 172.30.0.1 & ssh -fN root at 172.30.0.2 to set up
persistent SSH tunnels to be reused by future SSH sessions.
Then do ssh root at 172.30.0.1 pfctl -s Sources & ssh root at 172.30.0.2
pfctl -s Sources to generate the source tracking report from both
You can use nmblookup -A, avahi-browse -tar, OUI lookups, etc to find
out more about the computer with the IP address causing the problem.
Between throttling back tor on pony and getting someone to stop
torrenting at the space, the link characteristics of the sonic.net
line have improved significantly:
More information about the Rack