Performance problems on vlan bridge

We have been having some performance problems with our linux bridge. I have no idea where to look to start diagnosing the problem, so I'll explain what we've done and hope that you might tell me where to start looking. Almost everything appears to work fine, but when I try to stream music using WMP or mp3 streaming, I can only stream about 30 seconds before it cuts out. I used to be able to stream for hours.

We used to have a firewall which seperated our wireless network from the rest of the network. It is a dual 2.8ghz xeon with two Intel GB network cards. It used to have two interfaces, eth0 which was the default gateway and eth1 which was the wireless subnet.

The Problem:
We had a large amount of broadcast traffic on the network (often about 1Mpbs), and occasionally would have network storms where we would have 50-60Mbps of broadcast traffic (clearly not desirable for a network specifically for wireless clients.) We considered subnetting the network, but we'd like to allow people to register once with our wireless network and be able to roam anywhere we have wireless APs.

Our solution (well, if we can get it to work):

One day I had an epiphany: we could use a linux bridge and use iptables to block unwanted broadcast packets from going between buildings.

So I started setting up one vlan per building specifically for wireless and added them to br0. (I only got 2 buildings done before I noticed the problems though)

now the firewall has eth0 pointing to the internet and brctl show looks like this:

bridge name     bridge id               STP enabled     interfaces
br0             8000.00e0812a0540       no              eth1.307
                                                       eth1.336
                                                       eth1.6

where eth1.6 is everything I haven't done yet, and 307 and 336 are the buildings I separated off. I haven't added any iptables rules to block anything yet.

So, as I said in the beginning, everything but streaming music seems to work. The load average on the router/bridge is very low (ususally below .03) as is the cpu percentage. There isn't a lot of traffic this time of year, so I expect that the problem will get worse if we keep things as they are when students return in the fall.

The number of conntracks isn't anywhere near the limit we set. There is plenty of memory free...

Any ideas about how to find the problem would be appreciated.

Ethan Sommer
UNIX Systems Administrator
Gustavus Adolphus College

Maybe you are just hitting the limit of the transmit queue on the ethernet device. Look at the packet statistics to see if there are massive drops. Maybe the device has some watchdog or other function that is interfering. Some wireless devices do perodic scans and block all traffic.