It is quite often to use bridge to “connect” different interfaces (easiest example on many 5-ports devices - ether 2, that is master for ether3..ether5, and wlan1). It is good idea to set IP address on bridge since every interface communicates with it, so this is nice-looking approach. But I suspect this is bad idea when it comes to performance.
So the question is: when I use bridge to connect ether ports and wlan interface, should I set IP on bridge or on ether port as of performance? I think setting IP on ether will be faster that on bridge, but am I right?
It has to be set on the bridge, once you create one and add slave ports to it, you should “forget” about the slave interfaces, as now they belong to a new device, the bridge.
You’re right, a software bridge will consume more CPU resources than using a “switch” group. But IPs should always be assigned to the “master” device, no matter if software based (bridge) or hardware based (switch group).
The idea is clear. But after all, how much extra CPU will I spend on ‘make it look ideally’ approach, and won’t it increase network latency as well?
I do understand that have IP on bridge allows me to have LAN router address always up (like Loopback approach on Cisco devices), but I have no idea how significantly it impacts the router.
Surely it’s OK to spend extra CPU on these multi-cores but we talk about 2011, 951, 941 series (not to mention older one) we’d better save CPU cycles, am I wrong?
Well you can do the test and tell us if there is any difference (i don’t think there is any difference).
If you are concerned about performances, try to use a network setup avoiding bridge if possible (wlan1 as private IP + DHCP server + NAT/masquerade, for example).
The cpu itself will not be a bottleneck until you start firewall on the bridge. The problem will make the fact that there is normally just one single line between the switch chip and cpu that has to transfer all the traffic in and out…