Nic Bonding with bridge

If you make a bond of two interfaces the switch into which they are plugged needs it's interfaces to be bonded TOO, otherwise you just have two interfaces tied together.


you describe this:
            +-------TUN/TAP for virtual machine
            |
System |(br0)
            |                    +eth0-----+
            +---bond0 ---+               |
                                 +eth1-----+
                                                 |


What should have been:

            +-------TUN/TAP for virtual machine
            |
System |(br0)
            |                    +eth0-----+
            +---bond0 ---+               +bondX---+
                                 +eth1-----+               |
                                                                  |

For cisco these bonds are named port channels. Providing fallback (to less interfaces per channel) and more bandwidth then one interface can provide.  BOTH sides need to be bonded like this.

The reason you get two answers stems from the fact that your switch (or equipment at bondX 's site doesn't known those interfaces belong together.

You don't necessary need to have the switch bond the ports together.  It depends on what you are attempting to do.  

If you want to increase both inbound and outbound bandwidth, then both NIC's need to be connected to the same switch and the switch needs to bond them together.  However if the switch fails, you loose access to the computer.

If you are going for availability then each NIC should be connected to a different switch.  The computer will send traffic out on both NIC's, but only one NIC will be used for inbound traffic.  Say you have eth0 connected to sw0 and eth1 connected to sw1.  If either switch fails, you still have connectivity to the computer.

One setup gives you more inbound and outbound bandwidth, but availability is less.

The other setup gives you more outbound bandwidth only, but more availability.

Hi,

For giltjr :

I need to set up a bond on two NIC card and after that I need to bridge my bond.

I need the brige because my server is a VirtualBox server with many Guest Os on it. Every guest Os is configured with a Bridged interface because I need my Guest OS in the range of my current network.

For Noci :

I don't see the purpose of switch configuration.. I only need the configuration method for this



            +------- ETH0                        |--vbox0-- VboxGuest1 (OS)
            |                   |                       |--vbox1-- VboxGuest2 (OS)
System |                   |--bond0----br0-|--vbox2-- VboxGuest3 (OS)
            |                   |                       |--vbox3-- VboxGuest4 (OS)
            +------- ETH1                        |--vbox4-- VboxGuest5 (OS)

Note : vbox0 is the GuestOS interface that is affected to my bridge.



I need availability.

Thank you  

What I would suggest is that you use a different IP subnet for the guest OS's and route to them.  

What virtualization software are you using?

Ok...
If it is a hub it acts like a cable, so you will receive all data on all circuits.
A switch work differently, broadcast go to every interface, unicast go to any interface until the address of the interface is KNOWN in a cam-table/mac-table (name depends on switch manufacturer).

If the switch doesn't know it is a bonded link then the interface will flap to the interface that has seen the latest packet coming from your virtualised system(s).

If you use a switchcluster like two CISCO 3750 or likewise switches you reduce your SPOF on switching if you use one interface on one switch and another on the other switch. If one switch fails you can continue over the other.

(The upstreamlinks need to double too).

IF you don't want to use that than DONT bond them, hang them in a bridge and setup for spanning tree.
(brctl stp br0 on; ).

It is a pity the Fast Spanning tree is not supported on linux.  That would make failover near instantanous and forwards with wirespeed in stead of avg. 45 second delay delay.

This seems to be an RSTP daemon.

http://git.kernel.org/?p=linux/kernel/git/shemminger/rstp.git;a=commit;h=358260ffcc3985847f73f0882a09406f6d388dd9

Hum.. just to be sure that everybody is understanding me well.

My server is running on Archlinux dist. My Virtualization software is VirtualBox 2.0.0 Headless

I already configured my Archlinux server with NIC bonding (Without bridge) and everything is working fine.

So I really don't understand why you're talking about switches, switchcluster and hubs since I just want to BRIDGE my bonded interface (wich is working great)

Oh and I don't want to make any routing for my vm guest os.

Thank you for your help, it's appreciated.

You may need to check with Archlinux, I have seen that trying to using a bonded interface as a bridge has had problems on other Linux distributions.

What samples I have seen have had multiple interface defined on the bridge, the bonded interface and the virtual "tap" interfaces as VirtualBox calls them.    

You may want to look at VirtualBox 2.1, according to the documentation they have change the way you do this:

"With VirtualBox 2.1, net filter drivers were also added for the Windows and Linux hosts, replacing the mechanisms previously present in VirtualBox for these platforms; especially on Linux, the earlier method required creating TAP interfaces and bridges, which was complex and varied from one distribution to the next. None of this is necessary any more."

That is if upgrading is an option.

I will also say that the examples of networking configuration I have seen for VirtualBox is a routed enviroment.  I know you said you did not want to do this, but generally routing to virtual environments is more efficient as it cuts out all of the noise that you get from a bridged/switched network.  You may also need to look at running stp protocol because you basically are running a "switch" and it will appear to be a loop to the network.

All right thank you, I'll take a look at this