In my current topology, I have two locations linked by a trunk.

Everything on one side of the trunk is Site A, everything on the other is Site B

In the example above, for to talk to it must traverse the trunk because the gateway is on Side A.

My problem is that I have machines on the same subnet in both locations, but I dont want to traverse the trunk if my traffic stays within the site.

I'm sure there is something basic I'm missing here, and OSPF seems like it may be a more complicated solution that I need. Do I make both sides of the trunk gateways? Both are capable of layer 3.

  • Is it really necessary to have the same subnet/vlan on both sites? I understand changing may be a PITA, but the only options are really expensive ones.
    – Ron Trunk
    Nov 6, 2014 at 0:44
  • @Ron Currently yes it is necessary. Do you mean expensive $$$ wise? Cause they are currently a blackdiamond and a brocade vdx 67xx series (both layer 3)
    – Party Time
    Nov 6, 2014 at 1:46
  • What model of switches do you have now for your core and aggregation?
    – Ron Trunk
    Nov 6, 2014 at 1:47
  • Core is blackdiamond 8810, aggregation are Brocade vdx67xx series
    – Party Time
    Nov 6, 2014 at 1:49
  • I'm not that familiar with Extreme, and they may have some proprietary feature to help with this problem. But anything you do is basically a hack -- extending vlans between data centers is something you should always try to avoid, or you'll end up as you have with "tromboning" vlans (host on one side, gateway on the other). Basically, the only real fix is an architectural change.
    – Ron Trunk
    Nov 6, 2014 at 2:25

2 Answers 2


Create two gateways, .1 for Site A and .2 for Site B. Then setup all the machines in Site B to use the .2 gateway and all the machines in Site A to use the .1 gateway. The one rectangle with nothing in it would be the gateway and Then the one rectangle on the left labeled L3 gateway, this will stay the gateway for and

So in this setup you would have to re IP the circle in the top left with and to something else.

  • Thanks, this was my initial thought, two gateways and a static route between them. The problem is that there are hundreds of machines, and a dozen vlans in both sites, the diagram is just a vast simplification. Changing the default gateway for every machine in a site will introduce many headaches. Was hoping for an alternative.
    – Party Time
    Nov 5, 2014 at 22:56
  • I've never used any extreme hardware so I don't know if it is possible, but if your routers were Brocades you could setup vrrp-e and use the shortest-path forwarding feature. What this will do is allow both routers in both sites to act as the gateway at the same time. Arista also has this feature, but it is called virtual arp, here is a detailed description of how that works blog.ipspace.net/2013/06/…. I wouldn't be surprised if extreme had the same thing. Nov 6, 2014 at 4:58
  • You could also handle all the gateway changes with DHCP. List out all the machines mac address in one site to get specific IP address with the .1 gateway and then list out all the macs for the machines in the other site and have them get the .2 gateway. Nov 6, 2014 at 5:02

What you are asking will not work because the fundamental idea of a network segment is that everything in a network stays local, and everything outside the network gets route out via the gateway. You could make it work by overlaying a virtual layer 2 network on top of the existing routed network, but that is a painful, ugly hack to achieve what you are looking for.

Your choices are:

  • The two sides of the routed network necessarily need to be separate VLANs, subnets, etc, and route between them like normal (for example, 10.0.100/24 on the left, 10.0.200/24 on the right)
  • You create a switched network, which you can carve up, and assign layer 2 ports to VLANs. Your trunk is now a layer two link, and both backbone devices are operating at layer 2, with layer 2 down to the edge
  • You setup a routed network between the two routers and overlay a virtual switching fabric using OTV, tunnels, multicast and other complex technologies

Guess which one will be easiest (hint, keeping the two sides of the layer 3 boundary in separate network segments)

  • While wait Party time is asking certainly isn't common, granted VMs are making this more common, there is no reason for it not to work. Setting different gateways in different sites, while it would require extra server config, wouldn't require anything special on the network side. And some networking vendors are implementing features to vrrp which make this work as well. Nov 7, 2014 at 20:13

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.