Internet failover configuration between two MPLS sites for branch offices

[gqadirmirza]

Hi Everyone

I have come up with the requirement where i want to create the automatic failover for the internet between two of our datacenters. Here is my topology looks like:

1- Each datacenter has its own internet connection 

2- Each datacenter has its own core switch

3- Each data center has two CE routers

4- Both datacenter are connected to same MPLS cloud via EBGP peering with PE routers

5- All branch offices are also going to the same MPLS cloud via provider edge device at each location.

6- DC1 is the primary internet connection for branches right now.

7- OSPF is configured between core switch and the 2 CE routers for both Datacenters.

8- Redistribution is happening two way BGP into OSPF, OSPF into BGP. 

9- Right now DC1 is feeding the default route into BGP via OSPF. 

Requirement:

Want to make Datacenter2 (DC2) as backup of internet for all branch office in case DC1 internet fails. I want to make this failover automatic so that without doing any manual configurations i can redirect all branch office to DC2 in case of DC1 internet fails and vice versa.

Any ideas, configuration options will be highly appreciated. All of our WAN equipment is Cisco.

Thank you.

[Robert Hillcoat]

You can advertise your secondary default route with a longer AS path. The branch sites will continue to use the primary default route until this is no longer available. When you have a failure at your primary datacentre, the secondary default route will be available with the longer AS path. 

The only issue you might have is with your primary datacentre CE1 and CE2, when your OSPF default route is being advertised back from the core switch after a failure, CE1 and CE2 won't install the default route in the routing table because of the Administrative Distance (AD) of the secondary default route being advertised from datacentre 2.

You can fix this by changing the OSPF AD to a value lower than eBGP AD of 20, doing this on the CE1 and CE2 would be fine. 

Test this out on a lab environment before deployment. 

Hope this helps. 

[gqadirmirza]

Robert 

Thank you very much for your response. You are right  and these were the points which i was thinking on. i will try this out in the lab and if i get this config work in the lab then will try that in production. 

Appreciate your help again.

Thanks