BGP Redistributed Route Failover

[da.newman]

I'm looking for some routing design input/guidance on the best was to approach implementing routing failover for the following situation.

We have a primary and secondary datacentre which are connected to a private MPLS network which all of our edge sites connect too. In the primary and secondary data centres we ahve BGP advertising the relevant networks/subnets that are available in the two datacentres and the default route going to the primary datacentre. We also have a number of other WAN links to third parties in the primary datacentre behind a firewall.
Currently the primary and secondary datacentre routing cores (numbers 1 and 2 on the diagram) have a route map in BGP setup to resdistribute certain static routes which point to the firewalls into BGP based upon "tag ###" in the static route.

We recently deployed connectivity to an external MPLS network with a primary and backup MPLS link going to the primary and secondary datacentres. The backup external MPLS link is the first non internal MPLS link deployed to the secondary data centre which is not internal. The third party site is advertising BGP prefix's to our primary and secondary firewalls in our primary and secondary datacentres and we are advertising only selected /32 routes out to the third-parties sites via the MPLS.

Currently the prefix's advertised by the third-party site are advertised via the redistributing static routes for the destination subnets into BGP for the internal MPLS network.

Currently all the traffic in the network destined for the third party sites goes via the primary datacentre.

What would be the best way mechanism in BGP to advertise a secondary route to the third-party site via the secondary datacentre in the instance there is a failure with primary datacentre?

[trfinkenstadt]

Hello,

To me, it seems like you need some sort of failure detection in the primary datacenter for those selected /32 static routes.  You probably want to use IP SLA on the static routes to detect problems further downstream (ping the interface to the external MPLS network?).  In addition, you would announce the same prefixes from the secondary datacenter with a couple of as-prepends tacked on to the announcements.  Depending on your carrier's policies for communities and what not you could use that to change their local preferences.