[c-nsp] Internet in VRF

[Mark Tinka]

On 5/May/15 11:02, Adam Vitkovsky wrote:
>
>
>
> Hi folks,
>
> Assuming you have more than one AS-exit and you don't have full-mesh
> between all BGP speakers, then how do you get the alternate/backup
> AS-Exit paths for Internet prefixes to all the PEs please?
> Although I admit that the convergence times of Internet services might
> not be a cause for concern so a minute of downtime might be acceptable.

We have several transit interconnects as well as lots of peering in
Africa and Europe. These are all separate devices as unique PoP's are
involved.

Each PoP has an RR cluster, handling iBGP routing for that PoP. Of
course, full-mesh iBGP between the RR's across the network.

We allow the RR's to do best-path selection, and the edge routers end up
with the best path as seen from its point of view. This works very well,
and because we have a standard minimum interconnect bandwidth for
transit and peering, we never influence customer routing with prepends
or any of that (although we do give our customers BGP communities so
they can do this to their own routes if they want). In the end, we we
have evenly balanced traffic across all our transit and peering
interconnects, without doing any advanced traffic engineering in our
network. Admittedly, this works well because we peer most of our
traffic, and only a small portion is handled by transit interconnects.

It does not concern us to be able to hold the full table of every
peering relationship we have in the edge (which is where "Internet
VRF's" would come in handy). We offer our customers the best route
possible to any destination. If they want the full routing from a
specific AS or exchange point, we can help them arrange an EoMPLS
circuit between them and that AS or exchange point, and they can peer
directly without us being involved at Layer 3 - which is another service
we sell.

Mark.