Interconnection AS models for MPLS L3VPN

MPLS L3VPNs, based on well-known technologies such as MPLS and IP, are currently the most commercially widespread VPN offerings, thanks to their age and simplicity.

MPLS L3VPNs were first specified in 1999 in RFC2547 and subsequently updated by RFC4364 in 2006. In the latter version, paragraph 10 describes the different ways in which two VPN sites can be interconnected when they are geographically located in separate autonomous systems (AS). These autonomous systems may be networks belonging to different providers, or networks belonging to the same provider, but carrying heterogeneous services.

Previously, an MPLS VPN service could only be contained within a single AS. VPN inter-AS now enables multiple ASes to form a continuous, seamless network between VPN sites, overcoming geographical coverage gaps to connect customer sites.

In L3VPN MPLS, PE routers exchange VPN routes using the MP-iBGP protocol, either directly with each other, or via a route-reflector (RR) for scaling purposes. This protocol cannot be established between two ASes, so other techniques will have to be used, while respecting the choice criteria induced by AS connection and service constraints such as security, quality of service, convergence, availability or scaling.

This article describes the different models for inter AS VPN, i.e. interconnecting VPNs from different ASes. These models, called options, are 4 in number:

• Option A by direct connections between VRFs declared on the edge routers of each AS. This option is often referred to as "VRF to VRF";

• Option B involves redistributing VPN routes via MP-eBGP from one AS's edge routers to the neighboring AS;

• Option C uses a multi-hop MP-eBGP session to redistribute VPN routes, while end-to-end MPLS continuity is achieved between ASBRs using label exchange protocols;

• Option D is a hybrid solution that combines the advantages of options A and B.

A table of acronyms and abbreviations is provided at the end of this article.