Crafting optimal and resilient iBGP-IP/MPLS overlays for transit backbone networks

The Internet is a collection of interconnected Autonomous Systems (ASes) that use the Border Gateway Protocol (BGP) to exchange reachability information. In this regard, BGP stability and scalability in the inter-domain scope have been matters of major concern for many years, and network engineers have been applying several techniques to cope with these issues. BGP is also used intra-domain (internal BGP - iBGP), to disseminate reachability information inside each AS, and works together with the Interior Gateway Protocols (IGPs) such as OSPF or IS-IS, to build routing tables. Route reflection is a widely adopted technique to tackle BGP scalability in the intra-domain scope, and choosing which routers will play the reflector role and which BGP sessions will be established among reflectors and clients (i.e. the routers which are not elected as reflectors), building an overlay of iBGP sessions, is known as the iBGP overlay design problem. The design of an optimal iBGP overlay is known to be a NP-Hard problem, and we proposed solutions for pure IP networks (i.e. best effort traffic forwarding) in our previous work. However, most Internet providers implement their backbones by combining IP routing with MPLS (Multiprotocol Label Switching) for QoS-aware traffic forwarding. MPLS forwarding incorporates traffic engineering and more efficient failover mechanisms; under this traffic forwarding paradigm, the design of traffic-engineered Label Switched Paths (LSPs, also referred as MPLS tunnels) shall be combined with the aforementioned iBGP overlay design. The present work introduces a coordinated design of both the iBGP overlay and the IP/MPLS substrates. Our contribution is the proposal of an optimal and resilient topology design for an IP/MPLS Internet backbone, which takes advantage of traffic engineering features to optimize the demands, while guaranteeing iBGP overlay optimality. We present a complete solution for a real world scenario, and we study the scalability of the solution for synthetic topologies, achieving encouraging results.