sBGP: A hybrid SDN approach to interdomain routing

This PhD proposes an SDN based architecture for Internet routing based on BGP as the principal southbound interface protocol.
Today's Internet service providers face ever increasing customer demands for performance, capacity and availability, whilst ISP networks themselves are under increasing attack from both malicious and unintended sources. At the same time, the exhaustion of IPv4 addresses and the accelerating transition to IPv6 is a source of increased churn and complexity in IP address management. Well publicised recent major outages in backbone ISP networks underline the fragile state of current network architectures, and highlight the need for rigour and care in the management of the myriad of interconnections between ISPs. It is these challenges of complexity and scale that have driven much of the academic work on future network architectures, often under the broad heading of SDN (Software Defined Networking). Whilst SDN is currently a major research area within the networking community there are few instances of production quality network equipment based on SDN principles in deployment today which offer even basic technical capability to match the core routers used in the Internet. The predominant SDN paradigm today is OpenFlow, however it is not apparent that the flexibility of OpenFlow will properly address these issues any better than current solutions based on a conventional IP forwarding plane, coupled with classical control-plane semantics. In particular, the issues of Internet complexity highlighted above demand attention to the control-plane architecture, rather than increased richness in the forwarding plane semantics.

This proposal presents a new network architecture which incorporates aspects of both SDN and classical BGP based architectures. It extends other SDN work in two ways: a) by adopting an existing protocol ( BGP) as a new southbound protocol, and b) by implementing a 'hybrid'2 approach to SDN for routed networks at layer 3. It extends classical BGP based internet architectures by retaining the network elements (routers), and the associated control software without modification, whilst removing from those routers the responsibility for autonomously determining local forwarding policy, and collectively with other peer routers determining system-wide routing behaviour and communicating with external networks (other Autonomous Systems). In doing so it keeps the high performance and scale of current core router hardware whilst removing the dependency on core router control software to implement optimal network wide routing policy. The benefit of retaining BGP is that it provides exactly the level of control over the forwarding plane which is required for the task, and existing implementations are known to have the scaling and performance capacity to support at least the current internet demands. This proposal is based on the premise that an 'SDN like' analysis of Internet (routing) architecture is clearly applicable both from the standpoint of the SDN community - "if SDN is not applicable here, then where?", and also the Internet industry, where the principle of automating network operations ('devops') is central to most any strategies for improving Internet management, and what else is 'automating network operations, using software' than SDN by another name?
The novelty of this proposal is the explicit selection of BGP as a southbound SDN protocol. Implementations of network systems based on SDN and supporting BGP externally, or as a northbound interface (or east-west interface) can be found, albeit in small number, and in most cases not even in the classical role of core Internet routing, however the consideration of BGP as a first class SDN protocol alongside OpenFlow, and the consideration of hybrid SDN solutions is a very lightly touched topic in the field of SDN research.