Putting BGP on the right path: a case for next-hop routing

Abstract

BGP is plagued by many serious problems, ranging from protocol divergence and software bugs to misconfigurations and attacks. Rather than continuing to add mechanisms to an already complex protocol, or redesigning interdomain routing from scratch, we propose making BGP simpler. We argue that the AS-PATH, which lists the sequence of ASes that propagated the route, is the root of many of BGP's problems. We propose a transition from today's path-based routing to a solution where ASes select and export routes based only on neighboring ASes. We discuss the merits and limitations of next-hop routing. We argue that next-hop routing is sufficiently expressive to realize network operator's goals while side-stepping major problems with today's BGP. Specifically, we show that next-hop routing simplifies router implementation and configuration, reduces BGP's attack surface, makes it easier to support multipath routing, and provably achieves faster convergence and incentive compatibility. Our simulations show that next-hop routing significantly reduces the number of update messages and routing changes, and is especially effective at preventing the most serious convergence problems.