Enforcing iBGP convergence

BGP routers within an autonomous system (AS) exchange their inter-AS routing information via the internal border gateway protocol (iBGP). Within an AS, every BGP router needs to maintain an iBGP peering session with every border BGP router. This peering scheme fails to scale due to the large number of iBGP peering sessions required. Current solutions to this scalability limitation divide the AS into clusters, with a distinguished router, know as the reflector, acting as a representative of the cluster. Clustering, however, introduces routing anomalies, such as permanent routing loops and failure to reach a stable route to the destination. Furthermore, these anomalies are worsened by the multi-exit discriminator value used by BGP to differentiate multiple links connecting the same pair of AS'ms. In this paper, we present a simple enhancement to iBGP that prevents these routing anomalies. It requires minimal overhead, and contrary to other proposed solutions, preserves the efficiency of iBGP by having each reflector disseminate only a single path to each of its peers.