Comparison of IBGP topologies

Abstract

BGP is the de facto interdomain routing protocol in today’s Internet [1]. Its role is to allow ASes to exchange routing information with each other. It is divided in two components : eBGP rules the communications between ASes, while iBGP allows one router to transmit routing information received from its eBGP peers to the other BGP routers of its AS. The original specification of iBGP requires a fullmesh of sessions between BGP routers to prevent loops of the advertisements. This constraint means that there will be (n*n-1)/2 iBGP sessions maintained inside the AS, which does not scale well in the case of ASes. One alternative that has been proposed is to use route reflection [2]. In this scheme, one (or more) BGP router plays the role of route reflector for some of the others, which are its clients. This router reflects the routes it receives to its clients, and announces its routes and those of its clients to its non-client peers. With those rules, a client router must only maintain one session with its route reflector, and the iBGP topology is much lighter. However, this alternative has drawbacks. For example, as route reflectors have to go through their decision process before announcing routes to their clients, these only receive the routes selected by the reflector. The selection performed by the route reflector may differ from the selection that its clients would make in the case of an iBGP full-mesh, which can lead to sub-optimal routing. This paper presents a method to evaluate iBGP topologies with route reflection, and to measure their performances compared to the full-mesh, in order to choose the best one. First, we describe the procedure and the criteria used for the comparison. After that, we briefly present the result of an application of this method to the GEANT network [3]. This analysis allows us then to isolate some of the factors that are influencing the quality of an iBGP topology.