With the explosive growth of the Internet and the growing deployment of layer 3 virtual private networks (L3VPN), the size of the Border Gateway Protocol (BGP) routing table has dramatically increased over the past years (in the 750K range in early 2007 counting Internet and L3VPN routes in a typical major Service Provider). Studies have shown that dynamics of BGP may cause several minutes of packet loss during network failures. This duration increases further as the routing table grows, as the traditional convergence oper- ation scales with the number of prefixes. In this paper, we present an alternate solution to provide BGP convergence and demonstrate that it is possible to limit the traffic loss period to sub-second for any failure occurring within the network of a service provider (SP) or on peering links with redundantly-connected peers. This covers the vast major- ity (if not all) of business models involving tight BGP con- vergence requirements. We term this alternate solution as BGP Prefix Independent Convergence (PIC) since it works by triggering an immediate and prefix-independent dataplane rerouting of the BGP destinations via the alternate path at In- terior Gateway Protocol (IGP) convergence time. We present experimental results of the convergence behavior based on a benchmark of a commercially available carrier router that supports the BGP PIC solution, and on BGP data provided by a Tier-1 ISP.
François, Pierre
[UCL]
