RFC 3107 defines procedures for having Border Gateway Protocol (BGP) allocate labels to routes between BGP peers. This technique is useful in cases where Multiprotocol Label Switching (MPLS) networks must scale. RFC 3107 operation can be used to isolate much of the routing data that exists in an MPLS access domain from the core network. By implementing RFC 3107 at the aggregation point, where access networks are aggregated toward the core, BGP label allocation eliminates the need for core devices to learn all of the prefixes in the access domains as routes are summarized.

In the case of an RFC 3107 edge device used to scale deployment of MPLS services, the edge device receives a packet that has already had two labels applied. (These were appended by the device originating the MPLS service, such as a pseudowire.) The outer label again identifies the label-switched path (LSP), and the inner label identifies the MPLS service. In this case, the RFC 3107 edge device replaces the outer label with two labels, generating a three-label stack. The now outermost label is used to switch the packet across the core between RFC 3107 BGP peers. The second (middle) label is used to direct the packet towards the final edge device in the LSP (once it exits the core network), and the third (now innermost) label is the MPLS service label.

The BGP Unified MPLS iBGP Client feature extends the core MPLS across all access/aggregation and edge/core areas to allow for an end-to-end tunnel LSP, as shown in the figure below.

Figure 1. Unified MPLS iBGP Client

The figure above shows:

• RFC 3107 (internal BGP [iBGP] + label) is used as the inter-BGP MPLS domain protocol to carry all prefixes and labels.

• Unified MPLS clients (access and aggregation routers) will run iBGP + label toward the Area Border Router (ABR).

• Unified MPLS ABRs (edge routers) will be iBGP route reflectors set with next-hop self enabled on all sessions.

• A pair of ABRs separating two intra-MPLS domains will have a unique cluster ID.

• The BGP cluster list mechanism will prevent routing loops.

BGP Prefix Independent Convergence (PIC) is the technology that enables RFC 3107 procedures to be implemented with dramatically improved reconvergence characteristics. Prior to BGP PIC, BGP convergence was slow, potentially resulting in minutes of outage. BGP PIC brings convergence into the range of 50 to 300 milliseconds, depending on topology, with no additional configuration required. BGP PIC is an algorithm enhancement implemented entirely within one routing device, so there are no interoperability issues with non-BGP PIC devices, just improved performance.

The basis of operation for BGP PIC is that the BGP routing process is modified to calculate not only the primary (best) path, but also a repair path in case this primary path to the BGP next hop becomes unavailable. Once the route to a primary next hop fails, the forwarding mechanism of the router points all next hops to the new repair path by updating just a single pointer. This process is quicker than doing a prefix-by-prefix calculation, as with the new mechanism, merely a single pointer must be updated for all the paths that will use that new next-hop address. This function of updating a single pointer shared by all prefixes using the same next hop makes this feature prefix-independent.

A Cisco 7600 series router locally allocates labels for /32 prefixes learned from a remote provider edge (PE) router via a BGP send-label session, despite the prefixes not being advertised subsequently to any other BGP peers. This BGP label will then be used to stitch Layer 2 frames received on an ingress card on the Cisco 7600 series router and map an outgoing label to an egress card for transport through the MPLS cloud to the remote PE router. As a result, you can set up a pseudowire between two or more Layer 2 attachment circuits connected to either of the PE routers. Assigning the local label for an unadvertised BGP-learned /32 prefix is essential to setting up an LSP between the remote PE routers for routing Layer 2 frames.

Use the bgp mpls-local-label command on the Cisco 7600 series router to enable or disable local label allocation for unadvertised /32 prefixes.

Note	Toggling the bgp mpls-local-label command will cause a session flap of all peers configured under that address family. A warning message also will be displayed to notify you that additional labels are required for /32 prefixes.

1.    enable


2.    configure terminal


3.    router bgp as-number


4.    address-family ipv4


5.    bgp mpls-local-label


6.    end

Device> enable

Device# configure terminal

Device(config)# router bgp 100

Device(config-router)# address-family ipv4

Device(config-router-af)# bgp mpls-local-label

Device(config-router-af)# end