This chapter discusses MPLS operation over interfaces where labeled packets are sent encapsulated in Layer 2 frames (Frame-mode MPLS operation).
Label Switch Routers (LSRs) use Label Distribution Protocol (LDP, an IETF standard) or Tag Distribution Protocol (TDP, a Cisco pre-standard) to exchange IP prefix-to-label bindings. A Label Information Base (LIB, also called a Tag Information Base [TIB]) stores these bindings, which are used to build the Forwarding Information Base (FIB) entries in ingress Edge-LSRs as well as Label Forwarding Information Base (LFIB, also called Tag Forwarding Information Base [TFIB]) in all MPLS nodes. Cisco IOS supports both label distribution protocols, and you can use both in the same network, even on separate interfaces of the same LSR.
The tag-switching ip or mpls ip interface configuration command enables MPLS on a Frame-mode interface. In IOS releases supporting LDP, the desired label distribution protocol must be selected using the mpls label-distribution command. These commands start TDP or LDP on the specified interface. TDP/LDP finds other LSRs attached to the same subnet through TDP/LDP hello packets sent as UDP packets to broadcast or multicast IP addresses. When the neighboring LSRs are discovered, a TDP/LDP session is established using TCP as the transport protocol to ensure the reliable delivery of label mappings.
The IOS implementation of LSR on Frame-mode interfaces assigns labels to IP prefixes as soon as they appear in the routing table, even though the LSR hasn't received a corresponding label from its downstream neighbor, because it can always perform a Layer 3 lookup if needed. The router is thus working in independent control allocation mode, as opposed to ordered control allocation, where a device assigns only labels to those prefixes where a downstream label already exists in the LIB.
When running MPLS over Frame-mode interfaces, a Cisco router immediately propagates allocated labels to its TDP/LDP neighbors. This distribution method is called unsolicited downstream distribution, as opposed to downstream on demand distribution, where the upstream routers explicitly ask the downstream routers for specific labels.
A Cisco router acting as an LSR stores all label mappings received from its TDP/LDP neighbors. This storage method is called liberal retention mode as opposed to conservative retention mode where the LSR stores only labels received from its next hop downstream routers. The liberal retention mode uses more memory but enables instantaneous TDP/LDP convergence following the routing protocol convergence after a failure in the network.
After the LSRs in an MPLS network have exchanged label mappings, the ingress LSR can label the incoming data packets. The ingress LSR inserts a label stack header between the Layer 2 header and the IP header. For unicast destination-only IP routing, the label stack header usually contains only one label, but the MPLS architecture also supports stacked labels used by other MPLS applications, such as traffic engineering or Virtual Private Networks. The labeled packets are distinguished from the unlabeled IP packets by using different ethertype codes on LAN media and a different PPP Protocol field value.
Network designers usually consider MPLS only as a technology that allows seamless integration of IP routers and ATM switches or enables additional applications, such as MPLS Traffic Engineering or MPLS/VPN. They usually don't realize they can gain significant simplifications by deploying MPLS in any network that runs BGP as its exterior routing protocol. Deploying MPLS in a network running BGP allows you to remove BGP routing from core routers (non?Edge-LSRs), resulting in a network design that is more stable, requires less memory on the core routers, and prevents high CPU utilization due to BGP update processing on the core routers.