How Label Switching Works

The multiprotocol label-switching architecture is defined in RFC 3031.

In general terms, label switching in the data path is similar in concept to VC switching in traditional ATM switches. At each switching juncture, the label value is rewritten with an outgoing label, similar to ATM cell switching. The main difference is in the control plane, because LDP is used for label setup as opposed to ATM forum signaling.

The general label that is used for label swapping, also called a shim header, resides in the label stack, as defined in RFC 3032, "MPLS Label Stack Encoding." The shim header is a 32-bit header added in front of the IP header, as shown in Figure 4-2. All link types, including ATM and Frame Relay, insert this special shim header.

Figure 4-2. Label Stack Entry

The fields in the shim header shown in Figure 4-2 are as follows:

• Label The actual 20-bit label value.

• Exp Experimental use. Although this 3-bit field is defined as reserved for experimental use in RFC 3035, it is used in MPLS QoS support in the DiffServ model.

• S Bottom of stack. This bit is set to indicate the last entry label stack. In a multiple label stack application, such as MPLS VPN, only the innermost label stack entry has this bit set.

• TTL Time to Live. This field encodes the time to live.

The 20-bit label is used for label swapping. This MPLS header or stack of entries appears after the data link layer headers and before the network layer headers. For this reason, the MPLS header is normally called Layer 2-and-a-half (Layer 2 1/2). It is used in frame-based data link layers such as PPP, Ethernet, and Fast Ethernet.

However, ATM and Frame Relay already implement a label-switching forwarding mechanism, swapping identifiers (VPI/VCI and DLCI) on a link-by-link basis. Therefore, it makes sense to use these identifiers as the labels swapped in MPLS. On ATM MPLS, the VPI and VCI fields or the VCI field alone are used as the label. This is defined in RFC 3035, "MPLS Using LDP and ATM VC Switching." In Frame Relay-based MPLS, as defined in RFC 3034, "Use of Label Switching on Frame Relay Networks Specification," the DLCI is used as the label. Figure 4-3 shows the different label stack entries using different Layer 2 protocols.

Figure 4-3. Encapsulation in Different Layer 2 Protocols

VPI/VCI fields or VCI fields alone are used as the label. Labels are applied to each cell, which makes label swapping look like ATM switching from a forwarding perspective. Both Label Distribution Protocol (LDP, as defined in RFC 3036) and Tag Distribution Protocol (TDP, the Cisco proprietary prestandard distribution protocol) are used for label distribution in ATM on the control plane.

On label switching controlled ATM (LC-ATM) interfaces, label packets are sent using null encapsulation, as defined in Section 6.1 of RFC 2684, in which there is no need for protocol identification. The label stack header is always attached preceding the network layer header before segmenting the packet into cells. In this case, the label stack header carries a null label because, as you know, the top label is carried encoded in the VPI/VCI or VCI only fields. Why carry this shim header if the label is encoded in cell headers? The main reason is to allow a label stack of arbitrary depth, just as on non-ATM links. Another function of the label stack is to carry the TTL and Exp bits. You need both of these fields if a packet is to be label-switched further by a nonATM-LSR. This function also provides transparency for the Exp bits. In summary, labeled packets must always have a shim. The complete encapsulation of ATM MPLS labeled packets is shown in Figure 4-4.

Figure 4-4. MPLS Encapsulation on LC-ATM

The ATM MPLS packet shown in Figure 4-4 contains a stack of one label entry, the MPLS label. This would be a packet terminating in an LSR or eLSR. In an MPLS VPN application, the MPLS packet inside the AAL5 PDU has at least two entries in the label stack even when it is carried on a label virtual circuit (LVC): an implicit-null label (since the MPLS label is carried in the VPI/VCI fields), and the BGP VPN label. This will be discussed in detail in the upcoming section "IP Virtual Private Network Services."

MPLS Operation Example

Let's look briefly at a very basic and general MPLS operation example, as shown in Figure 4-5. At this stage, this example applies to both frame-based and cell-based MPLS.

Figure 4-5. How MPLS Works

We can divide the process in two planes well-separated by MPLS:

• Control plane It starts with a routing protocol such as OSPF, EIGRP, or IS-IS, establishing reachability across the network and thus determining the Layer 3 topology. As soon as reachability is established when the routing tables are built, the LDP assigns or binds labels to each route and propagates this information to its neighbors.

• Data plane A packet comes into the network. The packet is a normal IP packet. It hits the edge of the MPLS network (an eLSR). A router does a lookup based on the IP address. As soon as a match is found, a label is identified. That label is affixed to the packet, and the packet is forwarded to the appropriate interface. The packet then travels to the next-hop LSR. That LSR does a lookup based on the label only. It finds a match based on that label in its forwarding table. An outgoing label is identified. That outgoing label is swapped for the incoming label, and the packet is switched out the appropriate port. This process or procedure is repeated hop by hop until you get to the edge of the network. At that point, the exit eLSR does a lookup based on the label, finds a match, determines that there is no outgoing label, strips off the incoming label, and forwards the packet on its way as a normal IP packet.

The things you need to notice here are that labels have only local significance, meaning that they are meaningful in the link only and not network-wide. That is normally the difference between an identifier, which is locally significant, and an address, which is globally significant.

A key point to remember is that MPLS provides a clean separation between routing (in the control plane) and forwarding (in the data plane), as opposed to normal IP routing, in which both routing and forwarding are almost indivisible.

A frame-based MPLS implementation and a cell-based MPLS implementation have some important differences:

• Labels On frame-based interfaces, the MPLS label is carried in the shim header, at Layer 2 1/2, between the Layer 2 and Layer 3 headers. On ATM MPLS, VPI/VCI fields are used as MPLS labels, and in Frame Relay, DLCIs are used as labels.

• Label distribution With frame-based MPLS, label distribution is Downstream Unsolicited. With cell-based MPLS, the label distribution is Downstream on Demand. Downstream on Demand is also called upstream-controlled because the VPI and VCI used as the label are scarce resources.

• Label spaces LDP on frame-based interfaces uses a per-LSR label space. On the other hand, an ATM-LSR uses per-interface label spaces because two different label-controlled ATM interfaces can use the same VPI/VCI pair (the same label).

• LSPs On ATM MPLS, LSPs are called label virtual circuits (LVCs).

• Penultimate Hop Popping (PHP) On frame-based MPLS, the exit eLSR can ask the penultimate hop to pop the top label. This functionality does not exist in ATM MPLS because cells cannot be sent without VPI/VCI acting as the label. In LDPv1, it is assumed that all interfaces except LC-ATM and LC-Frame Relay are PHP-capable.