MPLS - Label Switching

Multi-Protocol Label Switching [MPLS] is similar to DiffServ in some respects, as it also marks traffic at ingress boundaries in a network, and un-marks at egress points. But unlike DiffServ, which uses the marking to determine priority within a router, MPLS markings (20-bit labels) are primarily designed to determine the next router hop. MPLS is not application controlled (no MPLS APIs exist), nor does it have an end-host protocol component. Unlike any of the other QoS protocols we describe in this paper, MPLS resides only on routers. And MPLS is protocol-independent (i.e., "multi-protocol"), so it can be used with network protocols other than IP (like IPX, ATM, PPP or Frame-Relay) or directly over data-link layer as well [MPLS Framework, MPLS Architecture].

MPLS is more of a "traffic engineering" protocol than a QoS protocol, per se. MPLS routing is used to establish "fixed bandwidth pipes" analogous to ATM or Frame Relay virtual circuits. The difference is arguable since the end-result is service improvement and increased service diversity with more flexible, policy-based network management control, all of which the other QoS protocols also provide.

MPLS simplifies the routing process (decreases overhead to increase performance) while it also increases flexibility with a layer of indirection. Here's a sketch of the process used by MPLS-enabled routers called a Label Switching Router (LSR):

• At the first hop router in the MPLS network, the router makes a forwarding decision based on the destination address (or any other information in the header, as determined by local policy) then determines the appropriate label value -- which identifies the Forwarding Equivalence Class (FEC) -- attaches the label to the packet and forwards it to the next hop.
• At the next hop, the router uses the label value as an index into a table that specifies the next hop and a new label. The LSR attaches the new label, then forwards the packet to the next hop.

The route taken by an MPLS-labeled packet is called the Label Switched Path (LSP). The idea behind MPLS is that by using a label to determine the next hop, routers have less work to do and can act more like simple switches. The label represents the route and by using policy to assign the label, network managers have more control for more precise traffic engineering.

Figure 4: MPLS label stack entry used to "encapsulate" IP Header

Label processing is actually a bit more involved than described above, since labels can be "stacked" (to allow MPLS "routes within routes"), and labeled packets have a time-to-live value (TTL), as shown in Figure 4. The TTL works essentially the same way TTL in an IP header works: each router hop decrements the value by one until it hits zero. The difference is that when an MPLS TTL reaches zero, the action is label dependent (so unlike with IP, the datagram may not be discarded and an ICMP "TTL Exceeded" message may not be generated). Nonetheless, label processing is the relatively simple aspect of MPLS.

A more complex aspect of MPLS involves the distribution and management of labels among MPLS routers, to ensure they agree on the meaning of various labels. The Label Distribution Protocol (LDP) [MPLS LDP] is specifically designed for this purpose, but it is not the only possibility. There are proposals to use RSVP [MPLS LSPS], BGP [MPLS BGP], and PIM [MPLS PIM] possibly "piggy-backing" label management information, so the use of more than one protocol for label distribution is expected.

Although infrastructure details such as label distribution are important to mention, for most network managers they will be transparent. More relevant to MPLS for most network managers is the policy management that determines which labels to use where, and not how the labels are actually distributed.

Wed, 23 July, 2003 13:07