MPLS has redefined the core of service provider networks by streamlining and expediting forwarding logic. Like layer 2 technologies before it, Frame Relay and ATM, it can carry any higher level technology such as IPv4, IPv6, Ethernet, HDLC and so on, via AToM, Any Tranposrt over MPLS. It shares the advantage with IP of being able to carry any payload and thus has become dominant, likewise.
An ingress PE, an MPLS ingress router on the provider's edge, attaches a label to an ingress packet and transports it over a common infrastructure to the egress PE, on it's way to it's destination IP. Since a label is used for forwarding across the core, BGP is needed only on edge routers to supply external prefixes such as customer or Internet routes. The cores need only the next-hop IP of the egress PE, which can be distributed via OSPF or IS-IS, requiring much less overhead!
As in routing, a forwarding table is used to store label mappings. The incoming label is swapped for an outgoing label at each MPLS router. This process is computationally simpler than analyzing a 4-octet IP address. Routers can push beyond 40Gbps of packets with the help of ASICs so speed isn't the main benefit of MPLS any more.
It also provides terrific support for IP over ATM integration. Prior to MPLS we had RFC 1483 define the AAL5 method, whereby each circuit had to be configured manually. There was LANE, LAN Emulation, which made the entire ATM core path between two LANs seem like a bridged Ethernet segment. As well there was MPOA, Multiprotocol Over ATM, defined by the ATM Forum, providing the tightest integration. And while they all provided a workable solution, all of them were quite cumbersome to deal with.
Prior to MPLS, a meshed customer set of sites needed a point to point link between all of the sites, requiring hundreds of links. MPLS provides optimal traffic flow, whereby the customer's edge router, CE, connects to the PE, and the MPLS core provides the meshing.
It also solves key problems for VPN users. Early VPNs used the overlay networking model, whereby a private network was created atop a common infrastructure. There was no peering with the provider's devices, each router in the customer's edge was peering with every other customer edge router. This required tedious and lengthy configuration of all sites for each new site. As well GRE tunnels can be used to provide an overlay network on layer 3 instead. The Peer to Peer model was an alternative, but it required tremendous amounts of access lists and route filtering to provide privateness required by VPNs. MPLS solves all this with the MPLS VPN application. It creates a PE-CE peering relationship, no other links are needed between sites. Customer routes are kept separate from others via the VRF (virtual routing/forwarding table) on PEs. Adding a new site requires the creation of a peering arrangement instead of reconfiguration of all the sites.
A final advantage of MPLS is traffic engineering. It allows for the utilization of under utilized links. Links that are not least-cost paths can now be used to forward traffic without the need to configure each router along the path. Only the head-end router is setup, which directs, via labels, the traffic down the entire path!
Isn't MPLS amazing? :)
