Multi-Protocol Label Switching
While almost all the words in its name have changed importance or meaning since multi-protocol label switching (MPLS) was invented, it has become an increasingly important technique in communications networks. Originally, it was seen as a faster means of packet forwarding than IP routing, using locally significant labels rather than globally significant addresses. Link-local means that it only has significance to the next LSR, which will swap to its own label and forward as indicated for that FEC.
Originally, IP was not dominant, so it could handle other protocols, but that need has largely disappeared except for telephony, which itself is evolving to Voice over Internet Protocol.
As MPLS developed, however, it has been displacing Asynchronous Transfer Mode (ATM), a method of forwarding fixed-length "cells" with locally significant identifiers, rather than variable-length packets with locally significant identifiers. It has been called "ATM without cells", as it does share the virtual circuit orientation of ATM as opposed to the datagram orientation of classic IP. ATM, however, had been growing increasingly complex, and offered little advantage over IP.
MPLS offered operational capabilities can be used to optimize the utilization of network resources and to enhance traffic oriented performance characteristics. 
Why MPLS vs. ATM or IP?
MPLS is not any faster to forward than other methods. What it offers is performance that is more predictable and failover than general IP. It also offers simplification by treating the members of a Forwarding Equivalence Class in the same way. Be aware that MPLS is not a simple technology, and is much more intended for carrier use than enterprise applications.
Establishing MPLS paths
Before MPLS paths (i.e., virtual circuits) can be established, IP connectivity must be running among the relevant nodes. MPLS path allocation overlays an IP network, whose topology is created with IP routing protocols. The MPLS software determines which traffic belongs to the same Forwarding Equivalence Class (FEC), which usually means the same destination, performance guarantees, and availability/failover requirements. Using a path setup protocol, usually an extension  of the Resource Reservation Protocol (RSVP)
MPLS does impose less workload on the control plane of routers, as the assignment to a FEC is done only once, in the ingress Label Edge Router (LER). The FEC is encoded in a "label" of link-local significance.
The ingress and egress LERs are the only ones that have to do the complex matching between FEC requirements and the first label passed to the first intermediate Label Switching Router (LSR) Any number of LSR can connect the ingress and egress LSR. The label is stripped at the egress LER.
MPLS has an extensive range of failover techniques, more versatile than SONET.  These can be more efficient than some more hardware-intensive techniques such as SONET dual ring, especially the dual ring variant where traffic is transmitted over redundant rings (i.e., 1:1).
MPLS does not require rings; it can have arbitrary topologies. MPLS does have failover modes equivalent to 1:1, and to the mode where there is an idle backup facility (i.e., "preestablished: 1+1). Going to lower availability but higher cost-performance, there is "preplanned" 1+1, where the protection path is planned to be available, but might not be for some reason specific to the path being backed up. Preplanned 1+1 still assumes a single backup path for each working path.
To go to even more cost-effective methods, reasonable given the inherently high reliability of optical facilities, 1:N allows one protection path to back up multiple working paths. 1:N is much harder to implement in a ring topology than an arbitrary one.
Even MPLS, however, failed to cover all technologies, and the Generalized MPLS (GMPLS) model was established to cover both packet and non-packet technologies. GMPLS can also set up switched paths for lambdas (optical wavelengths), time slots (as from a SONET/SDH optical multiplex system) or physical interface position (as from a cross-connect switch) A specific GMPLS method, for example, deals with the time slots of SONET and SDH.
For carriers or enterprises with ultrahigh bandwidth requirements, Optical Transmission Networks (OTN) now allows optical paths, usually of 10 Mbps increments, to be set up dynamically through the carrier, or even a campus, network. Remember that multiple 10 Mbps increments can be multiplexed onto the same fiber. Other names for OTN include the Automatic Switched Transport Network (ASTN) or Automatic Switched Optical Network (ASON). There are GMPLS specifications for transmission  and control.
- Awduche, D. et al. (September 1999), Requirements for Traffic Engineering Over MPLS, Internet Engineering Task Force, RFC2702
- Awduche D., Hannon A., and Xiao X. (December 2001), Applicability Statement for Extensions to RSVP for LSP-Tunnels, Internet Engineering Task Force, RFC3210
- Braden, R. et al. (September 1997), Resource ReSerVation Protocol (RSVP) -- Version 1, Functional Specification, Internet Engineering Task Force, RFC2205
- Sharma V. and Hellstrand F., ed. (February 2003), Framework for Multi-Protocol Label Switching (MPLS)-based Recovery, Internet Engineering Task Force, RFC3469
- Berkowitz, Howard C. (2002), Building Service Provider Networks, John Wiley & Sons pp. 460-464
- E. Mannie, ed. (October 2004), Generalized Multi-Protocol Label Switching (GMPLS) Architecture, RFC 3945
- G. Bernstein, E. Mannie, V. Sharma, E. Gray (December 2005.), Framework for Generalized Multi-Protocol Label Switching (GMPLS)-based Control of Synchronous Digital Hierarchy/Synchronous Optical Networking (SDH/SONET) Networks, RFC 4257
- D. Brungard, ed. (November 2005), Requirements for Generalized Multi-Protocol Label Switching (GMPLS) Routing for the Automatically Switched Optical Network (ASON), RFC 4258
- D. Papadimitriou, ed. (January 2006), Generalized Multi-Protocol Label Switching (GMPLS) Signaling Extensions for G.709 Optical Transport Networks Control, RFC 4328