MPLS has its roots in Ipsilon's IP Switching, Cisco's Tag Switching, IBM's ARIS technology and a few other proposals to bring the sort of traffic engineering found in connection-oriented Asynchronous Transfer Mode and frame relay networks to connectionless IP networks.
The idea is to steer IP traffic onto a variety of routes instead of the single one discovered by an interior gateway protocol such as Border Gateway Protocol, to avoid congestion or failures, or to enable a particular class of service or guaranteed service level.
MPLS switches and routers affix labels to packets based on their destination, type-of-service parameters, Virtual Private Network membership or other criteria. As a packet traverses a network, other switches and routers build tables associating packets and routes with labels. The MPLS switches and routers - dubbed label switch routers - assign each packet a label that corresponds to a particular path through the network.
All packets with the same label use the same path - a so-called label switched path (LSP). Because labels refer to paths and not endpoints, packets destined for the same endpoint can use a variety of LSPs to get there.
Also see: GMPLS
MPLS research center
The latest MPLS news and analysis from NetworkWorld.com.
MPLS discussions on NetworkWorld.com.
MPLS-VPLS Resource Center.
MPLS - An introduction to multiprotocol label switching
Reguires free online registration.
DiffServ vs. MPLS
Compares the two approaches.