5.5. Generalized MPLS (GMPLS)
5.5.1. GMPLS Overview
MPLS, as originally defined, applies to packet[2] networks. In essence, an MPLS LSR receives a packet from one interface, replaces the (incoming) label in the MPLS header by an outgoing label, and forwards the packet out of another interface. Thus, MPLS defines a virtual circuit capability. Now, consider a true circuit-switched network. Depending on the type of the network, the switching operations would differ:
[2] We loosely use the term packet to mean either frames or cells (as in ATM).
Time-Division Multiplexing (TDM) networks: In these networks, a node switches data from time slots in an incoming interface to time slots in an outgoing interface. An example is a SONET network in which an LTE switches the path layer payload from one channelized interface to another as described in Chapter 3.
Port or fiber switching networks: In these networks, a node switches the data received over an incoming interface onto an outgoing interface. An example is a SONET network in which an LTE switches the path layer payload from one unchannelized interface to another. Another example is an all-optical network in which a PXC switches light from one interface to another.
Lambda switching networks: In these networks, the nodes switch wavelengths. An example is an optical network with PXCs, which operate at the granularity of an individual wavelength. Each PXC in such a network would be able to switch a wavelength received over an incoming interface to an outgoing interface (with or without wavelength conversion).
From these examples, one can see a correlation between the MPLS label switching operation and different circuit switching operations. Whereas the labels in MPLS networks are explicit, the “labels” are implicit in the other networks. Table 5-1 illustrates the “labels” in the three types of networks described above:
| Network Type | “Label” |
|---|---|
| TDM | Time Slot |
| Port or Fiber Switching | Port Number |
| Lambda Switching | Wavelength |
Thus, in TDM networks, for instance, the switching of data from a time slot on an incoming interface to a time slot on an outgoing interface is similar to the act of MPLS label switching. The difference, of course, is that whereas MPLS labels are bound to packets, TDM switching applies to specific circuits as they traverse a switch. What this analogy leads to is the conclusion that the same control plane mechanisms used under MPLS for establishing label mappings in switches can be used for setting up cross-connect tables in circuit switched networks. This was first highlighted in a contribution to the IETF on MPλS (Multi-Protocol Lambda Switching) [Awduche+01a] in the context of lambda switching. Since then, it has been realized that the analogy applied to many different circuit switching networks, and the term Generalized MPLS (GMPLS) has been adopted to denote the generalization of the MPLS control plane to support multiple types of switched networks. The term “LSP” is used under GMPLS to denote different types of circuits such as a SONET/SDH connection, an optical light path, an MPLS LSP, and so on.
Under GMPLS, a suite of distributed control plane protocols is defined. These protocols cover three main functions:
Link management: This deals with neighbor discovery, maintenance of signaling control channels, and so on.
Topology and resource discovery: This deals with topology and resource information propagation within a GMPLS control domain. This function is also referred to as “GMPLS routing.”
Signaling: This deals with protocols for connection provisioning and restoration.
The protocols for topology and resource discovery and signaling are adapted from MPLS-TE routing and signaling protocols. Link management is a new function defined under GMPLS.
5.5.2. GMPLS Link Management
Unlike MPLS LSRs, the optical network nodes may have several tens or even hundreds of links to adjacent nodes. Neighbor discovery is a function that allows each network node to determine the identity of the neighboring nodes and the details of link connectivity to these nodes. This information is necessary for both topology discovery and signaling. Normally, in MPLS networks running IP routing protocols with TE extensions, such adjacencies will be discovered by the routing protocol. In networks with dense interconnections, it is inefficient to have the routing protocol run over multiple links between the same pair of nodes. Thus, GMPLS supports a separate link management function, which permits automatic neighbor discovery as a precursor to topology discovery. Link management is a function implemented between every pair of neighboring nodes, and supports the maintenance of the signaling control channels and the verification of configured parameters in addition to neighbor discovery. Link management is described in detail in Chapter 6.
5.5.3. GMPLS Routing
The MPLS-TE routing protocols have been extended to support topology and resource discovery under GMPLS. As with MPLS-TE routing, GMPLS routing permits the assignment of various attributes to links and the propagation of link connectivity and attributes (resource) information from one node to all others in a network. The attributes assigned to a link are specific to the underlying technology. To deal with dense interconnections, GMPLS routing allows multiple links between a pair of nodes to be bundled into a single TE link in topology descriptions. This is particularly useful in optical networks, where many tens of links may exist between neighboring nodes. GMPLS routing also incorporates the notion of Shared Risk Link Groups (SRLGs), used for indicating which physical facilities could be affected by a common failure event. GMPLS routing allows the partitioning of a large network into multiple, smaller areas. In this case, GMPLS routing extends the multiarea TE concepts used under MPLS-TE. As in MPLS-TE, the path computation algorithms themselves are not specified as part of GMPLS routing. GMPLS routing implementations are free to choose the path computation algorithms. GMPLS routing is described in detail in Chapter 10.
5.5.4. GMPLS Signaling
GMPLS signaling utilizes MPLS-TE signaling protocols (RSVP-TE and CR-LDP) with extensions for handling multiple switching technologies. The significant extensions are:
Generalized label: The MPLS notion of label is generalized to include different label formats corresponding to packet and circuit-switched technologies (see Table 5-1).
Common end-to-end label: This extension allows switches in a network to determine a common end-to-end “label.” This feature is useful in all-optical networks without wavelength conversion, where the same wavelength (label) must be used end-to-end.
Bidirectionality: GMPLS signaling extensions support the provisioning of both unidirectional and bidirectional connections. This is in contrast to MPLS-TE, which supports only unidirectional LSPs.
Separation of control and data planes: MPLS-TE signaling typically occurs over the same network node interfaces over which data flow. GMPLS signaling allows control and data interfaces to be distinct. Thus, a single control link between network elements can be used to control several data links between them. Furthermore, GMPLS signaling is designed such that failures in the control plane do not affect data transport over previously established connections.
Control plane restart procedures: These extensions allow network elements to recover their signaling control states after a node or link failure.
GMPLS signaling is described in detail in Chapter 7.