Bandwidth requirements in networks are constantly changing, driven by the deployment of new services and the increasing penetration of existing services. These requirements are further complicated by the migration to on-demand services and to IP-based services.

Factoring into this the need to maintain quality of service (QoS) across a wide range of services, effective bandwidth management has become critical to today's network. At the competitive level, pressures are growing to accelerate service delivery and network migration while lowering overall network costs (CapEx and OpEx). To meet all these demands, an overall bandwidth management mechanism is needed which substantially automates network operation while providing flexibility and ensuring adequate QoS. This is particularly true at the optical layer, where requirements are rapidly evolving. GMPLS (generalized multi-protocol label switching) provides such a mechanism and is available today.

MPLS (multi-protocol label switching) has been used in IP networks for many years to couple Layer 2 devices (for example, Ethernet switches) more tightly to IP routers at Layer 3. As such, MPLS allows these Layer 2 devices to serve as extensions of the routers at Layer 3, thereby giving much better control and integration of the overall IP network at both Layers. MPLS was initially developed to bring the speed of Layer 2 switching to Layer 3, but MPLS has found many wider applications and benefits and is now broadly standardized and in use in networks across the country.

In an MPLS network, edge routers assign a "label" to incoming packets. As these packets traverse the network, a label switched path (LSP) is created from the ingress to the egress point. These packets are then forwarded along the LSP by a label switch router (LSR), which makes forwarding decisions based on the contents of the label, rather than by the IP address. Attributes are assigned to the label that define how the LSR should handle the packets in the LSP.

LSPs can be created based upon a wide range of criteria and for a wide range of reasons. Typically, LSPs are used to route around or avoid network congestion, to guarantee a certain level of performance, or to create IP tunnels through a network for virtual private networks (VPNs). As such, LSPs look very much like circuit-switched paths in ATM networks. However, MPLS is not constrained to a particular Layer 2 technology and can create end-to-end circuits with defined performance over any transport layer. More typically today, MPLS is used within Ethernet networks.

Initial efforts to standardize MPLS began in 1997 at the IETF (Internet Engineering Task Force), which is still the primary standards body for MPLS, though several other organizations also create or contribute to MPLS standards and extensions. As MPLS evolved, it became widely recognized that the benefits of MPLS could be extended to Layer 1 at the optical level, and GMPLS was developed to address this need. The IETF is also primarily responsible for GMPLS standards. While both MPLS and GMPLS standards are still evolving and being developed, core standards are in place and in use today for both.

GMPLS extends MPLS into the Layer 1 optical network by allowing the creation of label switched paths in the optical network. Thus, GMPLS can serve as a control mechanism for devices such as optical add-drop multiplexers (OADMs) and optical cross-connect switches (OXCs), operating in both the wavelength (DWDM lambda) and spatial domains.

GMPLS supports both peer and overlay operational modes. In the peer mode, all devices in a given network domain interoperate over the same control plane. This provides true operational integration of OADMs, OXCs, and routers. Routers have visibility into the optical network topology and peer directly with OADMs and OXCs. In the overlay mode, the optical and routed IP layers are separate. In this mode, GMPLS is used to manage and optimize the optical layer without interaction with Layer 3. To date, most deployments of GMPLS use the overlay model, but provide a future migration path to the peered model.

Overlay GMPLS brings significant traffic engineering and management capabilities to optical networks and supports virtually full automation of optical network operations, including topology auto-discovery, span loss measurement, network element and inventory discovery, optical layer turn-up, dynamic network optimization, fault detection and correction, service turn-up, wavelength set-up and tear-down, node insertion, and network evolution. The benefits of a GMPLS-based optical network are obvious: a drastic reduction in operational costs and greatly accelerated network upgrades and delivery of new services.

GMPLS at the optical layer

While a broad discussion of GMPLS is beyond the scope of this article, a brief overview is useful to understand how GMPLS works at the optical layer. As already mentioned, MPLS works by creating label switched paths. In optical networks, these LSPs behave similarly to circuits and may be defined as anything sufficient to identify a traffic flow: a fiber, a lambda, a timeslot, etc.

The LSPs in the optical network are established through the use of routing and signaling protocols. In GMPLS, OSPF (Open Shortest Path First) and IS-IS (Intermediate System to Intermediate System) interior gateway routing protocols (IGPs) are used to exchange information between OADMs and OXCs about network topology, resource capabilities and availability, and network policies.

This information is then applied as input into a constraint-based algorithm that computes network paths based upon topology, resource availability, and service requirements. Constraint-based routing allows the network to automatically provision additional bandwidth based upon congestion, shifting service or content-on-demand requirements, or other network parameters. Once an appropriate path has been defined, a signaling protocol such as RSVP-TE (Resource Reservation Protocol for Traffic Engineering) or CR-LDP (Constraint-Based Label Distribution Protocol) is used to create the service connection along the path and reserve resources for it. Additional link management functions are implemented using LMP (Link Management Protocol), which runs between adjacent nodes. LMP provides control channel connectivity and failure detection. Once the path is established end-to-end, user traffic may flow through it.

Of course, GMPLS can only provide as many capabilities as are supported by the underlying OADMs and OXCs in the network. And since we are discussing automated reconfiguration of the network by GMPLS to control bandwidth, this implies the OADMs and OXCs in the network must also be fully reconfigurable. To achieve the full benefits of GMPLS, this means that all essential operating parameters of the OADMs and OXCs must be remotely configurable via software.

Figure 1
Figure 1: GMPLS architecture with reconfigurable optical
add/drop multiplexer (ROADM).

To address these requirements, a new class of reconfigurable OADMs (ROADMs) has been developed which provides fully automated and remote network configuration (see Figure 1). To provide this capability, four elements are essential to a ROADM. First, the ROADM must provide a means of selectively dropping wavelengths at nodes where local traffic is to be handed off (or passing these wavelengths directly through in the optical domain to their destination). Typically this is handled by a wavelength selective switch (WSS), which ideally should provide non-restrictive, single wavelength granularity. Secondly, the ROADM must support tunable lasers so that wavelengths may be remotely assigned or recycled at any time. Third, the ROADM should provide an optical backplane, which eliminates multiple manual optical jumpers and the need to roll a truck to re-configure these jumpers when network changes are made. Finally, the ROADM must support the GMPLS control plane which provides the intelligence and control for network automation.

Figure 2
Figure 2: GMPLS architecture with reconfigurable optical switching platform.

Similarly, a new class of reconfigurable OXCs has emerged to provide automated optical switching between fibers and networks, wavelength translation, and multicast service replication for services such as broadcast digital TV (see Figure 2).

As with ROADMs, these optical switching platforms provide fully remote reconfiguration and typically integrate wavelength transport in the platform along with GMPLS. The key element here, of course, is a non-blocking, redundant switch fabric to provide the cross-connect capabilities. This switching may be done in the optical or electrical domain, but doing so in the electrical domain allows additional traffic grooming at the protocol or packet level and enables additional regeneration, reshaping, and retiming of the signals at each node to simplify network engineering. These platforms are ideal for aggregation applications, mesh networks, and n-degree networks.

As networks migrate to peered GMPLS, significant additional capabilities can be realized. With routers peered directly with OADMs and OXCs, these Layer 1 devices can be controlled at Layer 3, which enables routers to reconfigure the optical network. This opens the door for assigning lambdas on-demand based upon congestion avoidance, increased bandwidth requirements driven by content-on-demand, or the need to create new protection paths through the network.

GMPLS is primarily focused today on automating the operation of optical networks, enabling automated network self-discovery, optical layer turn-up, dynamic real-time network optimization, and network evolution. This is particularly of value in WDM (wave division multiplexing) networks where manual engineering, configuration, and provisioning are expensive and time-consuming. GMPLS greatly reduces the operational complexity of WDM optical networks while accelerating network and service turn-up. Future implementations of GMPLS will allow greater interoperability at the optical layer between equipment in the network and will allow bandwidth to be assigned on-demand in real time at the optical layer (wavelength routing) to support dynamically shifting network requirements.


1. "The Intelligent Network: Dynamically Managing Bandwidth at the Optical Level," Gaylord Hart and Steven Robinson, NCTA 2005 Technical Papers: Proceedings of the 54th Annual Convention and International Exposition of the National Cable & Telecommunications Association, April 4, 2005, San Francisco, Calif.
2. "The Dynamic Network: Managing Bandwidth and Content on Demand at the Optical Level," Gaylord Hart and Zouheir Mansourati, Proceedings Manual: Collected Technical Papers of the 2004 SCTE Conference on Emerging Technologies, January 15, 2004, Dallas, Texas.
3. "The MPLS FAQ," maintained by Irwin Lazar, The MPLS Resource Center,