Redefining the boundaries: switched services for the wavelength core
SPECIAL REPORTS / Optical Networking/WDM
A new photonic services platform switches packet, TDM, and wavelength traffic over a Generalized MPLS network.
MARK MILINKOVICH, AcceLight Networks
The last five years have brought unprecedented investment in optical networks. Tens of thousands of miles of optical fiber were laid to meet the increasing demands of data and voice transport. One of the key questions service providers face today, especially in view of the recent economic downturn, is how best to take advantage of this enormous investment in optical transport infrastructure.
A recurring theme in discussions of this issue is the need to consolidate the network core. Consolidation will help service providers stay ahead by reducing the cost of networks and their management and by offering more opportunities to increase revenues through flexible service offerings. The question, then, is, what exactly does consolidation of the core mean and how best can it be achieved?
Consolidation of the network core eliminates the demarcation between services and transport. Operators will move away from traditional layered networks toward consolidating services and transport in peer-based networks where every network element has full knowledge of every other network element. That means bringing intelligence and dynamic circuit (or path) provisioning equally to the three essential services of the consolidated optical core: packet, TDM, and wavelengths.
The first requirement for core consolidation is a protocol for a common control plane. Generalized Multiprotocol Label Switching (GMPLS) is that protocol. However, GMPLS is not in itself a complete solution. By itself, GMPLS cannot reduce the number or diversity of network elements and hence the cost of deploying and managing networks. Consolidation of the core has a second requirement: a fast GMPLS-enabled network element-a photonic-switching element able to forward or switch traffic of all types across a GMPLS network (see Figure 1).
The traditional approach to network architecture is to separate services from transport. Whatever their form, services are switched; that is, a switch or router makes a processing decision from the packet, frame, cell, or label, and the service is directed through a node or network (see Figure 2).
Transport, on the other hand, is generally through multiplexers, crossconnects, add/drop multiplexers, repeaters, and amplifiers. This separation of function and equipment is characteristic of overlay networks, which have separate control planes for each network layer. The shortcomings of overlay networks, with their multiple layers of multiplexing, switching, and protection, are well known: no visibility across layers, inefficient use of resources, limited flexibility of services, difficult to scale, complex, and costly to build, manage, and maintain.
There is often confusion about what is meant by control plane and data plane. The data plane, also called the forwarding plane, can be described as a switching layer that carries the content of the message. The control plane is a series of mechanisms such as signaling, route table synchronization, and network management for controlling network topology and directing messages to their destinations. Consolidation requires both a common mechanism for a unified control plane and the equipment to handle different types of traffic on the data planes. Before the advent of GMPLS and fast photonic switching, these two essential enablers were not available.
Three approaches are currently the focus of much attention in discussions about how to improve network efficiency and scalability. All three promise consolidation of some sort. These approaches are IP over DWDM, MPλS, and GMPLS with photonic switching.
IP over DWDM uses IP addressing and routing over DWDM networks. Most implementations use packet over SONET (PoS) directly onto DWDM trunks to allow consolidation of the IP and data planes over wavelengths and fiber. There are many advantages to this approach: IP is a well understood and proven technology; PoS is widely deployed; IP interfaces are supported up to OC-192c/STM-64c; and IP is statistically multiplexed. Finally, IP convergence allows use of routers to perform switching, thereby providing opportunity for reducing equipment.
IP over DWDM also has some disadvantages that weigh against it as a strategy for consolidating the core, however. The core router acts as a large PoS crossconnect, hence sacrificing speed to gain intelligence. IP route convergence and network stability mechanisms can run to tens or even hundreds of seconds in the event of large trunk failures. IP over DWDM requires multiple provisioning, monitoring, and management tools. There is no mechanism for communication between the routers and other transport equipment, such as crossconnects or add/drop multiplexers (ADMs). IP over DWDM does not support integrated service creation, synchronous traffic, or wavelength conversion. Finally, since IP over DWDM uses expensive packet line cards for all traffic, network operators deploying this solution must pay a premium price for all transit traffic, including wavelengths.
MPλS proposes adding extensions to the mpλs control plane protocol (which is chiefly used with IP services) to bring signaling and labeling to wavelength services and offers intelligent interconnections between optical crossconnects (OXCs) for DWDM long-haul systems. This protocol brings control such as rerouting (local or node) and traffic engineering to wavelength traffic. Like IP over DWDM, MPλS builds on an existing strategy. And it has the support of a number of OXC vendors.
However, while it offers benefits specifically to long-lived connections set up for long-haul transmission, MPλS does not provide a means for consolidation of core networks. This approach assumes an overlay model based on a user-to-network interface and does not integrate the packet forwarding data plane. MPλS offers little more than optical patch panels, and MPλS networks still require core routers to process packets and ADMs and broadband digital crossconnects to process time slots on SONET/SDH channels.
As well, implementations of MPλS rely on OXCs. Many commercial OXCs use micro-electromechanical systems (MEMS) technology, which has well-known inherent problems: mechanical, vulnerability to environmental changes, and inability to handle multiplexing and grooming. More crucial, however, is that, whatever their underlying technology, commercially available OXCs offer relatively slow switching and cannot efficiently handle packet and time-slot traffic. Thus, an MPlS control plane provides dynamic provisioning of wavelength traffic but brings improvements to only a limited set of service types. This limitation makes it unlikely that MPλlS will gain acceptance for more than some specific uses for long-haul.
GMPLS combined with photonic switching presents the most promising approach for successful consolidation of core networks. GMPLS is the most recent and comprehensive addition to a series of standards arising from efforts by the Optical Internetworking Forum, the Optical Domain Service Inter connect consortium, and the Internet Engineering Task Force to develop a protocol usable for all traffic types. It offers an integrated control plane, which extends topology awareness and bandwidth management across all network layers, effectively permitting consolidation of services and transport.
This consolidation moves the traditional point of demarcation down. Services and transport are now together and separated from transmission. Long-haul transmission is left as the non-switched element (see Figure 3).
GMPLS is a suite of protocol extensions that provides common control to packet, TDM, wavelength, and fiber services. These extensions affect MPLS routing and signaling protocols for such activities as label distribution, traffic engineering, and protection and restoration and enable rapid provisioning and management of network services.
GMPLS can be used with traditional overlay network architectures in which each traffic type is managed through its own control plane. However, the great potential of GMPLS is that it makes possible evolution to peer-based networks in which all network elements have full information about all other elements and their link capabilities.
The overlay and peer models apply to both routing and signaling. The overlay model maintains separate network layers for each traffic type and separates different administrative domains. In contrast, peer-based networks are built of devices that have full information about all other devices at all layers in the network.
The overlay model is favored for network operations between network operators, because it allows each network operator's route information to be kept within its own administrative domain. The peer model is favored for network operations within a service-provider domain or between friendly service providers with compatible protocols, because it allows for greater flexibility in route optimization.
A key benefit of GMPLS is that it allows network operators the freedom to design their networks to best meet their specific needs and business objectives. GMPLS can be used with either overlay or peer networks or with some sort of hybrid that consolidates some, but not all, traffic types. Thus, GMPLS fulfills the first requirement for service providers to be able to initiate and complete smooth transitions to a consolidated control plane for the service and transport types of their choice.
The second requirement is a network element capable of handling packet, TDM, and wavelength traffic simultaneously and at optical speeds. This requirement is met by fast photonic switching, which uses the common control plane enabled by GMPLS and switches services and transport through a single optical core fabric. For efficiency, this network element should be complemented by an element- and network-management system designed to manage multiple layers of an intelligent network consolidated onto the GMPLS control plane.
As is so often the case in this industry, the key will be timing. Not just for the switch vendors, but for the network operators. As competitive forces continue to play out in an ever more demanding marketplace, the future belongs to those companies first able to reduce costs while maintaining revenues. Bets are on the network operators that are first off the block to consolidate and simplify their cores.
Mark Milinkovich is the director of strategic marketing at AcceLight Networks (Ottawa, Ontario).