Any service to any lambda accommodates metro uncertainty functionality

Metro Networks

When all-packet fabrics providing this are deployed in metro transport nodes, crossconnecting services between wavelengths becomes as easy as packet switching.

HAGAY KATZ and DR. MICHAEL MESH, PacketLight Networks Ltd.

Carriers face many new complexities in today's dynamic metro environment, including different types of interfaces, different cost structures, and different capacity needs. Traffic demands from business customers are evolving so rapidly that their metro market needs are becoming increasingly difficult to forecast. The frequent upgrade of business networks has given rise to an unpredictable bandwidth growth. In the face of rapidly advancing technologies, customers want more and varied services. As opposed to the traditional service mix of voice and data, today's customers want different types of data interfaces and technologies and would like them provided on their existing equipment. Along with these challenging customer demands, carriers are expected to supply rapid service provisioning.

As the optical metro network evolves, network planners must address the dilemma of how to best utilize optical technology while maintaining the simplest and most efficient metro transport network. They need to make the most of the services available today and plan for future metro services, without having to rebuild the entire network for each new service. Specifically for the metro environment, they must address several key issues:

• Maintaining network simplicity by unifying their metro-access and metro-core transport and DWDM technologies.
• Continued support for existing SONET services and the use of SONET protection schemes while migrating to Ethernet, Fibre Channel, and alternative restoration schemes.
• Ability to increase the bandwidth capacity of existing fiber infrastructure with minimal impact on operational cost, along with fast service provisioning.
• Need to interface efficiently with a new all-optical long-haul network.

While many architectures and technologies claim to address these issues, the approaches currently on the market or recently announced have proven costly or inefficient. However, a packet-based single-fabric approach can address metro requirements and help service providers meet customer needs.

Current and new architectures of optical transport systems offer costly solutions to the escalating problems faced in metropolitan area networks. The current architectures being introduced in the metro can be classified both formally and colloquially.

Multiple fabric systems: "Throw a fabric at the problem." These systems use multifabric-such as time-division multiplexing (TDM) crossconnect, ATM, and Internet Protocol (IP)-and high-layer (Layer 3 and above) processing at the edge to provide network efficiency and multiservice support. Although they achieve high efficiency, they charge a premium in terms of equipment, operational complexity, lifecycle cost, and overall expense of network infrastructure.

The multifabric systems use minimal integration to create a "few separate complicated boxes in one box." Consequently, they suffer from an increased fabric cost and a myriad of service cards tailored (typically by ASICs) to each fabric type. The system must double or triple the connections in the backplane and therefore requires more slot space in the chassis for fabric at the expense of service slots. That increases the cost of the common equipment, necessitating a high initial investment.

These systems provide numerous configuration options that complicate provisioning. In addition, service card inventory is very large, since there are different service cards for each type of protocol, resulting in multiple shelf types. This operational complexity is exacerbated by the need for equipment and facility protection and basically moves many of the traditional "core operational issues" over to the many network edge devices.

Although much effort has been invested in processing at the edge, the inherent architecture of the devices does not allow them to efficiently mix traditional SONET TDM traffic and data traffic on each lambda resource. They dedicate a lambda to a single traffic type; for example, all TDM traffic goes on one lambda and all IP traffic goes on another. Each node in the network requires more lambdas, since each node must have different "lambda pools" for each type of service.

When an entire network is deployed, the inefficiency in equipment cost, operational complexity, and lambda utilization at each node multiplies.

Multilayer SONET systems: "Throw a SONET layer at the problem." Multilayer SONET systems use a classical SONET STS-1 crossconnect that is not optimized to handle data and voice together. To efficiently mix packet traffic with TDM traffic, they add two types of functionality: time-slot concatenation and dynamic bandwidth allocation (DBA) protocols. Time-slot concatenation allows these systems to allocate bandwidth using more appropriate granularity. A typical SONET uses STS-1 granularity, where these systems can concatenate in VT1.5 and 512-kbit/sec circuits. DBA protocols resolve the inherent problem of rigid allocation of SONET bandwidth to users.

To provide concatenation in low granularity, the system requires the addition of costly "front-end" electronic layers before the SONET fabric. Also, network designers for this type of system must define specific limited "pipes" and "subpipes" within each lambda; they must also allocate specific services to these pipes.

Metro DWDM systems: "Throw a lambda at the problem." Metro DWDM systems focus primarily on DWDM transponders; these systems are used mainly to solve the problem of unpredictable bandwidth demand and better fiber utilization. The primary limitations of these systems are their inefficient use of lambdas and inability to provide SONET add/drop multiplexing (ADM) and aggregation functionality.

These systems attempt to resolve the inherent problem of "lambda burning" by adding another stage of multiplexing before transponding. They still burn lambdas because they cannot mix different types of services on the same lambda. Furthermore, their performance management is very limited. A notable advantage of these systems is their transparency, which, in turn, allows service providers simple operation and reduced inventory.

Optical networks need to incorporate a totally new type of system that allows carriers to plan their network in a dynamically changing environment. Service providers do not like the high cost and operational complexity of multifabric systems. They want better data functionality than that currently provided by multilayer SONET, and they want more efficient aggregation and flexibility from their current metro DWDM equipment.

A completely new optical Layer 1-2 packet-based network architecture provides an answer to current architecture limitations. This network involves innovative functionality both at the network level and the network-element level.

The aggregator will take those wavelengths from many collectors, sort the services by type, and aggregate them to the proper network connection. The number of wavelengths from each collector to the aggregator is proportional to bandwidth demand (see Figure 1). The transparent edge and the aggregator can be connected in protected rings or mesh topologies (see Figure 2). By simplifying the edge, the carriers reduce the overall cost of the network and provide aggregation only where needed.

The approach's "simple-transparent edge" is an alternative to models that advocate "data-aware high-layer processing at the edge." The simple edge, with a wide pipe, provides a solution to unpredictable bandwidth demands. Rather than use high-level processing at the edge, simplicity at the edge is increased. Low-layer adaptation to all types of traffic achieves this goal. In addition, the approach uses scalable fat packet pipes to transport the traffic. As shown in Figure 1, the "fat pipe" approach starts with a minimum of a single lambda pipe and provides the scalability to add more lambdas as demands increase. Since fiber has virtually unlimited capacity, and optical technology advances very fast, it is now feasible to use a fat-pipe approach as opposed to the conventional "data-aware" limited-bandwidth approach.

The main advantage of the fat-pipe approach is the operational simplicity it offers to carriers at the many edge sites. The benefits include fewer upgrades and less planning. Carriers currently upgrade their customer-site pipes from DS-3 to OC-3 to OC-12 to OC-48 to multiple OC-48s. Using this new approach, carriers will be able to provide the wide pipe once and continue serving the customer by simply adding service cards as the demands change. This is a viable solution that can be sustained over a long period of time. In addition, less planning is required, as there is no need for complex bandwidth analysis or Layer 3 and ATM traffic engineering on a network basis.

One of the important features of the simple edge is its transparency; that provides a solution to the problem of operational complexity. It allows carriers to use one type of service card for ATM, IP, or TDM. For example, on a single service card, one port may be connected to a router, another to an ATM DSL access multiplexer, and another to TDM channelized voice. Therefore, it provides a distinct advantage when compared to "data-aware" edge solutions that use a different card for each service, thereby quickly consuming their slot space in the chassis.

The transparency allows service providers to reduce their service card inventory and serves to maintain operational simplicity. Another advantage is that carriers are not forced to change hardware in their equipment each time a customer changes the interfaces in their customer-premises equipment. Also, they do not need to plan the relative traffic growth for each type of protocol.

Using a single point of aggregation is the key to operational simplicity and network efficiency. When traffic comes from many edge devices in a packet form, it is easy to sort the services and perform aggregation over these many streams. Only one point of processing is presented for the many feeding edge devices. The advantages are clear: service collection and service processing are separated, thus reducing the number of fabrics deployed in the network significantly; processing functions average many inputs signals; and ports that provide an interface to the network "cloud" are utilized more efficiently. (Figure 2 illustrates the network deployment architecture.)

Instead of having multiple switching and routing stages, this approach provides a single packet fabric that performs SONET ADM and crossconnect functions, in addition to handling native Ethernet and ATM. The single fabric is effective when it is used like a Layer 1-2 "service switch" instead of being an ATM/TDM/IP fabric. Thus, at the other end of the pipe, all services could be easily sorted according to their type, aggregated to the appropriate bit rate, and connected to the network "clouds": TDM, ATM, IP, WDM, or Multiprotocol Label Switching (MPLS). This all-packet approach represents a quantum leap from time-slot allocation to arbitrary placement of any service to any lambda. Thus, it solves one of the main drawbacks of multifabric systems: SONET and any other service can now be efficiently packed, thereby significantly reducing the number of lambdas required in the network. One lambda pool serves all.

All services can be mixed on one or more wavelengths in different combinations, limited only by the lambda bandwidth. These services could be crossconnected between different wavelengths. Different logical topologies could be mapped by using wavelengths on existing physical fiber infrastructure to achieve a flexible service delivery scheme.

Moreover, changing the allocation of services on each lambda is as easy as clicking on the browser. If a new lambda is added somewhere in the network to accommodate a new service (such as Gigabit Ethernet), the spare capacity on this lambda can be used to maximum efficiency. This layer allows any combination of services-for example, Fibre Channel with DS-3-ATM and OC-12 channelized TDM.

Current SONET and metro DWDM architectures see a lambda as raw bandwidth. Using a mathematical analogy, a lambda is viewed in current solutions as a "scalar"-a convenient way to describe a trunk's capacity.

When an all-packet fabric that provides any service to any lambda functionality is deployed in metro transport nodes connected in combinations of mesh and rings, a lambda becomes a network connection. Using a mathematical analogy, a lambda is viewed in this new architecture as a "vector"; it also provides a direction. Simply put, a lambda is described by its source and destination IP addresses.

The overall result is a dramatic improvement in cost, space, and power per lambda, as well as an entirely new flexibility in fiber planning and utilization. This novel architecture (see Figure 3) consists of four main layers:

• Layer 1, the wavelength adaptation layer, provides low-layer adaptation of the incoming services and ensures service transparency, as opposed to data awareness. It also ensures compliance with all stringent SONET timing and crossconnect requirements and preserves the SONET performance monitoring capabilities.
• Layer 2 provides integrated packet- and lambda-switching capabilities. This layer implements the any service to any lambda functionality and, in addition, allows switching of a whole wavelength as required.
• Layer 3 provides fat packet transport and DWDM/optical ADM, which together implement all the benefits of packet rings/trunks and classical metro DWDM systems. These benefits include DWDM, scalable growth in lambda capacity, fine service granularity, easy support of best-effort services, and simple end-to-end provisioning. Moreover, unlike the multilayer SONET solutions that need to define pipes and users of these pipes within a wavelength, this new architecture frees the planner from such duties. There is one flexible bandwidth pipe to share. Not only is it much simpler to plan and operate, it also performs better, since the pipe is fat, thus allowing improved dynamic use of bandwidth. Finally, while costly real-time DBA protocols are required to move and release SONET bandwidth, similar operations are done more easily in packet-based systems. True packet-switch-based systems provide similar functionality at lower cost and improved performance. That essentially renders the DBA protocols obsolete.
• Layer 4, the MPLS control-plane layer, serves as a technology enabler for this new architecture. Due to its basic nature, MPLS provides routing and forwarding, service allocation per lambda, and quality of service.

The single-fabric, packet-based approach to metro networks addresses the major challenge metro carriers are facing today. Unlike current architectures, this system approach enables metro carriers to cope with the unpredictable demand of the metro environment, while making optimal use of their resources: fiber, wavelengths, and core-network equipment. Moreover, this new approach enables interoperability with future all-optical networks, where a wavelength is a connection, not just raw bandwidth.

Hagay Katz is co-founder and director of product management and Dr. Michael Mesh is co-founder and chief technical officer of PacketLight Networks (Kfar Saba, Israel). Katz can be reached via e-mail at [email protected] or phone at +972-9-7645414.