Routing and conversion can be embedded components

Wdm93821 37

P. Harshavardhana

New challenges arise with the arrival of all-optical networks, which bring increased bandwith, reliability and cost benefits Specifically, signal or bandwith routing looms as a critical issue for network planners and managers who seek to ensure signal delivery and optimal network performance.

Telecommunications industry observers have said that there is almost a glut of bandwidth and that bandwidth costs are preparing to drop. Certainly costs have plummeted: an OC-192 link today might cost the same as a T1 line cost three to four years ago. While costs have decreased and available capacity has increased, user demand for bandwidth has soared even higher and faster. Given the need to meet this demand and maximize return on investments in optical networks, carriers must find a way to optimize network performance. Wavelength routing and conversion are key to this optimization.

Studies have shown that many carriers will experience a significant degradation in performance because of an inability to route lambdas-or wavelengths-across the available paths in a network. This degradation will manifest itself through capacity constraints that will slow network transmissions, or through capital expenditures for deploying new bandwidth.

To a certain extent, networks can sidestep this dilemma by incorporating wavelength conversion into network nodes. So, if a node has no available links to route a signal on a particular wavelength, then the signal can be optically converted to another wavelength for which there is an available link. Providing for wavelength conversion, however, is expensive and rarely available at all network nodes (see Fig. 1).

Intuitive software packages based on newly developed algorithms are addressing the complex challenges involved in optimizing a network subject to wavelength conversion and routing constraints.

The routing conundrum

With recent advances in technology, a single optical fiber equipped with WDM capability can transport dozens of wavelengths simultaneously. These wavelengths are usually spaced at 50-, 100-, and 200-GHz intervals.

A typical demand usually traverses multiple fiber-links while traveling from its originating node to its destination. The demand may be transported on a different wavelength on each of the fiberoptic links it traverses, depending on the availability of wavelengths on individual links.

To demonstrate, let`s suppose a demand from Los Angeles to New York traverses three links: a Los Angeles/Denver link, a Denver/Chicago link, and a Chicago/New York link. The transmission from Los Angeles to New York might begin as a red wavelength, then might be converted to a green wavelength on the Denver-to-Chicago link, and finally arrive as a yellow wavelength on the Chicago-to-New York link.

Here we assume that wavelength conversion was available at the Denver and Chicago nodes so that the changes from red to green in Denver and green to yellow in Chicago could be achieved. Such a conversion was necessary because the red wavelength was not available on the Denver-to-Chicago link and the green wavelength was not available on the Chicago-to-New York link. If one of those nodes did not support wavelength conversion, we would have no choice but to look for an alternate path, which would obviate the need for conversion. For instance, if the Denver node had no wavelength conversion, and the red wavelength was available on the Denver-to-Phoenix link, we would use it even though it means taking a longer and potentially more expensive path.

Conversion constraints

Many of today`s optical networks have an underlying electrical fabric. The electro-optical conversion makes it relatively easy to change the wavelength to which a given demand is assigned during its journey. But with newer all-optical networks, wavelength conversion is an expensive process, requiring the deployment of costly lasers.

The usual solution is for network operators to have spare lasers available. These are field-installed by technicians when a network planner determines that capacity and bandwidth-routing limitations are being reached. For network operators, there is a high cost for purchasing, inventorying, and installing spare lasers. These steps can result in an annual cost of thousands of dollars to carriers.

One solution to the wavelength-conversion challenge is to use tunable lasers. Placed at strategic locations in the network, these eliminate the need to install wavelength- or lambda-specific lasers. By technician intervention, these lasers can be tuned to handle anticipated wavelength-conversion needs. It is also possible to put enough intelligence in a network so that lasers can be software-controlled-tuned, on demand, to handle the wavelength conversion needed.

Recent breakthroughs in tunable laser technology are making wavelength conversion affordable. First-generation tunable lasers, known as distributed-feedback lasers (DFBs), were able to switch between two to four wavelengths. The newest tunable lasers can be tuned over the commonly used wavelengths in carriers` optical backbones.

Wavelength routing and conversion

The type of wavelength conversion available dictates the permissible options for wavelength routing. If a network has no wavelength-conversion capability, all traffic demands must be met by utilizing available wavelength-routing options. From the other extreme, if a network has full conversion capability, then any desired routing option can be supported as in today`s networks based on electrical fabrics.

The interesting scenarios for carriers will involve some wavelength-routing flexibility, along with limited wavelength conversion. For a given network, finding the optimum balance between full- and no-conversion options is a complex, yet solvable, mathematical problem.

Software solutions that optimize network performance through consideration of wavelength routing and conversion are becoming available. With embedded intelligence, applications can identify paths for new demands "on the run" and can configure backup paths. Placed in a distributed optical mesh, the software can support self-healing and self-provisioning.

The software can be housed on the desktop PC of a network planner for long-range or periodic planning. Network carriers can then integrate the software into their own network management capabilities.

Software also can be embedded in the element management systems of optical-networking products such as optical crossconnects, routers, or add/drop multiplexers. Optical-networking vendors, such as Corvis (Columbia, MD), Sycamore Networks (Chelmsford, MA), and Lucent Technologies (Murray Hill, NJ) are beginning to include such software in their products. Shipments of products including wavelength-routing/conversion-optimization models are expected by the middle or end of 2001.

In advance of deploying these optical-networking products, carriers also can use traffic-planning software to simulate the products` performance in the carriers` network model. This gives the carrier a better understanding of what the products` actual behavior will be when installed in the network.

Software solutions

Without an algorithm-based traffic-planning software solution, the question of optimizing network performance through the use of bandwidth routing and conversion has been too complex to solve. An engineer or network planner could spend weeks on such a problem and still be unable to address it or properly analyze all of the options. A software solution, using a combination of heuristics and combinatorial algorithms, is able to solve these problems in just seconds-allowing, for example, online provisioning.

The inputs to such a software solution include network topology, point-to-point demands, and wavelength-conversion assumptions (for instance, some conversion, no conversion). Typical outputs from software include a network design that delineates minimum-network-capacity wavelength routing-with and without network restoration. This capability includes assigning wavelengths to demands and routing of demands.

Another software output is centralized network planning. This option optimizes total network performance, assuming all demands are known. This type of analysis and resulting changes to the network are made during off-hours for the network. Optimization software is able to address network planning for all types of optical networks. Traffic-planning software with wavelength-routing and wavelength-conversion analysis capabilities can produce network designs that meet performance requirements, but cost as much as 20% less to build than designs created through conventional methods (see Fig. 2 and Fig. 3).

In the online, or provisioning application, optimization algorithms address random demands as they arise anywhere in the network. Tests have shown that distributed provisioning and restoration algorithms can result in network performance that is within 10% to 20% of the optimized performance levels achieved through a centralized planning approach.

Provisioning, in fact, is an application for which a software-based solution is particularly valuable. Along with operating and repair expenses, provisioning accounts for 80% or more of a network operator`s costs. Of these three costs, provisioning is the largest item. Capital costs typically account for the remaining 10% to 20%. Recent development work has produced software-based solutions that can reduce provisioning costs by an order of magnitude. And while these problems become more complex as a network grows, the speed and ease of use of software solutions allow them the ability to scale to network size.

Planning for survivable networks

Most optical networks today are migrating toward a distributed design, with Internet-Protocol-type (IP) provisioning and restoration architectures. Newer carriers are expected to begin deploying these architectures next year.

The optimization challenge becomes even greater when designing survivable networks. Here the network planner or the network operating system must find a primary path for each demand, as well as identify a restoration path to serve as a backup in the event of a node or link failure. For a restoration plan to work cost-effectively, potential restoration paths have to share network capacity.

As carriers deploy all-optical networks, they need effective ways to optimize network performance and reduce costs. Software solutions that streamline network performance through planning for wavelength routing and conversion give network planners a powerful new tool, and also enable networks to employ dynamic traffic-routing.


Y. Agarwal, P. Harshavardhana, M. Geyer, A. Mittal and J. Strand, NFOEC 2000 Proc.

P. Harshavardhana is group technical officer of IP optical networks at Virtual Photonics Inc., 943 Holmdel Rd., Holmdel, NJ 07733. He can be reached at 732-332-0233, x110, or at 37

FIGURE 1. A typical network topology has point-to-point demands. To optimize transmission, software must determine the minimum-capacity wavelength-routing and assign wavelengths to meet demands. Wdm93821 38

FIGURE 2. Traffic-planning and optimization software was applied to a model network to help develop a design that balances network cost against network performance. This SONET/WDM network included 58 nodes, 28 of which had wavelength-conversion capability via limited-range tunable lasers. The network included 78 single-fiber links. There were 281 wavelength demands.Wdm93821 39

FIGURE 3. The software produced a network design with 100% restoration that required 6,206 lambdas (wavelength windows) without using any wavelength conversion. Taking advantage of wavelength conversion, the network would need only 4,205 lambdas-a more than 30% decrease in the number of wavelengths that need to be routed.

More in DWDM & ROADM