All-optical and OEO switches will share the load

0301feat08 1

J. Robert Palmer

Optical-electrical-optical designs provide stepping-stones for growth and will remain the choice for many applications. All-optical switching promises to shave costs by eliminating the need for OEO conversions while providing fast, speed-independent service.

High-speed network switching has crossed the threshold into the real world as an in-service, revenue-generating technology for carriers. As carriers looked to scale their networks by dramatically increasing the capacity of their optical transport systems, it became clear that optical core switching must be based on at least OC-48 (2.5 Gbit/s per channel) and higher technology. This transmission speed is supporting the rapid growth in network traffic demand from customer IP routers, ATM switches, and SONET add/drop multiplexers.

The advent of high-speed optical switching has resulted in significant carrier and customer benefits, including substantial cost savings in building the network itself, service-failure recovery improvement, and dramatic reductions in service provisioning cycles. Cost savings derive from the elimination of a layer of infrastructure equipment such as costly SONET multiplexers and fiber-ring terminals. By combining mesh-based architectures with traditional ring-based architectures, optical switching permits rapid traffic restoration while using less equipment.

Supported by commercial-grade management and servicing software, carrier circuits are being rapidly provisioned and restored using optical-layer crossconnects with electronic switching fabrics at capacities up to 1.28 Tbit/s. In addition, the migration paths are coming into place for carriers to move in sensible steps farther up the technology curve to all-optical switching.

Optical-electrical-optical (OEO) switching technology is at the heart of currently deployed switches for two basic reasons. First, the electrical-switching technology layer is following silicon-technology design and bandwidth growth curves that allow carriers breathing room with increased capabilities into the near-term, while subsequent technologies are readied for commercial deployment.

The second reason for the OEO technology choice is likely more critical for the provider's business success: management software. Software manageability requires visibility into the optical signal stream. Available software now detects and uses optical stream information for managing faults, configurations, accounting, performance, and security.

Linked with traditional operational support-system software tools, the switch management software allows the carrier to operate, administer, maintain, and provision the whole network to meet customer-service-level commitments. It is important to remember that without these management capabilities, no switching technology—optical or otherwise—will contribute to a carrier's commercial success.

While OC-48-level switching and faster is highly suited for bulk, mainline transmission within the core, carriers also deploy STS-1 level (50 Mbit/s) switches, usually at the edge, for grooming customer traffic into larger streams running at core transmission capacities. OC-48 switches are more optimized—and needed—at the core because of their inherent scalability characteristics.

As a result, the fundamental OEO switching and routing technology is not likely to go away, despite the fact that the next generation of optical-switching technology is on the horizon. However, the placement of these OEO technologies in the network infrastructure will shift toward the network's edge for grooming transmission. Carriers will still need to take smaller data quantities and put them together for core network transmission at some point in the data path.

Already, however, the definition of grooming has moved up the chain. For example, up to four OC-48 streams are being packed into an OC-192 stream for 10-Gbit/s transmission. The reality is that not all communication is high-bandwidth or high-speed, so carriers will want and need the flexibility to serve different needs efficiently at all wavelengths.

The needs for carrier bandwidth are expected to grow significantly. Demand will clearly increase because of customers' requirements. In addition, more general market demand is expected, as bandwidth becomes more of a commodity, due to the proliferation of optical networks. This snowball effect will come from new classes of customers as the technology becomes more affordable.

The increased wavelength-traffic needs will cause a migration to 10 Gbit/s, then 40 Gbit/s and beyond. This move is driving the need for all-optical switching, primarily because electrical switching technologies have finite limits on scalability. For example, a 40-Gbit/s signal must be broken down to 16 individual 2.5-Gbit/s streams to make use of present technology. This capacity would absorb 16 ports in an electrical switch, which is not commercially practical given the current cost of electrical cores.

In an all-optical switch in an OC-768 system, this capacity takes only a single port for switching, and the whole 40-Gbit/s signal is kept intact at the wavelength level. In the same manner, with the increased implementation of DWDM, the greater numbers of wavelengths will overwhelm electrical cores.

Since all-optical switching is capable of handling such large signal volumes through a single port, one of its key promises is that OEO signal conversions within the core will be eliminated, streamlining the economic models for interconnection expense. Indeed, the ultimate result is streamlined provisioning time and faster time-to-market (see figure).

The technology short-list for all-optical switching has essentially come down to waveguide-based devices like lithium niobate and solid-state devices, or liquid crystal and microelectromechanical systems (MEMS) (see table).

MEMS-based optical switches currently appear to be emerging as the most viable means of building next-generation fabrics, when all factors are considered. (The lithium niobate design, for example, has been plagued by crosstalk problems and high loss, which limits its use to lower-scale applications.) MEMS is favored because of its inherent batch-fabrication economics, insensitivity to bit rate, and the ability to exploit high-density interconnection of optical beams in free space. In addition, variations on the architecture are emerging as strong contenders for wavelength add/drop multiplexing and other applications.

For scaling to large sizes (1000 ports and beyond), three-dimensional analog (3-D) MEMS are preferred to two-dimensional (2-D) digital MEMS. Representative 3-D MEMS vendors today are Calient Networks (San Jose, CA), Lucent Technologies (Murray Hill, NJ), Nortel Networks Corp. (Brampton, Ont., Canada), OMM (San Diego, CA), and Tellium Inc. (Oceanport, NJ). Three-dimensional MEMS scale to thousands of ports, which is where the carrier demands are going. In addition, 3-D is a single-stage design, exhibits low loss, and does not suffer from polarization impairments.

Even as the preferred technology, 3-D MEMS offer challenges to designers at the present time. One challenge is control: the switches are ultimately subject to analog feedback-control loops for multi-positioning. This structure represents an interesting irony. Early voice networks were completely analog. Current infrastructure is almost all digital. Now, the future of optics depends on analog system engineering and design for both switching and transmission. While bit-stream capacity has gone up by a factor of a million thanks to optical networking, the design cycle appears to be repeating itself.

Another challenge is reliability. In MEMS there are, of course, moving parts. The performance is highly promising—there is total agreement on that. But the reality is that no vendor has yet proved the reliability of this technology at any level. This is a major design and engineering challenge presently faced by all vendors.

Packaging—and therefore volume manufacturing—comprises a third challenge. These devices are subject to very stringent alignment requirements. For example, with a 1000 x 1000 port array, light from 1000 incoming fibers has to be focused on 1000 lenses, which have to be focused on 1000 mirrors, which in turn have to be flipped back and forth accurately by analog control. This all requires a perfectly aligned, tuned, and stable package.

Engineers are also faced with installation environment challenges. Will devices need to be hermetically sealed? If so, how will this be achieved? What happens to any dust, smoke, or other impairments, and how would this affect system performance? What is the tolerance to vibration and indeed to minor earthquakes? Will there be severe deployment constrictions when installing in existing buildings?

Another challenge is migration at the software layer that makes up the operating system controlling the optical layer, and at the management system that interfaces with provisioning tools. The software must be designed so that upgrades do not disturb in-service OEO switches. This migration path preserves carrier's switching investments and helps them stay competitive.

There is no doubt about the promise and excitement of new all-optical technologies. But the reality check is this: commercial volume-deployable all-optical technology probably will not make it to the market until 2002.

As a result, carriers will have technology breathing room in the near term. There is a commercially viable technology path from OEO to all-optical in terms of today's OEO platforms. Ten terabits per second of total throughput is possible using electrical fabrics. The necessary application-specific integrated circuits will be available to vendors in the near term. Consequently, a 512-port OC-48 switch, currently operating at 1.28-Tbit/s total capacity, could be scaled in steps by a factor of eight.

Looking ahead, all-optical switching promises to shave equipment investments by eliminating the need for numerous OEO conversions along the transmission path, while providing fast, transmission-speed-independent service. The delivery of this promise has taken a giant step—developmentally, all-optical designs have left the realm of physics and progressed into the realm of engineering. This is a key milestone on the road to commercial implementation.

J. Robert Palmer is a technology writer based in Holliston, MA. He can be reached at 508-429-5713 or

More in Home