The rollouts beg the question of whether a true all-optical packet-switching network will ever be necessary or even desirable. As Rajiv Ramaswami, vice president of systems architecture at OXC startup Xros Inc. (Sunnyvale, Calif.), noted, even if large, dense arrays of all-photonic switches prove viable, there is still no memory technology to use for buffering optical packets. Also, the need to read a packet header to perform true "routing" would seem to imply optoelectronic conversion.

At the very least, the new crop of transitional OXC and OADM architectures appear on track to enjoy a long life as telephony networks phase out first-generation Sonet and move to data-dominated networks.

Two vendors of metropolitan aggregators will introduce at this week's conference systems that are more than mere OADMs. Astral Point Communications Inc. (Chelmsford, Mass.) will show an Optical Services Node that combines a multiservice electronic switching fabric, an array of optical transponders and a passive OADM front end in a single rack-mounted system. And Chromatis Networks Inc. (Herndon, Va.) will show off its first Metropolis system, combining a configurable electronic switching fabric for packet traffic with ASICs that allow multiple protocols to be carried on one wavelength, and an interface to allow selective use of dense wave-division multiplexing (DWDM) on the system outputs.

On the OXC front, Xros (pronounced "Ki-ros") will demonstrate a massive 1,152-port cross-connect system based on the company's patented bulk-micromachined micromirrors, which allow very high-granularity traffic assignment using all-optical signaling in the data path. Ramaswami claimed that the system can scale better than OXCs based on multistage switches using liquid crystals and other technologies, such as those from Tellium. He also maintained that the micromirrors are easier to control and cheaper to manufacture than those in Lucent's LambdaRouter designs.

Fewer racks

Though the system's port count is dense, it occupies only three NEBS racks. Less-dense OXCs from competitors require four racks.

The rollouts will take place in an Optical Fibers Conference environment characterized by more OEM participants than in past years. A conference once seen as primarily for component makers has attracted nearly as much optical-system interest as the National Fiber Optics Engineers Conference last fall.

OADMs have only been in a position to replace Sonet add-drop multiplexers in the past two or three years. Nevertheless, optical-system developers already are being pushed to show that systems can do more than selectively add or drop wavelengths, a task now seen as merely a small step above passive DWDM equipment.

Bill Mitchell, vice president of marketing at Astral Point, said it wasn't just an issue of carriers demanding more and more for complex service provisioning. "The service providers were being told to yank out all their old TDM Sonet and replace one node with a few boxes," Mitchell said, "and a lot of them said, 'You've got to give me a better argument than that.' "

To do so, Astral Point first has to convince metropolitan carriers that a self-healing mesh, similar to the mesh topologies used in long-haul optical networks, is superior to a ring topology. Founder and CEO Raj Shanmugaraj said that the move to a mesh makes it possible to implement a path-protection algorithm, developed by Astral Point, which allows sub-50-ms protection switching in a distributed fashion. That, in turn, provides a hierarchy of protection levels and allows protection paths to extend over different media — for example, fiber and copper leased lines.

The hardware architecture supporting this mesh is a three-part system, configured as a modular single chassis in a seven-foot rack priced between $30,000 and $250,000. It integrates a configurable switching fabric that can handle ATM cells, Internet Protocol packets or TDM circuits; a bank of transponders that can support 16 ports of OC-3 fiber, four ports of OC-12, or a single port of OC-48 or Gigabit Ethernet; and a passive OADM mapping wavelengths to granularities down to T1 levels. Multiple ON 5000 Optical Service Nodes are linked through OC-48 trunks called Virtual Optical Tunnels.

The carriers can use the new optical nodes while TDM Sonet nodes remain in the network, yet the system offers the immediate advantage of guaranteed protection levels across a mesh rather than a bidirectional ring. It also lets the carrier provision pre-emptible services at will.

The Astral Point systems are entering beta trials this month. Advanced TelCom Group Inc. already has committed to buying up to $25 million worth of them.

At Chromatis, the rationale behind the company's Metropolis 2000 and 4000 systems was to keep the cost of optical links to an absolute minimum, said vice president of marketing Doug Green. Even in systems with ATM fabrics as large as typical broadband switches, he said, the optics still consume 80 percent of system costs.

Last summer, Chromatis touted the concept of bringing up DWDM-enabled links on a node-by-node basis within a ring, making wave multiplexing dynamically selectable. Not only do the Metropolis units implement this feature, they also allow the protocol multiplexing of multiple packet and circuit types across a single wavelength channel, further cost-optimizing any fiber connections.

Chromatis has not fully disclosed the protocol-multiplexing technology, but Green said it involves off-the-shelf TDM processor technology that allows encapsulation of traffic in "digital wrappers."

Redundant fabrics

The seven-foot Metropolis 4000 chassis is targeted for central offices and large carrier installations. The four-foot Metropolis 2000 chassis is designed for smaller venues, such as office buildings, apartment houses and small carrier locations. Both systems have redundant internal switching fabrics for ATM, TDM and packet traffic, with an aggregate bandwidth of 140 Gbits/second in the 4000 and 70 Gbits/s in the 2000.

Selective WDM implementation is carried out via shared OC-48 channels, which can handle up to 32 wavelengths in the 4000 (or four in the 2000). Selectivity concepts are designed around rings, though Green said that Metropolis can support ring, mesh or star topologies.

In optical cross-connects, meanwhile, Xros is leveraging four years of experience with micromirrors to offer a cross-connect of unprecedented size. Ramaswami, who joshed that Xros' founding in early 1996 made it "a real old-timer in the optical community," said the company perfected its manufacturing methods over the first three years of its existence. Xros closed its first round of financing in January 1999, with a decision to concentrate on optical-transmission end systems for service providers.

That means that Xros will not offer mirror arrays as component-level products, Ramaswami said, nor will it license its technology to others. The company hopes to move beyond OXCs by using mirror arrays in next-generation optical gear that includes active grooming and mapping of traffic, but will not do anything to destroy its proprietary edge in micromachined components.

The X-1000 system being shown at the Optical Fibers Conference is intended for backbone reassignment and aggregation, and cannot groom to TDM rates such as T3. Future products will groom traffic at 622 Mbits/s, but probably no lower.

The OXC inherently supports sub-50-ms protection switching, and can handle any optical protocol at unlimited bit rates across 1,152 nonblocking ports. Xros says it has developed unique nonintrusive test-access ports — not easy in a system with no electronic switching in the data path.

Mirror arrays

At the heart of the X-1000 are two mirror arrays of 1,152 mirrors each. These optically flat devices mount on two hinges that shift the mirrors along two axes. The mirrors use single-crystal silicon rather than polysilicon. Special sensors in the hinges enable an unprecedented degree of control, allowing the mirror to shift a laser beam to more than 1,000 different positions. The technology is capable of scaling to 4,000 ports over the next couple of years, according to the company.

OXCs do bandwidth provisioning and circuit protection, plus pass-through and add-drop functions on a backbone.

The OXC represents the final stage where interoperability of optical signaling is necessary, Ramaswami said. Xros belongs to the Optical Domain Service Interoperability coalition, but sees that group primarily as addressing interoperability from electronic-based service-layer equipment to the OXC. New efforts are needed in the Internet Engineering Task Force or Optical Interoperability Forum to ensure that the Xros OXCs can talk to those from other companies, Ramaswami said.