Metro networks were the slowpokes of the optical networking boom, and most analysts and executives say there's no way to go but up — partly because the MAN lagged in the great buildout of recent years. Service providers pushed hard to beef up their core and access networks, "but the nexus in the middle, the metro, has sort of been left behind," said Ramji Raghavan, chief technical architect for Coriolis Networks Inc. (Boxborough, Mass.).

The number of startups in the area — the figure kicked around at NFOEC was 80 — has kindled concern that the market is due for a collapse. But the metro market is broader than many people realize, said Peter Wagner, general partner with venture capital company Accel Partners (Menlo Park, Calif.).

The metropolitan market consists of four or more distinct segments that each could sustain winners, Wagner said. Metro core transport provided by such companies as ONI Systems Corp. (San Jose, Calif.) is a different sector than the packet-service switches being touted by Redback Networks Inc. (Sunnyvale, Calif.), for example.

Whatever the market conditions, startups continue to be enthused about the potential of the metro market, and several NFOEC papers addressed ways to prepare it for heavy amounts of data, specifically Internet Protocol (IP) traffic. The barrier seen by many was the voice-oriented nature of existing metro networks, which are laid out in rings and carry traffic over Sonet.

In addition, rings need to be replaced by a mesh network, some said. Emmanuel Nachum, vice president of marketing for Lightscape Networks (Herndon, Va.), presented a proposal for migrating ring networks to a mesh topology, which gives each node connections beyond its two "neighbors" in a ring.

Nachum noted that a mesh makes it possible to upgrade each node without affecting the rest of the network, a luxury unavailable with rings. In addition, a mesh arrangement allows nodes to be added with point-to-point connections, a process simpler than adding nodes to a ring, he said.

Another critical factor lies in the control plane, where some of Sonet's network awareness could be added to the IP network, some said.

Coriolis' Raghavan suggested that an entirely new control plane was needed for metro networks, one that handles signaling, routing protocols and automatic neighbor discovery. The latter "becomes more important when you're moving from ring networks, where Sonet already has this built in, to a mixture of ring and mesh networks," he said.

Claus Popp Larsen, manager of transmission design for Wavium AB (Stockholm, Sweden), noted in his paper that more intelligent control was necessary in order to overlay an IP network onto an existing optical mesh. Otherwise, the IP routers have to be configured manually to be aware of their neighbors, a process that can take weeks, he said.

Signaling standards such as generalized multiprotocol label switching (GMPLS) are intended to add that kind of control but are still being finalized. For now, Wavium is using proprietary signaling to get IP nodes to communicate with one another, Larsen said.

A next step would be to let customers demand bandwidth themselves by having the network configure itself automatically. It's a nice idea but impractical in current carrier setups: "I don't see that happening yet, because the whole provisioning procedure for the networks is tied in to the order-entry systems," Raghavan said.

On the topic of scalability, two researchers presented studies that suggested how the metro network needs to adapt to higher traffic volumes.

Colin Kelly, director of optical architecture for Tropic Networks Inc. (Ottawa and Andover, Mass.), outlined his company's finding that wavebands — wavelengths bundled together and routed as a unit — helped lower the cost of building and operating the network, but only to a point. Hypothetical networks using wavebands wherever possible got to be more complex and expensive than those that forced wavebands to contain more than two wavelengths, Kelly said.

His report showed that wavebands lower the number of transceivers required. At the same time, heavy use of wavebands forces the use of more amplifiers, which add expense on their own but also affect optical signals to the point that filters need to be added.

For ring-based metro networks, dense wave-division multiplexing (DWDM) may be the key to scaling. Ondrea Wasem, principal engineer for Network Design Tools Inc. (Eatontown, N.J.), presented a real-world case study in which DWDM became more cost-effective than Sonet as traffic volumes increased. The crossover for this particular case occurred when traffic reached three times its existing levels, she said.

Metropolitan rings do remain viable and have spawned movements such as the resilient packet ring standards effort. Even beyond RPR, suggestions continue to emerge for putting Ethernet on metro rings.

Systems architect Peeya Iwagoshi of LuxPath Networks (Dublin, Ohio) presented an Ethernet add-drop multiplexer to enable bidirectional Ethernet rings. Similar to a Sonet add-drop mux, the part would take in 10 lines of Gigabit Ethernet and merge them into a 10-Gbit Ethernet stream. On the receiving end, the stream would be broken into its original components with the channel number staying intact for each of them.

The device would be used in fiber rings so that in the event of a fiber cut, traffic could be routed the other way around the ring and still reach its destination, Iwagoshi said.

RPR isn't necessarily suitable for this use, said Iwagoshi, because it's a more complex way to arrive at a similar functionality. "RPR requires a lot more complex configuration and complex devices to implement its algorithms," he said.

Suresh Gopalakrishnan, executive vice president of engineering for Riverstone Networks (Santa Clara, Calif.), presented a proposal for combining 10-Gbit Ethernet and DWDM in metro rings. Like Iwagoshi and others, he saw 10-Gbit Ethernet as an aggregation vehicle for Gigabit Ethernet streams.

"This is what is going to drive the initial deployments of 10-Gig, and a lot of what the standards body is doing is to address this particular need," he said.

DWDM is still too expensive for mass deployment in the metro space, but it could improve the cost of 10-Gbit Ethernet rings if passive DWDM components, rather than active ones, are used to link routers to rings. This would remove the extra optical-to-electrical conversion between the DWDM equipment and the router, simplifying the network, Gopalakrishnan said.

One key to that implementation would be to retain some active components in the "passive" DWDM system for functions such as signal monitoring. "You could still have some amount of active components in there — not in the optical space, but to collect bit-error information, for example," he said.