OPTICAL NETWORKS PACKET UP

The network is collapsing at the hands of the equipment vendors that helped to build it. Facing an inevitable migration from circuit-switched networks to more cost-efficient packet-based ones, carriers are looking to replace the stacked equipment of their Sonet networks with packetized optical ones.

To help them create the optical Ethernet network of the future, equipment vendors are introducing a new class of products that some analysts are calling packet optical networking platforms, or packet ONPs. Though different products in this category have varying degrees of Sonet or SDH functionality, at its heart the packet ONP combines packet traffic, usually Ethernet, and DWDM. Though the space includes some very new gear, its importance is hard to overstate. According to Dell'Oro and Ovum-RHK, the market for converged Ethernet transport and switching equipment will grow from $3 billion last year to $5 billion next year.

“Packet ONPs are the next major evolution in optical networking,” said Michael Howard, principal analyst for Infonetics Research.

There are principally two kinds of packet ONPs: those that combine packet traffic (usually Ethernet) and wavelengths, and those that combine those two plus Sonet (or, in some geographies, SDH) so that carriers can gradually wean themselves off Sonet at their own pace. Packet ONPs also seek to grant packet traffic the predictability of circuit-switched traffic as it traverses the network. That can be accomplished in a number of ways, but the most popular choices are MPLS — with or without using pseudowires, transport-MPLS (T-MPLS) and provider backbone transport (PBT), also called provider backbone bridges-traffic engineering (PBB-TE).

As the packet optical space is comprised of combinations of previously separate gear, it is being approached from all directions: the Sonet segment, optical vendors, Ethernet switch-makers and more.

The latest entry in the packet ONP space is Fujitsu Network Communications' Flashwave 9500 platform. Available in this year's fourth quarter, the 9500 offers 480 Gb/s of Sonet or packet switching in a quarter of a shelf. In addition, a 2°, quarter-shelf reconfigurable optical add/drop multiplexing (ROADM) plugs directly into the chassis (upgradable to include more degrees) with a single 10 Gb/s transponder.

“Some ROADM vendors have Sonet [add/drop multiplexer] on a blade,” said Sam Lisle, Fujitsu's director of market development. “Some ROADM vendors have big Sonet [switch] fabrics. They have a ROADM shelf and a Sonet shelf that sits underneath. We do ROADM and a fully grooming ADM on the same chassis.”

Fujitsu doesn't want to be religious about which technology — PBT, T-MPLS or MPLS-based pseudowires — is the best one to use for connection-oriented transport and maintaining quality of service (QOS). But it is focusing its energy first on pseudowires because it is the furthest along in the standards process.

A year ago, Lucent Technologies unveiled its own packet optical platform called the Universal Packet Mux (UPM), a 280 Gb/s box with 40 Gb/s interfaces that would combine Sonet, packets and wavelengths. It would commence customer testing in the fall of 2007, Lucent said then, and become generally available early in 2008. Before the year was over, however, Lucent became Alcatel-Lucent, and the UPM joined the same portfolio as another packet ONP, Alcatel's 1850 Transport Service switch.

Having sold the SDH version overseas since late 2005, Alcatel brought a 100 Gb/s Sonet version of the 1850 to North America last fall. And more recently, the company has integrated some ROADM functionality into the 1850. With the 1850 and the UPM potential competitors under the same roof, analysts have had more doubts about the UPM than the more mature 1850. At press time, Alcatel-Lucent had not responded to queries on the subject.

Last fall, a new start-up vendor, Matisse Networks, emerged from stealth mode to make an end-run past Sonet networks to optical Ethernet. Matisse's approach, based on a technology called optical burst switching, relied on two boxes working together: a photonic switch that is not packet-aware, and an Ethernet switch that maps Ethernet traffic into optical packets of varying wavelengths. Traffic that enters the system is aggregated into wavelengths based on parameters such as QOS levels or destination. And Matisse's system rapidly retunes itself, in 50 nanoseconds or less, to send bursts of optical packets where they need to go. Because the system isn't designed to deal with legacy traffic, its designers imagine it to gain traction first among enterprise customers with a history of using private Ethernet networks. Internet service providers and CLECs could follow, the company believes, but as Bell companies move away from Sonet and other legacy technologies, Matisse expects to become increasingly relevant to them.

Having researched optical burst switching, Fujitsu determined it to be latency-sensitive and subject to concerns regarding reach. “You have to ask: What's the circumference of the ring?” Fujistu's Lisle said. “How big can I build the network — 10 kilometers or hundreds of kilometers across a metro?”

Cisco Systems planted its flag in the packet ONP space in late 2005, when it announced an IP-over-DWDM configuration for its CRS-1 core router. Though it's not clear how widely deployed that configuration is, the CRS-1 has become big business for Cisco, which recently reported an annual run rate approaching $1 billion in revenue for the CRS-1 alone.

While Cisco exhorted carriers to collapse their optical and IP networks, Meriton Networks has more recently taken the opposite approach, introducing an optical transport architecture that uses PBT to keep transport and service layers separate.

According to Infonetics' Howard, however, Cisco's ON 15454 multiservice transport platform, which has been around since the late 1990s, also competes in the packet ONP space, thanks to adaptations the vendor has made to later versions of the product.

As packet ONPs combine multiple functions, the lines defining this category can be blurry. For example, some analysts say Ciena entered this space two years ago last month with its CN 4200 multiservice transport platform — among its chief selling points was the flexibility that allowed carriers to remotely program any service to any port. The 4200, which today brings in tens of millions of dollars in revenue per quarter, has been particularly well received among European operators such as British Telecom and Swisscom and among North American cable companies. (It came through Ciena's 2004 acquisition of Internet Photonics, which had racked up several top cable companies as clients before it was acquired.)

However, other analysts say a newer Ciena product is the company's answer to packet ONPs: the CN 5060 multiservice carrier Ethernet platform, announced in February. Designed by the team that came to Ciena through its acquisition of Wavesmith Networks, the 5060 offers 60 Gb/s of switching capacity in 2.5 Gb/s and 10 Gb/s increments as well as 576 Gigabit Ethernet ports — or 96 10 GbE ports — per rack; it also supports up to 384,000 pseudowires per shelf.

Because packet ONPs pull more functionality into the same box, they will force network operators to learn how to master whichever part of the box is new to them.

“Each box is a little more complex, so carriers have to learn how to deal with that,” Howard said. “But it will be easier to manage once they get over initial learning period. It will be simpler because there will be fewer layers in the network.”