Voice as smooth as silk, stock quotes as steady as a heartbeat, firewalls as secure as a mothers love — its a lot to ask of a network built on threads of glass as thin as human hair.
But its where the Internet is going — to a simpler, optical architecture that marries technologies, layers and protocols that have always lived in isolation. This year and next year, optical networks will begin assuming the elegance and convergence dreamed up in company labs and demonstrated in gee-whiz technology trials.
For years, optic fiber was like the kid with unpolished talent who shows up for football practice. Yes, speed to die for. But until it learns how to feint and switch, groom, scale, massage and account, its nothing but a big, fat, dumb pipe that you wouldnt want operating by itself in your biggest money games.
But now its showing brains as well as brawn.
This month, WorldCom completed the first phase of its Terabit Challenge, sending data at 1.6 terabits per second over a hair-thin pair of fibers connecting New York and Washington, D.C., and laying a foundation for services such as Web hosting and virtual private networks. WorldCom achieved those speeds using Fujitsu Network Communications Flashwave Dense Wavelength Division Multiplexing (DWDM) equipment that divides light into dozens of wavelengths, each capable of carrying its own traffic, and Nortel Networks Optera optical switch, which is capable of supporting 160 wavelengths.
Also this month, Celox Networks introduced its Services Creation Switch, which operates at 80 gigabits per second on the networks edge and lets service providers carve out 16 different levels of service, from firewall protection to real-time billing.
Ma Bell isnt ready to leave the new game to her descendants. Earlier this month, AT&T announced that it was using Avici Systems router to connect six cities coast-to-coast on an Internet Protocol (IP) backbone carrying traffic at OC-192 (10-Gbps) speed. Avici in 1999 was the first to announce a router capable of achieving speeds of 1 terabit per second, but winning AT&T as a customer was an important validation of its high-speed technology.
At least a dozen other optical component and gear makers have dev ices in the works to help redesign the network architecture.
The aim is to be invisible to the end user, to achieve such reliability that the consumer or the office worker doesnt know or care how the architecture came together, or where the fiber starts and the copper takes over.
Building the optical network with smart components is essential, because without them providers cant offer the platinum-, gold- and silver-level options that are the roads to profits. If it costs $30 per month per subscriber to make and maintain a connection but there is just one level of service available, providers will submarine each other, offering prices a hairsbreadth over costs. But if end users are willing to pay, say, $70 per month for service that helps them secure and organize their home offices and participate in the weekly videoconference calls, thats a lot more profit with very little extra infrastructure cost.
The companies that will make it happen run the gamut from the gargantuan high jumpers such as Nortel to the lithe sprinters such as Amber Networks.
Mindful of Consumers
Keeping in mind what the consumer wants never hurts.
At Amber, the goal is equipment that gives the IP network the same 99.999 percent availability that the circuit-switching telephony world enjoys.
That “five-nines” availability means the network shouldnt be out of service for more than four minutes per year. Today, a single failure of a traditional router might mean that service is out for five, 10 or 15 minutes; it can take that long for the routers to relearn the routing information.
Amber has built an edge router from the ground up, with hardware and software capable of restoring services in the IP services layer in 50 milliseconds — as fast as Synchronous Optical Network (SONET) can do it. A single router does the jobs of both layer two and layer three of todays architecture.
“We build the instrument in such a way that the other peer routers are unaware of the failure,” says Amar Gupta, chief technology officer at Amber.
Overall, the goal is to simplify without sacrificing speed, to develop a box that can do multiple tasks and that is technology agnostic — that is, it keeps running as a network evolves from Asynchronous Transfer Mode (ATM) to SONET to IP and Multiprotocol Label Switching.
The first-generation optical network employed an optical fiber with terminal equipment, and required three devices that changed the signal from optical to electronic and back to optical for each switch. The second generation, featuring microelectronic switch fabric from the likes of Lucent Technologies and Nortel, still required one optical-electronic-optical (O-E-O) switch on either side of a space switch.
Thats a lot of boxes in the central office, layer after layer of cards to convert from one format to another. And each conversion from optical to electronic and back to optical adds jitter to the network.
“At the end of the line, youre going to want to make as few as possible O-E-O conversions,” says Chien-yu Kuo, Cintas founder and CTO. “What carriers were looking for were vendors who could simplify the functionality, and integrate wavelength conversion and space conversion into one box,” directing signals to different fibers.
Cintas new router integrates transport and wavelength switching into a single device with just one O-E-O conversion, switching to one of 40 wavelengths. The year-old San Jose company hopes to have its box carrying live traffic in networks by the end of the year.
“Were not simply consolidating the switch and the terminal together,” Kuo says. “Were actually making the components a lot smarter. Each component serves double duty. The wavelength conversion is part of the switch, and vice versa.” Only a ground-up architecture aimed at including two functions in one device can meet the function requirements of the future network, he says. Software resides on every node to detect any command along the pipes for reserving or provisioning wavelength. “And we make it happen very, very fast — as fast as 50 milliseconds for a mesh or ring configuration. Thats faster restore than SONET.”
The optical network continues to move closer to the customer. Now dominant in the core and long-haul parts of the network, over the next couple years it will likely also become dominant in the metro space, where bits and bytes crash, screech and make U-turns around office parks and campuses.
In March 2000, Zaffire announced the industrys first packet-enabled DWDM solution for the metro area. Zaffires architecture divides light into dozens of colors, each able to carry its own traffic. It gives service providers a new level of control over packets, and slashes network maintenance costs, Zaffire officials say.
The first DWDM technologies were developed by Lucent and Nortel for the long-haul networks — for example, Denver to Chicago, or New York to Atlanta. Long-haul vendors that tried to take that same technology and apply it to metro networks were putting a square peg in a round hole, says Zaffire CTO Rob Keil.
Zaffire put 15 doctoral-level engineers on the problem of creating a DWDM from scratch for the metro market.
“We looked at a brand new architecture and a dramatically lower price point,” Keil says. Zaffire offers its system for $12,000, compared with $50,000 to $80,000 for earlier systems.
Space in a central office in a large metro area is becoming a precious commodity, with carriers and providers scratching for market share, and vendors dreaming up cant-do-without services. So, along with putting more oomph and flexibility in switches, routers and processors, gear makers are faced with a need to make them smaller, too.
“In the metro, we dont have room or floor space to be able to handle each network element individually anymore,” says Dana Hartgraves, senior vice president for marketing at Metro Optics. “Youre beginning to see the equipment integrated.”
Metro Optics City Stream platform sits on one shelf, taking up a third of a 7-foot rack, and integrates SONET; Synchronous Digital Hierarchy; and ring, linear and digital cross-connects. It can scale from narrow — 44 megabits per second — to fat — 2.5 Gbps. Scheduled to be made widely available at the end of this month, it combines the capabilities of ATM and IP routers, and the grooming and handing-off functions of old M13 multiplexers.
“CLECs [competitive local exchange carriers] are tired of needing seven or eight racks of equipment to do a cross-connect in a central office,” Hartgraves says.
No one is going to tear out the legacy networks that can still earn money, but the regional carriers will converge their systems as quickly as makes fiscal sense.
“Two years from now, everything will be migrating to one box,” Hartgraves predicts. “Voice and data will be sent over the same networks, and you wont have to have different elements to accommodate data and voice.”
Products such as those offered by Metro Optics let the ATM switching and cross-connect grooming happen at the edge, closest to the end user. So local traffic from, say, office to office or neighbor to neighbor doesnt have to go back to the central office where another big cross-connect switches, grooms and sends it right back to the edge.
With convergence and smaller footprints will come the flood of dream-come-true services such as first-run movies to the home on demand. “Two years? Three years? It all depends on how fast they get these networks converged,” Hartgraves says.
Its not enough to dazzle with throughput. What service providers demand is a switch that can send light down paths toward firewall security, real-time billing and data storage.
Looking for a way to create higher speeds and faster pipes, PhotonEx gathered a team of 20 workers with doctorates from such varying disciplines as satellite and radio frequency technologies. Together, they “shattered boundaries that people felt were limited by the laws of physics. We created a super-high-bit-rate backbone that will be much more flexible and services-aware than the systems of today,” says Chief Executive Kristen Rauchenbach. Late last year, the company introduced a 40-Gbps all-optical switch for the core of the network — four times faster than anything deployed today.
“Going fast over the long haul is the major challenge,” Rauchenbach says. “You can supply a solution thats four times faster, but four times harder to use. Or you can create something thats four times faster and simpler to use.”
An industry working group is hammering out rules to standardize Resilient Packet Ring (RPR) technology — a packet-optimized access architecture — for the metro sector of the optical network.
Dynarc, a start-up founded two years ago that builds routers to channel traffic onto fiber in the metro ring, is counting on RPR technology becoming the industry standard. Dynarc makes a router that handles and analyzes packet traffic and reads the headers to determine their priority. So if an employee is downloading the Victorias Secret catalog while videoconferencing with his boss, neither task suffers in quality because theyre separated and treated on different channels. And with Dynarcs router, each channel of the fiber can offer a different quality of service and can guarantee, for example, billing on a per-application basis.
“The router understands these are separate applications and can guarantee different traffic performances,” says Frederick Hanell, vice president of marketing at Dynarc.
“If you double-click on a video-on-demand site, the network understands that youre asking for video to be supplied, and the right quality is allocated to that applications,” Hanell says. “Youre suddenly allocated the bandwidth you need to watch the video, with the quality of service you are paying for.”
So what are the rules of survival in the era of optical network architecture?
“Understanding the market space, having team members that understand service providers and their requirements,” says Alex Dobrushin, vice president of marketing at Amber. “The days of comparing products on price alone are gone. Service providers are looking for products and companies that give them a sustainable advantage in the marketplace.”
The casualties will be the vendors that focus on just one type of network, Hartgraves says. “You need a flexible platform that will grow as bandwidth demand grows, as protocols change. Those that have platforms that arent easily upgradable — those that cant grow without throwing away things —wont survive.”