Educating Routers

As carriers demand dynamic provisioning
along with greater aggregation
and granularity, it's time to speed up
the network's learning curve.

You can't teach an old dog new tricks, so the saying goes. But in this day and age, engineers are searching for ways to use the intelligence in routers to teach networks how to better meet customer needs. Carriers want flexibility to offer customers bandwidth-on-demand and to be able to shift network resources where needed in real-time. Most industry experts agree the way to get there is by adding more intelligence to routers or at least to better use the intelligence routers now possess.

"A router already knows where the packet is going, all the way down to the building destination, but then says, 'That's east.' The router is intelligent, but it has to learn how to use more of its intelligence," says Richard Norman, president and chief, technology office, Hyperchip, Montreal, Canada.

Carl Blume, director of product marketing for IronBridge Networks in Lexington, Massachusetts, agrees that the industry's view of what of a router should be has changed.

"A couple of years ago, the idea was that core routers would be fast and dumb with edge routers that are relatively small and intelligent. Now you have put on a different spin. You need speed, capacity, and some intelligence directly out of the backbone," Blume says. "By putting more intelligence into the router, it allows the hardware to be configured in novel ways to build virtual routers."

IronBridge, he says, has developed virtual routers, allowing a single terabit router to be subdivided so that smaller routers can be resold as services to wholesale customers who need large bandwidth but do not have the capital or fiber plant to build their own fiber networks.

Another factor driving the push toward smarter routers in the core network is the need for faster provisioning.

"The optical core is labor-intensive. It takes months to provision," says Amy Copley, senior product manager for Sycamore Networks in Chelmsford, Massachusetts. "Instead, carriers need to add provisioning of the optical core to reduce the time to hours and minutes to bring up a circuit by pointing and clicking. Then you have smart algorithms that say when you need bandwidth, you can ask the core."

"The Holy Grail is a self-provisioning, self-healing network," adds Leon Woo, chief technology officer for Tenor Networks in Acton, Massachusetts.

And today, everyone seems to want routers that are faster with more capacity, yet consume less power, take up less space, and can create new network paths instantaneously.

"To scale a network, the consensus is that the network has to be packet-based. In optical, people say the network should be lambda-based. But there is the same problem with circuit orientation. Data is bursty by nature. If it bursts beyond the circuit or lambda, you can't borrow bandwidth or at least it is not easy to do that. Some try to overlay the ability with bandwidth pooling or gain, but it has been tricky," Woo says.

"To some degree, when you try to give gain to a circuit network architecture, it is like putting a square peg in a round hole. We propose that you need bandwidth pooling points. It allows you to remove the complexity or at least ameliorate the complexity of a circuit problem. It allows you to refine bandwidth into a service billed on a differentiated basis."

"Broadband has changed things," notes Steve Akers, chief technical officer for the internetworking systems business unit of Lucent Technologies in Maynard, Massachusetts.

"There is so much more bandwidth in the core, you need something intelligent to direct it," Akers says. "Our view at Lucent is to use virtual routing at the edge with lots of intelligence and switching at the core. You want to reduce the multi-hop boxes across the country, signal the network for paths, and build them dynamically. This is a big problem people are trying to solve."

"Point and click gets you at the end of the day," adds Jeff Low, vice president of business development for Ericsson based in Burlington, Massachusetts.

"Operators want to deploy and deliver new services. They want new revenue streams and ways to derive revenue. And they want to dramatically reduce operations costs. Any operator wants to know, 'What are the new services I can offer?' and 'How can I bill for them?' " Low says.

In today's Internet economy, service provider customers also need an ability to quickly sense when a lot of traffic will hit Web sites and then bring up more bandwidth as required, Lucent's Akers says.

"The core has plenty of lambdas. It's getting at it when you need it. Historically, routers have not been flexible enough to know they are running out of bandwidth and need more or need to better prioritize traffic. Ours is aware and treats each application as a flow. It is a smart emerging network."

The model, Akers says, is to have the network recognize specific users and what they are allowed to do. "The user can go on a Web page maintained by a service provider and augment it. The user can upgrade with a click of a mouse and have a premium service for a given length of time, and then we tear it down. It doesn't tie up the resources, and the user gets a bill for a specific amount of time. That's a lot different from a router that blazes packets."

Interface Lessons

Another key advance is the work of several standards bodies to create a management interface between optical technologies and routers. The mission of the Optical Internetworking Forum, for example, is to foster the development and deployment of interoperable products and services for data switching and routing using optical networking technologies.

Why? If you build a management interface between the router and the optical layer, explains IronBridge's Blume, the unused bandwidth could be made available for other applications. The ultimate goal, then, is to give carriers more capacity and scalability along with bandwidth efficiency so that they can offer customers more services over that same bandwidth.

"This is important because routers have a dynamic requirement for bandwidth. Let's say there is a router in Boston that connects to a router in Chicago. It may need only one channel between 8 a.m. and 4 p.m., but between 4 p.m. and 8 p.m., traffic increases and it needs more. Right now, to satisfy that requirement, the carrier must provision for the peak and have it available all the time. That's very inefficient," Blume says.

"The Internet, application service providers (ASPs), and e-commerce are all things that are driving more traffic. Then you bring in storage area networks, 3G, and all these other technologies that will produce more demand for bandwidth. The market is becoming richer and richer," Blume says.

Fredrik Hanell, vice president of global marketing for Dynarc in Sunnyvale, California, says decision-making for the network also is being pushed into main aggregation points where terabit routers are positioned.

"Depending on the way architectures are implemented and the way the optical backbones are developing, there is a big need for terabit routers at ingress into the backbone before some of the WDM equipment. One dream is that every box and every router in the metro and access networks will support intelligence and be able to function independently," Hanell says.

"The problem is not about bandwidth only anymore," adds Paul Liesenberg, director of marketing for Zettacom in San Jose, California. "It is also about complexity to turn the Internet into a true infrastructure for telecommunications. That means not just forwarding based on a destination, but with more granularity and more services. What you will see happen is different people will focus on different core competencies. System designers will focus on services, manageability, and the ability to provision services."

Speed Sessions

As the market grows, so does the need for routers that handle more traffic faster. The already huge and growing demand for services created by the increased capacity and connection speeds for businesses and residential users worldwide means a greater carrier need for equipment that can scale to handle IP traffic beyond gigabits.

Nortel Networks calls the combined factors of more users, higher connectivity speeds, and higher transmission speeds a "bandwidth tornado" that will eventually create a high-performance Internet for everyone and a number of new opportunities for service providers to create more revenue.

That's why so many vendors are either delivering or plan to deliver terabit switch/routers soon and why some are even talking about a petabit, which is equal to a quadrillion bits.

But Tenor's Woo says it is interesting to note just how people define "terabit" routing and how numbers are aggregated within system architectures. "What people are doing to reach terabit is to aggregate forwarding engines with the appearance of other protocols aggregating together. The other confusion is that people don't talk about input/output (I/O). The internal architecture is a terabit, but the I/O is 64 Gb/s on a particular rack. It's hard to compare apples to apples."

Some, including Tenor, he says, count input and output to make it a bigger number.

"I haven't seen a terabit router, but it is good slideware," says Jeff Low, vice president of business development for Ericsson in Burlington, Massachusetts.

Woo also warns about the trade-off involved for more speed and capacity. "You might be able to design a faster product, but what if it takes two buildings of equipment or 4.5 megawatts of power? There are some pragmatic elements involved," he says.

While many are working on multiterabit and faster systems, proven architectures are operating today at about 160 Gb/s to 300 Gb/s, Zettacom's Liesenberg says.

"This is not just about wirespeed," he says. "Moving to higher speeds is nice, but if it is a very large network today, there are not many links that require OC-192. But the needs are more sophisticated, meaning a need for more administration, building in more manageability, more quality of service, and so on," he says.

But most in the industry agree that terabit routers are likely to meet the industry demands for at least the next few years. Avici Systems says its terabit switch/router architecture will allow for at least two more generations of products, while IronBridge says its latest router in development will handle 1.3 terabits.

"We have a machine on the drawing board that runs at 45 Tb/s (terabits per second)," says IronBridge's Blume. "There is gas left in this IP architecture by converging on a single standard. IP has the ability to move ATM, frame relay, and circuit-switched traffic."

Peter Chadwick, vice president of marketing for Avici in Billerica, Massachusetts, says the current technologies pointed toward terabit switching and routing are designed for scalability and should not portend a major technological shift in the near- or mid-term.

"Long-term, now that could be a lot of science fiction," Chadwick says. "Optical routers are a long way off. The next big push we will see is a closer coupling with the router plane aware of IP services and the optical plane looking at big pipes. We are working with the [standards bodies] to develop signaling between the optical and data layer."

The challenge, he explains, is how to map and how to get a control plane that can be used in the data and the transport world. "SS7 has been the control plane model of choice for the circuit world. In the data world, IP routers are running BGP [border gateway protocol] and interior gateway protocols. MPLS [multiprotocol label switching] has the potential for converging the two, allowing you to bridge the gap between the IP data world that is packet-oriented and to map closely to the circuit world that is connection-oriented. But how do you then converge the optical and data control plane to increase the speed to provision at all layers?"

A wavelength, Chadwick says, is fundamentally a circuit. That means a wavelength would only be able to offer a static high capacity OC-192. "Not many services require that. You aggregate at lower speed connections to try and fill up the pipe. You do that through routers. Routers are the devices with scalability and flexibility to optimize bandwidth between any two points," he says.

Tenor Networks sees its role as solving the mismatch in handling the increased bandwidth coming out of the network, citing research showing that 16 lambdas of OC-192 would take 96 racks of cross-connects at a huge cost in addition to 150,000 watts of power.

"We saw an opportunity to more evenly match what would be an explosion of optical growth bigger than Moore's Law. There wasn't any electronic equipment that could match that boundary. If one fiber has that many lambdas generated, the floor space would be enough to sink a ship," Tenor's Woo says.

"People forget they need to scale both dimensions -- data and the control plane. Siliconization is focused on the data plane. In reality, the control planes have to be controlled, too, with routing protocols. Most routing protocols still run in software while forwarding is in the hardware."

Another aspect, Woo says, is that electrical advances will only keep up with optical bandwidth growth for a few more generations of development.

All-optical Report Card

Just how fast the industry will get to the next step -- all-optical -- is a topic for lively debate. "The all-optical network is not there today," Sycamore's Copley says. "You still need to regenerate the signals. There will be a need for an optical core and for high-speed transport. Then there are the edges to put the combination for grooming and a combination of grooming and non-grooming. Everyone says OC-192 to the desktop will not happen overnight. But there will be an evolution."

The challenge, she says, is that wavelengths are not smart, meaning the technology is limited right now because it can't be groomed to offer sub-rate circuits.

"It sounds sexy to say provision wavelengths," adds Ericsson's Low. "But it took 10 years and only in the last couple of years to provision ATM [asynchronous transfer mode] ports in service provider networks. Provisioning wavelengths is a little ways out."

But there are several efforts under way designed to lead toward an all-optical network. Lucent's WaveStar LambdaRouter that uses a series of microscopic mirrors to direct and route optical signals from fiber to fiber in a network is commercially available.

Cisco has developed a wavelength routing protocol it calls WaRP. It's designed to enable point-and-click auto-provisioning of end-to-end wavelength paths.

"The electronic bottleneck is crunching the network now," says Dave Bishop, director of micromechanics research at Bell Laboratories for Lucent Technologies in Murray Hill, New Jersey.

"The paradigm to do switching at the packet level is just too slow. The amount of data on an optical fiber doubles roughly every nine months. No electronic box can keep up."

But Bishop says Lucent is capable of selling and shipping switches with a petabit of capacity today if a customer wants it. "In three to five years, there will be all-optical packet routers. The core switching and edge switching will be optical. Basically, electronics isn't going to keep up and eventually will not be there at all," he says.

Lucent has been selling its WaveStar LambdaRouter commercially for more than four months, and Bishop says micro-electro-mechanical systems (MEMS) is proving to be "robust, reliable, and scalable. It is nearly a perfect technology."

Others are not so sure. At least not yet.

Jost Spielvogel, chairman of Optisphere in Munich, Germany, and a member of the board of Siemens Information and Communications Networks, says the latest developments started with optical transport and the rapid expansion toward terabit pipes. "Then we talk about switching or routing. That's the normal way of evolution," he says.

Spielvogel mentions MEMS and bubble technology as showing promise, but adds the technologies are not yet mature because they are not manufactured in quantity and have not shown long-term stability.

"The vision is that people will buy bandwidth the same as a flight ticket. You use it and that's it. This vision will come true in the next couple of years," Spielvogel says.

Hyperchip's Norman agrees that it will just take time for fundamental technologies like mirrors and bubbles to gain reliability and become more cost-efficient. "Putting the pieces together is an engineering challenge, but people are close."

The forward issues, Lucent's Bishop argues, will be how to design and build networks, meaning software and standards issues.

"OC-48 is passe. Now people are talking about OC-192 and OC-768 is on the horizon. The problem is as you increase the channels -- an inexorable trend -- you need something to deal with it. In a big central office, each fiber could have a terabit on it, and the service providers then have to deal with a petabit in an office. They are not going to have 1,000 terabit routers," Bishop says.

"Our customers want to allow their customers to program their own network with rapid provisioning, self-healing, restoration, and VPNs. If a big customer wants an extra wavelength between headquarters, then the carrier wants to allow that customer to dial up and get the pipe. People want an adaptive network."

Wayne Walley (wayne_walley@intertec.com) is Global Telephony's Editor-in-Chief based in Chicago, Illinois.

Sidebar

Making the Math Work

Hyperchip, which bills itself as "the petabit routing company," says optical switching among other advances will help make petabit routers a reality by 2004.

Richard Norman, Hyperchip's president and chief technology officer in Montreal, Canada, explains the coming evolution as part of a mathematical formula that would result in dropping the cost of each bit moved by a factor of 280.

"That is a huge drop in the cost of a bit moved. Moore's law can account for a factor of 10 to 1, but you have to find other things," Norman says.

With optical switching, the per bit moved becomes less expensive than today's routing because the bit is only routed once, saving a factor of 5 to 1 on the number of bits now handled electronically, he says.

"That drops the cost 5 to 1. Then comes multiprotocol label switching. MPLS is like an open-book exam. It figures things out once per flow instead of once per packet, another 5 to 1 savings."

So, according to Norman's calculations, the factor of 10 to 1 per Moore's law (the doubling of computer chip capacity every 18 months), combined with a factor of 5 to 1 for optical switching and a factor of 5 to 1 for MPLS, means a factor of 250 to 1 in savings.

"The cost will drop by 250 to 1 in per bit moved, meaning we will get to petabits in a 2004 time frame," Norman says.

"Then you need another factor of 100 to 1 to get to exabits. It's not easy, and it can be done, but no one needs it yet. Understanding petabits will happen, then people will build for it. When people are comfortable with that, then comes exabits."

Wayne Walley

Sidebar

Defining Bits

More zeroes are coming. As networking equipment gets faster and adds more capacity, the industry vernacular expands to include words like terabits, petabits, and exabits. But where do these terms come from?

In its simplest form, the vocabulary derives from the metric system and the prefixes were adopted by the sciences to avoid confusion. For example, the word "billion" has a different meaning in England than it does in the United States. But both would agree that "giga-" would be equal to 1,000,000,000. Today, there is an agreed international system of units (SI) that sciences, including telecommunications, now use to make it easier to talk about very large and very small numbers.

Here are the numerical equivalents for the bit terms you already use and might be using in the future:

Kilobits = 1,000 bits

Megabits = 1,000,000 bits

Gigabits = 1,000,000,000 bits

Terabits = 1,000,000,000,000 bits

Petabits = 1,000,000,000,000,000 bits

Exabits = 1,000,000,000,000,000,000 bits

Zettabits = 1,000,000,000,000,000,000,000 bits

Yottabits = 1,000,000,000,000,000,000,000,000 bits

Wayne Walley