Switching at the Speed of Light
Optical packet router could facilitate ultrafast terabit computing.
The Integrated Router Interconnected Spectrally (IRIS) program aims to create an optical packet routing system. Lucent Technologies’ Bell Labs is leading the team that is researching a photonic router.
Researchers are taking optics to new levels by developing the architecture, components and prototype systems for all-optical packet routing that can send and receive up to 100 terabits of data every second. The research is based on the premise that photons can do more than just carry a signal from one point to another; they also can facilitate extremely high-speed switching.
The work is being conducted by Lucent Technologies’ Bell Labs, Murray Hill, New Jersey, and goes beyond the first layer of switching, referred to as cross connections and beyond the low-speed reconfiguration of legacy networks known as wavelength add/drop. Information flows on wavelengths from one point to another, and the capability exists to drop or switch out a wavelength as it reaches certain points. However, using current technology, this process occurs at a relatively slow rate that cannot be considered high-speed routing.
Rod C. Alferness, senior vice president, optical networking research division, Bell Labs, explains that now that data can be transmitted without intermediate electronics, a capability is required to switch short bursts or packets of data that represent file transfers, Web downloads or e-mail rapidly. As a result, Bell Labs is leading a team that is working on the Integrated Router Interconnected Spectrally (IRIS) program. The project is funded by a four-year award from the Defense Advanced Research Projects Agency. The U.S. Air Force Research Laboratory, Rome, New York, is administering the program.
IRIS is based on principles that Bell Labs has been developing during the past six years in its work in long-haul data transmission. The program features both a concept and a technology component.
Alferness relates that the concept is relatively simple and is based on the behavior of light passing through a prism. When a white light is shined through a prism, the light is separated into the rainbow of colors that appear at different angles on the other side. However, when a single color of light moves through a prism, it emerges from the prism at a specific angle. If the light is collected and put into a fiber, and a different color comes out and is collected in another fiber, the light would exit at a different angle, he explains.
The goal of the IRIS program is to develop a way to change the color of light so the output position changes and sends the light to a different destination. “If we have a prism together with a source whose wavelength we can change, we can affect the same kind of switching using optics that today we do using electronics,” he relates. The key, Alferness points out, is designing a means to accomplish this switch in nanoseconds.
“In today’s communications systems, electronic packet routers direct the individual parcels of traffic on a time scale that is very, very rapid. If we want to replace this function using optics, we have to be able to switch at very high speeds,” he says.
This approach could solve several problems that emerging computing requirements will pose in the future, Alferness contends. Electronic routers consist of a number of different components, including the switch fabric and the interconnectivity of signals coming into the router. “Those signals can be coming in at very high speeds—let’s say 10 billion bits per second. As we scale routers that carry all the traffic at the core of the business network and try to scale those to higher capacity—beyond the hundreds of gigabits per second into the hundreds of terabits per second—the interconnection of signals inside the switch gets difficult as the bit rates increase. So, just as we use fiber to transmit volumes of data in metropolitan networks and in national networks, people are looking to see if they can use optics to provide the interconnectivity of those high-speed signals inside the router,” he says.
Alferness allows that, if the switching function can be achieved optically, the conversions between optics and electronics could be avoided. “That’s cost and power dissipation that could be avoided,” he adds.
In the backbone network, fiber is required to carry the terabits of information on multiple wavelengths. “If we can avoid terminating those signals on electronics along the way in the network, we can save money. The same argument can be made for designing the processes inside routers. If I can replace not only the interconnect with optics but also actually achieve the switching with optics, perhaps I can reduce the cost, particularly as I have to scale to these higher rates. We believe that optics can play an important role in achieving those very high capacities but in a way that is still cost, power and space effective,” he says.
Alferness shares that another reason to explore optical switching is that many networks today already carry data traffic on wavelengths. Perhaps the backbone network of the future will not require electronics at all, he proposes, but instead will be one that features multiple wavelengths carrying terabits of information and wavelength-based high-speed switches that handle session-by-session packet switching.
“So the principle is very simple but very powerful. If we look at the way we do electronic switching, we have to build active elements to get inside the switches and make changes to the switches. Here’s the real magic done by the prism. At some entry point, we make a change—the color of the light—and the impact of that change is that this different color of light finds its way through the prism and comes out the right exit port. Rather than getting inside that prism and making some changes about where that light goes, I literally only have to make some change on the entry and I get change on exit. The ability to affect the change only on the entry point is something electronics is not able to do. It’s part of the magic of optics for switching,” Alferness states.
|Achieving optical routing requires several components, including (t-b) the optical delay line, wavelength converter and fast-tunable filter.|
The other aspect of developing this capability is moving the information from one wavelength to another very quickly but without changing the data itself. This is a difficult challenge, Alferness admits. “The light source has the information impressed on it, and those bits are on a specific color. What is needed is a device that accepts the bits but does not alter them as they move to a different color to route them in a different direction,” he explains. To accomplish this task in nanoseconds, the IRIS team is developing the wavelength converter.
“On top of all of that, we believe that by doing this we are going to develop switches that allow us to switch more information in a box size that is no larger, and potentially smaller, than we have with current electronics. So we have to develop the technology to allow us to build all of this on substrates in circuits, not necessarily with the density that you can get today in electronics but still with very, very high density. These are wavelength converters on a single chip, or in other words, a magical prism plus a magical wavelength changer to produce a high-speed router on a single chip,” he explains.
Alferness says the Air Force is interested in this capability because so many of its operations require high-capacity information services. A multiterabit router could replace aging lower speed electronic routers. In addition, it would not only benefit the Air Force but also other government agencies, including the intelligence community, he notes.
Although optics offers many advantages such as sending wide bandwidth signals over distances, Alferness admits that there are some limits—for example, photon storage. “It’s hard to slow photons down, so building memory chips to store optical bits for a short period of time is challenging. It’s not impossible, but it is difficult. And in this program, we will explore techniques to improve optical buffers. We will use optical buffers because as we make the decisions about how to change that color, we have to hold those photons for a little while. We’re putting optical buffers in those switches, but the flexibility of those buffers is challenging,” he states.
To address this challenge, the IRIS team has developed a new architecture for building the switch. This load-balancing architecture reduces the dependence on optical buffers while improving switching, Alferness explains.
The goal of the program is to demonstrate the feasibility of building a packet-routing capability that can handle hundreds of terabits of capacity. Once the research is complete, the next step will be to build a prototype.
“We’re talking about multiple orders of magnitude in routing capability beyond what is commercially available today, and of course, that’s appropriate because this is being funded as research to alleviate bottlenecks. This is based upon a philosophy that optics has advantages that go well beyond its current use today in transmission systems,” Alferness says.
Although information security is not one of the focuses of the project, Alferness notes that networks that can transmit and switch data at very high rates could be inherently more secure than today’s electronic systems. Extracting data that is moving so quickly would be difficult, he states.
If all goes well, Alferness believes the capability could be commercially available in approximately five years; however, program success depends on several factors, he adds. While the principle of the optical switching capability has been demonstrated, the size, number of wavelengths and ports, and integration work remain.
Other members of the IRIS team include the University of California–Santa Cruz, Lehigh University and Agility Communications.