The networking buzzwords of the day are "optical" and "all-optical," presaging the inevitable shift to communicating with laser light alone. While it's fun stuff to talk about, industry experts agree that a purely optical network-one where laser light isn't converted to an electrical signal until the very end of its journey-is years off. "It's going to be many, many years before you have light coming out of my computer on my desktop and never turning to electricity before coming out on a computer on somebody else's desktop," said David Skirmont, systems architect for Pluris Inc. (Cupertino, Calif.).
Still, optical networking is a booming industry, as evidenced by the stock market. E-business and Linux might have had their day, but IPOs from the likes of Sycamore and Avenex Corp. made Wall Street's Most Wanted list, too. Acquisitions have also been hot, with Lucent shelling out $3 billion for Ortel, and JDS Uniphase chewing its way through multiple mergers recently.
Likewise, products related to optical networking have been a bustling business, and not every part is aimed at some new magical feat of optics. The need to save space and money still runs strong in the optical world. This week's focus takes a look at the realities of all-optical networks with contributors discussing the changes, the technology and the components needed for an all-optical infrastructure.
Even now definitions are still being hammered out on exactly what an all-optical network is. "It depends on your scope of what the network is. For companies such as Ciena, Sycamore or Monterey, the network is the transport infrastructure, as opposed to when we at Pluris think of the network as a bunch of routers," Skirmont said.
Today fiber optics are consuming more of the point-to-point network. Once a backbone architecture only, fiber is expanding its reach in the network, moving closer to the home and office. And, the nature of the backbone itself is beginning to change. For example, today's existing networks run optical signals using synchronous optical network (Sonet), which is bracketed by the service layer and physical layer.
Network experts such as contributor Sri Nathan, vice president, network architecture at Qtera Corp. (Boca Raton, Fla.), predict that optical networking will play a major role in removing the Sonet layer, thus letting the physical layer and service layer speak to each other directly using optics. "A truly optical Internet will link the service layer to the transport or photonic layer without any intervening device or layer whatsoever. In this model, the transport of data packets is done over a purely photonic backbone," he writes.
Work that has progressed so far has brought certain sections of the network into all-optical territory. But it's taken some breakthroughs to get even that far. Micromirrors, for example, appear to be on the verge of supplying all-optical cross-connects. Developed by startups such as Optical Micro Machines (San Diego) or veterans such as Lucent Technologies and Texas Instruments, the devices use micromachining technology to produce tiny movable mirrors to switch light signals simply by reflecting them.
Lucent Technologies (Murry Hill, N.J.) recently introduced a commercial micromirror-based router, called the "lambda router" that can switch 256 optical channels. The device is designed to work with current Sonet networks, but will really come into its own in dense wavelength-division multiplexing (DWDM) systems that Lucent is also marketing.
Current optical networking vendors are skeptical of the micromirror approach. "These switches aren't clean. They don't scale well," said John Adler, director of marketing for Cisco's (San Jose, Calif.) optical transport group.
Micromirrors also can't do provisioning of channels, Skirmont said, because so far they can't move fast enough. There's still a nonzero setup time for the mirrors to get positioned for switching. An array fast enough to handle provisioning is still five years out, he said.
But optical switching so far "ignores the higher layers," Skirmont said. Micromirrors can bounce the light from one fiber to another. In order to peek into the data packets and discern information about their nature and destination, an optoelectronic conversion is still required. The process is called OEO in networking shorthand.
In addition to being used for switching, the OEO obstacle course is useful for altering wavelengths. Canoga Perkins Corp. (Chatsworth, Calif.), for example, makes parts that convert the optical signal, route it electrically and then convert it to a different optical wavelength for transport.
Complex routing decisions are based on packet information, and the only way to get packet information quickly enough is to convert the optical signal into an electrical one, then back into optical. "To be able to do that in optics requires a level of filtering that's impossible to consider right now," said Donald Bouchard, vice president of fiber-optical communications for Canoga Perkins. "The technology to do that in the optical world just doesn't exist in that volume in that capacity."
The newer DWDM systems are beginning to alleviate some of those bottlenecks by offering an all-optical routing capability. By multiplexing a number of channels-current state-of-the-art can squeeze 400 channels on a single fiber-that technique can multiply the raw capacity of fiber systems. But just as important, the different channels can be added or dropped from a fiber at any point, offering a routing function based solely on wavelength. That means that a data stream can find its destination without any intermediate decoding, either electronic or optical.
Such all-optical routing is not able to get to the fine-grained, node-to-node messaging required by current networks. It's also important to remember that electrical switches will be inside the network for years. They aren't getting the spotlight, but that doesn't mean that they are being ignored. "There's a lot more money being poured into electrical switches than there used to be," Adler said. "What we see is an acceleration of the electrical-switching world."
Even if the basic problem of optical switching is solved, Adler listed many other reasons why the network can't go all-optical just yet. Optical signals do need amplifiers to keep from dimming, and those amps cause noise, as does the fiber-optic cable itself. The only alternative is "regeneration," that is, terminating the signal and restarting it. So far, this can only be done in the electronic realm.
Regeneration still works
And, reliability is another reason to hang onto electronic regeneration, as carriers can't pinpoint degradation in optics. "You can't take a sample of that light and really find out the detailed kind of analysis you could for a telephone," Adler said. "Most carriers we have today are pragmatists. They want to build networks today that they can scale and make fault-tolerant. So they are still keeping regeneration in their networks."
Then there's the problem of managing "colors"-you can't put two of the same wavelength on a fiber. If a signal goes between points A and Z, you can't introduce a new wavelength at points B through Y; the entire cable is spoken for. Wave-division multiplexing has begun to alleviate this but is still pushing its limits at 100 wavelengths per fiber. "Someplace, you're going to get blocking," Adler said. So, networks powered solely by optics aren't quite ready for construction. "They are the dream of optical networking. Unfortunately, there are a lot of barriers to implementing those kinds of networks," he said.
But there are companies that are tackling the problems-even if it's one component at a time. For example, later in the section, Steven Veneman of Agilent Technologies' (San Jose) Semiconductor Group writes about a small-form-factor transceiver as a way to pack more network ports onto a card. Ken Herrity, director of product development at Blaze Network Products (Dublin, Calif.) discusses recent work with coarse wave-division multiplexing that involves less exotic parts and a standard manufacturing process, with the aim of cutting the price tag of WDM. And, Rob Plastow, chief technical officer at Swedish company Altitun AB (Stockholm) details a switching method using a tunable laser.
See related chart