Back when all-optical networks seemed right around the corner, figuring out how to route packets optically looked like the next major technical challenge. Economic realities have pushed "right around the corner" well down the street, yet researchers continue to believe they can remove electronics from packet processing and forwarding. And the resumption of new systems development efforts has caused some companies to ponder whether these researchers might be right.
While Ethernet has threatened IP's place as the end-to-end protocol of choice for future data networks, the fact that both transmission techniques use packets underscores the desirability of optical packet processing. Current research focuses on pure packet switching and a transitional approach called optical burst-mode switching. The former generally refers to processing packets individually; the latter, as is the case with current electronic burst-mode transmission, refers to a scheme where packets headed for the same destination are grouped together, then transmitted into the network (and through the optical switch) in bursts.
Both require extremely fast switching speeds measured in nanoseconds. Semiconductor optical amplifiers appear to be the preferred base technology for such switches, at least according to a review of recent conference proceedings. However, a combination of arrayed-waveguide gratings and fast-tuning lasers also has received study in Bell Labs and other places.
The trick, of course, is knowing where the packets are supposed to go and directing them there without converting the optical signal to electrical within the switch or router. Currently, that means electronics remain in the network somewhere. Some schemes propose separating the control packets from the rest of the signal and having the control packets arrive ahead of the payload for electrical processing; the payload is then switched in the optical domain. Other approaches, particularly for optical-label switching based on MPLS and Generalized MPLS techniques, call for electrical routers on the network edge that would feed all-optical routers in the core. The edge devices would perform traffic aggregation, generate optical labels, and help with label distribution and label switching path setup, as reported in a paper (S. Yoo, UC Davis) at last March's OFC.
The synchronization of edge routers, core routers, and the paths between them becomes a tricky proposition when making routing decisions at the packet level. Optical burst-mode switching and transmission ease this burden somewhat and therefore may prove closer to implementation. As described in another paper at OFC (C. Qiao and Y. Chen, SUNY Buffalo, and J. Staley, Brilliant Optical Networks LLC), optical burst switching builds on the techniques developed for its electrical counterpart. Packets are assembled at the edge of the network. The edge node sends a control packet over dedicated control channels for electrical processing as described above. The packet burst remains in the optical domain and is switched optically in the core.
Both burst and packet switching face the problem of resource contention. That is particularly true for burst-mode transmission, where the individual bandwidth requirements for each burst are varied and unpredictable. Fast-tuning lasers, which would enable burst transmission across several wavelengths within a single fiber, promise to help overcome this obstacle—good news for companies looking for applications for their tunable transmitters.
Intune Technologies (Dublin) is one such company. "The larger companies now are starting to go back and look at what their next-generation systems are going to look like," says Intune chief executive John Dunne. "And in effect, the people who traditionally have looked at circuit-switched systems are reexamining a lot of ideas that came up in the late 1990s, taking better advantage of the type of traffic, the bursty type of traffic that's starting to be generated within systems."
Because these larger companies have in many cases divested themselves of their component and subsystem operations, they're looking for partners to help develop systems that conform to bursty traffic patterns, Dunne says. "It's taking place, normally with as little cash injection as possible," he says of the current level of research. "They're looking for smaller innovative companies to partner with in development."
Dunne sees a noticeable difference in the timelines between the advent of burst-mode transmission and all-optical packet switching. Both are "long-term," in his opinion, with systems that support optical burst-mode transmission at least 18 months away. "But from a systems perspective, 18 months is actually quite short," he points out. "So I would divide up the 'longer-term view' into two parts: one is how can you tinker with the systems today to incorporate a better match with packet-type traffic, and the second issue is, in the much longer term, how can you actually incorporate optical switching into your entire telecommunications backbone. I think the second issue is very long-term—you're talking five years out from here. But to see intermediate steps, where people try and address the bottlenecks that occur, say, in aggregation between metro and local-area networks, that's going to happen within two years."
Some passive-optical-network systems already support burst-mode transmission. But as a precursor to the type of packet switching envisioned by researchers, Dunne says the enterprise will be the incubator. "You'll see new technologies coming into the server industry, server routers, before you see it transfer into telecommunications," he predicts.
Dunne sees fast-tuning transmitters such as those his company produces as building blocks within these routers. The laser within these transmitters must be electronically tunable, he believes, to ensure temperature stability. The transmitters also need wavelength locking for high reliability. Finally, Dunne says tuning speeds of 50–200 nsec will be required.
He reports that potential customers have asked for fairly wide tuning ranges: 20 channels minimum and as many as 80–100 across the C-band.