It's rare that vendors, experts and the press concur on something, but they all agree that dense wave division multiplexing (DWDM) is the way of the future. If it is, though, a lot of other so-called revolutionary technologies are going to fall flat on their faces.
DWDM is an optical technique for creating parallel information paths on a fibre strand by separating the information by the wavelength of the light used to carry it. Each wavelength, the electrical symbol for which is lambda, can carry as much as an entire, old-generation SONET fibre used to carry. Today's technology can support up to about 64 lambdas on a fibre; we may see 1000 or more lambdas per fibre within five years.
Initially, most of the attention given to DWDM focused on its ability to multiply fibre capacity. You can support 96Gbits of information on a SONET OC-192. If each lambda on the same fibre carried that payload of information, you'd have terabits of capacity. Recently, however, vendors such as Monterey Networks and Sycamore have introduced a new wrinkle called optical networking, and it's this new concept that threatens many other network technologies.
Optical networking works by what could be called "lambda hopping". With standard SONET or DWDM, each fibre is terminated in a big switch, router or optical cross-connect. Information paths must be interconnected with electrical devices at each fibre junction, and huge fibre capacity would mean huge switch/router products inside the optical core. With lambda hopping, information moves from fibre to fibre by jumping from "Lambda A" on the first fibre to "Lambda B" on the second, and so forth.
What's significant is the gear that makes these lambda-hopped connections: an ordinary optical laser and detector and an inexpensive electrical bridge. The contents of a lambda are switched from strand to strand at a very low cost, without terabit products to build the bridges. By using lambda hopping, a kind of optical virtual circuit can be built across the optical core.
DWDM makes it possible to spread the optical edge of the network -- the place where the last, deepest, switch/router is needed -- outward toward the users. As that happens, we find more devices in the network at that level. The more devices at a given place in the network, the more devices share the total traffic -- which means less traffic per device. Thus, terabit products become less useful. We'll probably never see any more than a few supercapacity switch/routers in the deep core because the deep core will be built with inexpensive lambda-hop components.
What holds us back in this evolution of DWDM is the break-even point between optical and electrical multiplexing. Right now, it's cost-effective to do DWDM at OC-48 levels, so we can push optical cores outward toward the user to the point at which we can't collect an OC-48's worth of traffic.
Currently, that's not very far. But as traffic in the network increases with the growth of data applications and improvements in optical networking reduce the cost of DWDM, we can expect to see a lot more optics and a lot fewer electronics in the networks.
There's more at stake than terabit routers. The whole IP/ATM convergence debate becomes irrelevant if we can make DWDM so inexpensive and effective that different network protocols can be multiplexed together at the lambda level. There's no advantage to changing equipment or making compromises in features to converge on one protocol if the network can support any set of protocols at the optical level with equal efficiency.
Cisco was smart to drop its big ATM switch; DWDM will make terabit switches and routers dinosaurs. Moreover, Cisco's investment in Monterey Networks, a DWDM network player, proves the company understands the direction of the market. Other players with big electrical-layer box plans will have to sell out to an unwary telecomms giant or focus on the real problem: the optical-electrical boundary.
We can't push lambdas to everyone's doorstep, so we have to construct service networks by combining electrical and optical technologies. The boundary between these two different network strategies is where the features of our networks will have to be created.
A dumb optical core can't provide phone service, Internet service or any other service. Smart electrical devices will have to absorb the service-feature mission of all our network services, present and future. It will take a new architecture to accomplish that. That's what we should be looking to electrical and optical switching vendors to provide.
Thomas Nolle is president of CIMI Corp, a technology assessment firm in Voorhees, New Jersey. He can be reached at +1-856-753-0004 or email@example.com.