An IP Traveler in the Land of Optics

Column Steve Vogelsang, CTO, IP Routing and Transport, Alcatel-Lucent 10/27/2014

There's a funny thing about the OSI stack: It's not just an abstraction of a communication system, but also a social hierarchy that creates a kind of class system in the engineering world.

Think about it. Those folks working on the layers below you -- what do you call them? If you're like me, you've spent most of your working life calling them "plumbers." If pushed, you would admit you couldn't get along without them, but there is still a kind of superiority implied: What you're working on is more valuable, more clever, and, obviously, a lot more interesting.

For me, the teams working on optics have always been the plumbers. So imagine my surprise when in a meeting between our SDN and IP router teams, the SDN folks, who work closer to the application layer, referred to us IP "infrastructure" folks as -- you guessed it -- "plumbers." That was putting the shoe on the other foot!

After getting over my surprise, I started thinking about how unhelpful this whole plumbing thing is. After all, the future of this industry requires a convergence of IP and optics: It's time to give the optics guys some overdue respect and dig into the world of optical transmission and switching.

Working with the experts who are pushing the limits of technology on a daily basis exposes the nuances of the technology that you could never learn from books or the Internet. One thing for sure is that optical networking is very different from IP networking. The base system designs and some of the underlying technology are similar, but the design goals and resulting optimizations are quite different.

Here are four key observations I'd like to share:

Optics, as we all know, is in the physical layer. That's physical as in "physics." As in the really tough math that most of us didn't want to go near in university. These guys are serious rocket scientists.

The coherent optical transmission world is no longer digital. The information being transported is digital, but it is encoded into analog signals using increasingly sophisticated algorithms. To reach thousands of kilometers, many stages of amplification are required, each of which degrades the signal, which is further compounded by the natural degradation that occurs when propagating signals over any distance of fiber. The signal barely resembles what was transmitted when it reaches its destination, yet the receiver somehow makes sense of it and recovers the original digital information. This is done, in large part, by a specialized digital signal processor (DSP) that takes samples of the received waveform and determines -- or you might say "guesses" -- the bit pattern each sample represents. This is not an exact science, but forward error correction (FEC) algorithms are used to reduce the number of errors, while more sophisticated "soft decision" FEC algorithms improve performance further by adding a "confidence" factor to the guess. This all translates into 1-2dB of coding gain. Optics people will be quick to point out that an extra 2dB of coding gain increases distance by 25%, which could be 1,000 kilometers or more.

To maximize the transmission capacity of each fiber, we're optimizing every part of the optical transmission system by turning constants into variables. Wavelengths are tunable. The grid that defines the "color" of wavelengths is now flexible (e.g. FlexGrid), allowing variable bandwidth allocations per signal (i.e., the colors of the wavelength rainbow are no longer all the same width). The modulation used to encode digital bits into analog is variable. The FEC algorithm and number of error correction bits is variable. And, the photonic switching layer is becoming more diverse, allowing signals to be routed in any direction and reused.

When you overlay an optical network with IP, how does a router know that the shorter hop is more efficient than the longer one? As we converge the IP and optical layers, these are exactly the kinds of problems that need to be solved. Routers can't simply be black boxes running on top of optical plumbing. What looks like an optimal routing decision to a router might not work out that way at the optical layer.

The first problem to solve is automating the optical layer, because much of what happens, even today, involves hands-on setup. ROADMs were a great start, but they only allow automation of the middle of the route, but not the ingress and egress points. Next-generation ROADMs solve a lot of these issues by making them colorless, directionless, contentionless (CDC) and, for networks over 100 Gbit/s, flexible (CDCF). But the key will be getting the routing layer to talk intelligently to the optical control layer and vice versa.

Again, there is a lot of sophisticated and tricky maneuvering happening at the optical layer, which few IP engineers recognize or understand. While that was OK in the past, it is entirely insufficient today. We're on the cusp of a major shift in how networks are architected and there are just too many opportunities to build those networks better and more efficiently by bringing together IP and optical technologies.

— Steve Vogelsang, VP Strategy and CTO, IP Routing and Transport Business Division, Alcatel-Lucent