What it will take to shift the Net
page 2:  So the backbone becomes...

Well, there isn't enough time to do that at a terabit per second. So we can't do packet-switching at that speed. We have to break down the line into slower capacity lengths, maybe 80 gigabits per second. I realize "slow" and "80 gigabits" seems like an odd thing to say, but 1.6 terabits is 20 80-gigabit-per-second circuits.

So, Michael O'Dell, the chief technologist for UUNet, has been architecting the network so that it's possible to do optical switching down in the lowest levels of the Net.

UT: Could you describe how that would work?

Cerf: You've got your fiber layer, and you're going to carry some number of wavelengths on each fiber -- the term that is used is lambda for wavelength, so lambda means color, really. One fiber can carry a number of colors -- there could be as many as a hundred, maybe even more. Each color might be transporting as much as a terabit of capacity.

But instead of having to look at each packet when it's down at that level in the network, you might just literally switch things optically. You switch simply by saying, "If the photon is green it goes this way, and if it's blue it goes that way, and if it's red it goes that way."

You don't know anything about packet boundaries. All you know is the color of the light.

UT: But if you're not inspecting individual packets, how do you perform more complex operations on the packets? Say, quality-of-service?

Cerf: Well, at the edge of the Net you have traffic coming in at lower data rates like OC-12 [622 megabits per second] or OC-48 [2.4 gigabits per second]. At 2 gigabits a second you have a microsecond to look at a 1,000-bit packet, and a microsecond is a long time. I mean, that could be a thousand instructions. Hell, we've got all the time in the world.

So you can do a lot to classify a packet. I look at the various protocols, I look at where they're going, I look at other priority bits, and then I decide, "This packet is going to go out on that fiber in the color blue."

And once it's a blue packet, the rest of the network takes it along whatever the blue path turns out to be. I'm hiding the fact that we may actually change colors as it goes through the Net, and that's just for switching reasons.

But the fact is that once it's destined for a given color, it just goes through the optical part of the Net and it pops out in another router on the other end, which now is taking it out at lower data rates, deciding what to do, and handing it to the target.

So you've got this optically switched layer just above basic fiber, and above that you have a set of routers. But the Net looks like one hop, from one router to the other. The interior stuff is all being handled optically.

UT: So the backbone becomes...

Cerf: An optically switched system. Now, today's optical technology may not actually allow you to avoid going up from optics to electronics and back down. The reason is that if you want to change the color of the packet on the fiber, you may have to go up into electronic mode and back down again to run a laser with the different wavelength. But we're seeing future technologies that will allow us to actually switch all optically without having to go back up and do electronic modes.

UT: Just from an equipment standpoint, this sounds like a massive reengineering project.

Cerf: No, it isn't, actually. The fiber is still the same fiber. You interface to it the same way you would otherwise. And so it's really a question of sliding in some switching capability that might not be there today.

Today, if you look at the structure of the Net, you see a router, an ATM [asynchronous transfer mode] switch, frequently, then a Sonet [9.95-gigabit-per-second optical] layer with add-and-drop multiplexes, and then the underlying dense-wavelength division multiplexing fiber.

We'd like to pull the ATM out; we'd like to pull the Sonet out; we'd like to slide in an optical switch, and then drop the router right down on top of that.

Now, you don't have to do all of this everywhere all at once, and the optical switching is almost invisible to the router. All it knows is that it handed this packet out to the optical interface.

UT: Is this a change that's likely to happen soon, or are we talking 5 years down the line?

Cerf: I think it's something that we will wind up doing some of either next year or the year after that. It's very near-term, as these things tend to go.

UT: Is this something that an individual company, like MCI WorldCom, could do on its own, or would everybody sort of have to do it at once?

Cerf: You don't have to do it everywhere at once because there's a well-defined interface between networks. I don't foresee the switching, optical switching, going between networks in the near-term.

Additional Article Pages:
1. What it will take
2. So the backbone becomes...
3. How broadband will change the Net
4. The promise of V6

Printable Version

    | Home  | Newsroom  | Magazine  | About  | Advertise  | Feedback  | Privacy Statement  |
  Copyright 1993-2002 UMAC Inc. All rights reserved.
Powered by Apache with Thunderstone