LANs /

Tester's Choice: The all-optical myth

By David Newman
Network World, 04/30/01

Some day soon, a sales representative for a metropolitan- or wide-area service provider will pitch you on the benefits of its "all-optical" network. Show said sales rep the door, fast.

All-optical is a myth. It exists only in simple configurations where it can't support niceties such as redundancy and intelligent path selection.

Don't get me wrong - I've got nothing against optical networking. I'd love to test optical routing. But it looks like it may be a while before I have the chance.

The premise of all-optical networking is simple: Data will travel much faster because it will be encoded only as light.

"Only" is the key word. Optical equipment today accepts pulses of light from fiber cabling, converts the pulses to electrical signals for processing, and changes the signals back to light for transmission. Even if processing time were zero, this conversion would add delay.

Optical advocates say eliminating the conversion will push data rates into the terabit range. An oft-quoted statistic is that fiber has a theoretical capacity of 25 to 75 terabit/sec. Compare that with copper, for which data rates are usually measured in megabits.

But this argument doesn't address two fundamental requirements for all-optical networks: routing and buffering.

There is no such thing as a Layer 1 routing protocol. Optical devices operate today in point-to-point or ring topologies. Point-to-point means light pulses get from Box A to Box B or they don't. There's no backup in the event of a cable cut. Ring technologies like SONET's automatic protection switching offer a bit more redundancy: In the event of a cable cut, a ring can wrap around itself.

Any more-complex topology requires routing. Imagine a network with nodes in Boston, Miami and San Diego. If I present a packet to the Boston node with a destination of Miami, how does that node know to send the packet directly, instead of the long way around? If the Boston-Miami link goes down, how does the Miami node know it can still reach San Diego? The answer is routing. And it doesn't exist at Layer 1, at least not yet.

Some optical advocates say routing decisions belong at the edges of the optical network. That's true, as long as all-optical networks remain small and simple. If switch makers are serious about increasing sales, they'll need more intelligence in their boxes.

The other major roadblock to all-optical networks is figuring out a way to buffer light. No optical device can slow or store light, the way electrical devices buffer packets.

This makes all-optical devices impractical anywhere congestion exists. Imagine an optical switch on which two light pulses arrive at the same instant, bound for the same destination output interface. The switch will simply discard one of the pulses.

This may change. In January, two research groups published papers indicating they'd found ways to slow and even store light. Impressive stuff, but the leader of one of the groups estimated it would be 10 to 50 years before the research would be put to practical use.

Optical devices already play an important role. But until the industry addresses some key challenges, all-optical remains a neat idea but a dubious sales pitch.