If the rate limiting step has now become switching out of the fiber network then it sounds good for LUMM. From today's WSJ editorial page...
<<Bad Bets on Bandwidth
By Bret Swanson. Mr. Swanson, an analyst at the Gilder Technology Report, invests in some of the technologies discussed in this article.
With Lucent and Nortel firing workers and bleeding cash -- at about $19.2 billion, Nortel's second-quarter loss will be one of the biggest losses ever by any company -- two of the telecom-gear companies that drove the '90s market boom are running out of time to get themselves on the right side of the latest digital divide.
The last great digital divide played out in the mid-1980s. On one side were companies achieving billions of operations per second in a single computer; on the other were companies enabling production of hundreds of millions of computers. Companies such as Control Data, Cray and IBM that focused on aggregating computation in a single, costly box lost to companies that focused on enabling millions of people to use computers: Intel, Microsoft and Dell.
Abundant Bandwidth
Today we've got a similar divide in communications networks. It also revolves around the key abundances and scarcities of an industry. The computer era's most abundant and cheapest resource -- and thus the one you want to waste in order to conserve other, more precious resources -- was transistors on chips. Today's key abundance -- as the companies that spent billions in recent years laying currently unused fiber-optic cable are finding out -- is bandwidth. In computers we wasted transistors to give everyone a computer -- that is, unshared processing resources. In communications, the new mandate is to waste bandwidth to give everyone an unshared connection. At the moment, Lucent and Nortel are betting on the wrong technology.
Wave division multiplexing is the process of combining many colors of light, each carrying separate streams of data, onto a single fiber-optic strand and sorting them at the other end. It is the most crucial technology of the Internet economy, and it has increased the carrying capacity -- the bandwidth -- of a single optical fiber to almost 3 terabits (3 trillion bits) per second.
Every six to nine months the number of wavelengths (colors) on a single fiber thread doubles, from 4 in 1996 to 320 in 2001 -- all with no increase in cost. The amount of data that can be carried on each color of light is also increasing, but only one-half or one-third as quickly.
The key decision in the industry -- the one that marks the great divide -- is whether to increase the data rate on each wavelength of light by pushing the limits of silicon technology, or to make more wavelengths of light (more colors, more communication paths), each carrying data at a slower rate.
Think of a how a railroad compares to a superhighway. A railroad track supports a single train carrying a large payload. On a large highway, instead of a single track, you have many lanes. The automobiles traveling in each lane carry much less than the train, but if you add all the contents of all the autos in all the lanes, you can match the train's total capacity.
The difference is the manageability of the traffic. All the cars in the train have the same starting point and the same destination. If you want to add or drop off a car at a point in between, you take the train off the main track, disassemble it, hook or unhook the car in question, reassemble and remount the track.
On a superhighway, autos exit via off-ramps, leaving the rest to speed on by. You know which car needs to exit by the lane in which it is driving.
In a single-track optical network, the work of disassembling the stream of data (the train) and finding and rerouting the correct bits (railroad cars) is performed by expensive and slow microchips. In a multilane optical network, the off-ramp is just a cheap mirror that steers the bits (the autos) traveling on a particular color of light (the lane) onto another fiber (a side road). A train is a shared method of transportation. Autos are autonomous.
In the communications world, Nortel is the leading proponent of railroads. It plans to offer 80 gigabits per second on each of 80 wavelengths early next year. That's 6.4 terabits of information. Not to be outdone, Lucent proclaims the laboratory feat of putting 160 gigabits per second on a single wavelength.
Lucent is doing so because it was roundly criticized for not upgrading its 2.5 gigabit-per-second product to 10 gigabits per second in time to compete with Nortel two years ago. Lucent has learned the wrong lesson. Sixty or 80 wavelengths in a world where thousands are soon possible is analogous to the single-track train. Fewer wavelengths, and more data per wavelength, require more sharing of wavelengths, and thus more electronic processing to sort out the correct bits of information for the right customers. If you send 6.4 terabits down a single thread in a second -- roughly as much as the total average traffic of the world's telecom infrastructure -- you have accomplished nothing unless you can distribute the bits to the correct users.
By adding to network complexity and exacerbating the dearth of last-mile connectivity, this focus on raw bandwidth has been key to the meltdown of the fiber-optic carriers and service providers. People will not pay for raw bandwidth any more than for raw transistors; people will pay for connections and easy-to-use services.
The obvious alternative is to multiply colors. Instead of 80 colors of 80 gigabits each, why not make 640 colors of 10 gigabits each, or 6,400 colors of one gigabit each? The consumers of optical bandwidth -- servers, routers, storage devices, and someday PCs -- would much rather deal in one- and 10-gigabit chunks. It is more than enough for any application you can imagine, and it more closely conforms to the needs of the end user.
Cornucopian Connectivity
So the new digital divide is between the companies focused on big, costly bandwidth such as Nortel and Lucent, and companies focused on cornucopian connectivity such as Avanex, Corvis and ONI. The color-blind, big-bandwidth players will lose just as surely as the big supercomputer players lost to the PC.
Yet Lucent is now trying to pay off debt by selling its most important division, the one best positioned to capitalize on this new trend in connectivity -- its optical fiber business. The Lucent-Alcatel merger talks that failed recently, in fact, grew out of Alcatel's bid for that Lucent division.
Neither Lucent nor Nortel is going to save itself with outside deals. They need the right inside technology. The big, bold bet is against big bandwidth, against "big iron." Bill Gates, Andy Grove and Michael Dell have already shown how it can work.>> |
