Many people are beginning to think that all-optical networking using wavelength division ought to become eventually the next networking generation after ATM (Asynchronous Transfer Mode). Why would anyone in his right mind think this - that networks made out of glass and using the oldest, least sophisticated transmission format (frequency division, a.k.a. wavelength division) would be that interesting? There are several reasons, capacity (the main one), transparency, and cost.
The 25,000 GHz of capacity in the passband is not only astonishingly large, but is highly likely to be needed before we know it. Consider the very thing you are using at this moment to interact with data objects or programs: the WorldWide Web. WWW can be though of as the second wave of easy computing for homes or businesses. In the first wave, you point, click and get an immediate response from some local object within your machine - a file, a program, a picture, a sound bite, and so forth. In the second (WWW) generation you get the same thing from the distributed world of objects that don't necessarily have to be in your machine. If you are lucky enough (or rich enough) to have a high speed LAN plus a T-carrier connection to the telephone company backbone (telco), the distributed point/click works just as fast as the local one; if not, you are probably waiting for such a connection to evolve. Let's look at when and how this will happen.
The number of users loading up the world's phone companies and cable companies with WWW traffic is growing rapidly, but the traffic per user per unit time is growing even more rapidly, in fact by a factor of 8 per year. This means that the portion of the available communication infrastructure devoted to descendents of the Web must undergo a capacity growth of about one billion over the next decade just to keep up with demand, and even more than that if everyone is to get the response time performance they want.
There is no communication technology in place today whose bandwidth is evolving by a factor of 8 per year; a factor of 1.5 is more typical. This is true of the pipe between the CPU and the outside world (the fastest I/O bus in the PC), the local environment (the LAN), the access to the telco backbone (T-carrier or SONET), and the modems that connect local premises to the telco. (Note that all these technologies are based on time division multiplexing - TDM). An even worse problem, as most of the readers of this note will realize (since they can't afford a T-carrier or SONET link into the telco) is that the use of voice grade lines and modems constitutes a huge bandwidth bottleneck between the local premises and the telco. This is the infamous "last mile" bottleneck.
There is only one transmission technology that can fill the bill in a future-proof way, whether we're talking about the telco backbone, the local environment, or especially the bottleneck in between: optical fiber. Without optical wavelength division, but using the traditional TDM instead, each fiber can conveniently carry up to around a gigabit per second; with WDM the number could be 10,000 times even larger. Clearly what the world needs is fiber to the user premises, exploited first in TDM mode, and when that proves insufficient (as it soon will) in WDM mode. The problem with TDM is that each node spends a lot of time looking at bits that have nothing to do with that node.
The dream of "fiber to the home", "fiber to the premises" or "fiber to the desktop", call it what you want, has been very slow in coming. If you plot the number of homes (or offices) served by the nearest fiber end as a function of time, you find that for the ten percent of the homes/offices that were closest to a fiber end, on the average these amounted to 100,000 homes served per fiber end in 1984. By 1995, the number had decreased to 100, because the telcos and cable companies have been steadily extending fiber toward user premises in trying to solve the last mile bottleneck. In other words, for the ten percent of premises that were closest to a fiber end, the nearest fiber served roughly 100 such premises on the average by 1995. The interesting but discouraging thing about these trends is that when you extrapolate to predict the date at which ten percent of the homes/offices will each be served by a single fiber, the most optimistic predictions say it will not happen until 2005. Cable and telco experts predict that if optoelectronic component costs are greatly reduced, it could happen faster. Increased competition between carriers could also speed things up.
We conclude that only fiber can satisfy future WWW performance needs and that by the time the needed fiber is in place, the all-optical technology should be also.
Since each path on a physical route between end users at a given wavelength (usually called a "lightpath") acts like an independent fiber having its own bit rate maximum, any sort of traffic can be sent over it, independently of the others, provided the lightpath's bandwidth maximum is not exceeded.
Thus, all-optical networks that put lightpaths into play for user-defined intervals (i.e. circuit-switched such networks) offer the user protocol transparency, the freedom to send different bitrates, framing conventions, etc. (and even analog) over different lightpaths. Protocol transparency is one of the things our group's customers have found so interesting about IBM 9729 Optical Wavelength Division Multiplexer.
Lest you think that protocol transparency is only of academic interest, reflect on the fact that it is the protocol transparency of the traditional voice grade telco connection that has made today's communication richness possible. A phone connection has proved to be a potent mediium, being able to send ASCII, binary, teletype, fax, voice, compressed image, 300 baud, 28.8 Kbaud, pretty much anything that doesn't exceed 3.5 KHz bandwidth and 50 volts across the line. Think of what a world can be built on a medium - fiber - having ten orders of magnitude greater bandwidth (and also typically ten orders of magnitude better error probability; 10-15 versus 10-5. Also, think of the growing number of different protocols in use today that must be somehow accomodated, plus the historical fact that the reason the Internet proliferated in the first place was that IP datagrams were so easily usable by almost anything from LANs to packet radio to... you name it.
As you will read in the description of our work on the optoelectronic chipset, the optoelectronic component art is these days undergoing a slowly accelerating transition from bulk optics to the kind of structure that can be fabricated lithographically. This involves planar waveguides and planar realization of such things as coupled lasers and detectors, gratings, magnetooptic isolators, interferometers, and so forth.
While the lithography revolution in analog and digital electronics is some thirty years old, the optoelectronic equivalent is scarcely one tenth that old. It is very difficult to predict how far this will take us, but as far as the passive components are concerned, one can take encouragement from the compact audio disk, which is produced today very cheaply in polymer form, in huge quantities, and to tolerances no less exacting than those for optical structures.
[ IBM home page | Order | Search | Contact IBM | Help | (C) | (TM) ]