Alan, this is to pay off my IOU to you about intelligence in an all-optical world. This discussion is particularly timely now because of something said at the Analyst Day that needs amplification.
Today, the optical backbone to the Internet and traditional telephony consists mostly of long-distance optical circuits connected by optical switches from one end of an optical fiber circuit to the other. While most current optical switches require decoding and encoding using microelectronics to redirect (switch) from one optical fiber to another, the latest technology delicately positions tiny mirrors to redirect light along a chosen path. Thus connected, data can flow from one end to the other of an optical circuit completely bypassing today's electronic bottlenecks. Each mirror can reflect each distinct color of data-carrying light made possible through Dense Wavelength Division Multiplexing (DWDM), making possible a large number very high-bandwidth circuits. The next advance will enable optical circuits to be "provisioned" though software that controls the mirrors rather than requiring technicians to patch a maze of fiber optic cables.
Surely something about this scenario should be making you uncomfortable. If you haven't put your finger on what it is, ask yourself why I'm talking about circuits and not packets? Clearly the intention of fiber optics is not to connect everything to everything else with fixed or even on-the-fly circuits. That's old telephone technology. Bits travel inexpensively over the Internet precisely because they are packetized and do not require the fixed resources of a circuit, optical or otherwise. The number of possible circuits is equal to the DWDM factor (say 100) times the number of optical fiber cables. While growing rapidly, this number is tiny when compared to the number of nodes on the Internet, and will only get tinier. Circuits represent an extremely inefficient means of connecting nodes for most types of communication.
Relax, no one is thinking in such antiquated terms. The Internet uses, and will continue to use, packets that must be routed from an entry point to its destination, and routers function by interpreting each packet's IP address. The Internet merely takes advantage of available optical circuits to hitch a ride to a point generally closer to the final destination. If you funnel in enough packets fast enough, a circuit can be made extremely efficient for communications between two nodes.
At its most basic level, routing requires intelligence to interpret the IP address. There is nothing on the drawing board that can substitute photons for electrons to supply that intelligence. Microphotonics remains the Holy Grail of the photon set, and has for years. The optical computer remains a dream that I doubt will be realized in my lifetime. As a more practical first step, fiber optics probably will find its way inside the traditional computer, substituting photons for slower-moving electrons over wire interconnects. After that, if you are young now, look for photons to substitute electronic interconnects inside the microprocessor. If microphotonics, or any use of photons, can advance intelligence to the realm of optics, then so much the better. Tornado for Managed Switches will run on optical computers, or computers enhanced with optics, or today's plain vanilla microelectronic computers. But the point is that intelligence is required throughout the Internet, now and in the future -- forever. For the foreseeable future, satisfactory intelligence can only be provided by electrons.
I visualize the "cloud", or the telephony backbone to the Internet as an uninteresting maze of high-bandwidth dumb circuits. I would expect WIND's RTOS has been deployed in most types of switches used in the cloud, and probably will be deployed to keep those tiny mirrors properly aligned. But so what? The cloud pales when compared to number of interconnects, mostly intelligent, occurring on the "edge" of the cloud. The edge of the cloud is where packets surface after grabbing a ride on a dumb circuit. Immediately upon surfacing, each packet must be routed toward its destination. By the way, the route may entail another ride on a dumb optical circuit, say to the innards of an office building or it may just merrily bump along from one router to another until it finds its destination. The increasing number of combinations of long-distance and local optical circuits made possible with DWDM and innumerable metropolitan optical cables being laid make the edge the new sweet spot of the network equipment market. Why else would Cisco pay $7 billion for a switch company making $10 million in revenue, if not for the explosive ramp occurring on the edge? Having dominated routing within the enterprise, Cisco intends to extend its dominance to the next big thing for IP routing, the edge of the cloud.
Today, the edge of the communications backbone physically is likely to be a rack closet in a telephone company switching station. A treasured feature of Cerent's switches is their diminutive size that conserves this valuable real estate, thereby minimizing rental fees for space. But clearly, the edge is getting fuzzy, as some circuits extend to lots of non-traditional locations, like office buildings. In the future, the edge will get fuzzier yet, as optical circuits extend to homes and small offices, providing a rich mixture of broadcast information and IP packet handling capability. In my mind, and apparently Cisco's and WIND's, the edge offers market opportunities that are staggering. IP packets will play a primary role in that opportunity, at every spot on the edge, and this implies not just basic routing, but a host of other exotic functions that can be layered on the process of handling packets -- all intelligent. These include VoIP, VPNs, firewalls, encryption, intrusion detection and QoS to name a few. New categories of devices will emerge that switch and route and do all sorts of intelligent things, made practical by intelligent network processors and vertical software like TMS.
Yesterday, I heard a news commentator claim that the edge was growing at a 50% rate. I instantly thought, "50% is high, but I equipment sales on the edge must be doubling at least." Then the commentator finished his sentence, "per quarter for at least the next five quarters." That corresponds to over 500% growth in the coming year!
One thing that came out of WIND's Analyst Day, that I wasn't able to emphasize properly without this background, was confirmation that optical switches/routers on the edge is an exploding market for WRS Networks. Dave Fraser made this point emphatically in one of his slides. I would not be surprised to learn that this lily pond exceeds all others in terms of near-term implications, including both an old favorite I2O and certainly DSL. I anxiously await published metrics for this market along with WIND's market position and guidance on ASP's.
As I interpret Gilder, he seems to imply that network intelligence in unnecessary. Because optical circuits are virtually unlimited, he seems to suggest that everything simply should be sent everywhere, without electronic intervention. To the extent that there will be mainstream broadcast circuits, he is right. To the extent the future rests on the packet architecture of the current Internet, he is wrong. His views are certainly misleading, or at least confusing, as they pertain to the current explosive activity occurring today on the edge.
As long as IP packets must be intelligently routed using electrons, there will be pressure to minimize the necessary number of packets floating around the Internet. Moving bits will continue to get cheaper as bandwidth increases and microelectronics improves with Moore's Law, but in my lifetime they will NEVER be free. The gating factor will always be the variable cost of required intelligent processing, bounded by the fixed cost of provisioning a fixed circuit. Neither complete circuits nor microelectronic processing are destined to be free.
Only if the microphotonic computer becomes a reality will the IP packet truly become one with optical communication links. Should this occur, the variable cost of transmitting a packet becomes infinitesimal. However, even then, intelligence like TMS still will be required. It is not that photons obviate the need for intelligence; it is that intelligence might move to photons.
Be warned. This is just my opinion.
Allen |
