John’s vision 3
EXTRACT: MOBILITY
[…]
From Faster to Closer
Those of us who were enthralled by the potential of the Internet in its early years once hoped that teleshopping would replace trips to the mall, that air travel would give way to teleconferencing, and that digital transmission would replace the physical delivery of books and videos. Each has happened to a certain degree—but with technology serving as additions to, not replacements for, other kinds of mobility. The Internet has increased transport intensity in the economy as a whole more than it has displaced individual acts of movement. It continues to stimulate more mobility than it replaces in much the same way that roads built to relieve congestion often end up increasing traffic. Rhetorics of a “weightless” economy, the “death of distance,” and the “displacement of matter by mind” sound ridiculous, in retrospect.
There is an alternative. The speed-obsessed computer world, in which network designers rail against delays measured in milliseconds, are years ahead of the rest of us in rethinking space-time issues. They can teach us how to rethink the real-world mobility dilemma. Embedded on microchips, computer operations entail carefully accounting for the speed of light. A 600-megahertz Pentium II processor, for example, executes one computing instruction a nanosecond; this is the time needed for a signal to move nine inches on a metal wire—and a leading-edge chip today houses as much as seven miles of wires. On the ground, network delays stem chiefly from the distances between Internet routers. Across the Internet, the average message flows through seventeen routers, and sometimes as many as forty. Many of these routers are thousands of miles—or tens of milliseconds—apart.
The problem the geeks are struggling with is called latency—the delay caused by the time it takes for a remote request to be serviced or for a message to travel between two processing nodes. Another key word, attenuation, describes the loss of transmitted signal strength as a result of interference—a weakening of the signal as it travels farther from its source (much as the taste of strawberries grown in Spain weakens as they are trucked to faraway places like Amsterdam).
The inevitability of latency and attenuation prompt serious talk of a “light-speed crisis” in microprocessor design. Optical computing guru George Gilder, for example, says that “the chip faces a light speed crisis that requires a radical change in the time-space relations of processors and memories. Money will not change it: you can’t bribe God.” The only way to combat the limits of light speed is by moving closer to the data—and moving the data closer to you. Hence the emphasis now being placed by network designers on geodesic distance. Gilder describes the Internet as a “computer on the planet. Like a computer on a mother board, it faces severe problems of memory access.”
The search for geometric efficiency now dominates all scales of information processing and distribution. This search has prompted the emergence of the so-called storewidth paradigm or “cache and carry”—a focus on copying, replicating, and storing Web pages as close as possible to their final destination—at so-called content access points. If you go to retrieve a large software update from an online file library, you are often given a choice of countries from which to download it. The technique is called “load balancing”—even though the loads in question, packets of information, don’t actually weight anything in real-world terms. Choosing a nearby country will usually result in a faster transmission. Firms optimize the delivery of data to customers by storing lumps of popular and heavy data in caches sprinkled around the world. Akamai, a cache-and-carry market leader, maintains eleven thousand such caches in sixty-two countries. By monitoring demand for each item downloaded and making more copies available in its caches when demand rises and fewer when demand falls, Akamai’s network can help to smooth out huge fluctuations in traffic.
Other companies have not given up on distributed computing. Kontiki’s approach, for example, combines Akamai’s cache-and-carry approach with smart file sharing similar to that in the system invented by Napster: Users’ own computers, anywhere on the Internet, are used as caches so that recently accessed content can be delivered quickly when needed to other users nearby on the network. The light-speed crisis favors specialized distributed processors doing their work on location—the network’s example of sending the acorn, not the tree.
While investigating the subject of distributed computing, I received a flyer for a report called Colocation 2002: A Telegeography Guide to Power and Space. The explosion of global networks, opened markets, evolving information transmission methods, and competing information carriers has introduced a new problem: Where should the multitude of new carriers and content providers interconnect their networks? The book promised to identify and evaluate 350 “colo sites” in fifty cities around the world. The cover price of the report was $1,795—so I cannot tell you where they are—but I am able to conclude that colo sites are the information equivalent of the intersections between road and rail networks, inelegantly named “transferiums” in some countries, which—like airports—are now being developed as destinations in their own right.
Where the Internet actually is, is in cities. Anthony Townsend, an urban planner at the Taub Urban Research Institute at New York University, says that just as cities are often railway or shipping hubs, they are also the logical places to put network hubs and servers, the powerful computers that store and distribute data.
Cities are already vast information storage and retrieval systems in which different districts are organized by activity or social group. A mobile Internet device can be come a way to probe local information resources. Distance between two points is one thing (and even that matters, according to Gilder)—but where those points are still matters a lot. Says Townsend, “mobile devices reassert geography on the internet.”
The Law of Locality
People and information want to be closer. When planning where to put capacity, network designers are guided by the law of locality; this law states that network traffic is at least 80 percent local, 95 percent continental, and only 5 percent intercontinental. Between 1997 and 1999, for example, 30 percent of all U.S. Internet traffic never crossed the national infrastructure but stayed within a local metropolitan network. Someone should have mentioned the law of locality to investors before they dumped some seventy billion dollars into projects for long-haul Internet infrastructure. Only a tiny fraction of these costly fibers are currently“lit”—as little as 3 percent by some estimates. According to research firm Probe Research, only 14 percent of the fiber-optic cable laid across the Atlantic to support Internet traffic may ever be needed.
This is not the “death of distance” that the companies who laid the fiber had in mind. The assumption driving the money spent on this long-haul infrastructure was that the need for more capacity on the Internet would grow exponentially through the widespread adoption of bandwidth-sucking applications such as virtual private networks and videoconferencing. The enduring popularity of the telephone is proof that high-value connectivity is not bandwidth-dependent. At the height of Napster’s popularity, in 2000, the service was using about 5 percent of the available network capacity in the United States—but no other Internet-based service has ever come near that level of usage. High-capacity networks are a fabulous technology chasing applications that do not yet exist—and may never exist. The designers of communication networks use another design rule that we can learn from: “The less the space, the more the room.” In silicon, the trade-off between speed and heat generated improves dramatically as size diminishes: Small transistors run faster, cooler, and cheaper. Hence the development of the so-called processor-in-memory (PIM)—an integrated circuit that contains both memory and logic on the same chip.
This design principle—“the less the space, the more the room”—is nowhere better demonstrated than in the human brain. Edward O. Wilson describes the brain’s custardlike mass as “an intricately-wired system of a hundred billion nerve cells, each a few millionths of a metre wide, and connected to other nerve cells by hundreds of thousands of endings. It comprises the equivalent of one hundred billion squids linked together. Overall the human brain is the most complicated thing in the known universe—known, that is, to itself.” Information transfer, Wilson explains, is improved when neuron circuits, filling specialized functions, are placed together in clusters. Examples of such clusters so far identified by neurobiologists are sensory relay stations, integrative centers, memory modules, and emotional control centers. “The ideal brain case is spherical, or close to it,” Wilson says. “One compelling reason is that a sphere has the smallest surface relative to volume of any geometric form—and hence provides the least access to its vulnerable interior. Another reason is that a sphere allows more circuits to be placed close together. The average length of circuits can thus be minimised, raising the speed of transmission while lowering the energy cost for their construction and maintenance.
Comprar Plano B na Livraria Cultura