Europe may be outclassed by the US and Japan in supercomputers, but
not in supercomputing. How so? The answer is grid computing. The US is starting to catch
up in grid development, in part through “the biggest, fastest network ever undertaken for
scientific research,” which will lead not only to “some truly extraordinary
discoveries” but also to “extreme multimedia,” such as “real telepresence,”
enabling scientists to collaborate across the world as easily as with colleagues
across the hall.

Competition is as hot at the Lilliputian as at the giant end of the computing
devices scale. Japanese and US researchers have independently devised ways to
read and write data to individual molecules, promising massive molecular memory storage in tiny
packages. But if the promise is not kept, there will always be plastic to fall
back on. It may be big compared to molecular substrates, but plastic memory is so easy and cheap to
manufacture that size may be an acceptable sacrifice.

If you prefer something more exotic than plastic to satisfy your
computing aesthetic, consider circuits built by DNA, or crystal photonic
circuits, or “adiabatic” circuits that recycle their own electrons. You may have
to wait a few years for those, but not for something possibly even more exotic:
unbreakable quantum encryption, on
the market now for US$50-100,000.

The New York Times‘ John Markoff notes three significant shifts in
computing this year.

• First: While attendance at Comdex, the leading annual show for the computing
  industry, was down, attendance at the annual supercomputing expo (SC2003)
  surged, stimulated by the completion of NEC’s Earth Simulator, a
  supercomputer an order of magnitude more powerful than any other.
• Second: The supercomputing exhibition “highlighted the extent to which the
  industry has seized on grid computing.” (See next article for recent
  developments in grid computing, which is essentially about achieving
  supercomputing power without supercomputers.)
• And third: There is evidence of a distinct trend to the adoption of
  “reconfigurable logic” chips whose circuits can be reconfigured on the fly to
  run different kinds of application. Indeed, there is speculation that the reason
  Microsoft chose IBM chips over Intel’s for the next generation of its
  Xbox video game machine is because IBM is going the reconfigurable logic
  route.But the low end of supercomputing is generating as much buzz as the high end,
  now that Virginia Tech’s “Big Mac” supercomputer has been officially declared
  the world’s third-fastest. Here’s a brief run-down on supercomputers making the
  headlines:
• Big Mac: Uses 1,100 dual-processor Power Mac G5 chips. Cost
  US$5.2 million. Performs 10.28 trillion floating-point operations a second
  (teraflops). Big Mac II to begin construction in 2006, will be shooting
  for 50 teraflops.
• Earth Simulator: Uses 5,120 processors; would fill four tennis
  courts. Estimated cost US$350 – $400 million. Performs 35 teraflops.
• Blue Gene/L: 130,000 PowerPC 440GX processors, would fill half
  a tennis court. Will perform 360 teraflops to become easily the world’s fastest
  supercomputer when it comes online next year. The machine will likely run on
  Linux. IBM is due to deliver that machine plus the 100-teraflop ASCI
  Purple supercomputer to the Lawrence Livermore Laboratories by early 2005
  for a combined cost of $267 million.
• A dishwasher-sized prototype of Blue Gene/L just unveiled is the
  73rd-fastest computer in the world in its own right and may be produced for sale
  to corporate data centers. It will be air-cooled and use no more power than the
  average home. Performs about 1.4 teraflops.The US Defense Department has set a goal of a petaflop (1,000 teraflop)
  computer by the end of the decade. Battelle, the firm that manages the Oak Ridge
  National Laboratory in Tennessee, is working to deliver just that by 2008 for
  about US$500 million. It is intended in part specifically for medical research
  applications. But Battelle could be beaten to the punch by IBM, which is already
  planning the petaflop BlueGene/P for completion as early as 2006.

  Among the first applications IBM is exploring to harness Blue Gene‘s
  power is to model the “folding” of proteins in human cells, which will lead to a
  better understanding of diseases and potential cures. Already in use for medical
  research is Arizona State University’s TGen supercomputer, which cost
  US$4.4 million and gets 1.8 teraflops out of 1,024 Intel Xeon microprocessors.
  TGen started work in September, searching for genetic markers of cancer,
  Alzheimer’s, and other diseases. It uses genetic data from the Human Genome
  Project and from the cells of cancer patients who did and did not respond well
  to specific treatments. Demand is expected to “saturate the system power” within
  a year, based on the flood of requests for use of the machine by researchers.

  Reference: Markoff, John (2003). “Switching
  Allegiances in Computers.” New York Times, November 24.

  Reference: Unknown (2003). “Mac
  Supercomputer Joins Elite.” Wired News/AP, November 15.

  Reference: Markoff, John (2003). “IBM Says
  Supercomputer to Be Suitable for Businesses.” New York Times, November
  14.

  Reference: Unknown (2003). “Supercomputer
  Could Be Used to Analyze Medical Data.” iHealthBeat, November 19.
  

  Reference: Wichner, David (2003). “Powerful computer to
  aid in research.” Arizona Daily Star, 20 November.

  Grid Computing: US Lags EU

  We reported last month on the grid-based National Digital Mammography
  Archive. Now, to expedite researchers’ access to “key bioinformatics platforms,”
  the National Cancer Institute (NCI) plans to build a “cancer biomedical
  informatics grid (caBIG),” a “common, extensible informatics platform that
  integrates diverse data types . . . supports interoperable analytic tools . . .
  [and] will allow research groups to tap into the rich collection of emerging
  cancer research data while supporting their individual investigations.” These
  are welcome developments, if a little tardy compared to Europe, where the
  potential of grid computing was recognized and acted upon much faster than in
  the United States — as we have noted in recent issues of HFD.

  “Europe has decided that this is a real competitive advantage, and they are
  going after it,” a US National Science Foundation director told John Markoff and
  Jennifer Schenker of the New York Times. The fact that Europe’s lead in
  grid computing may be as much as 18 months, and that the European Union has a
  five- to ten-year strategic plan in place, is “a slap in the face and a wake-up
  call that things have gone global,” another US expert said. However, Europe is
  going after it with US-made technology.* The US National Science Foundation is
  in discussions with European Commission officials to try to bring about more US
  participation in the European grid activities.

  The University of California at San Francisco, meanwhile, appears to be
  heading in the opposite direction. Assisted by IBM — one of the chief
  proponents of grid computing — UCSF plans to move all the medical data from 150
  databases into one. Perhaps the US, or at least IBM, is hedging its bets.

  * For example, Swiss pharmaceutical company
  Novartis used software by United Devices of Austin, Texas, to link 2,700 desktop
  PCs to help create drugs. The grid has already helped discover several promising
  new chemical molecules, and Novartis now plans to expand the grid to all 70,000
  PCs in its corporate network.

  Reference: caBIG website

  Reference: Markoff, John, and Jennifer L. Schenker
  (2003). “Europe Exceeds
  U.S. in Refining Grid Computing.” New York Times, November 10.

  Reference: Kirby, Carrie (2003). “UCSF,
  IBM to team up: Medical info to go into one database.” San Francisco
  Chronicle, November 17, p. E-1.

  Warp Speed Networks

  The first leg of the US National LambdaRail — “the biggest, fastest
  network ever undertaken for scientific research” — has gone live between the TeraGrid
  facility in Chicago and the Pittsburgh Supercomputing Center. Hundreds of
  research institutions around the United States could be linked when the NLR is
  completed by the end of 2004.

  It is being created from 10,000 miles of unused or “dark” fiber.* It will use
  about 40 wavelengths (“lambdas”) operating at 10 Gbps each. If all wavelengths
  can be combined for a single transmission, that would make a bandwidth of 400
  Gbps and do to network speeds what IBM’s Blue Gene/L is about to do to
  existing supercomputers: blow them away. Meanwhile, the Oak Ridge National
  Laboratory has been awarded a contract by the US Department of Energy to design
  a 10-40 Gbps network called Science UltraNet. We’re not sure if this is
  intended as a leg in the NLR, but it certainly could become one.

  Grid computing puts a new premium on bandwidth, since trying to accommodate
  petaflops of processing power and petabytes of data over the existing Internet
  would be like trying to operate a jumbo jet from a dirt airstrip. Quoting from a
  US National Science Foundation report, Wired‘s Leander Kahney notes: “The
  amounts of calculation and the quantities of information that can be stored,
  transmitted and used are exploding at a stunning, almost disruptive rate.
  Powerful data-mining techniques operating across huge sets of multidimensional
  data open new approaches to discovery. Global networks can link all these
  together and support more interactivity and broader collaboration.”

  An NLR director told her: “We’re going to have some truly extraordinary
  discoveries and data-mining capabilities, but we need these kinds of network
  connections to allow the scientists to trawl through these enormous amounts of
  data.” As if that were not enough to revolutionize science, the new global
  networks will also enable “extreme multimedia,” such as “real telepresence,”
  enabling scientists to collaborate across the world as easily as with colleagues
  across the hall.

  * Perhaps the fiber laid by over-optimistic
  interexchange carriers during the dot com boom.

  Reference: Kahney, Leander (2003). “Fast Track
  for Science Data.” Wired News, November 17.

  Reference: Associated Press (2003). “Tenn.
  lab to design high-speed network.” USA Today, November 25.

  Molecular Memory

  Japanese researchers have found a way to write (store) and read (retrieve) a
  bit of data electronically on a single molecule of photochromatic diarylethene.
  The method may be compatible with existing electronics, works at room
  temperature, and requires very little power. Molecular memory could store very
  large amounts of information in a very small space, and is inexpensive enough to
  be disposable. The researchers think inexpensive disposable memory circuits
  could become practical into three years, and ultra-high density molecular memory
  systems in five to ten years.

  Not to be outdone, US researchers have bonded a molecule of photochromic
  fulgimide with a fluorescent dye molecule to form a single molecule with two
  states — the basis for binary computing. The molecule “resists accidental
  erasure,” switches quickly, and could lead to terabyte floppies. This method
  uses photons rather than electrons to read and write data to the molecule, and
  could become practical in five to seven years.

  The race is on.

  Reference: Unknown (2003). “Molecular Memory
  is Electric.” Technology Research News, November 14.

  Reference: Unknown (2003). “Paired Molecules
  Store Data.” Technology Research News, November 3.

  See also “Memory“; “Molecular
  Memory”

  Plastic Memory

  Princeton University and HP Labs researchers have combined a conducting
  polymer commonly applied to photographic film as an antistatic coating with foil
  and silicon diodes to make a WORM (write-once, read many times) memory material.
  Production would be simpler and per-megabyte costs should be lower than silicon
  memory, because manufacturing won’t need lithography, expensive clean rooms,
  vacuum chambers, and high temperatures, and because layers of the material could
  be stacked.

  Reference: Fordahl, Matthew (2003). “New
  Plastic Memory Technology Unveiled.” Associated Press, November 12.
  

  Reference: Biever, Celeste (2003). “Plastic memory
  promises cheap, dense storage.” New Scientist, November 13.

  Reference: Unknown (2003). “Layers Promise
  Cheap Storage.” Technology Research News, November 24.

  Exotic Computing

  Progress in DNA Computing

  Israeli scientists have harnessed DNA to “self-assemble” a nanoscale
  electronic device made of gold-plated carbon nanotubes. The DNA is not involved
  in operation of the circuitry, only in its construction.

  The scientists attached the nanotubes to a protein that helps construct DNA
  as part of a natural biological process called “recombination,” which cells use
  to repair damaged DNA and to swap genes. The protein moves to an exact location
  along the DNA strand, taking its attached nanotube along for the ride, rather
  like construction workers carrying building materials to a specific spot
  according to the architect’s plan. The nanotubes are then connected to gold
  wires at each end, enabling current to flow when an electric field is applied —
  in other words, they have become transistors.

  The next step will be to build an actual circuit by stretching DNA across a
  surface to provide a template, a process that has worked in simulated
  experiments.

  Reference: Chang, Kenneth (2003). “Smaller Computer
  Chips Built Using DNA as Template.” New York Times, November 21.

  Photonic Computing Comes a Step Closer

  Stanford and MIT researchers have designed (but not yet made) a simple
  optical switch and an optical transistor. To be constructed from man-made
  photonic crystal, both would be smaller than a micron, require little power, and
  be capable of manufacture in existing chip fabrication plants. The devices could
  be in practical use in photonic computing in two to five years.

  Reference: Unknown (2003). “Switch Promises
  Optical Chips.” Technology Research News, November 20.

  Reversible Computing

  We noted in “Screeching Halt?” (in the Acceleration section) that heat
  dissipation will be a show-stopper for Moore’s Law when transistor gates on
  computer chips reach the five-nanometer level in about 2018 — unless a way is
  found to recycle the electrons.

  University of Florida researchers have found a way. They propose to
  incorporate tiny oscillators that allow the chips to recapture the energy
  expended in one calculation and re-use it for the next calculation — a process
  known as reversible, or adiabatic, computing. A startup company, Adiabatic
  Logic, has already designed a commercial reversible processor. (It is not clear
  from the Wired article, but the inference is that Adiabatic Logic is
  using the University of Florida method.) Yet AMD, IBM, and Intel are showing
  little interest, apparently because the technology is not yet developed enough
  to be competitive with current technology. Adiabatic Logic thinks it is only a
  matter of time before the chipmakers will have no choice but to turn to
  reversible computing.

  Reference: Asaravala, Amit (2003). “Chip Design
  Reverses a Hot Trend.” Wired News, November 13.

  Quantum Encryption For Sale

  Startup firm MagiQ Technologies has begun selling what may be the first
  commercially available quantum encryption system. It uses photons to transfer
  the encryption keys used to protect documents from prying eyes and is
  unbreakable, because if the photons — quantum particles — are observed, they
  change, scrambling the key and also alerting sender and recipient to the
  attempted intrusion. The system, called Navajo after the Navajo language
  code used with great success in World War II, is contained in small boxes that
  generate and read the signals over a fiber-optic line. The boxes can be up to 70
  miles apart, and more boxes add more distance. Navajo costs
  $US50,000-100,000, affordable to banks, insurers, government agencies,
  pharmaceutical companies, and other organizations that transmit sensitive
  information worth considerably more. Navajo also changes the keys ten
  times a second, so is impervious to the human carelessness with keys that makes
  other otherwise invulnerable systems vulnerable.

  The device has significance beyond encryption, because it represents a step
  towards practical quantum computing. And the beauty of it is that even a quantum
  computer has been shown theoretically to be incapable of decoding a
  quantum-encrypted message.

  Reference: Bergstein, Brian (2003). “Quantum
  leap for encryption.” Associated Press/Australian IT, November 17.