Special Report: What To Expect In Scientific Computing

Since personal computers first came into widespread use in the early 1980s, they have become essential for virtually all scientists. The best thing about these now-ubiquitous machines, aside from their ability to revolutionize research, is that as time goes by, they continue to get faster, smaller, and--relative to their power--cheaper. And with this increasing power, and the new software that it fuels, will come additional fundamental changes in the way scientists work. Advancing technology already has blurred the distinction among types of computers. As the machines are refined even further, it will become less important to scientists whether they are working on small or large computers; the operations will be the same from the user's end. For the next few years, at least, scientists will continue using desktop computers that can be connected to larger computers. However, Bill Joy, vice president for research and development at Sun Microsystems Inc. of Mountain View, Calif., predicted in Byte magazine (15[2]:30, February 1990) that five years from now, those desktop computers will be replaced by laptops that are smaller and more powerful than those out now. With portability added to the list of the PC's attributes, anyplace will be able to become an office or lab. In personal computers, emerging microprocessor technologies are promising to create entirely new abilities for scientists. Intel Corp., headquartered in Santa Clara, Calif., is chief supplier of microprocessors to the IBM family of PCs. The company has doubled the speed of its microprocessors every 18 months for the last 10 years, and Intel staffers expect this rate to continue for the next decade. Motorola Inc. of Schaumburg, Ill., supplier of microprocessors for Macintosh computers, is also developing faster and more powerful chips. These more powerful microprocessors will allow scientists with PCs to do tasks such as visualization, the conversion of complex or multidimensional data sets into visual forms. Currently, larger, faster supercomputers are required for advanced visualization tasks. Additionally, the smaller machines will incorporate more RISC (reduced instruction set computing) chips: processors, currently in use in larger workstations, that simplify how computers handle instructions, adding even greater speed. These developments will advance electronic communications as well. With computers, high-speed communications, and increasingly powerful video capabilities, says Terrence R. Smith, professor of geography and computer science at the University of California, Santa Barbara, "we will have the ability to have teleconferencing among different university research locations, and we will be able to talk on screen and use [the screens] as if they were blackboards ... just as if everyone was in the same office together." One prototype of this teleconferencing for scientists, Smith says, is the National Aeronautics and Space Administration's Earth Observing System (EOS) project, which will eventually contain data measured in petabytes--millions of gigabytes--collected over a 15-year period from a series of orbiting satellites. NASA plans to launch the first satellite in 1998. The data, measuring such things as vegetative cover, temperature, and atmospheric moisture, will be contained in a central location, but scientists all over the United States will tap in through workstations and, using special software, will be able to access the data as if the information were stored down the hall. Visualization, the ability to pre-sent vast quantities of otherwise incomprehensible data in visual forms that make them understandable, is now one of the hottest areas of computer science. With this technique, concepts such as the evolution of galaxies, the movement of tectonic plates, and the shapes of molecules can be shown graphically. Over the next few years, scientists will be able to produce increasingly sophisticated graphics on mainframes, workstations, and even PCs, instead of working directly on the supercomputers currently required for complex visualization tasks. Much of this visualization requires larger super- computers, says Dan Sulzbach, executive director of the San Diego Supercomputer Center (SDSC), operated by San Diego-based General Atomics, a high-tech R&D company, at UC-San Diego for the National Science Foundation. Machines generally used for these operations are produced by the Eagan, Minn.-based Cray Research Inc. For example, Sulzbach says, a simulation on a Cray Y-MP supercomputer calculating the trajectory of 100,000 particles in a fluid dynamics problem would produce thousands of numbers on hundreds of pages describing the position of different particles. Calculations on the Cray allow visualization of the data, turning numbers into pictures on the screen that show the trajectories of the particles, turbulence, densities, and other patterns not apparent otherwise, obviating the need to print out hundreds of incomprehensible pages. This ability soon will extend beyond supercomputers, says Sulzbach. Smaller computers are already capable of carrying out some visualization tasks, such as molecular modeling (The Scientist, Dec. 10, 1990, page 30). As they gain power and speed, PCs will have the potential to perform increasingly complex visualization functions. Moreover, computers will be able to share visualization tasks. The visualization itself could be displayed on the lab workstation, although the hard number-crunching would take place on a distant supercomputer. At IBM's T.J. Watson Research Center in Yorktown Heights, N.Y., scientists are already using visualization for a full range of scientific research with the new, relatively small--yet mighty--IBM RISC 6000 computer. "The human mind works better using pictures than columns and columns of numbers," says Denis Arvay, manager of supercomputing information at the lab. The computers of the next decade also will make searching vast amounts of data easier, according to Rick Belew, an assistant professor of computer sciences and engineering at UC-San Diego. Belew and others are working on new ways of collecting data using neural network algorithms, keywords, and artificial intelligence to search huge databases. The programs are based on reality: When a scientist wants to find something, the first place searched is his or her own files, then the department files, then, finally, the university library. This can be complicated because some terms, such as "neural network" or "fractional crystallization," can mean different things to different scientists. The software Belew and others are designing will emulate techniques that a scientist in "real life" would use, taking into account different definitions of words, and remembering what it does to make subsequent searches faster. Belew's software reflects yet another innovation that will further erase the distinction among machines. The scientist sits at a PC to begin the search. If the computer has to go to the department file and then to the library file to find what the scientist seeks, it does so invisibly--the scientist neither knows nor cares where the computer is searching. If the computation suddenly requires more memory than is available on the PC, the computer will go immediately to a larger computer to get what it needs. To the scientist in the lab it will all look as if it's happening on the desktop, not in another room or another state. The world of supercomputers is changing in yet another way that was unanticipated just a few years ago. At that time, computers such as Cray's Y-MP or Cyber, by the Minneapolis-based Control Data Corp., seemed to be the trend of the future. These machines, known as "vector computers," have several extremely fast processors. Each works independently, from its own set of instructions, and the data are handled--for the most part--in single file. But another family of machines, called parallel computers, may surpass them for special tasks. "You start to reach a limit on how fast a single processor can go," says SDSC's Sulzbach. Parallel computers such as the Connection Machine, made by the Cambridge, Mass.-based Thinking Machines Corp., have thousands of processors working simultaneously. While these machines solve many of the problems of non-parallel computers, their usefulness tends to be quite narrow--a parallel computer designed for a certain task can do that task phenomenally well, but is unable to do anything else. Sulzbach notes that while part of this limitation is due to a lack of software for the parallel machines, computer scientists are also finding that "not all problems parallelize well;" some tasks are better solved sequentially. These vast and pervasive changes challenge the very definition of supercomputing. In many cases, scientists won't need supercomputers, provided they can get the machines they already have to operate in conjunction with each other. The average computer science department at an average university already has that computer power; all it needs is the hardware and software to let the computers work synergistically. Many scientists think this ability to connect smaller computers could spell the end of the large machines. Regardless of whether or not this prediction comes true, however, scientists will find that their computers will soon be plugged into a global community. Joel N. Shurkin is currently a science writer for Stanford University, and is soon to be a freelance writer based in Santa Cruz County, Calif.

Special Report: What To Expect In Scientific Computing

Interested in reading more?

Become a Member of