Computer Networks: Priming For High-Speed Applications

The High Performance Computing Act, which became law last December, provides a boost in federal funds to improve and integrate the confusing collection of computer networks used by scientists to transmit data and electronic mail. The ultimate goal is to create a high-speed national computer network, analogous to the interstate highway system, with a capacity to transmit data at one gigabit (1 billion bits) per second, a speed 700 times greater than today's system. Such a system would provide enough data transfer capacity to allow supercomputers to effectively converse or for multimedia video conferencing to take place. Proponents argue that federal investment in high-speed networking for scientists will be a key to United States competitiveness in the information age. The National Science Foundation and the Defense Advanced Research Projects Agency (DARPA) have been designated the lead agencies to create, in collaboration with five other government agencies, the gigabit network, to be called the National Research and Education Network (NREN). And, to that end, federal support for networking rose from $65 million in 1991 to $93 million in 1992 and is to be funded at a proposed $123 million in 1993. Freeman Gilbert, a seismologist at the Scripps Institution of Oceanography in San Diego, Calif., says gigabit networking will offer new ways to conduct research in many fields. Citing an example, Gilbert says, "It's not beyond the realm of imagination that a medieval Latin scholar would want to look at a manuscript in the Vatican." A gigabit network would simplify the task of shipping its digitized image to the scholar's computer screen. "I think it's a way for people to do things with their minds that would have once required physical travel," Gilbert says. The first goal of this NSF/DARPA gigabit initiative is to integrate today's loosely organized networks into NREN. The agencies are also charged with directing research in advanced data switching and transmission to make gigabit networking possible in the mid-1990s. Today, scientific computer networking takes place over Internet. Internet is not a single network but rather a collection of many subnetworks that have established linkages, as scientists have sought ways to exchange electronic mail, participate in research forums, or download computer data from remote machines. Highways To Success In a recent article in Scientific American (265:150-3, September 1991), Sen. Al Gore (D-Tenn.), NREN's congressional champion, likens modern computer networks to the highways, bridges, and canals that were the keys to economic success in the industrial age. "Today," he writes, "transportation is less important compared with other factors, such as the ability to move information and to increase the value of this information." The ancestor of all computer networks was ARPANET, created by the Defense Advanced Research Projects Agency in 1969 as an experimental linking of its computer research projects. As computers proliferated in the 1970s and 1980s, university and industrial laboratories improved upon the ARPANET technology to form local area networks, or LANs. These LANs linked all the computers on a given campus. Users then sought ways to exchange information with colleagues on different campuses, and so LANs branched out via fiber optic cable to create wide area networks, or WANs, like the popular Bitnet, which came into being in 1981. Today many WANs exist, and each has its own backbone, or main circuit, which is usually composed of fiber optic cable. To exchange message traffic, these WANs are linked by switches, routers, and bridges--hardware devices that direct data from one fiber optic backbone to another. The sum of all these interconnected WANs, plus the software that directs them, is called Internet--a chaotic but workable system for exchanging computer data among thousands of industrial and scientific laboratories worldwide. --T.A. Gore says the federal investment in NREN will stimulate domestic research in the networking technology that will be one key to future competitiveness. Toward this end, the legislation creating NREN calls for the U.S. to upgrade its network from the present data transfer rate of 1.44 megabits per second to a gigabit per second as the technology becomes available. In so doing, Gore hopes the U.S. will pave the way for the information highways of the future. "No country in the world has the level of research in gigabit technology that we do," says Stephen Wolff, NSF's director of network research and infrastructure. Wolff says more than half of this year's $93 million NREN budget will be used to upgrade NSF's current network, called NSFNET, which operates at a speed of 1.44 megabits per second. The upgrades will bring the rate of data transmission to 45 megabits per second by 1993, requiring new fiber optic lines, as well as new connecting devices called switches, routers, and bridges, that can operate at these accelerated rates. He says NREN won't take the next leap, to gigabit speeds, in a single bound but will take smaller steps to get there. "The technology is beautiful because you don't have to upgrade the network all at once," Wolff says. "It can exist as a collection of subsystems operating at vastly different speeds." NSFNET was created in 1986 to help scientists tap into six supercomputing centers established the previous year. Eric Aupperle is president of Merit Network Inc., which manages both NSFNET and a regional network for the University of Michigan. He says there's "a misconception" that NSFNET serves only supercomputer users. "Traffic has become increasingly pervasive," Aupperle says. He cites figures showing that, in addition to 600 research universities that belong to NSFNET, more than half the nation's four-year colleges and 20 percent of the liberal arts institutions are linked to the network. Aupperle says traffic on NSFNET grows 11 percent each month. About half of the academic scientists in the U.S. access NSFNET, which allows them to send messages to 3,200 networks in the U.S. and 1,600 networks abroad. Aupperle says about half of the traffic on NSFNET involves database transfers or supercomputer downloads. He says electronic mail makes up another 20 percent of the traffic, with the remainder given over to a variety of applications. Work Simplified Astrophysicist Robert Kurucz of the Harvard-Smithsonian Astrophysical Observatory in Cambridge, Mass., says NSFNET has simplified his work. "We don't have a very big computer in Cambridge," he says. "So, my whole career I've been traveling around the country looking for free computer time." The observatory is now linked to the supercomputer in San Diego, which he uses to analyze the composition of gases burning in stars. "I can sit in my office in Cambridge and get the output faster than in San Diego," says Kurucz, because the printer in Cambridge is on the floor below his office, while at San Diego, he must walk down two flights to retrieve the printout. Janet Price, director of computer services at Kalamazoo College in Michigan, says administrators at her small liberal arts institution now sign onto the network. "If our grants development officer is interested in guidelines, NSF grant information is available on the Internet," Price says. In her capacity as a computer graphics teacher, Price once needed a special computer graphics routine to run on a PC. So, she posted a message on Internet. "A person in Germany responded with the program," she recalls. "Generally speaking, younger scientists tend to make more use of the network," says Stephen Levene, a biophysicist at the University of Texas. He says electronic mail is the most commonly used network feature. "Older scientists, even when they are computer literate, tend not to use E-mail as heavily," says Levene, who began using NSFNET to download supercomputer data about protein and DNA complexes. Congressional Inspection In March, the House committee on Space, Science, and Technology, chaired by Rep. Frederick Boucher (D-Va.), held hearings to look at how NSF was managing the transition to NREN. "The purpose of the hearing was to review the current management of NSFNET and look at the bid for a new management contract," says committee staffer Jim Wilson. NSFNET has been managed in partnership with IBM and MCI under a five-year, $14 million contract that will expire soon. According to NSF's Wolff, other companies interested in managing NREN want to make sure that a forthcoming bid request will give them a chance to compete. At stake is the management of the nation's future information highway, and the spoils that go to the victor. "As NSFNET becomes a more important part of the education and research effort, it is only to be expected that Congress will lift the corner of the blanket and look inside from time to time," Wolff says. In the addition to funding being used to upgrade the present network, Wolff says NSF will spend $2 million, and DARPA about $33 million, to develop the software and hardware necessary to achieve gigabit speeds in the future. NSF engineer Darleen Fisher supervises a gigabit research project cosponsored by DARPA. This three-year, $15 million effort began in the spring of 1990 with the establishment of five research "test beds." Each test bed was intended to solve technical problems in gigabit networking and develop computer applications that take advantage of high-speed data transfer. "I believe all the test beds will be running at a gigabit by the end of the year," Fisher says. Through the Aurora test bed, the University of Pennsylvania's Distributed Systems Laboratory is linked to the Massachusetts Institute of Technology's Laboratory for Computer Science by parallel networks created with the help of engineers from IBM Corp.'s Thomas J. Watson Research Center in Hawthorne, N.Y.; MCI Telecommunications Corp. in Washington, D.C.; Bell Atlantic Corp. of Philadelphia; NYNEX Corp. in New York; and Bellcore's Research and Engineering Laboratory in Morristown, N.J. Fisher says the Aurora project will test two different methods of transmitting data at gigabit speeds. "It is important to emphasize that the industry folks are coming in on their own nickel," says Fisher, noting that the federal investment in gigabit research has stimulated a much larger private contribution. The point that industry has a big investment in high-speed networking is echoed by David Farber, a computer scientist at the University of Pennsylvania. In a paper recently presented to an international workshop on high-speed networking in Munich, Germany, Farber estimated that "for the approximately $15 million of funding supplied by the government, industry will be spending, we estimate, over $400 million" for fiber optic cables, switches, and routers, in addition to engineering brainpower. (See story on page 11.) Another gigabit project, called the CASA test bed, will link a supercomputer at San Diego to one at Los Alamos. Physicist Larry Smarr, director of the National Center for Supercomputing Applications at the University of Illinois at Urbana-Champaign, says the two supercomputers will cooperate on a global climate modeling program that neither machine could handle alone. "The two supercomputers will talk to each other," Smarr says. Satellite Link Even as the test beds develop gigabit networking on the ground, DARPA and NASA hope to launch a satellite in January that could become a portable link in configuring temporary gigabit networks. Paul Mockapetris, a computer scientist who manages DARPA's network research, says the high-speed satellite link will ensure wide access to scarce scientific resources such as supercomputers, even when users are not directly linked by fiber optic cables. "The research world is now in the preliminary stages of reaching gigabit speeds," says Mockapetris. By 1995 or 1996, he expects campus local area networks (LANs) to be operating at near-gigabit speeds. "Historically, what has happened is that people set up faster LANs first because it is cheaper to do things locally," he says. "Once the applications develop locally, people want to do them remotely," driving demand for faster regional and global networks. Tom Abate is a freelance science writer based in Brooklyn, N.Y.

Computer Networks: Priming For High-Speed Applications

Interested in reading more?

Become a Member of