sign in sign up
<<
Home , Computer network

Steps Toward a National Research Network

Gordon Bell

Introduction

Modern science depends on rapid communica- In response to provisions in Public Law tions and information exchange. Today, many major 99-383, which was passed 21 June 1986 by national and international networks exist using the 99th Congress, an inter-agency group some form of to interconnect under the auspices of the Federal Coordin- . State and regional networks are ating Council for Science, Engineering, and proliferating. NSFNET, an "" designed initially (FCCSET) for Research to improve access to centers, has and Applications was formed to study the in the space of two years, forged links among 17 following issues: the networking needs of state, regional, and federal agency networks. the nation's academic and federal research In the early , the lack of access to super- computer programs, including supercomputer power by the research community caused programs, over the 15 years, addressing the formation of the NSF Office of Advanced Sci- requirements in terms of volume of , entific Computing, which funded five centers for reliability of transmission, software . Given the highly distributed loca- compatibility, graphics capabilities, and tion of users, the need for a national wide area transmission ; the benefits and network for computer access and for the inter- opportunities that an improved computer change of associated scientific information (such network would offer for electronic mail, as mail, files, ) became clear. , and remote access and com- Further, it immediately became obvious that munications; and the networking options existing agency networks both lacked the inherent available for linking academic and research capacity and were overloaded. In fact, the current computers, including supercomputers, with a federal wide area networks still operate at the particular emphasis on optics. Bell speed (56 Kbits/sec) of the original ARPAnet (circa reports on the and recommendations 1972), despite the fact that computers have both associated with the committee's work, and increased in number (by a factor of 100) and speed suggests a means for accomplishing the - (by a factor of 30). Moreover, the very nature work objectives addressed by its report. of what is communicated has changed with the power of computers. Computer simulations of physical Bell is Vice President for Research and De- processes output expressed as three-dimensional velopment at Ardent Computer, Inc., Sunnyvale, dynamic graphics. Computer-based research tools California; he was previously Assistant Director are becoming more interactive, and the problems of the Computing and Information Science Direc- under study often require the active collaboration torate at the National Science Foundation. of researchers who are distributed in various research

ISSUE 21 33 institutions. eluding supercomputer programs, over the Today's fiber optic links offer next 15 years, including requirements in terms the ability for a factor of 1,000 to 100,000 of volume of data, reliability of transmission, increase in capacity and speed over traditional software compatibility, graphics capabilities, cable and satellite channels. The price of fiber and transmission security; optic links is based on equivalent voice grade cir- the benefits and opportunities that an improved cuits, and as such remains high, despite a vast computer network would offer for electronic amount of unused capacity. Switching equipment mail, file transfer, and remote access and to exploit even DS3 (45 Mbits/sec) is unavailable. communications; and While several proposals for DS3 fast packet pro- the networking options available for linking tocols and exist, no technology exists academic and research computers, including to packets at fiber optic speeds. supercomputers, with a particular emphasis Most federal agencies support an average of on fiber optics. two, independent wide area networks. One may be used to couple researchers to one another for The Computer Network Study Planning Group, mail, collaboration, and file transfer. The other composed of participants from the Department of connects researchers to large, central systems that Defense (DoD) (including DARPA), Department of support major research initiatives. Often these Energy (DOE), Department of Health and Human networks go to the same institution (an academic Services (HHS), NASA, National Bureau of Standards campus, federal facility, or industrial lab), but to (NBS), and the National Science Foundation (NSF), a different building, wasting resources in the pro- produced and delivered a three-volume report for cess. While a few years ago all wide area net- OSTP in early August 1987, consisting of recommenda- works used different protocols and could not com- tions, results of a workshop, and background papers. municate with one another, today most networks are migrating toward DARPA's TCP/IP protocol Summary of Recommendations to OSTP (for with a commitment to use the IS0 protocols when Congress) for the National Research Network available. Campuses are being wired to interconnect The Subcommittee on Networking, Infrastruc- the array of Local Area Networks (LANs)--a wiring ture and Digital Communications (NIDC) recommend- scheme for a single building or small cluster of ed the following: buildings--to form Campus Area Networks (CANS). At present, these Campus Area Networks form net- "The US. should undertake, as a national worked islands. goal, the establishment of a National Research Net- work in a staged approach that supports the upgrade Congressional Action and Executive of current facilities, and development of needed Branch Response new capabilities. Achievement of this goal would foster and enhance the US. position of world leader- Some in Congress have recognized the tele- ship in computer networking. communications problem and opportunity. In June "As rapidly as feasible, the National Research 1986, Senator Gore introduced the Supercomputer Network should be designed, deployed and maintained Network Study Act of 1986. Public Law 99-383, as an advanced computer network. This network passed 21 June 1986 by the 99th Congress, charged should interconnect substantially every academic, the Office of Science and Technology Policy industrial, and government research establishment (OSTP) with conducting "a study of critical prob- and unique scientific resource to encourage scientific lems and current and future options regarding collaboration unhindered by and to permit communications networks for research computers, the of unquiet research faculties and resour- including supercomputers at universities and federal ces. Since security of the network is a vital concern, research facilities in the United States." appropriate policies should be adopted to protect At OSTP's direction, an inter-agency group the information in the network from threats, vul- under the auspices of the Federal Coordinating nerabilities and risks, and to assure a uniform level Council for Science, Engineering, and Technology of security. (FCCSET) for Computer Research and Applications "Until the National Research Network can --the Computer Network Study Planning Group replace the current system, existing networks should --was formed to carry out the study of the follow- be maintained and modified as they join the national ing issues: network. Since supercomputer systems comprise a special and valuable national research resource the networking needs of the nation's academic with very high performance requirements, the re- and federal research computer programs, in- sponsibility for network access to supercomputers

34 HI TECH should be vested in the supercomputer centers will require an additional funding of ap- themselves until the advanced computer network, proximately $50,000,000 to operate (GB: assumed capable of offering the requisite service level, is the price of T1 line would decline by a factor operational. of 2 over the next five years, whereas, in "Industry should be encouraged through spe- the six months since the report, line charges cial incentives to participate in research, develop- for many T1 lines have already dropped more ment, and deployment of the National Research than a factor of two due to over-supply.) Network. Tariff schedules which have been set "In the third stage, a vigorous focused program for voice transmission should be reexamined in of research and development for the National light of the requirements for transmission of data Research Network should be immediately es- through computer networking. tablished. A total of $400 million is needed "To meet the goal for the National Research over ten years to advance the knowledge base Network and to set an agenda for the future, the and technology of computer network capabilities following actions are recommended: in order to achieve data communications and switching capabilities to support transmission "The Subcommittee on Computer Networking, of three billion bits per second within fifteen Infrastructure and Digital Communications years. This will support a network 100,000 which was established by the Office of Sci- times more capable than currently available ence and Technology Policy on 15 May 1987, and will be essential to foster scientific col- should oversee the first stage in development laboration and sharing of research resources. and operation of the National Research Net- When fully deployed, the cost of operating work--a coordinated internetwork that would this advanced network is estimated to be $400 include the Federal agencies that currently million per year, given the current commercial operate research supporting networking. tariffs for data communications. "The FCCSET Subcommittee on Computer Net- working, Infrastructure and Digital Com- "Support should be given to the development munications should identify a lead agency of standards and their harmonization in the interna- which would be responsible for requesting tional arena. Aggressive action is needed to increase funds for the National Research Network, participation in the standards development and eventually for selecting a contractor to process, to get requirement for standards expressed manage the network. The manager would early in the development process, and to speed be responsible for Stage 2 (45 Mbit/sec links) the implementation of standards in commercial off- and Stage 3 (multi-Gigabit/sec links). the-shelf products. It is essential that standards "As a first stage in the development of the development be carried out within the framework National Research Network, the current Inter- of overall systems requirements to achieve inter- net system developed by DARPA and networks operability, common user interfaces to systems and supported by agencies should be intercon- enhanced security." nected over the next two years. These fa- cilities, if coordinated and centrally managed, Motivation for the National Research Network have the capability to interconnect many computer networks into a single virtual com- If we use the original ARPAnet as a predictor puter network. The Federal government of the future, it is safe to assume that the National should encourage and assist research facilities Research Network is likely to have both more direct and academic institutions to establish local impact on the research and education community, and campus area networks to connect to the and more indirect impact through the construction Internet systems. The estimated cost for of a modern communications network than almost this proposed upgraded service is $5 million any other single program that can be identified. per year, and should be implemented through ARPAnet was initially justified on the basis the shared resources of NSF, DOE, DARPA, of being able to facilities including particular NASA, and HHS. programs, databases, and files, and being able to "In the second stage, new funding for devel- promote the use 'of idle computer capacity. What opment should be requested at $5 million per happened was that a completely different style year over the next five years to upgrade and of interaction developed based on being able to expand the nation's computer networks, which send mail and large documents. Extensive public support research programs, to achieve data forums and conferences evolved through bulletin communications at 1.5 Mbits/sec to 200-400 board and computer conferencing. Remote terminal U.S. research institutions. It is estimated access is negligible, and file interchange was rela- that these expanded and upgraded facilities tively smaller than expected.

ISSUE 21 35 With supercomputing, remote access is criti- Given the clouded record of very large pro- cal. Few institutions can support supercomputers grams (e.g., Space Station, Wars, Superconducting and, given the limited amount of time available Super-Collider, Superconductors, Human Genome on them, it becomes necessary to send large pro- Mapping) in either achieving funding or meaningful grams and databases throughout the network in results, one can hardly be optimistic about the and from smaller computers including . outcome of large scale recommendations encompassing A recent NSF report on pointed out virtually every facet of , technology, the need for research that would enable users to use, and networking. deal with the prodigious amount of data that comes The community of research and educational from modern computers and applications. Also, computing users and network providers must under- we would expect a new form of video and com- stand the ramifications of the National Research puter conferencing to be possible utilizing high Network initiative so that work on it can proceed speed networks. Already researchers are consider- as rapidly as possible, independent of the success ing how libraries might share information. Thus, or failure of a much larger set of goals for higher it is fairly easy to predict that the future use of performance computing. It is imperative that the the research network will be a different, and vast- network be built as soon as possible, and that we ly expanded, version of the past. not lose precious time in its development due to The inter-connection among government lab- inaction or coupling of the project with other objec- oratories, universities, and industrial research or- tives in a much larger program. ganizations provided by the Network should facil- Unlike the super-highway system, any one itate technology transfer unlike almost any other or any combination of existing telecommunications facility. It will enable the instantaneous transfer suppliers could preempt fhe National Research Net- of much of our knowledge, which we are increas- work activity by simply building the network and ingly embedding in databases and programs. Unlike offering the service for sale. A highly aggressive, earlier networks, the National Research Network imaginative industry could view the Network as is aimed at connections with many more institu- the major, large scale, social experiment of this tions. century--an experiment that could strategically In addition to the direct benefit resulting position that industry for dynamic growth in the in the use of the Network, we believe the Network twenty-first century. An effort to achieve Phase is vital for maintaining a competitive telecom- I1 (45 Mbits/sec) in three years would cost a small munications industry. By interacting with the aca- fraction of this industry's current research, de- demic community to build and develop the velopment, and operations budget. Government Network, we expect to create the same vitality should encourage efforts on the part of private and synergy that the computer industry enjoys. industry to build this system, but, based on history, Other countries are actively building high we certainly can't depend on industry. Government speed (fiber optic) computer networks aimed at must take the initiative to oversee development Gigabit data transmission. For example, the Euro- of the Network, an initial step being the development pean programs of ESPRIT and Eureka have signifi- of NSF's . cant research underway at all levels including fiber The National Research Network can and must optics, switching, standards, and applications. proceed using the current inter-agency coordinating committee. Even without the impetus of a major Onward Toward Our National Research Network government initiative and outlay, funds can and must be made available on a shared basis utilizing OSTP asked Congress for a time extension existing agency, industrial, and academic resources until November 1987 in order to incorporate the in a much more propitious fashion. It is highly proposal for a National Research Network into a possible that coordination of the development and much larger report encompassing the support of use of one major network would save money for high performance computing, computer science, all agencies, which would not have to operate dupli- and revamped infrastructure for computer applica- cate networks. Agencies and other affected organiza- tions in science and engineering. This new report tions could build the Network from their own coor- from the FCCSET Committee on Computer Research dinated resources and related savings in telecom- and Applications was based on an integration of munications costs. Hence, it is extremely important reports from its subcommittees: Science and En- that these organizations commit themselves to the gineering Computing; Computer Research and De- coordinated development and use of the Network. velopment; and Computer Networking, Infrastruc- And our federal government should take a leadership ture, and Digital Communications. The resulting step in committing itself to the development of report is entitled "A National R&D Strategy for a Network that will provide positive benefits and High Performance Computing." savings for its constituents (the taxpayers). BD

36 LIBRARY HI TECH