Appendix A: OSTP Proposals for a National Research Network
NETWORKING
-
FINDING: The U.S. faces serious challenges in networking technology which could become a barrier to the advance and use of computing technology in science and engineering.
-
Current network technology does not adequately support scientific collaboration or access to unique scientific resources. U.S. commercial and government sponsored networks presently are not coordinated, do not have sufficient capacity, do not interoperate effectively, and do not ensure privacy.
-
Europe and Japan are aggressively moving ahead of the U.S. in a variety of networking areas with the support of concentrated government and industry research and implementation programs.
-
Computer network technology provides the means to develop large scale distributed approaches to the collaborative solution of computational problems in science, engineering, and other applications areas. Today, researchers sharing a local area network are able to exploit nearly instantaneous communication and sharing of data, creating an effect of linking their workstations and high performance servers into a single large scale heterogeneous computing facility. This kind of capability is now appearing in larger scale campus-wide computer networks, enabling new forms of collaboration. National networks, on the other hand, have low capacity, are overloaded, and fail to interoperate successfully. These have been expanded to increase the number of users and connections but the performance of the underlying network technology has not kept pace with the increased demands. Therefore, the networks which in the 1970s had significant impact in enabling collaboration, are now barriers. Only the simplest capabilities, such as electronic mail and small file transfers, are now usable. Capacity, for example, is orders of magnitude less than the rates required, even if the network is used only for graphics.
Other countries have recognized the value of national computing networks, and, following the early U.S. lead, have developed and installed national networks using current technology. As a result, these countries are now much better prepared to exploit the new opportunities provided by distributed collaborative computing than the U.S. is at the present time. The basic technologies for later generations are also being developed in the U.S., but there have been no major efforts to apply them to address the needs.
Applications include (1) distributed access to very large databases of scientific, engineering, and other data, (2) high bandwidth access to and linking among shared computational resources, (3) high bandwidth access to shared data generation resources, (4) high bandwidth access to shared data analysis resources, such as workstations supporting advanced visualization techniques.
A longer term goal is the creation of large scale geographically distributed heterogeneous systems that link multiple superworkstations and high performance supercomputers to provide service to scientists and engineers distributed across the country. A well-coordinated national network could link these resources together when required on an ad hoc basis to provide rapid response to computational needs as they arise. This could reduce the number of sites needed for the physical presence of supercomputers. Present access to computer networks by researchers is dependent upon individual funding or location. There is unnecessary duplication in the links from various agencies to each campus. The development of improved networking facilities could greatly stimulate U.S. research and provide equitable access to resources.
Many scientific research facilities in the U.S. consist of a single, large, and costly installation such as a synchrotron light source, a supercomputer, a wind tunnel, a particle accelerator, or a unique database. These facilities provide the experimental apparatus for groups of scientific collaborators located throughout the country. Wide area networks are the logical mechanism for making data from such facilities more easily accessible nationwide. An important issue is that of computer and network security to ensure privacy and trustworthiness in a heterogeneous network environment. At present, responsibility for privacy and the assurance of trust are vested principally in the computers and switching nodes on the network.
Existing government-supported wide-area networks include ARPANET, HEPNET, MFENET, NSFNET, NASNET, MILNET, and SPAN, as well as private and commercial facilities such as TYMNET, TELENET, BITNET, and lines leased from the communication carriers. Longer-range estimates vary, but it is expected that by the year 1995 the nation’s research community will be able to make effective use of a high capacity national network with capacity measured in billions of bits per second. Without improved networks, speed of data transmission will be a limiting factor in the ability of researchers to carry out complex analyses. The digital circuits most widely available today with transmission speeds of 56 kilobits per second (kb/s) are impediments to leading edge research and to optimal remote high performance computer use.
Point-to-point connections require interconnects through multiple vendors with cumulative costs. Greater network speed can reduce the time required to perform a given experiment and increase both the volume of data and the amount of detail that can be seen by researchers. Scientists accessing supercomputers would benefit because access speed is often critical in their work. Improved functionality frees scientists to concentrate directly on their experimental results rather than on operational details of the network. Increased network size extends these opportunities to thousands of individuals at smaller academic institutions throughout the nation. These modernization measures would significantly enhance the nation’s position in scientific research. A national network would help maintain the U.S. leadership position in computer architectures,
microprocessors, data management, software engineering, and innovative networking facilities, and promote the development of international networking standards based on U.S. technology.
Integrated Systems Digital Networks (ISDN-voice and data) have been installed abroad on a national or regional scale. Research abroad is being conducted on service up to 1 Gb/s. Within the next five years, Integrated Services Digital Network (ISDN) circuits ranging from 64 kb/s to 1.5 Mb/s will be available in the larger metropolitan areas of the U.S. However, these services will fall short of the requirements for computer networks. By 1988 more than fifty Campus Area Networks will be operational at speeds approaching 100 Mb/s. Wide area networks operating at 1.5 Mb/s or less will not be able to handle the data volume expected.
Japan and Europe have extensive efforts with experimental nets in intermediate (40Mb) and high (gigabit) range. Japan is studying operational aspects of fiber nets using their national research network as a testbed, which includes exploring the feasibility of fiber optic services to residences.
To estimate the network bandwidth needed to support research at a major installation, the kinds and volume of traffic that would be used have been estimated at a representative campus, extrapolated ten years into the future. Three models were used to compute three independent estimates of the requirements for bandwidth needed by type of work, information needs by type of user, and information flowing at the installation boundary. In each model, the peak bandwidth was estimated for each type of service. For example, in the Task model, the need is dominated by that of at least one researcher to receive full color and full-motion high resolution images. A high-resolution color image contains about 30 megabits of information, so that a display rate of 30 frames per second requires a bandwidth of nearly one gigabit per second (Gb/s). In the User model, a research university with 35,000 students and 3,000 faculty and research staff using a mix of bandwidths again requires an aggregate bandwidth of approximately one Gb/s. In the Edge of the Installation model, bandwidth is estimated by the types of remote facilities being accessed and the expected number of simultaneous users; typical facilities include particle accelerators, supercomputers, and centers for imaging and/or animation. The aggregate bandwidth needed is one Gb/s. Thus three independent means of estimating bandwidth arrive at nearly the same requirement for a large research installation, and one Gb/s can confidently be used as a lower bound on the bandwidth of a national research network.
-
RECOMMENDATION: U.S. government, industry, and universities should coordinate research and development for a research network to provide a distributed computing capability that links the government, industry, and higher education communities.
A research network should be established in a staged approach that supports the upgrade of current facilities and development of needed new capabilities. Achievement of this goal would foster and enhance the U.S. position of world leadership in computer networking as well as provide infrastructure for collaborative research. The FCCSET Committee on Computer Research and Applications should provide a forum for interagency cooperations. Elements of the plan should include:
-
Stage 1. Upgrade existing facilities in support of a transition plan to the new network through a cooperative effort among major government users. The current interagency collaboration in expanding the Internet system originated by DARPA should be accelerated so that the networks supported by the agencies are interconnected over the next two years.
-
Stage 2. The nation’s existing networks that support scientific research should be upgraded and expanded to achieve data communications at 1.5 Mb/sec for 200 to 300 U.S. research institutions.
-
Stage 3. Develop a system architecture for a national research network to support distributed collaborative computation through a strong program of research and development. A long-term program is needed to advance the technology of computer networking in order to achieve data communication and switching capabilities to support transmission of three billion bits per second (3 Gb/s) with deployment within fifteen years.
-
Develop policy for long term support and upgrading of current high performance facilities, including timetables for backbone and connection development, industry participation, access, agency funding, tariff schedules, network management and administration. Support should be given to the development of standards and their harmonization in the international arena.
-
Until the national research network can replace the current system, existing networks should be maintained and modified as they join the national network. Remedial action should be initiated as soon as possible. Upgrading the backbone to at least 1.5 Mb/s should be accomplished by 1990. This will ensure that the new generation of high performance computing can be effectively interconnected.
Industry should be encouraged to participate in research, development, and deployment of the national research network. Telecommunication tariff schedules
which have been set for voice transmission should be reviewed in light of the requirements for transmission of data through computer networking.
Prompt effective coordination is needed to increase user participation in the standards development process, to get requirements for standards expressed early in the development process, and to speed the implementation of standards in commercial off-the-shelf products. It is essential that standards development be carried out within the framework of overall systems requirements to achieve interoperability, common user interfaces to systems, and enhanced security.
