Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.
OCR for page 25
Toward A National Research Network 4 ISSUES, OBSERVATIONS, AND RECOMMENDATIONS In this chapter, the committee summarizes its findings and makes specific recommendations to lead to a successful NRN. Four areas are explored. First, the committee addresses the technical considerations of importance and identifies where the technology is and is not sufficient to meet the three-phase plan presented by OSTP. Second, it focuses on funding as a critical consideration that was improperly addressed by OSTP. Third, it identifies some of the key management issues that must be dealt with at the outset; this was virtually ignored by OSTP. Lastly, the committee considers the role of commercial and specialty networks in the NRN development. TECHNICAL CONSIDERATIONS Phase 1 is happening now; it needs coordination. Phase 2 will present critical challenges in network resource management. Phase 3 requires extensive networking research, and this research must be immediately funded and initiated. Issues and Observations The technology is available for the early phases of the proposed NRN; that is, only commitment and funding prevent achieving the benefits conveyed by greater interconnection and a significant increase in communication speed. With that commitment and funding, achieving those benefits will require some engineering development. Recently announced plans for upgrading the NSFnet are a move in the right direction, and interconnecting existing networks, as anticipated in phases 1 and 2, will itself be a valuable step. The OSTP report emphasizes only one aspect of network technology, namely, speed. The committee believes this emphasis is misleading, since requirements for phase 2 trunk and access transmission rates can be achieved without a special development effort (indeed, industry will probably meet the bandwidth requirements
OCR for page 26
Toward A National Research Network more rapidly than the phase 2 time table) and phase 3 bandwidth is also on the horizon, as noted below. (Note that high bandwidth serves two functions: it accommodates an increase in the number of users (aggregate bandwidth), and it makes possible applications that individually require high-bandwidth and/or rapid response.) Also, techniques for data compression, including video compression, will contain growth of bandwidth demand. The critical and underemphasized need of phases 2 and 3 is the development of a new set of tools for service-oriented management of network resources. These tools will permit control of access, allocation of resources, and support of service; however, they are not now a part of the protocol architecture and must be developed. The time table for achieving phase 2 goals will be set by the time necessary for management software design, implementation, and deployment. This deployment will be very difficult because it will require single-vendor-supported software to be changed to accommodate a multivendor environment. Designing the migration path to this new functionality will be a central part of achieving phase 2. Part of this difficulty comes also from the realization that different parts of the phase 2 internetwork will be managed by different organizations, and these organizations will require tools to enforce policies related to their parts of the internetwork. This multiorganization internetwork exists today, but without effective tools to manage it, and so it functions poorly at present; further growth of the internetwork will only make the problem worse. Since most vendor products do not focus on the development of a large multiorganization network, industry is not likely to provide these management tools on their own. Funding and coordination will be required. One vehicle for managing the phase 2 internetwork is to bring the majority of the resources under the control of one organization. This committee feels that one organization should take the lead in the phase 2 transit service, but a partially centralized administration will not avoid the need for developing much more sophisticated tools for network management, as noted above. Further, moving from today’s environment with fragmented responsibility to a single lead organization will be difficult, requiring decisions and changes regarding control, responsibility, authority, funding, and other factors. The move from phase 2 to phase 3 is a totally different proposition from that of the earlier transitions. Phase 3 presumes a new design based on research. The move from megabit per second networks to gigabit per second networks is revolutionary in a number of dimensions. Without appropriate research, phase 3 will not achieve a fraction of its intended impact. The critical issues involve a complex interaction of the following, all of which are closely coupled: switching technology, processor interfaces, protocols, connection-oriented communications, routing, and layered architectures as well as coexistence with carrier environments. Note that the carriers, unlike the network research community, understand how to handle billing and accountability, and they have developed effective standards for naming and numbering that accommodate hundreds of millions of users. Some of these issues are discussed below.
OCR for page 27
Toward A National Research Network Switching technology, protocols, and connectionless communications. Current connectionless services create significant overhead per packet, and this overhead implies severe limitations on packet rates in the switch. The source of the current connectionless world in which our networks and gateways find themselves was the tradeoffs among bandwidth, storage, and switch complexity. However, these tradeoffs are changing dramatically, and certainly there will be a significant change by the time the phase 3 networks arrive. The information available with connection-oriented communication can be extremely valuable in simplifying the processing requirements of the switch; this is especially important since the switch is likely to become the bottleneck in phase 3. Indeed, the tradeoff of current networks (using processing in the switch to save on communication bandwidth) is rapidly shifting to a new paradigm (using communications bandwidth to reduce the cost of the switch); this has implications for hybrid forms of packet and circuit switching. Reconsideration of connections, along with the development of faster switching processors, is critical since the OSTP goals in phase 3 are at the edge of what specialized technology might permit in the context of existing architectures. The creation of protocol suites to allow for robust network operation and management at the higher signaling and packet rates is also essential. Consider the following: (1) industry has been developing packet switches at T1 rates; however, even these rates are hopelessly inadequate for the gigabit per second speeds of phase 3; (2) it is altogether likely that government support for DS-3 links will be provided in the SDI research network development; and (3) photonic switching is another promising area for revolutionary switch improvements. These and other developments emphasize the need for an early start on the research needed to usher in phase 3. Processor interfaces. Under the early assumptions of fast, large, efficient disk systems, there arose a need for channels and I/O controllers to control these devices. When local area networks were developed, they were interfaced to computers just as were the disks, namely, as I/O devices. This led to long software paths between the communicating programs and the actual transmission of data out of the processor. The need for efficient and reliable transmission over poor long-haul lines led to many of the connectionless protocols in use today (along with their burden of overhead). We are now anticipating transmission speeds in excess of the data rates on the processor buses connected to the network and are anticipating error rates as low as those seen on processor buses. Thus it is interesting to view this networking environment as similar to a multiprocessor system (with a very long bus). However, unlike a multiprocessor bus, network protocols based on memory bus techniques would suffer from propagation delays associated with the speed of light. A number of trends that affect the directions of network technology development can be predicted. For example, language and systems support at the workstation nodes (with a reduced dependency on system effort at the supercomputer sites) is a likely development; and this development is badly in
OCR for page 28
Toward A National Research Network need of some standardization. There will be a definite need for the development of a standard graphics language for transmission of programs to display data. It is important for nodes to be able to deposit code and data at the supercomputers and then to be able to control job processing without heavy data transfer across the network. The use of supercomputers for processing of semiintelligent data bases is a possible scenario, and this implies the development of very large data stores. There will be an emerging role for expert systems in network management. Recommendations Phase 1 is already happening. It is crucial that the proper management functions be put in place rapidly to ensure that the steps being taken fit into the overall NRN goals, particularly with regard to its utility. Network management tools are under aggressive development by industry, and the NRN should rapidly embrace the latest of all these tools in its development. Management tools must be addressed as part of the detailed planning required for the NRN. The 1-Mbit/s service (phase 2) must be done well and administered well; it can be achieved and will set the tone for any future acceptance of the NRN. The research necessary to achieve phase 3 goals should be launched as soon as proper planning is completed. The phase 3 research program should be coordinated between the government and industry, so that the government can benefit from the vast experience gained in industrial research laboratories (for example, laser development, photonics, multiprocessors, VLSI tools, semiconductor technologies, switching technologies, network management, and so on) and in network service markets (for billing, accountability, and addressing). The basic architecture and technology issues should be settled by the network researchers in cooperation with the potential user community, not the forces of the free market. FUNDING This committee believes that unless a stable source of funding is provided, the NRN will never achieve the impact of which it is capable. Issues and Observations The dollar figures suggested by OSTP were assumed by this committee to be rough estimates as opposed to firm calculations. Clearly, a complete analysis is required to properly calculate NRN costs; that analysis must distinguish dollar allocations for network research from dollar allocations for NRN deployment, operation, administration, and maintenance. The dollar order of magnitude suggested (see Figure 2) is not unreasonable, given that the annual U.S.
OCR for page 29
Toward A National Research Network value-added network market is about $0.5 billion, the U.S. local area network market is roughly that or larger, and the total data communications market is roughly $3 to $4 billion. An NRN spanning the country and serving perhaps one-half million researchers with suitable capabilities would likely have an annual operating budget (including depreciation) of at least $250 million and possibly more than $1 billion (especially if the number of users grows to include students as well as faculty). An important issue that historically has been avoided in the world of research networking (and was not addressed by OSTP) is how to make sure that research network resources-which, however expanded with the NRN, will nevertheless be finite—are used efficiently and with maximum benefit to the research community and the nation. Most communities of network users (e.g., business users) pay for network services. Payment allows network service providers to recover their costs (and expand service offerings). It also constrains the level and nature of network service used; as in other settings, user charges serve to ration limited resources. Thus, relatively high rates for voice communications presumably limit the number of long-distance calls among researchers, as among other groups of users. To most researchers, access to research networks, such as Arpanet, has been free. One consequence of free access, when combined with inattention to network load management, is severe overloading of the network; others are limited user support (in terms of training, documentation, and troubleshooting as well as access) and limited growth in network and support capability. Another way to look at the problems created by the absence of market mechanisms is to consider the choice between government-funded and commercial network services. Historically, when an agency of the government has established its own research network, it has usually absorbed all costs for users. Years after the research network has become largely operational, there may be no incentive for industry to offer such services on a commercial basis to those users. Even if a comparable commercial service develops, there is no incentive for users to switch from a research net to a commercial one as long as the former is provided “free” (i.e., at government expense) when the other is not. However, should the free service disappear, the user may not be able to switch to an equivalent commercial service without a supportive research funding structure. While the high bandwidth proposed for the NRN would support an increase in the user population, user charges or some other means of allocating bandwidth would be necessary to avoid either the recurrence of overloading or the crowding out of applications that cannot exist without high bandwidth (for example, real-time dynamic displays and timely transfer of large data files) by others that are less sensitive (for example, routine e-mail or transmission of files that are unnecessarily large). Peer review is useful for setting priorities for the funding of research, whereas pricing is most efficient for guiding choices of researchers from among suppliers and offerings of network services. If, as discussed below, user charges are eschewed, then the problems here envisioned can only be overcome by technical solutions that may themselves be
OCR for page 30
Toward A National Research Network expensive (transferring the costs to the research stage, perhaps) or by allocation schemes that may be both politically and technically difficult. Many of the organizations whose resources will be sought by researchers through the NRN may have to restructure their services charges to meet expanded demand associated with NRN connections. However appealing user charges may be to those who design, build, and manage networks, they cannot be imposed unilaterally on the research community unless the research funding infrastructure adapts to accommodate them. Note that the research funding infrastructure already accommodates voice communications charges; telephone charges are typically covered in research grants. Similarly, research funding accommodates charges for computing resources; obviously, such charges are a relatively recent phenomenon, and they suggest that there should be some flexibility within the research funding environment. Nevertheless, current modes of networking support for universities and the research community in general are inadequate to meet the research community’s present needs (let alone its future requirements). While implementing an economically sound NRN poses challenges, the committee concludes that it is possible to institute market mechanisms that are applicable to the short-term and long-term needs of the science community. It cautions that both structuring of incentives for efficient use and provision for fairness of access are important considerations here. In developing mechanisms to fund an NRN, a central concern is to lessen the (perceived) tradeoff between dollars for research and dollars for the NRN. The greater that tradeoff appears to be, the greater will be the resistance that is likely to emerge from the research community, especially from those segments new to research networking, and the more difficult it will be to achieve a truly national research network. Recommendations National research network technologies should not be developed from scratch where the technology base exists within the private sector (as the committee has found that it does for at least the early phases of the NRN). Economic incentives should be provided to encourage industry to undertake the necessary development to bring this base to the NRN. This may reduce the funding required. Government should fund the long-range research implied by the phase 3 proposals and should consider applying the fruits of that research first in a (government-funded) seed network for testing by a subset of the research community. Once a seed network has reached a certain stage of development, it would acquire a quasi-operational status as a supplier of advanced network services. The system may be operated as a maturing research network while the nature of its capabilities and the market for them become understood by
OCR for page 31
Toward A National Research Network commercial suppliers. If funding mechanisms are revised, industry will be able to recognize the direct economic benefit of offering such services commercially (even if only for the research community), and the services of the maturing research network would then be expected to be incorporated into commercial networks. Whether the seed network approach or another approach is pursued, NRN planning should be specific as to how and when coordination with commercial network service providers, including carriers, is achieved. Will the providers be involved from the outset, or will a seed network be developed by the data processing/data communications research community outside of the carrier environment? Answers to such questions will influence the level of effort, funding, and timing of research as well as the ease with which the NRN may be interconnected with other, commercial network services. In addressing such questions, the government should be mindful of the fact that the carrriers and other providers are expert at moving bits and managing that movement. By contrast, the data processing/data communications research community has relative strength in the area of interfaces with computer systems to support high-level applications, including associated protocol aspects. Funding mechanisms should be structured to allow researchers to exercise choice when offered substantially similar services. On the other hand, it should be within the province of the management function (see below) to encourage use of (and provide incentives for) one network service offering when available; this will allow the NRN to take advantage of economies of scale. Once the NRN is running, it should be managed by an appropriate entity that understands how to handle the substantial problems of facilities management and customer service that are implied by a network with an annual cash flow in the $250 million to $1 billion range serving a very large number of users (500,000 and up). Funds for network research should be kept separate from funds for network operation, administration, and maintenance, including associated management. Operational funding should be decentralized as much as possible to make users accountable for their utilization of the NRN and to overcome the difficulties inherent in the federal budget process of amassing the necessary funds in the budget of a single agency or group of agencies. This will require agencies that provide research grants to provide for network charges in the grant process and to allow network charges to be added into the indirect costs charged by research organizations. Any such funds for network use should be earmarked. A plan is needed for the transition to user charges. One example of a mechanism for easing this transition is to provide the equivalent of vouchers for network services. Vouchers could be provided with research grants for network use (like stamps for postal service), helping users to see the link between their usage and the cost of providing that usage. This would also be a means for implementing the ability for researchers to exercise choice (see the third recommendation above).
OCR for page 32
Toward A National Research Network Billing of users should be done in real dollars and should be separable from research funding. People may get bills for $0, and billing rates may vary, but users should see bills. (Billing would also provide a valuable data collection and evaluation mechanism.) The carriers and other network-service providers have a lot of experience with and expertise in billing. Additional development work would be needed to support billing if the NRN were to be built separately from the carriers, without benefit from their expertise. A basic (minimal level) universal service should be provided to all users at a low charge. Such a service might, for example, be low- to moderate-speed access lines for e-mail service (in pure text mode). MANAGEMENT Competent management is absolutely essential for maximizing the effectiveness of the investment in an NRN. It is a critical issue that requires early (and continuing) attention and action, and it is a nontrivial element of the total cost of providing network service. The overall problem is to engineer facilities and services to match demand, and tune them to the needs of users as expressed by users. Issues and Observations The private sector offers experience with and expertise in planning, building, and operating large, reliable data communications networks serving millions of users. Planning elements include the following: strategic planning business planning requirements analysis preliminary design RFP development proposal evaluation final design implementation acceptance and testing system augmentation operations, administration, and maintenance: —planning and design —traffic and circuit analysis —work order generation —configuration control —moves, adds, and changes —user support —billing and administration —fault detection —failure recovery
OCR for page 33
Toward A National Research Network Other considerations for planning should include establishing and enforcing connectivity standards in areas such as: LAN gateways graphics interfaces database servers format and content of compound electronic documents All of these elements must be used in the NRN development. While the OSTP proposals focus on national connectivity, planning is also required for lower-level distribution of network access. Local connectivity and switching affect the quality of service at both institutional and individual levels. The management problem has many dimensions. Most management issues fall under the headings of network facilities management, operational service management, and process management. Issues within these three categories are as follows: network facilities management: —Monitoring the performance of links and network equipment is a critical management function. Maintenance and repair follow from this function. —Measurement of activity on the network, quantitatively and qualitatively, is necessary to support management. A reporting mechanism to disseminate, and provide access to, this measurement database is also necessary. —A rational mechanism is necessary for network installation, improvement and augmentation as load grows and technology invites. operational service management: —Fair and effective utilization of network resources is essential. This implies the creation of means for allocating network resources among competing uses and users. It also implies means for users to express concerns and for management to address and resolve those concerns. —Privacy and security are issues that are especially important to consider early on. Privacy considerations include confidentiality of data and messages. Security pertains not only to protecting classified information, but also to the more basic issues of protecting data and system integrity and access control. The supercomputer centers, which are accessed by users connected to Ethernet systems (over which passwords are broadcast in the clear), are particularly vulnerable. This suggests that privacy and security in data communications have been underappreciated and underprotected to date. Propagation of failures and viruses in the network (especially one as large as suggested here) is a growing risk. These issues gain significantly in importance as the number of users and, in particular, the number of international
OCR for page 34
Toward A National Research Network connections increase. Solutions may reside at the computer rather than in the network, but solutions must be developed and utilized. process management: —The development of the NRN, as OSTP noted, would involve multiple stages. At one or more points in time, mechanisms would be required to assist users in transferring from older or separate network systems and services. Mechanisms are also necessary to facilitate coordination between the NRN and other nets used by researchers, including regional, campus, and even local nets. —Developing and testing the NRN on the one hand, and operating it on the other hand, will probably require different management structures. —It is important to recognize that management of technical and user problems may be confounded by the political aspects of network access and connectivity. Certainly the structure of the current research networks reflects past political as well as technical considerations. Among the issues that may be counted as political are decisions about how to increase access to smaller research entities (in terms of geographic deployment and in terms of charging schemes—see above) and decisions about where to bound central, federal control (down to the region level, building on current federal initiatives, or down to local spurs). The structures and mechanisms for addressing these many issues are not necessarily the same. Recommendations A working group should be established immediately to develop the plans and procedures for a management strategy. Funding plans should explicitly provide funding for management tools and structures at the three levels identified above. Mechanisms should be established to obtain user input and feedback at both planning and operational stages to assure network reliability and predictability. Mechanisms should be established to inform researchers about the network, NRN services, and any problems that arise, as well as to provide operational support. This is a critical function. Privacy, security, and equitable access should be priority issues in planning (as well as research). The process of gathering appropriate data should begin immediately, and other concerned organizations should be invited to assist in creating a list of issues and vulnerabilities.
OCR for page 35
Toward A National Research Network Private sector expertise and resources should be tapped at every stage, from planning through operation. The long-term management structure should consist of (a) a means for allocation of funding via normal government channels, (b) an oversight committee from a variety of disciplines to coordinate and oversee the government’s investment, and (c) an experienced network executive in the private sector to handle day-to-day management of the operational aspects of the NRN (in conjunction with individual network managers) and to provide continuity of technical guidance. The network executive would organize or coordinate subordinate activities to consider or facilitate (a) network research or pilot projects that may affect the NRN, (b) relevant architecture and standards, and (c) liaison with industrial activities. A major goal of these activities would be to assist in the transfer of relevant information. One or more user groups should be established to identify the needs of the users and to ensure their representation in the other parts of the management structure. The user perspective should be represented in each of the above structures in the management organization. COMMERCIAL AND SPECIALTY NETWORKS The success and continuing existence of the NRN depend upon the existing carriers. Unless we work with them in developing this service, we may find that the NRN eventually diverges from the mainstream, thereby encouraging the existing carriers to develop networks that are incapable of supporting the research and development community. This possibility could be catastrophic for U.S. research and development. Development of the NRN must proceed in conjunction with the development of common carrier services as they evolve to meet the broader needs of education, industry, and government. Issues and Observations Researchers currently use a variety of network services to meet their various needs. It is not likely, at least in the near term, that researchers with special needs, such as geophysicists, will abandon specialized networks, nor is it clear that they should. Research networks at present build on layers of local networks, campus networks, regional networks, and national networks. In many cases, each layer is itself underfunded. Moreover, growing investment at the local, campus, regional, and national network layer heightens concern about the cost of research networking and its incidence. That concern will grow if, as some predict, the baby bust leaves universities with fewer students and overcapitalized. Note that campus networking costs are typically borne by administrations, encouraging researchers to view such networks as a right akin to library access. If paying for relatively local networks poses problems, paying for an NRN may be an even larger problem.
OCR for page 36
Toward A National Research Network Overall, data communications networking in this country builds on a fragmented base of common carriers and enhanced or value-added service providers. They may not provide in the public/commercial network system all of the capabilities sought for the NRN in the proposed timeframe, but they do offer tremendously relevant expertise and facilities. It is essential that the government leverage its investment in the NRN by structuring its development in a way that takes advantage of products and services being developed and offered by the carriers. The nation would thereby benefit significantly from synergy between commercial offerings and government-sponsored network services. In terms of carrier offerings, the national communications infrastructure is rapidly becoming digital. Already over 1.3 million miles of glass fiber are installed, most of which operates at 400 Mbits/s. Many of these facilities are being upgraded to 1.7 Gbits/s merely by replacing electronics at the fiber ends, and further upgrades to higher speeds on the backbone or trunk lines are anticipated. Meanwhile, progress is occurring in the integration of voice and data, often discussed in the context of so-called integrated services digital network (ISDN) services, although ISDN implementation plans will not support NRN goals in the near future. ISDN promises broad access to 64-kbits/s digital service, among other things, which will eventually enhance network communications among researchers working from home and others. Broadband ISDN (B-ISDN) using the 12-cell Synchronous Optical Network (SONET) promises 600 Mbits/s. There are some important limitations on the contributions of the commercial networks, particularly those operated by common carriers, arising from the divestiture of AT&T and a variety of regulations that affect pricing and types of services available from both local exchange and interexchange carriers. Many restrictions preventing carriers from providing information services have been relaxed recently, and, if past experience holds, the bulk of these restrictions may ultimately disappear in the 1990s. Such restrictions do not, however, prevent the carriers from playing an important role in developing and implementing an NRN. Recommendations Design and implementation of the NRN should involve cooperation with the telecommunications and computer industries in the development of standards and services. The NRN should be developed and managed in such a way as to allow the public/commercial network to contribute to the maximum extent to meeting NRN goals over time. Researchers should be encouraged to use commercial services where suitable, if this can be accomplished without compromising the NRN. The way to do this probably involves maximizing interconnections between the NRN and commercial services. While the intent of the NRN is to diminish the proliferation of networks (to reap economies of scale and maximize communication possibilities), planners and managers of the NRN should take into account specialized or competing
OCR for page 37
Toward A National Research Network services currently or potentially used by researchers and act to minimize redundancy and maximize interconnectivity. Detailed plans for the NRN should evaluate the many layers of networks that exist and that might contribute to an NRN and consider whether the NRN should be built as a single, integrated physical entity or whether it should be cast as a family of services that builds on a loose confederation of (physical) networks that may share bandwidth and management services at some level and that operates with a single addressing scheme. The design of the NRN should accommodate a range of services for users and a range of delivery modes. At one extreme, a federally sponsored service should provide researchers with the maximum speeds available. At the other extreme, researchers with less demanding requirements should have access to a basic, universal service (as, for example, e-mail) analogous to basic telephone service. Such a basic service would require a fee to prevent overloading and the other problems that have plagued such networks as Arpanet. Differential pricing would be a key means for sustaining a mix of services, channeling premium services to premium applications.
OCR for page 38
Toward A National Research Network This page in the original is blank.
Representative terms from entire chapter: