C.1 Network Components and Protocols

Principal Authors:

David Clark and Mark A. Wegleitner

Additional Contributors:

Charles Brackett, Vinton Cerf, David Charlton, Thomas Cheatham, Alexander G. Fraser, Barry K. Gilbert, Paul Green, Gary Herman, Rick Hronicek, H. T. Kung, Lawrence Landweber, Debasis Mitra, Han Quang Nguyen, Raymond Pickholtz, M. Niel Ransom, Raphael Rom, S. Howard Shimokura, W. David Sincoskie, John W. Sweitzer, Satish Tripathi, Jonathan S. Turner, Mario P. Vecchi and K.C. Wang


1. Background and Motivation

The National Information Infrastructure will provide applications and services via an interconnected set of networks and systems. The communications infrastructure will be derived from networks that were preceded by telephony, television and computer communications. Past technological limitations and economic trade-offs have led to today's separate networks, each with a unique architecture optimized to provide its intended applications and services. A challenge that lies ahead is to create a single infrastructure that embodies the attributes that have made these separate networks successful in their original contexts, and yet brings voice, video and text to a single focus that can be exploited by consumers in any combination.

The accumulation of advances in many key enabling technologies has made it possible to conceive the implementation of an advanced NII today. Among these are advances in materials that underlie electronic components and optical/opto-electronic components, including optical fibers; advances in electronic integrated circuits, including both the speed at which these circuits can perform their functions, as well as the achievable complexity that allows a single chip to perform complex tasks; advances in signal processing techniques that use electronic circuits and software to convert information and information-carrying signals into forms suitable for transport over distance and for storage; and advances in systems that process, store and route information through networks. Even advances in technologies that many engineers and scientists take for granted, like batteries, have played a key enabling role. While advances in these enabling technologies have made it possible to conceive the implementation of the NII, significant further progress is required before the vision of affordable multimedia communication and information services for all citizens can be realized. In the following Sections 2.1 through 2.3, we will identify areas where further research in foundational, transmission and switching technologies is most needed.

A goal of the research on internetworking for the NII should be to leverage the knowledge and experience from existing telephone, cable television and computer communications networks to identify a key set of interfaces and protocols that will allow seamless and transparent uses of a heterogeneous NII by a diversity of applications and users. For example, as one reviews the success of the Internet, one can identify certain features of the existing research Internet architecture that have contributed materially to its success to date. In particular, one of the noticeable characteristics of the Internet design is its internetworking approach, which is based on an hourglass-shaped communication protocol hierarchy.

A wide variety of application and transport layer protocols rest atop the common Internet Protocol (IP) that forms the "waist" of the hourglass. Similarly, below this IP layer a wide variety of physical layer transmission and switching services has been incorporated into the global Internet over its 20-year history. This successful Internet feature should be taken into account in defining the internetworking architecture of the emerging NII. On the other hand, the wide diversity of applications and services envisioned in the NII, which go well beyond those of the current Internet, reveal areas where today's internetworking approaches are deficient and where there is basic uncertainty as to how to proceed.

Accepting the hourglass model as the organizing principle of the NII internetworking architecture, we state the following as a grand challenge: to maintain the hourglass architectural model demonstrated in the Internet, and at the same time embrace the full range of applications as well as the full range of communication networks and services envisaged for the NII. In Section 2.4, we offer further observations about those aspects of the Internet (and other network architectures) that have lent it longevity, as well as speculation on areas of deficiency that need repair. It is in these areas that new research initiatives in internetworking may bear fruit.

2. Research and Development Recommendations

2.1 Foundations

Materials

Everything from the plastics used to build enclosures for telephones and computers, to the silicon wafers that are used to produce as many electronic circuit chips as grains of sand on a beach, to the ultra-transparent glass contained in optical fibers depends upon modern materials and materials science. As will be discussed further below, there is a need for faster electronic circuit chips, new kinds of optical components to support multiwavelength optical networking, and components that can perform combinations of optical and electronic functions. While some advances in these areas can be obtained using existing materials, breakthroughs are needed in creating new materials and materials systems (combinations of compatible materials) that have higher electron transport speeds, higher optical non-linearities and higher electro-optic coefficients; that can be used to amplify light at wavelengths where current material systems do not provide such amplification; and that are practical to use in long-lived, low-cost applications employing real environments.

High-Speed, Low-Cost, Low-Power Electronic Circuits

The digital electronics used in the highest-speed serial fiber links will need to be capable of multiplexing (combining) and demultiplexing (separating) digital substreams, including clock bits and/or framing bits at 10 gigabits per second or more. Candidate device technologies to support such high bit rates include gallium arsenide and perhaps eventually indium phosphide heterojunction bipolar transistors (HBTs). Additional research on these materials and device technologies--and perhaps others that emerge as candidates to achieve higher production yields, higher achievable complexity in integrated circuits (in the range of 1,000-5,000 gates), lower power requirements and decreased production costs--is needed.

All-Optical Networking

Even with advances in electronic technology, the achievable speed at which electronic components can perform their functions remains a bottleneck, which can be alleviated by all-optical networking techniques in the highest-speed portions of networks. All-optical networking, which usually includes the use of multiple wavelengths (wavelength division multiplexing), can reduce costs in networks, increase the flexibility with which networks can be reconfigured and can provide transport that is independent of (and thus transparent to) the formats with which information is packaged into the optical signals. The payoff of this research will be fewer complex electronic circuits/processors, higher-reliability networks and systems, and lower-cost networks and systems.

Powering Technologies

Telephone subscribers expect their telephones to work during emergencies, which include commercial power outages. Classical telephones continue to operate during commercial power outages because a moderate level of powering is provided over the copper wire pairs that provide telephone service.

However, some kinds of modern electronic telephones that plug into outlets in residences and businesses will not continue to function during a commercial power outage because they do not include battery back-up subsystems. Furthermore, various types of multiplexing systems that may be employed on customers' premises in emerging multimedia applications may not provide even basic telephone services during commercial power outages unless they are designed properly. Fibers cannot carry power into homes and businesses except perhaps in very small amounts.

Alternative solutions are possible, including co-installed power buses to carry power into homes and residences from telecommunications networks, local commercial powering with battery back-up and lower-priced interruptible services. With presently available technologies, these solutions do not meet the cost and performance objectives implicit within the vision of the emerging NII. In addition, there is a significant need to improve power storage and use parameters for wireless (mobile) terminal equipment. Powering considerations severely limit today's wireless appliances in communications range, available bandwidth and usage time between battery charges.

Research and corresponding advances are needed in battery technologies (to increase energy stored per unit weight and per unit volume, to increase recharging speeds and the number of recharging cycles before the end of useful battery life, etc.) and in alternative approaches to meeting powering challenges.

Coding and Modulation Techniques

The degree of effectiveness of the digital technologies deployed in the NII will be in large part the result of recent advances in coding and modulation, and those still to come. The efficiencies resulting from advances in video and image compression coding in converting such information types into digital form have made it possible to transport video and images over physical media and networks that have limited bandwidth capabilities (e.g., wire pairs and wireless networks), and have increased the capabilities of other media and networks (e.g., cable and satellite systems) to deliver more choices and on-demand access to information. Information storage has also been greatly facilitated by these compression techniques. In addition, advances in modulation and related signal processing technology has increased the information-carrying capacity of existing telephone, cable, satellite, wireless and terrestrial broadcast networks. As a result, the cost of implementing the NII and the timetable for its implementation have been greatly improved.

Further advances in compression technologies may result from such approaches as "wavelets" and "model-based coding." Transport and switching breakthroughs in the future may very well be based, or at least depend, on advances in these areas.

Survivability and Reliability

New algorithms and approaches to survivability are required in the gigabit-per-second NII of the future. The speed of transport, coupled with complex connection and session establishment and management, will present problems not seen in current networks. "Total network" approaches will be required. Perhaps most importantly, humans will require significant assistance (e.g., artificial intelligence, learning systems, etc.) just to manage the NII at expected quality-of-service levels.

Installation of Friendly Terrestrial Transmission Technologies

To make the most immediate contribution to the national economy, the NII should be deployed as rapidly as possible. While extensive use of the existing technology can assist in this objective, a critical constraint in reaching the full benefits of the NII over broadband systems is the time and cost associated with installing the fiber-optic cables needed to provide the broadband infrastructure to the neighborhood, curb, home, etc. Improved splicing methods and technology, lower-cost placing techniques, low-cost acceptance and repair testing methods, and improved training could significantly lower the cost of installation and more importantly, relieve the perceived human resource constraint impacting fiber infrastructure deployment.

Simulation, Design and Test Tools

Neither the electromagnetic (EM) simulation nor the design and test tools are in hand at present to provide reliable support to the designers of the electrical portions of the highest-speed fiber links. Presently available EM modeling tools for digital systems have upper frequency limits at approximately 5 GHz. Although initial theoretical work has been conducted to simulate digital circuits operating above 5 GHz, there are essentially no true CAD tools that implement even these first-pass algorithms. The costs to extend the additional EM tools will not be large when measured against the costs of hardware development, but a considerable elapsed time will be required to develop and prove out these tools. A much more dedicated effort will be required in this support discipline than has been undertaken to date.

2.2 Transmission Research

End-User Premises Network (megabits per second, plus)

The successful deployment of the NII will only be realized if network access from the home and office is mass-market affordable. Network access must be scalable and support several media, including coaxial cable, fiber, copper and wireless, using both the switched and shared-media approaches. Research should focus on the most cost-effective client access mechanisms (ATM, cordless, legacy LAN, etc.). Premise access must support the multiplexing of client video, voice and data sources requiring varied quality-of service-levels (QOS) and various bandwidths.

Access Network (gigabits per second, plus)

Near-term solutions for residential access must be developed. Today, a twisted-pair copper network exists into literally 100 percent of U.S. homes. Asymmetric Digital Subscriber Line (ADSL) technology has been identified as a method of achieving 6 megabits per second (one direction) over existing twisted pair to the home. Research on driving down the cost of components and the introduction of ADSL "standard" chips into home units can dramatically speed the deployment of 6 megabits per second into every home.

The future (more than five years away) requirement for the NII is to provide interactive broadband access to the home capable of providing two-way, multimedia service from any point to any point, on an affordable basis. The main barrier to deploying broadband to the home is cost. Therefore, research efforts must be focused on technologies leading to lower costs. Examples of specific enabling subgoals are a fiber trunking architecture allowing the cost of the fiber termination to be shared over many subscribers; low-cost laser technology enabling the use of the high bandwidth of the fiber (compared to LEDs) and the installation of longer fiber loops (due to higher coupled power); and wavelength division multiplexing providing high-bandwidth, transparent connectivity in a signal/transmission-format-independent way.

Research in the area of PCS/cellular will ensure full personal communication capability in the NII with mobility, and would be targeted at extension beyond current PCS/cellular applications to 1.5-megabit-per-second transmission rates.

Research in the broadband wireless area could lead to elimination of the need for cable infrastructure for the last mile to the subscribers. Specifically, high-quality entertainment video and data communications with very large file transfer capabilities to residential and business subscribers are needed, especially in certain environments (e.g., urban). The transmission capacity would likely be asymmetric: hundreds of megabits per second downstream, hundreds of kilobits per second upstream per subscriber.

Interactive bidirectional satellite communications to residential and business subscribers (clients) for data communications (and possibly voice and video telephony) are also deserving of further research. This would supplement DBS technology, which already has been developed for entertainment video. Transmission rates would be 64 kilobits per second through 1.5 megabits per second per subscriber. Perhaps the most significant usage of this technology would be in rural and remote regions. Research is needed in the areas of antenna design, signal processing, and coding, modulation and multiplexing methods for very large numbers of subscribers per satellite transponder.

Backbone Network (terabits per second, plus)

Backbone trunking systems capacity and cost must be improved dramatically. Several avenues to such improvement are available. They can be broadly separated into time division techniques and wavelength division techniques. Determining the potential of each technology would be a significant contribution.

In the time domain the limits are determined by the speed of the electro-optic transducers, the speed of required buffers and memory, and the speed of the switching and control logic required to manage the system. In addition, high-speed regeneration technologies play a pivotal role in delivering the benefits of the time division techniques to the system.

In long-distance systems, fiber properties such as loss and dispersion in the fiber limit the capabilities of the fiber span. Devices such as optical amplification and dispersion compensators redress performance impairments induced by the fiber properties themselves. Wavelength division multiplexers, wavelength converters, and wavelength filters and routers enable use of more of the available capacity of the fiber. Optical regeneration techniques permit clock recovery and lead to full regenerative capabilities in the optical domain, avoiding unnecessary optical-to-electrical conversion.

2.3 Switching Research

Switching Fabric Technology

Switching fabrics must have the ability to support general multipoint virtual circuits (both one-to-many and many-to-many), must be robust in the face of overload-induced traffic congestion and must be sufficiently reliable to ensure very high availability. They should achieve switching delays substantially smaller than digital telephone switches and will likely require mechanisms for distinguishing between high-priority, reservation-oriented traffic and lower priority, on-demand traffic. All these capabilities must be achieved in the context of significant cost constraints. Success will require research on alternative switching fabric topologies and performance evaluation, as well as design and construction of large-scale experimental systems. Key areas for research include electro-optic switching, all-optical switching and optical backplanes.

NII networking in the five- to 10-year time frame will likely depend on hybrid electronic/optical switching systems with electronics and optics each utilized in areas where they have advantages. For example, optical switching fabrics could be combined with electrical control. Research areas would include low-cost electrical-to-optical conversion; techniques for combining lasers, optical receivers and digital electronics on the same substrate; and studies of architectures that make optimal trade-offs of electronics and optics.

Despite the fact that electro-optic switching is here today, and is likely to be the initial switching technology path for the NII, end-to-end optical transport and switching will eliminate the inefficiencies introduced by electrical-to-optical conversion. Preliminary research is under way, but much more is needed.

The high bit rates demanded of switching systems supporting the NII will require high-speed interconnections between circuit boards. These demands will quickly exceed the ability of electrical backplanes to carry this traffic cost-effectively. A solution may be the use of optical backplanes with electrical-to-optical conversion performed on the individual circuit boards. This would follow the evolutionary trend of fiber optics first being used to interconnect switching systems, then used to interconnect modules within switching systems, and next used to interconnect individual circuit boards within modules. Research areas in optical backplanes would include low-cost electrical-to-optical conversion, guided wave techniques and hologram techniques.

Switch Control Technology

Signaling systems for switch control must support a much richer model of communication than previous generations of switching systems. User channels can operate at any rate, from a few bits per second to a gigabit per second and beyond. Multipoint communication channels (both one-to-many and many-to-many) are needed for applications including video distribution and multimedia conferencing. This requires a signaling and control system that supports a rich multipoint call model, where a call may include multiple virtual circuits, each with its own individual characteristics. Some applications may place extreme demands on the signaling subsystem (for example, in video distribution applications, a robust signaling architecture and corresponding protocol capability will be required to accommodate the flexible selection of service provider(s) for each service request), while other applications may be stable for long periods with much simpler demands.

A crucial issue in enabling the NII is designing switch control systems that can be used in conjunction with a variety of underlying switch fabrics. Research will be required in areas such as identification of appropriate switch fabric/control boundaries and design of protocol sets for multi-switch synchronization. This decoupling of the switch fabric from switch control will enable these components to be upgraded independently, to take advantage of successive generations of technology. It will also allow the capacity of control systems to be engineered separately from switch fabrics. This is essential, given the unpredictable nature of NII traffic and the need to maintain maximum flexibility in matching control capabilities to user demand.

Switch Capacity

The backbone network will require switches of tremendous capacity. Switches of this scale are not commercially available today, and considerable research must be done to make them so. Total system throughputs of 15 terabits per second and more are expected to be required. The overall challenge for switching systems is to achieve systems for the access network that can be scaled to support between 10,000 and 100,000 users with high capacity (greater than or equal to 150 megabits per second), high reliability and low cost per user. Historically, satellites have been deployed for transmission functions with all switching functions performed in terrestrial facilities. With the availability of low-power, highly reliable integrated electronics, satellite-based switching has now become an area for active research. An appropriate target would be cost-effectively deploying a 1-gigabit-per-second switch in a satellite.

ATM Switching

Asynchronous transfer mode switches are expected to be the mainstream switching technology for the NII in the next 10 years. Because of its novelty and importance, ATM switching is an area that calls for substantial research investment. In particular, architectural research is needed in the areas of congestion control, highly scalable architectures, low-cost switching, multicast/broadcast support, and reliable and survivable architectures. An overall goal of the research is to make ATM switching as ubiquitous and scalable as conventional telephone switching. In addition, ATM switching should provide full networking services for multimedia applications and high-speed computer communication.

2.4 Internetworking Research

The NII will be an Internet. It will comprise a multiplicity of separately administered networks, both private and public, incorporating an increasing diversity of underlying network technologies. This diversity is a natural outcome of a vigorously competitive environment, presenting information providers and consumers with a range of service options. A consequence of this competition-based heterogeneity is that, unlike the public telephone network since the time of Vail, the NII will not be designed from a single technology and architecture blueprint. This raises serious concerns about how applications and services can operate seamlessly across such a heterogeneous network infrastructure. The technical challenge, then, is to achieve the transparency and seamlessness that have been characteristic of "common architecture" networks like the Public Switched Telephone Network and the Internet, while allowing for the heterogeneity of technology and services that reflects the future competitive environment in the NII.

"Seamlessness" and "transparency" as desired properties of the NII raise a large set of issues, ranging from interoperability across subnetworks, to infrastructure mechanisms that support "plug-and-play" service portability. From the perspective of an application or consumer, achieving seamlessness requires defining a set of critical interfaces along each of a set of dimensions of internetwork service management. Those dimensions could include naming, addressing, accounting, billing, authentication and access control, resource discovery and management, and performance/QOS management, in addition to the recognized need for common internetwork transport protocols/interfaces. We postulate that, for each of these service management dimensions, key interfaces must be defined that achieve adequate economy of scope. Economy of scope results when an interface supports or hides a usefully large set of technology options below it (as IP operates over a large set of underlying network technologies) and when the interface supports a usefully large set of applications and application users above it.

2.4.1 Research Objectives

The research objectives for internetworking are:

2.4.2 The Future of the Protocol Hourglass

The IP layer of the Internet architecture provides a uniform space of identifiers ("IP addresses") shared by all addressable components of the Internet, and a common service abstraction (the "IP datagram service") used for communication between addressable components. The success of the IP in defining a common fixed point or "waist" in the Internet protocol suite must be appreciated in order to shape an effective vision of a general multiservice NII.

This same hourglass notion applies in part to the Public Switched Telephone Network (PSTN). The PSTN has a global numbering plan and a simple service abstraction, the 3-KHz analog channel. The original application of this channel was limited to human-to-human communication in the beginning but has been extended to support applications like fax and modem transmissions in modern times. At lower layers, the 3-KHz service abstraction has been implemented out of a continually changing set of technologies, from direct analog transmission and switching through today's modern fully digital switched and transmitted ISDN networks. Unfortunately, the service abstraction of the PSTN is an analog channel, which does not map well into the sorts of applications that are realizable over end-to-end digital channels. It is clear that the service abstraction of the NII will have to be some sort of digital service.

Why do "hourglass" network architectures appear in such seemingly different networks as the Internet and PSTN, and is this type of architecture a good abstraction for the NII? There is some belief that an hourglass is an important network simplifying tool. Consider the problem of trying to make N different applications work atop N different network technologies. One can develop a common intermediate abstraction, the N mappings from applications to intermediate abstraction, and the N mappings from intermediate abstraction to technologies. For the price of 2N mappings, full N2 communication is possible. Furthermore, the addition of new technologies and applications requires only constant (unit) work.

Assuming there is a central "waist" in the NII protocol hourglass, one essential challenge is to define what functionality is already understood to be needed, what functionality might be added at this level to provide full NII capability, and then what techniques are needed for implementing these functions. The fixed and uniform functionality (i.e., the service abstraction) of the Internet layer, which has been projected over all supporting networks, might be extended in three ways: new basic features and functionality required of all underlying networks; new optional features and functionality supported by some but not all networks; and evolution over time of the service abstraction (i.e., a time-varying service abstraction).

Indeed, some networks may support some features and other networks other features. This makes for a vexing interoperability problem. If we break the uniformity of function at the Internet layer, we will have to introduce mechanisms to negotiate the achievable commonality between communicating end points (and intervening networks). This has been a very difficult area to address in the past, and there is no reason to think this is going to be any easier with today's technology.

A contemporary phenomenon is the rapid growth of commercial networks providing data communication services. The planning for future networks providing integrated services would benefit from knowledge, based on analyses of network models, on whether there are fundamentally important reasons to justify coexistence of diverse networks. Is it possible that the benefits of resource sharing and economy of scale are outweighed by the overhead of managing a network that must satisfy a very broad spectrum of quality-of-service requirements? Topics that need to be examined are the consequences on network design and operations of diversity in quality of service as measured by guarantees, response times and flexible allocation of bandwidth, reliability and error rates.

Another research question is whether or not a single NII protocol hourglass is necessary and sufficient. If there are multiple NII protocol stacks, can interworking between the stacks be accomplished at another level, and does this interworking lead to cost efficiencies, increased flexibility or better backwards compatibility and evolution?

The basic datagram service on the Internet is a best-effort service to deliver a variable-length packet from source to destination--without guarantee of delivery. Internet's longevity is due, in part, to the ability to derive this simple service over virtually any transmission and switching architecture. This notion is largely based on the notion of "encapsulation" of datagrams in lower-level network frames. We believe that encapsulation will continue to be fundamental to NII implementation.

While this service has served the current population of the Internet well, it is clearly insufficient for an NII that must carry all forms of communication. The NII will need to carry not only the computer communications traffic prevalent on today's Internet, but also real-time traffic such as voice and video. Many services we would like to offer within the NII framework will require some form of resource assurance or allocation. Internet's simple best-effort's datagram service can handle only limited amounts of packet voice and video, and only in an uncongested network.

Commercial networks have devised innovative algorithms for providing bandwidth on demand, i.e., for adaptively allocating uncommitted bandwidth to active users on the basis of feedback on the congestion levels at distant nodes. These algorithms are optimized for local traffic and network conditions. Interworking has been given little thought, partly because of the lack of means and partly because of a lack of information. In fact, heterogeneous conditions, and also history, have inevitably led to a variety of protocols and algorithms that sometimes interact destructively. A subject of research is to understand better the interactions of these dynamic processes and to develop tools for their analyses and specifications, in which the end goal is to make these processes cooperative.

A final research topic relates to the operation of IP-style protocols, with full addressing and routing, over networks that themselves provide these functions.