RFC 1077






Network Working Group                              Gigabit Working Group
Request for Comments: 1077                             B. Leiner, Editor
                                                           November 1988


              Critical Issues in High Bandwidth Networking


Status of this Memo



   This memo presents the results of a working group on High Bandwidth
   Networking.  This RFC is for your information and you are encouraged
   to comment on the issues presented.  Distribution of this memo is
   unlimited.

ABSTRACT



   At the request of Maj. Mark Pullen and Maj. Brian Boesch of DARPA, an
   ad-hoc working group was assembled to develop a set of
   recommendations on the research required to achieve a ubiquitous
   high-bandwidth network as discussed in the FCCSET recommendations for
   Phase III.

   This report outlines a set of research topics aimed at providing the
   technology base for an interconnected set of networks that can
   provide highbandwidth capabilities.  The suggested research focus
   draws upon ongoing research and augments it with basic and applied
   components.  The major activities are the development and
   demonstration of a gigabit backbone network, the development and
   demonstration of an interconnected set of networks with gigabit
   throughput and appropriate management techniques, and the development
   and demonstration of the required overall architecture that allows
   users to gain access to such high bandwidth.


















Gigabit Working Group                                           [Page 1]

RFC 1077                                                   November 1988


   1.  Introduction and Summary





   1.1.  Background




   The computer communications world is evolving toward both high-
   bandwidth capability and high-bandwidth requirements.  The recent
   workshop conducted under the auspices of the FCCSET Committee on High
   Performance Computing [1] identified a number of areas where
   extremely high-bandwidth networking is required to support the
   scientific research community.  These areas range from remote
   graphical visualization of supercomputer results through the movement
   of high rate sensor data from space to the ground-based scientific
   investigator.  Similar requirements exist for other applications,
   such as military command and control (C2) where there is a need to
   quickly access and act on data obtained from real-time sensors.  The
   workshop identified requirements for switched high-bandwidth service
   in excess of 300 Mbit/s to a single user, and the need to support
   service in the range of a Mbit/s on a low-duty-cycle basis to
   millions of researchers.  When added to the needs of the military and
   commercial users, the aggregate requirement for communications
   service adds up to many billions of bits per second.  The results of
   this workshop were incorporated into a report by the FCCSET [2].

   Fortunately, technology is also moving rapidly.  Even today, the
   installed base of fiber optics communications allows us to consider
   aggregate bandwidths in the range of Gbit/s and beyond to limited
   geographical regions.  Estimates arrived at in the workshop lead one
   to believe that there will be available raw bandwidth approaching
   terabits per second.

   The critical question to be addressed is how this raw bandwidth can
   be used to satisfy the requirements identified in the workshop: 1)
   provide bandwidth on the order of several Gbit/s to individual users,
   and 2) provide modest bandwidth on the order of several Mbit/s to a
   large number of users in a cost-effective manner through the
   aggregation of their traffic.

   Through its research funding, the Defense Advanced Research Projects
   Agency (DARPA) has played a central role in the development of
   packet-oriented communications, which has been of tremendous benefit
   to the U.S. military in terms of survivability and interoperability.
   DARPA-funded research has resulted in the ARPANET, the first packet-
   switched network; the SATNET, MATNET and Wideband Network, which
   demonstrated the efficient utilization of shared-access satellite
   channels for communications between geographically diverse sites;



Gigabit Working Group                                           [Page 2]

RFC 1077                                                   November 1988


   packet radio networks for mobile tactical environments; the Internet
   and TCP/IP protocols for interconnection and interoperability between
   heterogeneous networks and computer systems; the development of
   electronic mail; and many advances in the areas of network security,
   privacy, authentication and access control for distributed computing
   environments.  Recognizing DARPA's past accomplishments and its
   desire to continue to take a leading role in addressing these issues,
   this document provides a recommendation for research topics in
   gigabit networking.  It is meant to be an organized compendium of the
   critical research issues to be addressed in developing the technology
   base needed for such a high bandwidth ubiquitous network.


   1.2.  Ongoing Activities




   The OSTP report referred to above recommended a three-phase approach
   to achieving the required high-bandwidth networking for the
   scientific and research community.  Some of this work is now well
   underway.  An ad-hoc committee, the Federal Research Internet
   Coordinating Committee (FRICC) is coordinating the interconnection of
   the current wide area networking systems in the government; notably
   those of DARPA, Department of Energy (DoE), National Science
   Foundation (NSF), National Aeronautics and Space Administration
   (NASA), and the Department of Health and Human Services (HHS).  In
   accordance with Phases I and II of the OSTP report, this activity
   will provide for an interconnected set of networks to support
   research and other scholarly pursuits, and provide a basis for future
   networking for this community.  The networking is being upgraded
   through shared increased bandwidth (current plans are to share a 45
   Mbit/s backbone) and coordinated interconnection with the rest of the
   world.  In particular, the FRICC is working with the European
   networking community under the auspices of another ad-hoc group, the
   Coordinating Committee for Intercontinental Research Networks
   (CCIRN), to establish effective US-Europe networking.

   However, as the OSTP recommendations note, the required bandwidth for
   the future is well beyond currently planned public, private, and
   government networks.  Achieving the required gigabit networking
   capabilities will require a strong research activity.  There is
   considerable ongoing research in relevant areas that can be drawn
   upon; particularly in the areas of high-bandwidth communication
   links, high-speed computer switching, and high-bandwidth local area
   networks.  Appendix A provides some pointers to current research
   efforts.






Gigabit Working Group                                           [Page 3]

RFC 1077                                                   November 1988


   1.3.  Document Overview




   This report outlines a set of research topics aimed at providing the
   technology base for an interconnected set of networks that can
   provide the required high-bandwidth capabilities discussed above.
   The suggested research focus draws upon ongoing research and augments
   it with basic and applied components.  The major activities are the
   development and demonstration of a Gigabit Backbone network (GB) [3],
   the development and demonstration of an interconnected set of
   networks with gigabit throughput and appropriate management
   techniques, and the development and demonstration of the required
   overall architecture that allows users to gain access to such high
   bandwidth.  Section 2 discusses functional and performance goals
   along with the anticipated benefits to the ultimate users of such a
   system.  Section 3 provides the discussion of the critical research
   issues needed to achieve these goals.  It is organized into the major
   areas of technology that need to be addressed: general architectural
   issues, high-bandwidth switching, high-bandwidth host interfaces,
   network management algorithms, and network services.  The discussion
   in some cases contains examples of ongoing relevant research or
   potential approaches.  These examples are intended to clarify the
   issues and not to propose that particular approach.  A discussion of
   the relationship of the suggested research to other ongoing
   activities and optimal methods for pursuing this research is provided
   in Section 4.


   2.  Functional and Performance Goals




   In this section, we provide an assessment of the types of services a
   GN (four or five orders of magnitude faster than the current
   networks) should provide to its users.  In instances where we felt
   there would be a significant impact on performance, we have provided
   an estimate of the amount of bandwidth needed and delay allowable to
   provide these services.


   2.1.  Networking Application Support




   It is envisioned that the GN will be capable of supporting all of the
   following types of networking applications.







Gigabit Working Group                                           [Page 4]

RFC 1077                                                   November 1988


   Currently Provided Packet Services

      It is important that the network provide the users with the
      equivalent of services that are already available in packet-
      switched networks, such as interactive data exchange, mail
      service, file transfer, on-line access to remote computing
      resources, etc., and allow them to expand to other more advanced
      services to meet their needs as they become available.

   Multi-Media Mail

      This capability will allow users to take advantage of different
      media types (e.g., graphics, images, voice, and video as well as
      text and computer data) in the transfer of messages, thereby
      increasing the effectiveness of message exchange.

   Multi-Media Conferencing

      Such conferencing requires the exchange of large amounts of
      information in short periods of time.  Hence the requirement for
      high bandwidth at low delay.  We estimate that the bandwidth would
      range from 1.5 to 100 Mbit/s, with an end-to-end delay of no more
      than a few hundred msec.

   Computer-Generated Real-time Graphics

      Visualizing computer results in the modern world of supercomputers
      requires large amounts of real time graphics.  This in turn will
      require about 1.5 Mbit/s of bandwidth and no more than several
      hundred msec.  delay.

   High-Speed Transaction Processing

      One of the most important reasons for having an ultra-high-speed
      network is to take advantage of supercomputing capability.  There
      are several scenarios in which this capability could be utilized.
      For example, there could be instances where a non-supercomputer
      may require a supercomputer to perform some processing and provide
      some intermediate results that will be used to perform still
      further processing, or the exchange may be between several
      supercomputers operating in tandem and periodically exchanging
      results, such as in a battle management, war gaming, or process
      control applications.  In such cases, extremely short response
      times are necessary to accomplish as many as hundreds of
      interactions in real time.  This requires very high bandwidth, on
      the order of 100 Mbit/s, and minimum delay, on the order of
      hundreds of msec.




Gigabit Working Group                                           [Page 5]

RFC 1077                                                   November 1988


   Wide-Area Distributed Data/Knowledge Base Management Systems

      Computer-stored data, information, and knowledge is distributed
      around the country for a variety of reasons.  The ability to
      perform complex queries, updates, and report generation as though
      many large databases are one system would be extremely powerful,
      yet requires low-delay, high-bandwidth communication for
      interactive use.  The Corporation for National Research
      Initiatives (NRI) has promoted the notion of a National Knowledge
      base with these characteristics.  In particular, an attractive
      approach is to cache views at the user sites, or close by to allow
      efficient repeated queries and multi-relation processing for
      relations on different nodes.  However, with caching, a processing
      activity may incur a miss in the midst of a query or update,
      causing it to be delayed by the time required to retrieve the
      missing relation or portion of relation.  To minimize the overhead
      for cache directories, both at the server and client sites, the
      unit of caching should be large---say a megabyte or more.  In
      addition, to maintain consistency at the caching client sites,
      server sites need to multicast invalidations and/or updates.
      Communication requirements are further increased by replication of
      the data.  The critical parameter is latency for cache misses and
      consistency operations.  Taking the distance between sites to be
      on average 1/4 the diameter of the country, a one Gbit/s data rate
      is required to reduce the transmission time to be roughly the same
      as the propagation delay, namely around 8 milliseconds for this
      size of unit.  Note that this application is supporting far more
      sophisticated queries and updates than normally associated with
      transaction processing, thus requiring larger amount of data to be
      transferred.


   2.2.  Types of Traffic and Communications Modes




   Different types of traffic may impose different constraints in terms
   of throughput, delay, delay dispersion, reliability and sequenced
   delivery.  Table 1 summarizes some of the main characteristics of
   several different types of traffic.












Gigabit Working Group                                           [Page 6]

RFC 1077                                                   November 1988


                Table 1: Communication Traffic Requirements

   +------------------------+-------------+-------------+-------------+
   |                        |             |             | Error-free  |
   | Traffic                | Delay       | Throughput  | Sequenced   |
   | Type                   | Requirement | Requirement | Delivery    |
   +------------------------+-------------+-------------+-------------+
   | Interactive Simulation | Low         |Moderate-High| No          |
   +------------------------+-------------+-------------+-------------+
   | Network Monitoring     | Moderate    | Low         | No          |
   +------------------------+-------------+-------------+-------------+
   | Virtual Terminal       | Low         | Low         | Yes         |
   +------------------------+-------------+-------------+-------------+
   | Bulk Transfer          | High        | High        | Yes         |
   +------------------------+-------------+-------------+-------------+
   | Message                | Moderate    | Moderate    | Yes         |
   +------------------------+-------------+-------------+-------------+
   | Voice                  |Low, constant| Moderate    | No          |
   +------------------------+-------------+-------------+-------------+
   | Video                  |Low, constant| High        | No          |
   +------------------------+-------------+-------------+-------------+
   | Facsimile              | Moderate    | High        | No          |
   +------------------------+-------------+-------------+-------------+
   | Image Transfer         | Variable    | High        | No          |
   +------------------------+-------------+-------------+-------------+
   | Distributed Computing  | Low         | Variable    | Yes         |
   +------------------------+-------------+-------------+-------------+
   | Network Control        | Moderate    | Low         | Yes         |
   +------------------------+-------------+-------------+-------------+

   The topology among users can be of three types: point-to-point (one-
   to-one connectivity), multicast (one sender and multiple receivers),
   and conferencing (multiple senders and multiple receivers).  There
   are three types of transfers that can take place among users.  They
   are connection-oriented network service, connectionless network
   service, and stream or synchronous traffic.  Connection and
   connectionless services are asynchronous.  A connection-oriented
   service assumes and provides for relationships among the multiple
   packets sent over the connection (e.g., to a common destination)
   while connectionless service assumes each packet is a complete and
   separate entity unto itself.  For stream or synchronous service a
   reservation scheme is used to set up and guarantee a constant and
   steady amount of bandwidth between any two subscribers.








Gigabit Working Group                                           [Page 7]

RFC 1077                                                   November 1988


   2.3.  Network Backbone




   The GB needs to be of high bandwidth to support a large population of
   users, and additionally to provide high-speed connectivity among
   certain subscribers who may need such capability (e.g., between two
   supercomputers).  These users may access the GN from local area
   networks (LANs) directly connected to the backbone or via high-speed
   intermediate regional networks.  The backbone must also minimize
   end-to-end delay to support highly interactive high-speed
   (supercomputer) activities.

   It is important that the LANs that will be connected to the GN be
   permitted data rates independent of the data rates of the GB.  LAN
   speeds should be allowed to change without affecting the GB, and the
   GB speeds should be allowed to change without affecting the LANs.  In
   this way, development of the technology for LANs and the GB can
   proceed independently.

   Access rate requirements to the GB and the GN will vary depending on
   user requirements and local environments.  The users may require
   access rates ranging from multi-kbit/s in the case of terminals or
   personal computers connected by modems up to multi-Mbit/s and beyond
   for powerful workstations up to the Gbit/s range for high-speed
   computing and data resources.


   2.4.  Directory Services




   Directory services similar to those found in CCITT X.500/ISO DIS 9594
   need to be provided.  These include mapping user names to electronic
   mail addresses, distribution lists, support for authorization
   checking, access control, and public key encryption schemes,
   multimedia mail capabilities, and the ability to keep track of mobile
   users (those who move from place to place and host computer to host
   computer).  The directory services may also list facilities available
   to users via the network.  Some examples are databases,
   supercomputing or other special-purpose applications, and on-line
   help or telephone hotlines.

   The services provided by X.500 may require some extension for GN.
   For example, there is no provision for multilevel security, and the
   approach taken to authentication must be studied to ensure that it
   meets the requirements of GN and its user community.






Gigabit Working Group                                           [Page 8]

RFC 1077                                                   November 1988


   2.5.  Network Management and Routing




   The objective of network management is to ensure that the network
   functions smoothly and efficiently, and consists of the following:
   accounting, security, performance monitoring, fault isolation and
   configuration control.

   Accounting ensures that users are properly billed for the services
   that the network provides.  Accounting enforces a tariff; a tariff
   expresses a usage policy.  The network need only keep track of those
   items addressed by the tariff, such as allocated bandwidth, number of
   packets sent, number of ports used, etc.  Another type of accounting
   may need to be supported by the network to support resource sharing,
   namely accounting analogous to telephone "900" numbers.  This
   accounting performed by the network on behalf of resource providers
   and consumers is a pragmatic solution to the problem of getting the
   users and consumers into a financial relationship with each other
   which has stymied previous attempts to achieve widespread use of
   specialized resources.

   Performance monitoring is needed so that the managers can tell how
   the network is performing and take the necessary actions to keep its
   performance at a level that will provide users with satisfactory
   service.  Fault isolation using technical control mechanisms is
   needed for network maintenance.  Configuration management allows the
   network to function efficiently.

   Several new types of routing will be required by GN.  In addition to
   true type-of-service, needed to support diverse distributed
   applications, real-time applications, interactive applications, and
   bulk data transfer, there will be need for traffic controls to
   enforce various routing policies.  For example, policy may dictate
   that traffic from certain users, applications,  or hosts may not be
   permitted to traverse certain segments of the network.
   Alternatively, traffic controls may be used to promote fairness; that
   is, to make sure that busy link or network segment isn't dominated by
   a particular source or destination.  The ability of applications to
   reserve network bandwidth in advance of its use, and the use of
   strategies such as soft connections, will also require development of
   new routing algorithms.


   2.6.  Network Security Requirements




   Security is a critical factor within the GN and one of those features
   that are difficult to provide.  It is envisioned that both



Gigabit Working Group                                           [Page 9]

RFC 1077                                                   November 1988


   unclassified and classified traffic will utilize the GN, so
   protection mechanisms must be an integral part of the network access
   strategy.  Features such as authentication, integrity,
   confidentiality, access control, and nonrepudiation are essential to
   provide trusted and secure communication services for network users.

   A subscriber must have assurance that the person or system he is
   exchanging information with is indeed who he says he is.
   Authentication provides this assurance by verifying that the claimed
   source of a query request, control command, response, etc., is the
   actual source.  Integrity assures that the subscriber's information
   (such as requests, commands, data, responses, etc.) is not changed,
   intentionally or unintentionally, while in transit or by replays of
   earlier traffic.  Unauthorized users (e.g., intruders or network
   viruses) would be denied use of GN assets through access control
   mechanisms which verify that the authenticated source is authorized
   to receive the requested information or to initiate the specified
   command.  In addition, nonrepudiation services can be offered to
   assure a third party that the transmitted information has not been
   altered.  And finally, confidentiality will ensure that the contents
   of a message are not divulged to unauthorized individuals.
   Subscribers can decide, based upon their own security needs and
   particular activities, which of these services are necessary at a
   given time.


   3.  Critical Research Issues




   In the section above, we discussed the goals of a research program in
   gigabit networking; namely to provide the technology base for a
   network that will allow gigabit service to be provided in an
   effective way.  In this section, we discuss those issues which we
   feel are critical to address in a research program to achieve such
   goals.


   3.1.  General Architectural Issues




   In the last generation of networks, it was assumed that bandwidth was
   the scarce resource and the design of the switch was dictated by the
   need to manage and allocate the bandwidth effectively.  The most
   basic change in the next generation network is that the speeds of the
   trunks are rising faster than the speeds of the switching elements.

   This change in the balance of speeds has manifested itself in several
   ways.  In most current designs for local area networks, where



Gigabit Working Group                                          [Page 10]

RFC 1077                                                   November 1988


   bandwidth is not expensive, the design decision was to trade off
   effective use of the bandwidth for a simplified switching technique.
   In particular, networks such as Ethernet use broadcast as the normal
   distribution method, which essentially eliminates the need for a
   switching element.

   As we look at still higher speed networks, and in particular networks
   in which the bandwidth is still the expensive component, we must
   design new options for switching which will permit effective use of
   bandwidth without the switch itself becoming the bottleneck.

   The central thrust of new research must thus be to explore new
   network architectures that are consistent with these very different
   speed assumptions.

   The development of computer communications has been tremendously
   distorted by the characteristics of wide-area networking: normally
   high cost, low speed, high error rate, large delay.  The time is ripe
   for a revolution in thinking, technology, and approaches, analogous
   to the revolution caused by VCR technology over 8 and 16 mm. film
   technology.

   Fiber optics is clearly the enabling technology for high-speed
   transmission, in fact, so much so that there is an expectation that
   the switching elements will now hold down the data rates.  Both
   conventional circuit switching and packet switching have significant
   problems at higher data rates.  For instance, circuit switching
   requires increasing delays for FTDM synchronization to handle skew.
   In the case of packet switching, traditional approaches require too
   much processing per packet to handle the tremendous data flow.  The
   problem for both switching regimes is the "intelligence" in the
   switches, which in turn requires electronics technology.

   Besides intelligence, another problem for wide-area networks is
   storage, both because it ties us to electronics (for the foreseeable
   future) and because it produces instabilities in a large-scale
   system.  (See, for instance, the work by Van Jacobson on self-
   organizing phenomena for self-destruction in the Internet.)
   Techniques are required to eliminate dependence on storage, such as
   cut-through routing.

   Overall, high-speed WANs are the greatest agents of change, the
   greatest catalyst both commercially and militarily, and the area ripe
   for revolution.  Judging by the attributes of current high-speed
   network research prototypes, WANs of the future will be photonic,
   multi-gigabit networks with enormous throughput, low delay, and low
   error rate.




Gigabit Working Group                                          [Page 11]

RFC 1077                                                   November 1988


   A zero-based budgeting approach is required to develop the new high-
   speed internetwork architecture.  That is, the time is ripe to
   significantly rethink the Internet, building on experience with this
   system.  Issues of concern are manageability, understanding
   evolvability and support for the new communication requirements,
   including remote procedure call, real-time, security and fault-
   tolerance.

   The GN must be able to deal with two sources of high-bandwidth
   requirements.  There will be some end devices (computers) connected
   more or less directly to the GN because of their individual
   requirements for high bandwidth (e.g., supercomputers needing to
   drive remote high-bandwidth graphics devices).  In addition, the
   aggregate traffic due to large numbers of moderate rate users
   (estimates are roughly up to a million potential users needing up to
   1 Mbit/s at any given time) results in a high-bandwidth requirement
   in total on the GN.  The statistics of such traffic are different and
   there are different possible technical approaches for dealing with
   them.  Thus, an architectural approach for dealing with both must be
   developed.

   Overall, the next-generation architecture has to be, first and
   foremost, a management architecture.  The directions in link speeds,
   processor speeds and memory solve the performance problems for many
   communication situations so well that manageability becomes the
   predominant concern.  (In fact, fast communication makes large
   systems more prone to performance, reliability, and security
   problems.)  In many ways, the management system of the internetwork
   is the ultimate distributed system.  The solution to this tough
   problem may well require the best talents from the communications,
   operating systems and distributed systems communities, perhaps even
   drawing on database and parallelism research.


   3.1.1.  High-Speed Internet using High-Speed Networks




   The GN will need to take advantage of a multitude of different and
   heterogeneous networks, all of high speed.  In addition to networks
   based on the technology of the GB, there will be high-speed LANs.  A
   key issue in the development of the GN will be the development of a
   strategy for interconnecting such networks to provide gigabit service
   on an end to end basis.  This will involve techniques for switching,
   interfacing, and management (as discussed in the sections below)
   coupled with an architecture that allows the GN to take full
   advantage of the performance of the various high-speed networks.





Gigabit Working Group                                          [Page 12]

RFC 1077                                                   November 1988


   3.1.2.  Network Organization




   The GN will need an architecture that supports the need to manage the
   system as well as obtain high performance.  We note that almost all
   human-engineered systems are hierarchically structured from the
   standpoint of control, monitoring, and information flow.  A
   hierarchical design may be the key to manageability in the next-
   generation architecture.

   One approach is to use a general three-level structure, corresponding
   to interadministrational, intraadministrational, and cluster
   networks.  The first level interconnects communication facilities of
   truly separate administrations where there is significant separation
   of security, accounting, and goals.  The second level interconnects
   subadministrations which exist for management convenience in large
   organizations.  For example, a research group within a university may
   function as a subadministration.  The cluster level consists of
   networks configured to provides maximal performance among hosts which
   are in frequent communication, such as a set of diskless workstations
   and their common file server.  These hosts are typically, but not
   necessarily, geographically collocated.  For example, two remote
   networks may be tightly coupled by a fiber optic link that bridges
   between the two physical networks, making them function as one.

   Research along these lines should study the interorganizational
   characteristics of communications, such as those being investigated
   by the IAB Task Force on Autonomous Networks.  Based on current
   results, we expect that such work would clearly demonstrate that
   considerable communication takes place between particular
   subadministrations in different administrations; communication
   patterns are not strictly hierarchical.  For example, there might be
   intense direct communication between the experimental physics
   departments of two independent universities, or between the computer
   support group of one company and the operating system development
   group of another.  In addition, (sub)administrations may well also
   require divisions into public information and private information.


   3.1.3.  Fault-Tolerant System




   Although the GN will be developed as part of an experimental research
   program, it will also serve as part of the infrastructure for
   researchers who are experimenting with applications which will use
   such a network.  The GN must have reasonably high availability to
   support these research activities.  In addition to facilitate the
   transfer of this technology to future operational military and



Gigabit Working Group                                          [Page 13]

RFC 1077                                                   November 1988


   commercial users, it will need to be designed to become highly
   reliable.  This can be accomplished through diversity of transmission
   paths, the development of fault-tolerant switches, use of a
   distributed control structure with self-correcting algorithms, and
   the protection of network control traffic.  The architecture of a GN
   should support and allow for all of these things.


   3.1.4.  Functional Division of Control Between Network Elements




   Current protocol architectures use the layered model of functional
   decomposition first developed in the early work on ARPANET protocols.
   The concept of layering has been a powerful concept which has allowed
   dramatic variation in network technologies without requiring the
   complete reimplementation of applications.  The concept of layering
   has had a first-order impact on the development of international
   standards for data communication---witness the ISO "Reference Model
   for Open Systems Interconnection."

   Unfortunately, however, the powerful concept of layering has been
   paired, both in the DoD Internet work and the ISO work, with an
   extremely weak concept of the interface between layers.  The
   interface designs are all organized around the idea of commands and
   responses plus an error indicator.  For example, the TCP service
   interface provides the user with commands to set up or close a TCP
   connection and commands to send and receive datagrams.  The user may
   well "know" whether they are using a file transfer service or a
   character-at-a- time virtual terminal, but can't tell the TCP.  The
   underlying network may "know" that failures have reduced the path to
   the user's destination to a single 9.6 kbit/s link, but it also can't
   tell the TCP implementation.

   All of the information that an analyst would consider crucial in
   diagnosing system performance is carefully hidden from adjacent
   layers.  One "solution" often discussed (but rarely implemented) is
   to condense all of this information into a few bits of "Type of
   Service" or "Quality of Service" request flowing in one direction
   only---from application to network.  It seems likely that this
   approach cannot succeed, both because it applies too much compression
   to the knowledge available and because it does not provide two-way
   flow.

   We believe it to be likely that the next-generation network will
   require a much richer interface between every pair of adjacent layers
   if adequate performance is to be achieved.  Research is needed into
   the conceptual mechanisms, both indicators and controls, that can be
   implemented at these interfaces and that, when used, will result in



Gigabit Working Group                                          [Page 14]

RFC 1077                                                   November 1988


   better performance.  If real differences in performance can be
   observed, then the implementors of every layer will have a strong
   incentive to make use of the mechanisms.

   We can observe the first glimmers of this sort of coordination
   between layers in current work.  For example, in the ISO work there
   are 5 classes of transport protocol which are supposed to provide a
   range of possible matches between application needs and network
   capabilities.  Unfortunately, it is the case today that the class of
   transport protocol is chosen statically, by the implementer, rather
   than dynamically.  The DARPA Wideband net offers a choice of stream
   or datagram service, but typically a given host uses all one or all
   the other---again, a static rather than a dynamic choice.  The
   research that we believe is needed, therefore, is not how to provide
   alternatives, but how to provide them and choose among them on a
   dynamic, real-time basis.


   3.1.5.  Different Switch Technologies




   One approach to high-performance networking is to design a technology
   that is expected to work as a stand-alone demonstration, without
   addressing the need for interconnection to other networks.  Such an
   experiment may be very valuable for rapid exploration of the design
   space.  However, our experience with the Internet project suggests
   that a primary research goal should be the development of a network
   architecture that permits the interconnection of a number of
   different switching technologies.

   The Internet project was successful to a large extent because it
   could incorporate a number of new and preexisting network
   technologies: various local area networks, store and forward
   switching networks, broadcast satellite nets, packet radio networks,
   and so on.  In this way, it decoupled the use of the protocols from a
   particular technology base.  In fact, the technology base evolved
   rapidly, but the Internet protocols themselves provided a stability
   that led to their success.

   The next-generation architecture must similarly deal with a diverse
   and evolving technology base.  We see "fast-packet" switching now
   being developed (for example in B-ISDN); we see photonic switching
   and wavelength division multiplexing as more advanced technologies.
   We must divorce our architecture from dependence on any one of these.

   At the host interface, we must divorce the multiplexing of the medium
   from the form of data that the host sees.  Today the packet is used
   both as multiplexing and interface element.  In the future, the host



Gigabit Working Group                                          [Page 15]

RFC 1077                                                   November 1988


   may see the network as a message-passing system, or as memory.  At
   the same time, the network may use classic packets, wavelength
   division, or space division switching.

   A number of basic functions must be rethought to provide an
   architecture that is not dependent on the underlying switching model.
   For example, our transport protocols assume that data will be lost in
   units of a packet.  If part of a packet is lost, we discard the whole
   thing.  And if several packets are systematically lost in sequence,
   we may not recover effectively.  There must be a host-level unit of
   error recovery that is independent of the network.  This sort of
   abstraction must be applied to all the aspects of service
   specification: error recovery, flow control, addressing, and so on.


   3.1.6.  Network Operations, Monitoring, and Control




   There is a hierarchy of progressively more effective and
   sophisticated techniques for network management that applies
   regardless of network bandwidth and application considerations:

      1.  Reactive problem management

      2.  Reactive resource management

      3.  Proactive problem management

      4.  Proactive resource management.



   Today's network management strategies are primarily reactive rather
   than proactive:  Problem management is initiated in response to user
   complaints about service outages; resource allocation decisions are
   made when users complain about deterioration of quality of service.
   Today's network management systems are stuck at step 1 or perhaps
   step 2 of the hierarchy.

   Future network management systems will provide proactive problem
   management---problem diagnosis and restoral of service before users
   become aware that there was a problem; and proactive resource
   management---dynamic allocation of network bandwidth and switching
   resources to ensure that an acceptable level of service is
   continuously maintained.

   The GN management system should be expected to provide proactive
   problem and resource management capabilities.  It will have to do so
   while contending with three important changes in the managed network
   environment:



Gigabit Working Group                                          [Page 16]

RFC 1077                                                   November 1988


      1.  More complicated devices under management

      2.  More diverse types of devices

      3.  More variety of application protocols.

   Performance under these conditions will require that we seriously
   re-think how a network management system handles the expected high
   volumes of raw management-related data.  It will become especially
   important for the system to provide thresholding, filtering, and
   alerting mechanisms that can save the human operator from drowning in
   data, while still permitting access to details when diagnostic or
   fault isolation modes are invoked.

   The presence of expert assistant capabilities for early fault
   detection, diagnosis, and problem resolution will be mandatory.
   These capabilities are highly desirable today, but they will be
   essential to contend with the complexity and diversity of devices and
   applications in the Gigabit Network.

   In addition to its role in dealing with complexity, automation
   provides the only hope of controlling and reducing the high costs of
   daily management and operation of a GN.

   Proactive resource management in GNs must be better understood and
   practiced, initially as an effort requiring human intervention and
   direction.  Once this is achieved, it too must become automated to a
   high degree in the GN.


   3.1.7.  Naming and Addressing Strategies




   Current networks, both voice (telephone) and data, use addressing
   structures which closely tie the address to the physical location on
   the network.  That is, the address identifies a physical access
   point, rather than the higher-level entity (computer, process, human)
   attached to that access point.  In future networks, this physical
   aspect of addressing must be removed.

   Consider, for example, finding the desired party in the telephone
   network of today.  For a person not at his listed number, finding the
   number of the correct telephone may require preliminary calls, in
   which advice is given to the person placing the call.  This works
   well when a human is placing the call, since humans are well equipped
   to cope with arbitrary conversations.  But if a computer is placing
   the call, the process of obtaining the correct address will have to
   be incorporated in the architecture as a core service of the network.



Gigabit Working Group                                          [Page 17]

RFC 1077                                                   November 1988


   Since it is reasonable to expect mobile hosts, hosts that are
   connected to multiple networks, and replicated hosts, the issue of
   mapping to the physical address must be properly resolved.

   To permit the network to maintain the dynamic mapping to current
   physical address, it is necessary that high-level entities have a
   name (or logical address) that identifies them independently of
   location.  The name is maintained by the network, and mapped to the
   current physical location as a core network service.  For example,
   mobile hosts, hosts that are connected to multiple networks, and
   replicated hosts would have static names whose mapping to physical
   addresses (many-to-one, in some cases) would change with time.

   Hosts are not the only entities whose physical location varies.
   Users' electronic mail addresses change.  Within distributed systems,
   processes and files migrate from host to host.  In a computing
   environment where robustness and survivability are important, entire
   applications may move about, or they may be redundant.

   The needed function must be considered in the context of the mobility
   and address resolution rates if all addresses in a global data
   network were of this sort.  The distributed network directory
   discussed elsewhere in this report should be designed to provide the
   necessary flexibility, and responsiveness.  The nature and
   administration of names must also be considered.

   Names that are arbitrary or unwieldy would be barely better than the
   addresses used now.  The name space should be designed so that it can
   easily be partitioned among the agencies that will assign names.  The
   structure of names should facilitate, rather than hinder, the mapping
   function.  For example, it would be hard to optimize the mapping
   function if names were flat and unstructured.


   3.2.  High-Speed Switching




   The term "high-speed switching" refers to changing the switching at a
   high rate, rather than switching high-speed links, because the latter
   is not difficult at low speeds.  (Consider, for example, manual
   switching of fiber connections).  The switching regime chosen for the
   network determines various aspects of its performance, its charging
   policies, and even its effective capabilities.  As an example of the
   latter, it is difficult to expect a circuit-switched network to
   provide strong multicast support.

   A major area of debate lies in the choice between packet switching
   and circuit switching.  This is a key research issue for the GN,



Gigabit Working Group                                          [Page 18]

RFC 1077                                                   November 1988


   considering also the possibility of there being combinations of the
   two approaches that are feasible.


   3.2.1.  Unit of Management vs. Multiplexing




   With very high data rates, either the unit of management and
   switching must be larger or the speed of the processor elements for
   management and switching must be faster.  For example, at a gigabit,
   a 576 byte packet takes roughly 5 microseconds to be received so a
   packet switch must act extremely fast to avoid being the dominant
   delay in packet times.  Moreover, the storage time for the packet in
   a conventional store and forward implementation also becomes a
   significant component of the delay.  Thus, for packet switching to
   remain attractive in this environment, it appears necessary to
   increase the size of packets (or switch on packet groups), do so-
   called virtual cut-through and use high-speed routing techniques,
   such as high-speed route caches and source routing.

   Alternatively, for circuit switching to be attractive, it must
   provide very fast circuit setup and tear-down to support the bursty
   nature of most computer communication.  This problem is rendered
   difficult (and perhaps impossible for certain traffic loads) because
   the delay across the country is so large relative to the data rate.
   That is, even with techniques such as so-called fast select,
   bandwidth is reserved by the circuit along the path for almost twice
   the propagation time before being used.

   With gigabit circuit switching, because it is not feasible to
   physically switch channels, the low-level switching is likely doing
   FTDM on micro-packets, as is currently done in telephony.  Performing
   FTDM at gigabit data rates is a challenging research problem if the
   skew introduced by wide-area communication is to be handled with
   reasonable overhead for spacing of this micro-packets.  Given the
   lead and resources of the telephone companies, this area of
   investigation should, if pursued, be pursued cooperatively.


   3.2.2.  Bandwidth Reservation Algorithms




   Some applications, such as real-time video, require sustained high
   data rate streams over a significant period of time, such as minutes
   if not hours.  Intuitively, it is appealing for such applications to
   pre-allocate the bandwidth they require to minimize the switching
   load on the network and guarantee that the required bandwidth is
   available.  Research is required to determine the merits of bandwidth



Gigabit Working Group                                          [Page 19]

RFC 1077                                                   November 1988


   reservation, particular in conjunction with the different switching
   technologies.  There is some concern to raise that bandwidth
   reservation may require excessive intelligence in the network,
   reducing the performance and reliability of the network.  In
   addition, bandwidth reservation opens a new option for denial of
   service by an intruder or malicious user.  Thus, investigations in
   this area need to proceed in concert with work on switching
   technologies and capabilities and security and reliability
   requirements.


   3.2.3.  Multicast Capabilities




   It is now widely accepted that multicast should be provided as a
   user-level service, as described in RFC 1054 for IP, for example.
   However, further research is required to determine the best way to
   support this facility at the network layer and lower.  It is fairly
   clear that the GN will be built from point-to-point fiber links that
   do not provide multicast/broadcast for free.  At the most
   conservative extreme, one could provide no support and require that
   each host or gateway simulate multicast by sending multiple,
   individually addressed packets.  However, there are significant
   advantages to providing very low level multicast support (besides the
   obvious performance advantages).  For example, multicast routing in a
   flooding form provides the most fault-tolerant, lowest-delay form of
   delivery which, if reserved for very high priority messages, provides
   a good emergency facility for high-stress network applications.
   Multicast may also be useful as an approach to defeat traffic
   analysis.

   Another key issue arises with the distinction between so-called open
   group multicast and closed group multicast.  In the former, any host
   can multicast to the group, whereas in the latter, only members of
   the group can multicast to it.  The latter is easier to support and
   adequate for conferencing, for example.  However, for more client-
   server structured applications, such as using file/database server,
   computation servers, etc. as groups, open multicast is required.
   Research is needed to address both forms of multicast.  In addition,
   security issues arise in controlling the membership of multicast
   groups.  This issue should be addressed in concert with work on
   secure forms of routing in general.









Gigabit Working Group                                          [Page 20]

RFC 1077                                                   November 1988


   3.2.4.  Gateway Technologies




   With the wide-area interconnection of local networks by the GN,
   gateways are expected to become a significant performance bottleneck
   unless significant advances are made in gateway performance.  In
   addition, many network management concerns suggest putting more
   functionality (such as access control) in the gateways, further
   increasing their load and the need for greater capacity.  This would
   then raise the issue of the trade-off between general-purpose
   hardware and special-purpose hardware.

   On the general-purpose side, it may be feasible to use a general-
   purpose multiprocessor based on high-end microprocessors (perhaps as
   exotic as the GaAs MIPS) in conjunction with a high-speed block
   transfer bus, as proposed as part of the FutureBus standard (which is
   extendible to higher speeds than currently commercially planned) and
   intelligent high-speed network adaptors.  This would also allow the
   direct use of hardware, operating systems, and software tools
   developed as part of other DARPA programs, such as Strategic
   Computing.  It also appears to make this gateway software more
   portable to commercial machines as they become available in this
   performance range.

   The specialized hardware approach is based on the assumption that
   general-purpose hardware, particularly the interconnection bus,
   cannot be fast enough to support the level of performance required.
   The expected emphasis is on various interconnection network
   techniques.  These approaches appear to require greater expense, less
   commercial availability and more specialized software.  They need to
   be critically evaluated with respect to the general-purpose gateway
   hardware approach, especially if the latter is using multiple buses
   for fault-tolerance as well as capacity extension (in the absence of
   failure).

   The same general-purpose vs. special-purpose contention is an issue
   with operating system software.  Conventionally, gateways run
   specialized run-time executives that are designed specifically for
   the gateway and gateway functions.  However, the growing
   sophistication of the gateway makes this approach less feasible.  It
   appears important to investigate the feasibility of using a standard
   operating system foundation on the gateways that is known to provide
   the required security and reliability properties (as well as real-
   time performance properties).







Gigabit Working Group                                          [Page 21]

RFC 1077                                                   November 1988


   3.2.5.  VLSI and Optronics Implementations




   It appears fairly clear that gigabit communication will use fiber
   optics for at least the near future.  Without major advances in
   optronics to allow effectively for optical computers, communication
   must cross the optical-electronic boundary two or more times.  There
   are significant cost, performance, reliability, and security benefits
   for minimizing the number of such crossings.  (As an example of a
   security benefit, optics is not prone to electronic surveillance or
   jamming while electronics clearly is, so replacing an optic-
   electronic-optic node with a pure optic node eliminates that
   vulnerability point.)

   The benefits of improved technology in optronics is so great that its
   application here is purely another motivation for an already active
   research area (that deserves strong continued support).  Therefore,
   we focus here in the issue of matching current (and near-term
   expected) optronics capabilities with network requirements.

   The first and perhaps greatest area of opportunity is to achieve
   totally (or largely) photonic switches in the network switching
   nodes.  That is, most packets would be switched without crossing the
   optics-electronics boundary at all.  For this to be feasible, the
   switch must use very simple switching logic, require very little
   storage and operate on packets of a significant size.  The source-
   routed packet switches with loopback on blockage of Blazenet
   illustrate the type of techniques that appear required to achieve
   this goal.

   Research is required to investigate the feasibility of optronic
   implementation of switches.  It appears highly likely that networks
   will at some point in the future be totally photonically switched,
   having the impact on networking comparable to the effect of
   integrated circuits on processors and memories.

   A next level of focus is to achieve optical switching in the common
   case in gateways.  One model is a multiprocessor with an optical
   interconnect.  Packets associated with established paths through the
   gateway are optically switched and processed through the
   interconnect.  Other packets are routed to the multiprocessor,
   crossing into the electronics domain.  Research is required to marry
   the networking requirements and technology with optronics technology,
   pushing the state of the art in both areas in the process.

   Given the long-term presence of the optic-electronic boundary,
   improvements in technology in this area are also important.  However,
   it appears that there is already enormous commercial research



Gigabit Working Group                                          [Page 22]

RFC 1077                                                   November 1988


   activity in this area, particularly within the telephone companies.
   This is another area in which collaborative investigation appears far
   better than an new independent research effort.

   VLSI technology is an established technology with active research
   support.  The GN effort does not appear to require major new
   initiatives in the VLSI area, yet one should be open to significant
   novel opportunities not identified here.


   3.2.6.  High-Speed Transfer Protocols




   To achieve the desired speeds, it will be necessary to rethink the
   form of protocols.

      1.  The simple idea of a stateless gateway must be replaced by a
          more complex model in which the gateway understands the
          desired function of the end point and applies suitable
          optimizations to the flow.

      2.  If multiplexing is done in the time domain, the elements of
          multiplexing are probably so small that no significant
          processing can be performed on each individually.  They must
          be processed as an aggregate.  This implies that the unit of
          multiplexing is not the same as the unit of processing.

      3.  The interfaces between the structural layers of the
          communication system must change from a simple
          command/response style to a richer system which includes
          indications and controls.

      4.  An approach must be developed that couples the memory
          management in the host and the structure of the transmitted
          data, to allow efficient transfers into host memory.

   The result of rethinking these problems will be a new style of
   communications and protocols, in which there is a much higher degree
   of shared responsibility among the components (hosts, switches,
   gateways).  This may have little resemblance to previous work either
   in the DARPA or commercial communities.


   3.3.  High-Speed Host Interfaces




   As networks get faster, the most significant bottleneck will turn out
   to be the packet processing overhead in the host.  While this does



Gigabit Working Group                                          [Page 23]

RFC 1077                                                   November 1988


   not restrict the aggregate rates we can achieve over trunks, it
   prevents delivery of high data rate flows to the host-based
   applications, which will prevent the development of new applications
   needing high bandwidth.  The host bottleneck is thus a serious
   impediment to networked use of supercomputers.

   To build a GN we need to create new ways for hosts and their high
   bandwidth peripherals to connect to networks.  We believe that
   pursuing research in the ways to most effectively isolate host and
   LAN development paths from the GN is the most productive way to
   proceed.  By decoupling the development paths, neither is restricted
   by the momentary performance of capability bottlenecks of the other.
   The best context in which to view this separation is with the notion
   of a network front end (NFE).  The NFE can take the electronic input
   data at many data rates and transform it into gigabit light data
   appropriately packetized to traverse the GN.  The NFE can accept
   inputs from many types of gateways, hosts, host peripherals, and LANS
   and provide arbitration and path set-up facilities as needed.  Most
   importantly, the NFE can perform protocol arbitration to retain
   upward compatibility with the existing Internet protocols while
   enabling those sophisticated network input sources to execute GN
   specific high-throughput protocols.  Of course, this introduces the
   need for research into high-speed NFEs to avoid the NFE becoming a
   bottleneck.


   3.3.1.  VLSI and Optronics Implementations




   In a host interface, unless the host is optical (an unlikely prospect
   in the near-term), the opportunities for optronic support are
   limited.  In fact, with a serial-to-parallel conversion on reception
   stepping the clock rate down by a factor of 32 (assuming a 32-bit
   data path on the host interface), optronic speeds are not required in
   the immediate future.

   One exception may be for encryption.  Current VLSI implementations of
   standard encryption algorithms run in the 10 Mbit/s range.  Optronic
   implementation of these encryption techniques and encryption
   techniques specifically oriented to, or taking advantage of, optronic
   capabilities appears to be an area of some potential (and enormous
   benefit if achieved).

   The potential of targeted VLSI research in this area appears limited
   for similar reasons discussed above with its application in high-
   speed switching.  The major benefits will arise from work that is
   well-motivated by other research (such as high-performance
   parallelism) and by strong commercial interest.  Again, we need to be



Gigabit Working Group                                          [Page 24]

RFC 1077                                                   November 1988


   open to imaginative opportunities not foreseen here while keeping
   ourselves from being diverted into low-impact research without
   further insights being put forward.


   3.3.2.  High-Performance Transport Protocols




   Current transport protocols exhibit some severe problems for maximal
   performance, especially for using hardware support.  For example, TCP
   places the checksum in the packet header, forcing the packet to be
   formed and read fully before transmission begins.  ISO TP4 is even
   worse, locating the checksum in a variable portion of the header at
   an indeterminate offset, making hardware implementation extremely
   difficult.

   The current Internet has thrived and grown due to the existence of
   TCP implementations for a wide variety of classes of host computers.
   These various TCP implementations achieve robust interoperability by
   a "least common denominator" approach to features and options.  Some
   applications have arisen in the current Internet, and analogs can be
   envisioned for the GN environment, which need qualities of service
   not generally supported by the ubiquitous generic TCP, and therefore
   special purpose transport protocols have been developed.  Examples
   include special purpose transport protocols such as UDP (user
   datagram protocol), RDP (reliable datagram protocol), LDP
   (loader/debugger protocol), NETBLT (high-speed block transfer
   protocol), NVP (network voice protocol) and PVP (packet video
   protocol).  Efforts are also under way to develop a new generic
   transport protocol VMTP (versatile message transaction protocol)
   which will remedy some of deficiencies of TCP, without the need to
   resort to special purpose protocols for some applications.  Research
   is needed in this area to understand how transport level protocols
   should be constructed for a GN which provide adequate qualities of
   service and ease of implementation.

   A new transport protocol of reasonable success can be expected to
   last for ten years more.  Therefore, a new protocol should not be
   over optimized for current networks and must not ignore the
   functional deficiencies of current protocols.  These deficiencies are
   essential to remedy before it is feasible to deploy even current
   distributed systems technology for military and commercial
   applications.

   Forward Error Correction (FEC) is a useful approach when the
   bandwidth/delay ratio of the physical medium is high, as can be
   expected in transcontinental photonic links.  A degenerate form of
   FEC is to simply transmit multiple copies of the data; this allows



Gigabit Working Group                                          [Page 25]

RFC 1077                                                   November 1988


   one to trade bandwidth for delay and reliability, without requiring
   much intelligence.  In fact, it is generally true that reliability,
   bandwidth, and delay are interrelated and an improvement in one
   generally comes at the expense of the others for a given technology.
   Research is required to find appropriate operating points in networks
   using transmission components which offer extremely high bandwidth
   with very good bit-error-rate performance.


   3.3.3.  Network Adaptors




   With the promised speed of networks, the future network adaptor must
   be viewed as a memory interconnect, tying the memory in one host to
   another, at least if the data rate and the low latency made possible
   by the network is to be realized at the host-to-host or process-to-
   process level.  The challenge is too great to be met by just
   implementing protocols in custom VLSI.

   Research is required to investigate the impact of network
   interconnection on a machine architecture and to define and evaluate
   new network adaptor architectures.  Of key importance is integration
   of network adaptor into the operating system so that process-to-
   process communications performance matches that offered by the
   network.  In particular, we conjecture that the transport level will
   be implemented largely, if not entirely, in the network adaptor,
   providing the host with reliable memory-to-memory transfer at memory
   speeds with a minimum of interrupt processing bus overhead and packet
   processing.

   Drawing an analogy to RISC technology again, maximal performance
   requires a well-designed and coordinated protocol, software, and
   hardware (network adaptor) design.  Current standard protocols are
   significantly flawed for hardware compatibility, suggesting a need
   for considerable further research on high-performance protocol
   design.


   3.3.4.  Host Operating System Software




   Conventionally, communication has been an add-on to an operating
   system.  With the GN, the network may well become the fastest
   "peripheral" connected to most nodes.  High-performance process-to-
   process (or application to application) communication will not be
   achieved until the operating system is well designed for fast access
   to and from the network.  For example, incorporating templates of the
   network packet header directly in the process descriptor may allow a



Gigabit Working Group                                          [Page 26]

RFC 1077                                                   November 1988


   process to initiate communications with minimal overhead.  Similarly,
   memory mapping can be used to eliminate copies between data arriving
   from the network and it being delivered to the applications.  With a
   GN, an extra copy forced by the operating system may easily double
   the perceived transfer time for a packet between applications.

   Besides matching data transfer mechanisms, operating systems must be
   well-matched in security design to that supported by the host
   interface and network as well.  Otherwise, all but the most trivial
   additional security actions by the operating system in common case
   communication can easily eliminate the performance benefits of the
   GN.  For example, if the host has to do further encryption or
   decryption, the throughput is likely to be at least halved and the
   latency doubled.

   Research effort is required to further refine operating systems for
   the level of performance offered by the GN.  This effort may well be
   best realized with coupling existing efforts in distributed systems
   with the GN activities, as opposed to starting new separate efforts.


   3.4.  Advanced Network Management Algorithms




   An important emphasis for research into network management should be
   on decentralized approaches.  The ratio of propagation delay across
   the country to data rates in a GN appear to be too great to deal
   effectively with resource management centrally when traffic load is
   bursty and unstable (and if it is not, one might argue there is no
   problem).  In addition, important principles of fault containment and
   minimal privilege for reliability and security suggest that a
   centralized management approach is infeasible.  In particular,
   compromising the security of one portion of the network should not
   compromise the security of the whole network.  Similarly, a failure
   or fault should affect at most a local region of the network.

   The challenge is clearly to provide decentralized management
   techniques that lead to good global behavior in the normal case and
   acceptable behavior in expected worst-case failures, traffic
   variations and security intrusions.


   3.4.1.  Control Flow vs. Data Flow




   Network operational communications can be separated into flow of user
   data and flow of management/control data.  However, the user data
   must contain some amount of control data.  One question that needs to



Gigabit Working Group                                          [Page 27]

RFC 1077                                                   November 1988


   be explored in light of changes in communications and computing costs
   and performance is the trade-off between these two flows.  An example
   of a potential approach is to use data units which contain predefined
   path indicators.  The switch can perform a simple table look-up which
   maps the path indicator onto the preferred outbound link and
   transmits the packet immediately.  There is a path set-up packet
   which fills in the appropriate tables.  Path set-up occurs before the
   first data packet flows and then, while data is flowing, to improve
   the routes during the lifetime of the connection.  This concept has
   been discussed in the Internet engineering group under the name of
   soft connections.

   We note that separating the data flow from the control flow in the GN
   has security and reliability advantages as well.  We could encrypt
   most of the packet header to provide confidentiality within the GN
   and to limit the ability of intruders to perform traffic analysis.
   And, by separating the control flow, we can encrypt all the control
   exchanges between switches and the host front ends thereby offering
   confidentiality and integrity.  No unauthorized entity will be able
   to alter or examine the control traffic.  By employing a path set-up
   procedure, we can assure that the GN NFE-to-NFE path is functioning
   and also include user-specific requirements in the route.  For
   example, we could request a certain bandwidth allocation and simplify
   the job of the switches in handling flow control.  We could also set
   up backup paths in case the output link will be busy for so many
   microseconds that the packet cannot be stored until the link is
   freed.


   3.4.2.  Resource Management Algorithms




   Most current networks deliver one quality of service.  X.25 networks
   deliver a reliable byte-stream.  Most LANs deliver a best-effort
   unreliable service.  There are few networks today that can support
   multiple types of service, and allocate their resources among them.
   Indeed, for many networks, such as best-effort unreliable service,
   there is little management of resources at all.  The next generation
   of network will require a much more controlled allocation of
   resources.

   There will be a much wider range of desired types of service, with
   current services such as remote procedure call mixing with new
   services such as video streams.  Unless these are separately
   recognized and controlled, there is little reason to believe that
   effective service can be delivered unless the network is very lightly
   loaded.




Gigabit Working Group                                          [Page 28]

RFC 1077                                                   November 1988


   In order to support multiple types of service, two things must
   happen, both a change from current practice.  First, the application
   must describe to the network what type of service is required.
   Second, the network must use this information to make resource
   allocation decisions.  Both of these practices present difficulties.

   Past experience suggests that application code is not prepared to
   know or specify what service it needs.  By custom, operating systems
   provide a virtual world, and the applications in this world are
   unaware of the relation between this and the reality of time and
   space.  Resource requests must be in real terms.  Allocation of
   resources in the network is difficult, because it requires that
   decisions be made in the network, but as network packet throughput
   increases, there is less time for decisions.

   The resolution of this latter conflict is to observe that decisions
   must be made on larger units than the unit of multiplexing such as
   the packet.  This in turn implies that packets must be visible to the
   network as being part of a sequence, as opposed to the pure datagram
   model previously exploited.  As suggested earlier in this report,
   research is required to support this more complex form of switch
   without compromising robustness.

   To permit the application to specify the service it needs, it will be
   necessary to propose some abstraction of service class.  By clever
   design of this abstraction, it should be possible to allow the
   application to describe its needs effectively.  For example, an
   application such as file transfer or mail has two modes of operation;
   bulk data transfer and remote procedure call.  The application may
   not be able to predict when it will be in which mode, but if it just
   describes both of them, the system may be able to adapt by observing
   its current operation.

   Experimentation needs to be done to determine a suitable service
   specification interface.  This experimentation could be done in the
   context of the current protocols, and could thus be undertaken at
   once.


   3.4.3.  Adaptive Protocols




   Network operating conditions can vary quickly and over a wide range.
   This is true of the current Internet, and is likely to affect the GN
   too.  Protocols that can adapt to changing circumstances would
   provide more even and robust service than is currently possible.  For
   example, when error rates increased, a protocol implementation might
   decide to use smaller packets, thus reducing the burden caused by



Gigabit Working Group                                          [Page 29]

RFC 1077                                                   November 1988


   retransmissions.

   The environment in which a protocol operates can be described in
   terms of the service it is getting from the next lower layer.  A
   protocol implementation can adapt to changes in that service by
   tuning its internal mechanisms (time-outs, retransmission strategies,
   etc.).  Therefore, to design adaptive protocols, we must understand
   the interaction between protocol layers and the mechanisms used
   within them.  There has been some work done in this area.  For
   example, the SATNET measurement task force has looked at the
   interactions between the protocol used by the SIMP, IP, and TCP.
   What is needed is a more complete characterization of the
   interactions at various layer boundaries, and the development of
   appropriate protocol designs and mechanisms to provide for necessary
   adaptations and renegotiations.


   3.4.4.  Error Recovery Mechanisms




   Being large and complex, GNs will experience a variety of faults such
   as link or nodal failure, excessive buffer overflow due to faulty
   flow and congestion control, and partial failure of switching fabric.
   These failures, which also exist in today's networks, will have a
   stronger effect in GNs where a large amount of data will be "stored"
   in transit and, to expedite the switching, nodes will apply only
   minimal processing to the packets traversing them.  In source
   routing, for example, a link failure may cause the loss of all
   packets sent until the source is notified about the change in
   topology.  The longer is the delay in recovering from failures, the
   higher is the degradation in performance observed by the users.

   To minimize the effects of failures, GNs will need to employ error
   recovery mechanisms whereby the network detects failures and error
   conditions, reconfigures itself to adapt to the new network state,
   and notifies peripheral devices of the new configuration.  Such
   protocols, which have to be developed, will respond quickly, will be
   decentralized or distributed to minimize the possibility of fatal
   failures, and will complement, rather than replicate, the error
   correction mechanisms of the end-to-end protocols, and the two must
   operate in coordinated manner.  To this end, the peripheral devices
   will have to be knowledgeable about the intranet recovery mechanisms
   and interact continuously with them to minimize the effect on the
   connections they manage.







Gigabit Working Group                                          [Page 30]

RFC 1077                                                   November 1988


   3.4.5.  Flow Control




   As networks become faster, two related problems arise.  First,
   existing flow control mechanisms such as windows do not work well,
   because the window must be opened to such an extent to achieve
   desired bandwidth that effective flow control cannot be achieved.
   Second, especially for long-haul networks, the larger number of bits
   in transit at one time becomes so large that most computer messages
   will fit into one window.  This means that traditional congestion
   control schemes will cease to work well.

   What is needed is a combination of two approaches, both new.  First,
   for messages that are small (most messages generated by computers
   today will be small, since they will fit into one round-trip time of
   future networks), open-loop controls on flow and congestion are
   needed.  For longer messages (voice or video streams, for example),
   some explicit resource commitment will be required.


   3.4.6.  Latency Control and Real-Time Operations




   Currently, there are several distinct approaches to latency control.
   First, there are some networks which are physically short, more like
   multiprocessor buses.  Applications in these networks are built
   assuming that delays will be short.

   Second, there are networks where the physical length is not
   constrained by the design and may differ by orders of magnitude,
   depending on the scope of the network.  Most general purpose networks
   fall in this category.  In these networks, one of two things happens.
   Either the application takes special steps to deal with variable
   latency, such as echo suppression in voice networks, or these
   applications are not supported.

   For most applications today, the latency in the network is not an
   obvious issue so long as the network is not overloaded (which leads
   to losses and long queues), because the protocol overhead masks the
   variation in the network latency.  This balance will change.  The
   latency due to the speed of light will obviously remain the same, but
   the overhead will drop (of necessity if we are to achieve high
   performance) which will leave speed of light and queueing as the most
   critical sources of delay.

   This conclusion implies that if queueing delay can be controlled, it
   will be possible to build networks with stable and controlled
   latency.  If applications exist that require this class of service,



Gigabit Working Group                                          [Page 31]

RFC 1077                                                   November 1988


   it can be supported.  Either the network must be underloaded, so that
   queues do not develop at all, or a specific class of service must be
   supported in which resources are allocated to stabilize the delay.

   If this service is provided, it will still leave the application with
   delays that can vary by several orders of magnitude, depending on the
   physical size of the network.  Research at the application level will
   be required to see how applications can be designed to cope with this
   variation.


   3.4.7.  High-Speed Internetworking and Administrational Domains




   Internetworking recognized that the value of communication services
   increases significantly with wider interconnection but ignored
   management and the role of administrations.  As a consequence we see
   that:

      1.  The Internet is more or less unmanageable, as evidenced by
          performance, reliability, and security problems.

      2.  The Internet is being stressed by administrators that are
          building networks to match their organization rather than the
          geography.  An example is a set of Ethernets at different
          company locations operating as a single Internet network but
          geographically dispersed and connected by satellite or leased
          lines.

   The next generation of internetworking must focus on administration
   and management.  Internetworking must support cohesion within an
   administration and a healthy separation between administrations.  To
   illustrate by analogy, the American and Soviet embassies in Mexico
   City are geographically closer to each other than to their respective
   home countries but further in administrational distance, including
   security, accounting, etc.  The emerging revolution in WANs makes
   this issue that much more critical.  The amount of communication to
   exchange the state of systems is bound to increase enormously.  The
   potential cost of failures and security violations is frightening.

   A promising approach appears to be high-level gateways that guard
   between administrations and require negotiations to set up access
   paths between administrations.  These paths are set up, and labeled
   with agreements on authorization, security, accounting, and possible
   resource limits.  These administrative virtual circuits provide
   transparency to the physical and geographical interconnection, but
   need not support more than datagram packet delivery.  One view is
   that of communication contracts with high-level gateways acting as



Gigabit Working Group                                          [Page 32]

RFC 1077                                                   November 1988


   contract monitors at each end.  The key is the focus on controlled
   interadministrational connectivity, not the conventional protocol
   concerns.

   Focus is required on developing an (inter)network management
   architecture and the specifics of high-level gateways.  The
   structures of such gateways will have to take advantage of advances
   in multi-processor architectures to handle the processing load.
   Moreover, a key issue is being able to optimize communication between
   administrations once the contract is in place, but without losing
   control.  Related is the issue of allowing high-speed interconnection
   within a single administration, although geographical dispersed.
   Another issue is fault-tolerance.  High-level gateways contain state
   information whose loss typically disrupts communication.  How does
   one minimize this problem?

   A key goal of these administrational gateways has to be failure
   containment: How to protect against external (to administration)
   problems and how to prevent local problems imposing liability on
   others.  A particular area of concern is the self-organizing problems
   of large-scale systems, observed by Van Jacobson in the Internet.
   Gateways must serve to damp out oscillations and control wide load
   swings.  Rate control appears to be a key area to investigate as a
   basis for buffer management and for congestion control, as well as to
   control offered load.

   Given the speed of new networks, and the sophistication of the
   gateways suggested above, another key area to investigate is the
   provision of high-speed network interface adaptors.


   3.4.8.  Policy-Based Algorithms




   Networks of today generally select routes based on minimizing some
   measure such as delay.  However, in the real world, route selection
   will commonly be constrained at the global level by policy issues,
   such as access rights to resources and accounting and billing for
   usage.

   It is difficult for connectionless protocols such as Internet to deal
   with policy controls, because a lack of state in the gateway implies
   that a separate policy decision must be made for each packet in
   isolation.  As networks get faster, the cost of this processing will
   be intolerable.  One possible approach, discussed above, is to move
   to a more sophisticated model in which there is knowledge in the
   gateways of the ongoing flows.  Alternatively, it may be possible to
   design gateways that simply cache recent policy evaluations and apply



Gigabit Working Group                                          [Page 33]

RFC 1077                                                   November 1988


   them to successive packets.

   Routing based on policy is particularly difficult because a route
   must be globally consistent to be useful; otherwise it may loop.
   This implies that the every policy decision must be propagated
   globally.  Since there can be expected to be a large number of
   policies, this global passing of information might easily lead to an
   information explosion.

   There are at least two solutions.  One is to restrict the possible
   classes of policy.  Another is to use some form of source route, so
   that the route consistent with some set of policies is computed at
   one point only, and then attached to the packet.  Both of these
   approaches have problems.  A two-pronged research program is needed,
   in which mechanisms are proposed, and at the same time the needed
   policies are defined.

   The same trade-off can be seen for accounting and billing.  A single
   accounting metric, such as "bytes times distance", could be proposed.
   This might be somewhat simple to implement, but would not permit the
   definition of individual billing policies, as is now done in the
   parts of the telephone system.  The current connectionless transport
   architectures such as TCP/IP or the connectionless ISO configuration
   using TP4 do not have good tools for accounting for traffic, or for
   restricting traffic from certain resources.  Building these tools is
   difficult in a connectionless environment, because an accounting or
   control facility must deal with each packet in isolation, which
   implies a significant processing burden as part of packet forwarding.
   This burden is an increasing problem as switches are expected to
   operate faster.

   The lack of these tools is proving a significant problem for network
   design.  Not only are accounting and control needed to support
   management requirements, they are needed as a building block to
   support enforcement of such things as multiple qualities of service,
   as discussed above.

   Network accounting is generally considered to be simply a step that
   leads to billing, and thus is often evaluated in terms of how simple
   or difficult it will be to implement.  Yet an accounting and billing
   procedure is a mechanism for implementing a policy considered to be
   desirable for reasons beyond the scope of accounting per se.  For
   example, a policy might be established either to encourage or
   discourage network use, while fully recovering operational cost.  A
   policy of encouraging use could be implemented by a relatively high
   monthly attachment charge and a relatively low per-packet charge.  A
   policy of discouraging use could be implemented by a low monthly
   charge and a high per-packet charge.



Gigabit Working Group                                          [Page 34]

RFC 1077                                                   November 1988


   Network administrators have a relatively small number of variables
   with which to implement policy objectives.  Nevertheless, these
   variables can be combined in a number of innovative ways.  Some of
   the possibilities include:

      1.  Classes of users (e.g., large or small institutions, for-
          profit or non-profit).

      2.  Classes of service.

      3.  Time varying (e.g., peak and off-peak).

      4.  Volume (e.g., volume discounts, or volume surcharges).

      5.  Access charges (e.g., per port, or port * [bandwidth of
          port]).



      6.  Distance (e.g., circuit-miles, airline miles, number of hops).



   Generally, an accounting procedure can be developed to support
   voluntary user cooperation with almost any single policy objective.
   Difficulties most often arise when there are multiple competing
   policy objectives, or when there is no clear policy at all.

   Another aspect of accounting and billing procedures which must be
   carefully considered is the cost of accumulating and processing the
   data on which billing is based.  Of particular concern is collection
   of detailed data on a per-packet basis.  As network circuit data
   rates increase, the number of instructions which must be executed on
   a per-packet basis can become the limiting factor in system
   throughput.  Thus, it may be appropriate to prefer accounting and
   billing policies and procedures which minimize the difficulty of
   collecting data, even if this approach requires a compromise of other
   objectives.  Similarly, node memory required for data collection and
   any network bandwidth required for transmission of the data to
   administrative headquarters are factors which must be traded off
   against the need to process user packets.


   3.4.9.  Priority and Preemption




   The GN should support multiple levels of priority for traffic and the
   preemption of network resources for higher priority use.  Network
   control traffic should be given the highest priority to ensure that
   it is able to pass through the network unimpeded by congestion caused
   by user-level traffic.  There may be additional military uses for
   multiple levels of priority which correspond to rank or level of



Gigabit Working Group                                          [Page 35]

RFC 1077                                                   November 1988


   importance of a user or the mission criticality of some particular
   data.

   The use of and existence of priority levels may be different for
   different types of traffic.  For example, datagram traffic may not
   have multiple priority levels.  Because the network's transmission
   speed is so high and traffic bursts may be short, it may not make
   sense to do any processing in the switches to deal with different
   priority levels.  Priority will be more important for flow- (or
   soft-connection-) oriented data or hard connections in terms of
   permitting higher priority connections to be set up ahead of lower
   priority connections.  Preemption will permit requests for high
   priority connections to reclaim network resources currently in use by
   lower priority traffic.

   Networks such as the Wideband Satellite Network, which supports
   datagram and stream traffic, implement four priority levels for
   traffic with the highest reserved for network control functions and
   the other three for user traffic.  The Wideband Network supports
   preemption of lower priority stream allocations by higher priority
   requests.  An important component of the use of priority and
   preemption is the ability to notify users when requests for service
   have been denied, or allocations have been modified or disrupted.
   Such mechanisms have been implemented in the Wideband Network for
   streams and dynamic multicast groups.

   Priority and preemption mechanisms for a GN will have to be
   implemented in an extremely simple way so that they can take effect
   very quickly.  It is likely that they will have to built into the
   hardware of the switch fabric.


   3.5.  User and Network Services




   As discussed in Section 2 above, there will need to be certain
   services provided as part of the network operation to the users
   (people) themselves and to the machines that connect to the network.
   These services, which include such capabilities as white and yellow
   pages (allowing users to determine what the appropriate network
   identification is for other users and for network-available computing
   resources) and distributed fault identification and isolation, are
   needed in current networks and will continue to be required in the
   networks of the future.  The speed of the GN will serve to accentuate
   this requirement, but at the same time will allow for new
   architectures to be put in place for such services.  For example,
   Ethernet speeds in the local environment have allowed for more usable
   services to be provided.



Gigabit Working Group                                          [Page 36]

RFC 1077                                                   November 1988


   3.5.1.  Impact of High Bandwidth




   One issue that will need to be addressed is the impact on the user of
   such high-bandwidth capabilities.  Users are already becoming
   saturated by information in the modern information-rich environment.
   (Many of us receive more than 50 electronic mail messages each day,
   each requiring some degree of human attention.) Methods will be
   needed to allow users to cope with this ever-expanding access to
   data, or we will run the risk of users turning back to the relative
   peace and quiet of the isolated office.


   3.5.2.  Distributed Network Directory




   A distributed network directory can support the user-level directory
   services and the lower-level name-to-address mapping services
   described elsewhere in this report.  It can also support distributed
   systems and network management facilities by storing additional
   information about named objects.  For example, the network directory
   might store node configurations or security levels.

   Distributing the directory eases and decentralizes the administrative
   burdens and provides a more robust and survivable implementation.

   One approach toward implementing a distributed network directory
   would be to base it upon the CCITT X.500/ISO DIS 9594 standard.  This
   avoids starting from ground zero and has the advantage of
   facilitating interoperability with other communications networks.
   However, research and development will be required even if this path
   is chosen.

   One area in which research and development are required is in the
   services supplied by the distributed network directory.  The X.500
   standard is very general and powerful, but so far specific provisions
   have been made only for storing information about network users and
   applications.  As mentioned elsewhere, multilevel security is not
   addressed by X.500, and the approach taken toward authentication must
   be carefully considered in view of DoD requirements.  Also, X.500
   assumes that administration of the directory will be done locally and
   without the need for standardization; this may not be true of GN or
   the larger national research network.

   The model and algorithms used by a distributed network directory
   constitute a second area of research.  The model specified by X.500
   must be extended into a framework that provides the necessary
   flexibility in terms of services, responsiveness, data management



Gigabit Working Group                                          [Page 37]

RFC 1077                                                   November 1988


   policies, and protocol layer utilization.  Furthermore, the internal
   algorithms and mechanisms of X.500 must be extended in a number of
   areas; for example, to support redundancy of the X.500 database,
   internal consistency checking, fuller sharing of information about
   the distribution of data, and defined access-control mechanisms.


   4.  Avenues of Approach


   Ongoing research and commercial activities provide an opportunity for
   more rapidly attacking some of the above research issues.  At the
   same time, there needs to be attention paid to the overall technical
   approach used to allow multiple potential solutions to be explored
   and allow issues to be attacked in parallel.


   4.1.  Small Prototype vs. Nationwide Network




   The central question is how far to jump, and how far can the current
   approaches get.  That is, how far will connectionless network service
   get us, how far will packet switching get us, and how far do we want
   to go.  If our goal is a Gbit/s net, then that is what we should
   build.  Building a 100 Mbit/s network to achieve a GN is analogous to
   climbing a tree to get to the moon.  It may get you closer, but it
   will never get you there.

   There are currently some network designs which can serve as the basis
   for a GN prototype.  The next step is some work by experts in
   photonics and possibly high-speed electronics to explore ease of
   implementation.  Developing a prototype 3-5 node network at a Gbit/s
   data rate is realistic at this point and would demonstrate wide-area
   (40 km or more) Gbit/s networking.

   DARPA should consider installing a Gbit/s cross-country set of
   connected links analogous to the NSF backbone in 2 years.  A Gbit/s
   link between the east and west coasts would open up a whole new
   generation of (C3I), distributed computing, and parallel computing
   research possibilities and would reestablish DARPA as the premier
   network research funding agency in the country.  This will require
   getting "dark" fiber from one or more of the common carriers and some
   collaboration with these organizations on repeaters, etc.  With this
   collaboration, the time to a commercial network in the Gbit/s range
   would be substantially reduced, and the resulting nationwide GN would
   give the United States an enormous technical and economic advantage
   over countries without it.




Gigabit Working Group                                          [Page 38]

RFC 1077                                                   November 1988


   Demonstrating a high-bandwidth WAN is not enough, however.  As one
   can see from the many research issues identified above, it will be
   necessary to pursue via study and experiment the issues involved in
   interconnecting high-bandwidth networks into a high-bandwidth
   internet.  These experiments can be done through use of a new
   generation of internet, even if it requires starting at lower speeds
   (e.g., T1 through 100 Mbit/s).  Appropriate care must be given,
   however, to assure that the capabilities that are demonstrated are
   applicable to the higher bandwidths (Gbit/s) as they emerge.


   4.2.  Need for Parallel Efforts/Approaches




   Parallel efforts will therefore be required for two major reasons.
   First is the need to pursue alternative approaches (e.g., different
   strategies for high-bandwidth switching, different addressing
   techniques, etc).  This is the case for most research programs, but
   it is made more difficult here by the costs of prototyping.  Thus, it
   is necessary that appropriate review take place in the decisions as
   to which efforts are supported through prototyping.

   In addition, it will be necessary to pursue the different aspects of
   the program in parallel.  It will not be possible to wait until the
   high-bandwidth network is available before starting on prototyping
   the high-bandwidth internet.  Thus, a phased and evolutionary
   approach will be needed.


   4.3.  Collaboration with Common Carriers




   Computer communication networks in the United States today
   practically ignore the STN (the Switched Telephone Network), except
   for buying raw bandwidth through it.  However, advances in network
   performance are based on improvements in the underlying communication
   media, including satellite communication, fiber optics, and photonic
   switching.

   In the past we used "their" transmission under "our" switching.  An
   alternative approach is to utilize the common-carrier switching
   capabilities as an integral part of the networking architecture.  We
   must take an objective scientific and economic look and reevaluate
   this question.

   Another place for cooperation with the common carriers is in the area
   of network addressing.  Their addressing scheme ("numbering plan")
   has a few advantages such as proven service to 300 million users [4].



Gigabit Working Group                                          [Page 39]

RFC 1077                                                   November 1988


   On the other hand, the common carriers have far fewer administrative
   domains (area codes) than the current plethora of locally
   administered local area networks in the internet system.

   It is likely that future networks will eventually be managed and
   operated by commercial communications providers.  A way to maximize
   technology transfer from the research discussed here to the
   marketplace is to involve the potential carriers from the start.
   However, it is not clear that the goals of commercial communications
   providers, who have typically been most interested in meeting the
   needs of 90+ percent of the user base, will be compatible with the
   goals of the research described here.  Thus, while we recommend that
   the research program involve an appropriate amalgam of academia and
   industry, paying particular attention to involvement of the potential
   system developers and operators, we also caution that the specific
   and unique goals of the DARPA program must be retained.


   4.4.  Technology Transfer




   As we said above, it is our belief that future networks will
   ultimately be managed and operated by commercial communications
   providers.  (Note that this may not be the common carriers as we know
   them today, but may be value-added networks using common carrier
   facilities.) The way to assure technology transfer, in our belief, is
   to involve the potential system developers from the start.  We
   therefore believe that the research program would benefit from an
   appropriate amalgam of university and industry, with provision for
   close involvement of the potential system developers and operators.


   4.5.  Standards




   The Internet program was a tremendous success in influencing national
   and international standards.  While there were changes to the
   protocols, the underlying technology and approaches used by CCITT and
   ISO in the standardization of packet-switched networks clearly had
   its roots in the DARPA internet.  Nevertheless, this has had some
   negative impact on the research program, as the evolution of the
   standards led to pressure to adopt them in the research environment.

   Thus, it appears that there is a "catch-22" here.  It is desirable
   for the technology base developed in the research program to have
   maximal impact on the standards activities.  This is expedited by
   doing the research in the context of the standards environment.
   However, standards by their very nature will always lag behind the



Gigabit Working Group                                          [Page 40]

RFC 1077                                                   November 1988


   research environment.

   The only reasonable approach, therefore, appears to be an occasional
   "checkpointing" of the research environment, where the required
   conversions take place to allow a new plateau of standards to be used
   for future evolution and research.  A good example is conducting
   future research in mail using X.400 and X.500 where possible.


   5.  Conclusions


   We hope that this document has provided a useful compendium of those
   research issues critical to achieving the FCCSET phase III
   recommendations.  These problems interact in a complex way.  If the
   only goal of a new network architecture was high speed, reasonable
   solutions would not be difficult to propose.  But if one must achieve
   higher speeds while supporting multiple services, and at the same
   time support the establishment of these services across
   administrative boundaries, so that policy concerns (e.g., access
   control) must be enforced, the interactions become complex.






























Gigabit Working Group                                          [Page 41]

RFC 1077                                                   November 1988


                                 APPENDIX

A. Current R and D Activities



   In this appendix, we provide pointers to some ongoing activities in
   the research and development community of which the group was aware
   relevant to the goal of achieving the GN.  In some cases, a short
   abstract is provided of the research.  Neither the order of the
   listing (which is random) nor the amount of detail provided is meant
   to indicate in any way the significance of the activity.  We hope
   that this set of pointers will be useful to anyone who chooses to
   pursue the research issues discussed in this report.

      1.  Grumman (at Bethpage) is working on a three-year DARPA
          contract, started in January 1988 to develop a 1.6 Gbit/s LAN,
          for use on a plane or ship, or as a "building block".  It is
          really raw transport capacity running on two fibers in a
          token-ring like mode.  First milestone (after one year?) is to
          be a 100 Mbit/s demonstration.

      2.  BBN Laboratories, as part of its current three-year DARPA
          Network-Oriented Systems contract, has proposed design
          concepts for a 10-100 Gbit/s wide area network.  Work under
          this effort will include wavelength division multiplexing,
          photonic switching, self-routing packets, and protocol design.

      3.  Cheriton (Stanford) research on Blazenet, a high-bandwidth
          network using photonic switching.

      4.  Acampora (Bell Labs) research on the use of wavelength
          division multiplexing for building a shared optical network.

      5.  Yeh is reserching a VLSI approach to building high-bandwidth
          parallel processing packet switch.

      6.  Bell Labs is working on a Metropolitan Area Network called
          "Manhattan Street Net."  This work, under Dr. Maxemchuck, is
          similar to Blazenet.  It is in the prototype stage for a small
          number of street intersections; ultimately it is meant to be
          city-wide.  Like Blazenet, is uses photonic switching 2 x 2
          lithium niobate block switches.

      7.  Ultra Network Technologies is a Silicon Valley company which
          has a (prototype) Gbit/s fiber link which connects backplanes.
          This is based on the ISO-TP4 transport protocol.

      8.  Jonathan Turner, Washington University, is working on a
          Batcher-Banyan Multicast Net, based on the "SONET" concept,





Gigabit Working Group                                          [Page 42]

RFC 1077                                                   November 1988


          which provides 150 Mbit/s per pipe.

      9.  David Sincowskie, Bellcore, is working with Batcher-Banyan
          design and has working 32x32 switches.



      10. Stratacom has a commercial product which is really a T1 voice
          switch implemented internally by a packet switch, where the
          packet is 192 bits (T1 frame).  This switch can pass 10,000
          packets per second.

      11. Stanford NAB provides 30-50 Mbit/s throughput on 100 Mbit/s
          connection using Versatile Message Transaction Protocol (VMTP)
          [see RFC 1045]



      12. The December issue of IEEE Journal on Selected Areas in
          Communications, provides much detail concerning interconnects.



      13. Ultranet Technology has a 480 Mbit/s connection using modified
          ISO TP4.



      14. At MIT, Dave Clark has an architecture proposal of interest.



      15. At CMU, the work of Eric Cooper is relevant.



      16. At Protocol Engines, Inc., Greg Chesson is working on an XTP-
          based system.



      17. Larry Landweber at Wisconsin University is doing relevant
          work.



      18. Honeywell is doing relevant work for NASA.



      19. Kung at CMU is working on a system called "Nectar" based on a
          STARLAN on fiber connecting dissimilar processors.



      20. Burroughs (now Unisys) has some relevant work within the IEEE
          802.6 committee.



      21. Bellcore work in "Switched Multimedia Datanet Service" (SMDS)
          is relevant (see paper supplied by Dave Clark).



      22. FDDI-2, a scheme for making TDMA channel allocations at 200
          Mbit/s.



      23. NRI, Kahn-Farber Proposal to NSF, is a paper design for high-
          bandwidth network.



      24. Barry Goldstein work, IBM-Yorktown.





Gigabit Working Group                                          [Page 43]

RFC 1077                                                   November 1988


      25. Bell Labs S-Net, 1280 Mbit/s prototype.



      26. Fiber-LAN owned by Bell South and SECOR, a pre-prototype 575
          Mbit/s Metro Area Net.



      27. Bellcore chip implementation of FASTNET (1.2 Gbit/s).



      28. Scientific Computer Systems, San Diego, 1.4 Gbit/s prototype.



      29. BBN Monarch Switch, Space Division pre-prototype, chips being
          fabricated, 64 Mbit/s per path.



      30. Proteon, 80 Mbit/s token ring.



      31. Toronto University, 150 Mbit/s "tree"--- really a LAN.



      32. NSC Hyperchannel II, reputedly available at 250 Mbit/s.



      33. Tobagi at Stanford working on EXPRESSNET; not commercially
          available.



      34. Columbia MAGNET-- 150 Mbit/s.



      35. Versatile Message Transaction Protocol (VMTP).



      36. ST integrated with IP.



      37. XTP (Chesson).



      38. Stanford Transport Gateway.



      39. X.25/X.75.



      40. Work of the Internet Activities Board.



















Gigabit Working Group                                          [Page 44]

RFC 1077                                                   November 1988


B. Gigabit Working Group Members



Member                  Affiliation

Gordon Bell             Ardent Computers
Steve Blumenthal        BBN Laboratories
Vint Cerf               Corporation for National Research Initiatives
David Cheriton          Stanford University
David Clark             Massachusetts Institute of Technology
Barry Leiner (Chairman) Research Institute for Advanced Computer Science
Robert Lyons            Defense Communication Agency
Richard Metzger         Rome Air Development Center
David Mills             University of Delaware
Kevin Mills             National Bureau of Standards
Chris Perry             MITRE
Jon Postel              USC Information Sciences Institute
Nachum Shacham          SRI International
Fouad Tobagi            Stanford University

































Gigabit Working Group                                          [Page 45]

RFC 1077                                                   November 1988


End Notes

     [1] Workshop on Computer Networks, 17-19 February 1987, San Diego,
         CA.

     [2] "A Report to the Congress on Computer Networks to Support
         Research in the United States: A Study of Critical Problems and
         Future Options", White House Office of Scientific and Technical
         Policy (OSTP), November 1987.

     [3] We distinguish in the report between development of a backbone
         network providing gigabit capacity, the GB, and an
         interconnected set of high-speed networks providing high-
         bandwidth service to the user, the Gigabit Network (GN).

     [4] Incidentally, they already manage to serve 150 million
         subscribers in an 11-digit address-space (about 1:600 ratio).
         We have a 9.6-digit address-space and are running into troubles
         with much less than 100,000 users (less than 1:30,000 ratio).
































Gigabit Working Group                                          [Page 46]