RFC 1126

Network Working Group                                          M. Little
Request for Comments:  1126                                         SAIC
                                                            October 1989

                 Goals and Functional Requirements for
                    Inter-Autonomous System Routing

Status of this Memo

   This document describes the functional requirements for a routing
   protocol to be used between autonomous systems.  This document is
   intended as a necessary precursor to the design of a new inter-
   autonomous system routing protocol and specifies requirements for the
   Internet applicable for use with the current DoD IP, the ISO IP, and
   future Internet Protocols.  It is intended that these requirements
   will form the basis for the future development of a new inter-
   autonomous systems routing architecture and protocol.  This document
   is being circulated to the IETF and Internet community for comment.
   Comments should be sent to: "open-rout-editor@bbn.com".  This memo
   does not specify a standard.  Distribution of this memo is unlimited.

1.  Introduction

   The development of an inter-autonomous systems routing protocol
   proceeds from those goals and functions seen as both desirable and
   obtainable for the Internet environment.  This document describes
   these goals and functional requirements.  The goals and functional
   requirements addressed by this document are intended to provide a
   context within which an inter-autonomous system routing architecture
   can be developed which will meet both current and future Internet
   routing needs.  The goals presented indicate properties and general
   capabilities desired of the Internet routing environment and what the
   inter-autonomous system routing architecture is to accomplish as a

   The goals are followed by functional requirements, which address
   either detailed objectives or specific functionality to be achieved
   by the architecture and resulting protocol(s).  These functional
   requirements are enumerated for clarity and grouped so as to map
   directly to areas of architectural consideration.  This is followed
   by a listing and description of general objectives, such as
   robustness, which are applicable in a broad sense.  Specific
   functions which are not reasonably attainable or best left to future
   efforts are identified as non-requirements.

   The intent of this document is to provide both the goals and
   functional requirements in a concise fashion.  Supporting arguments,

Little                                                          [Page 1]

RFC 1126            Inter-Autonomous System Routing         October 1989

   tradeoff considerations and the like have been purposefully omitted
   in support of this.  An appendix has been included which addresses
   this omission to a limited extent and the reader is directed there
   for a more detailed discussion of the issues involved.

   The goals and functional requirements contained in this document are
   the result of work done by the members of the Open Routing Working
   Group.  It is our intention that these goals and requirements reflect
   not only those foreseen in the Internet community but are also
   similar to those encountered in environments proposed by ANSI, ECMA
   and ISO.  It is expected that there will be some interaction and
   relationship between this work and the product of these groups.

2.  Overall Goals

   In order to derive a set functional requirements there must be one or
   more principals or overall goals for the routing environment to
   satisfy.  These high level goals provide the basis for each of the
   functional requirements we have derived and will guide the design
   philosophy for achieving an inter-autonomous system routing solution.
   The overall goals we are utilizing are described in the following

2.1  Route to Destination

   The routing architecture will provide for the routing of datagrams
   from a single source to one or more destinations in a timely manner.
   The larger goal is to provide datagram delivery to an identifiable
   destination, one which is not necessarily immediately reachable by
   the source.  In particular, routing is to address the needs of a
   single source requiring datagram delivery to one or more
   destinations.  The concepts of multi-homed hosts and multicasting
   routing services are encompassed by this goal.  Datagram delivery is
   to be provided to all interconnected systems when not otherwise
   constrained by autonomous considerations.

2.2  Routing is Assured

   Routing services are to be provided with assurance, where the
   inability to provide a service is communicated under best effort to
   the requester within an acceptable level of error.  This assurance is
   not to be misconstrued to mean guaranteed datagram delivery nor does
   it imply error notification for every lost datagram.  Instead,
   attempts to utilize network routing services when such service cannot
   be provided will result in requester notification within a reasonable
   period given persistent attempts.

Little                                                          [Page 2]

RFC 1126            Inter-Autonomous System Routing         October 1989

2.3  Large System

   The design of the architecture, and the protocols within this
   architecture, should accommodate a large number of routing entities.
   The exact order of magnitude is a relative guess and the best designs
   would provide for a practical level of unbounded growth.
   Nevertheless, the routing architecture is expected to accommodate the
   growth of the Internet environment for the next 10 years.

2.4  Autonomous Operation

   The routing architecture is to allow for stable operation when
   significant portions of the internetworking environment are
   controlled by disjoint entities.  The future Internet environment is
   envisioned as consisting of a large number of internetworking
   facilities owned and operated by a variety of funding sources and
   administrative concerns.  Although cooperation between these
   facilities is necessary to provide interconnectivity, it is viewed
   that both the degree and type of cooperation will vary widely.
   Additionally, each of these internetworking facilities desires to
   operate as independently as possible from the concerns and activities
   of other facilities individually and the interconnection environment
   as a whole.  Those resources used by (and available for) routing are
   to be allowed autonomous control by those administrative entities
   which own or operate them. Specifically, each controlling
   administration should be allowed to establish and maintain policies
   regarding the use of a given routing resource.

2.5  Distributed System

   The routing environment developed should not depend upon a data
   repository or topological entity which is either centralized or
   ubiquitous.  The growth pattern of the Internet, coupled with the
   need for autonomous operation, dictates an independence from the
   topological and administrative centralization of both data and
   control flows.  Past experience with a centralized topology has shown
   that it is both impractical for the needs of the community and
   restrictive of administrative freedoms.  A distributed routing
   environment should not be restrictive of either redundancy or
   diversity.  Any new routing environment must allow for arbitrary
   interconnection between internetworks.

2.6  Provide A Credible Environment

   The routing environment and services should be based upon mechanisms
   and information that exhibit both integrity and security.  The
   routing mechanisms should operate in a sound and reliable fashion
   while the routing information base should provide credible data upon

Little                                                          [Page 3]

RFC 1126            Inter-Autonomous System Routing         October 1989

   which to base routing decisions.  The environment can be unreliable
   to the extent that the resulting effect on routing services is
   negligible.  The architecture and protocol designs should be such
   that the routing environment is reasonably secure from unwanted
   modification or influence.

2.7  Be A Managed Entity

   Provide a manger insight into the operation of the inter-autonomous
   system routing environment to support resource management, problem
   solving, and fault isolation.  Allow for management control of the
   routing system and collect useful information for the internetwork
   management environment.  Datagram events as well as the content and
   distribution characteristics of relevant databases are of particular

2.8  Minimize Required Resources

   Any feasible design should restrain the demand for resources required
   to provide inter-autonomous systems routing.  Of particular interest
   are those resources required for data storage, transmission, and
   processing.  The design must be practical in terms of today's
   technology.  Specifically, do not assume significant upgrades to the
   existing level of technology in use today for data communication

3.  Functional Requirements

   The functional requirements we have identified have been derived from
   the overall goals and describe the critical features expected of
   inter-autonomous system routing.  To an extent, these functions are
   vague in terms of detail.  We do not, for instance, specify the
   quantity or types for quality-of-service parameters.  This is
   purposeful, as the functional requirements specified here are
   intended to define the features required of the inter-autonomous
   system routing environment rather than the exact nature of this
   environment.  The functional requirements identified have been
   loosely grouped according to areas of architectural impact.

3.1  Route Synthesis Requirements

   Route synthesis is that functional area concerned with both route
   selection and path determination (identification of a sequence of
   intermediate systems) from a source to a destination.  The functional
   requirements identified here provide for path determination which is
   adaptive to topology changes, responsive to administrative policy,
   cognizant of quality-of-service concerns, and sensitive to an
   interconnected environment of autonomously managed systems.

Little                                                          [Page 4]

RFC 1126            Inter-Autonomous System Routing         October 1989

      a) Route around failures dynamically

         Route synthesis will provide a best effort attempt to detect
         failures in those routing resources which are currently being
         utilized.  Upon detection of a failed resource, route synthesis
         will provide a best effort to utilize other available routing
         resources in an attempt to provide the necessary routing

      b) Provide loop free paths

         The path provided for a datagram, from source to destination,
         will be free of circuits or loops most of the time.  At those
         times a circuit or loop exists, it occurs with both negligible
         probability and duration.

      c) Know when a path or destination is unavailable

         Route synthesis will be capable of determining when a path
         cannot be constructed to reach a known destination.
         Additionally, route synthesis will be capable of determining
         when a given destination cannot be determined because the
         requested destination is unknown (or this knowledge is

      d) Provide paths sensitive to administrative policies

         Route synthesis will accommodate the resource utilization
         policies of those administrative entities which manage the
         resources identified by the resulting path.  However, it is
         inconceivable to accommodate all policies which can be defined
         by a managing administrative entity.  Specifically, policies
         dependent upon volatile events of great celerity or those which
         are non-deterministic in nature cannot be accommodated.

      e) Provide paths sensitive to user policies

         Paths produced by route synthesis must be sensitive to policies
         expressed by the user.  These user policies are expressed in
         terms relevant to known characteristics of the topology.  The
         path achieved will meet the requirements stated by the user

      f) Provide paths which characterize user quality-of-service

         The characteristics of the path provided should match those
         indicated by the quality-of-service requested.  When

Little                                                          [Page 5]

RFC 1126            Inter-Autonomous System Routing         October 1989

         appropriate, utilize only those resources which can support the
         desired quality-of-service (e.g., bandwidth).

      g) Provide autonomy between inter- and intra-autonomous system
         route synthesis

         The inter- and intra-autonomous system routing environments
         should operate independent of one another.  The architecture
         and design should be such that route synthesis of either
         routing environment does not depend upon information from the
         other for successful functioning.  Specifically, the inter-
         autonomous system route synthesis design should minimize the
         constraints on the intra-autonomous system route synthesis
         decisions when transiting (or delivering to) the autonomous

3.2  Forwarding Requirements

   The following requirements specifically address the functionality of
   the datagram forwarding process.  The forwarding process transfers
   datagrams to intermediate or final destinations based upon datagram
   characteristics, environmental characteristics, and route synthesis

      a) Decouple inter- and intra-autonomous system forwarding

         The requirement is to provide a degree of independence between
         the inter-autonomous system forwarding decision and the intra-
         autonomous system forwarding decision within the forwarding
         process.  Though the forwarding decisions are to be independent
         of each other, the inter-autonomous system delivery process may
         necessarily be dependent upon intra-autonomous system route
         synthesis and forwarding.

      b) Do not forward datagrams deemed administratively inappropriate

         Forward datagrams according to the route synthesis decision if
         it does not conflict with known policy.  Policy sensitive route
         synthesis will prevent normally routed datagrams from utilizing
         inappropriate resources.  However, a datagram routed abnormally
         due to unknown events or actions can always occur and the only
         way to prohibit unwanted traffic from entering or leaving an
         autonomous system is to provide policy enforcement within the
         forwarding function.

Little                                                          [Page 6]

RFC 1126            Inter-Autonomous System Routing         October 1989

      c) Do not forward datagrams to failed resources

         A datagram is not to be forwarded to a resource known to be
         unavailable, notably an intermediate system such as a gateway.
         This implies some ability to detect and react to resource

      d) Forward datagram according to its characteristics

         The datagram forwarding function is to be sensitive to the
         characteristics of the datagram in order to execute the
         appropriate route synthesis decision.  Characteristics to
         consider are the destination, quality-of-service, precedence,
         datagram (or user) policy, and security.  Note that some
         characteristics, precedence for example, affect the forwarding
         service provided whereas others affect the path chosen.

3.3  Information Requirements

   This functional area addresses the general information requirements
   of the routing environment.  This encompasses both the nature and
   disbursal of routing information.  The characteristics of the routing
   information and its disbursal are given by the following functional

      a) Provide a distributed and descriptive information base

         The information base must not depend upon either centralization
         or exact replication.  The content of the information base must
         be sufficient to support all provided routing functionality,
         specifically that of route synthesis and forwarding.
         Information of particular importance includes resource
         characteristics and resource utilization policies.

      b) Determine resource availability

         Provide a means of determining the availability of any utilized
         resource in a timely manner.  The timeliness of this
         determination is dependent upon the routing service(s) to be

      c) Restrain transmission utilization

         The dynamics of routing information flow should be such that a
         significant portion of transmission resources are not consumed.
         Routing information flow should adjust to the demands of the
         environment, the capacities of the distribution facilities
         utilized, and the desires of the resource manager.

Little                                                          [Page 7]

RFC 1126            Inter-Autonomous System Routing         October 1989

      d) Allow limited information exchange

         Information distribution is to be sensitive to administrative
         policies.  An administrative policy may affect the content or
         completeness of the information distributed.  Additionally,
         administrative policy may determine the extent of information

3.4  Environmental Requirements

   The following items identify those requirements directly related to
   the nature of the environment within which routing is to occur.

      a) Support a packet-switching environment

         The routing environment should be capable of supporting
         datagram transfer within a packet-switched oriented networking

      b) Accommodate a connection-less oriented user transport service

         The routing environment should be designed such that it
         accommodates the model for connection-less oriented user
         transport service.

      c) Accommodate 10K autonomous systems and 100K networks

         This requirement identifies the scale of the internetwork
         environment we view as appearing in the future.  A routing
         design which does not accommodate this order of magnitude is
         viewed as being inappropriate.

      d) Allow for arbitrary interconnection of autonomous systems

         The routing environment should accommodate interconnectivity
         between autonomous systems which may occur in an arbitrary
         manner.  It is recognized that a practical solution is likely
         to favor a given structure of interconnectivity for reasons of
         efficiency.  However, a design which does not allow for and
         utilize interconnectivity of an arbitrary nature would not be
         considered a feasible design.

3.5  General Objectives

   The following are overall objectives to be achieved by the inter-
   autonomous routing architecture and its protocols.

      a) Provide routing services in a timely manner

Little                                                          [Page 8]

RFC 1126            Inter-Autonomous System Routing         October 1989

         Those routing services provided, encapsulated by the
         requirements stated herein, are to be provided in a timely
         manner.  The time scale for this provision must be reasonable
         to support those services provided by the internetwork
         environment as a whole.

      b) Minimize constraints on systems with limited resources

         Allow autonomous systems, or gateways, of limited resources to
         participate in the inter-autonomous system routing
         architecture.  This limited participation is not necessarily
         without cost, either in terms of responsiveness, path
         optimization, or other factor(s).

      c) Minimize impact of dissimilarities between autonomous systems

         Attempt to achieve a design in which the dissimilarities
         between autonomous systems do not impinge upon the routing
         services provided to any autonomous system.

      d) Accommodate the addressing schemes and protocol mechanisms of
         the autonomous systems

         The routing environment should accommodate the addressing
         schemes and protocol mechanisms of autonomous systems, where
         these schemes and mechanisms may differ among autonomous

      e) Must be implementable by network vendors

         This is to say that the algorithms and complexities of the
         design must be such that they can be understood outside of the
         research community and implementable by people other than the
         designers themselves.  Any feasible design must be capable of
         being put into practice.

4.  Non-Goals

   In view of the conflicting nature of many of the stated goals and the
   careful considerations and tradeoffs necessary to achieve a
   successful design, it is important to also identify those goals or
   functions which we are not attempting to achieve.  The following
   items are not considered to be reasonable goals or functional
   requirements at this time and are best left to future efforts. These
   are non-goals, or non-requirements, within the context of the goals
   and requirements previously stated by this document as well as our
   interpretation of what can be practically achieved.

Little                                                          [Page 9]

RFC 1126            Inter-Autonomous System Routing         October 1989

      a) Ubiquity

         It is not a goal to design a routing environment in which any
         participating autonomous system can obtain a routing service to
         any other participating autonomous system in a ubiquitous
         fashion.  Within a policy sensitive routing environment, the
         cooperation of intermediate resources will be necessary and
         cannot be guaranteed to all participants.  The concept of
         ubiquitous connectivity will not be a valid one.

      b) Congestion control

         The ability for inter-autonomous system routing to perform
         congestion control is a non-requirement.  Additional study is
         necessary to determine what mechanisms are most appropriate and
         if congestion control is best realized within the inter-AS
         and/or intra-AS environments, and if both, what the dynamics of
         the interactions between the two are.

      c) Load splitting

         The functional capability to distribute the flow of datagrams,
         from a source to a destination, across two or more distinct
         paths through route synthesis and/or forwarding is a non-

      d) Maximizing the utilization of resources

         There is no goal or requirement for the inter-autonomous system
         routing environment to be designed such that it attempts to
         maximize the utilization of available resources.

      e) Schedule to deadline service

         The ability to support a schedule to deadline routing service
         is a non-requirement for the inter-autonomous routing
         environment at this point in time.

      f) Non-interference policies of resource utilization

         The ability to support routing policies based upon the concept
         of non-interference is a not a requirement.  An example of such
         a policy is one where an autonomous system allows the
         utilization of excess bandwidth by external users as long as
         this does not interfere with local usage of the link.

Little                                                         [Page 10]

RFC 1126            Inter-Autonomous System Routing         October 1989

5.  Considerations

   Although neither a specific goal nor a functional requirement,
   consideration must be given to the transition which will occur from
   the current operational routing environment to a new routing
   environment.  A coordinated effort among all participants of the
   Internet would be impractical considering the magnitude of such an
   undertaking.  Particularly, the issues of transitional coexistence,
   as opposed to phased upgrading between disjoint systems, should be
   addressed as a means to minimize the disruption of service.  Careful
   consideration should also be given to any required changes to hosts.
   It is very unlikely that all hosts could be changed, given historical
   precedence, their diversity and their large numbers.

Appendix - Issues in Inter-Autonomous Systems Routing

A.0  Acknowledgement

   This appendix is an edited version of the now defunct document
   entitled "Requirements for Inter-Autonomous Systems Routing", written
   by Ross Callon in conjunction with the members of the Open Routing
   Working Group.

A.1  Introduction

   The information and discussion contained here historically precedes
   that of the main document body and was a major influence on its
   content.  It is included here as a matter of reference and to provide
   insight into some of the many issues involved in inter-autonomous
   systems routing.

   The following definitions are utilized:

      Boundary Gateway

            A boundary gateway is any autonomous system gateway which
            has a network interface directly reachable from another
            autonomous system.  As a member of an autonomous system, a
            boundary gateway participates in the Interior Gateway
            Protocol and other protocols used for routing (and other
            purposes) between other gateways of this same autonomous
            system and between those networks directly reachable by this
            autonomous system.  A boundary gateway may also
            participate in an Inter-Autonomous System Routing Protocol.
            As a participant in the inter-autonomous system routing
            protocol, a boundary gateway interacts with other boundary
            gateways in other autonomous systems, either directly or
            indirectly, in support of the operation of the

Little                                                         [Page 11]

RFC 1126            Inter-Autonomous System Routing         October 1989

            Inter-Autonomous System Routing Protocol.

      Interior Gateway

            An interior gateway is any autonomous system gateway which
            is not a boundary gateway.  As such, an interior gateway
            does not have any network interfaces which are directly
            reachable by any other autonomous system.  An interior
            gateway is part of an autonomous system and, as such,
            takes part in the Interior Gateway Protocol and other
            protocols used in that autonomous system. However, an
            interior gateway does not directly exchange routing
            information with gateways in other autonomous systems via
            the Inter-Autonomous System Routing Protocol.

   The following acronyms are used:

      AS -- Autonomous System

            This document uses the current definition of "Autonomous
            System": a collection of cooperating gateways running a
            common interior routing protocol. This implies that networks
            and hosts may be reachable through one or more Autonomous

            NOTE: The current notion of "Autonomous System" implicitly
            assumes that each gateway will belong to exactly one AS.
            Extensions to allow gateways which belong to no AS's
            and/or gateways which belong to multiple AS's, are beyond
            the scope of this discussion. However, we do not preclude
            the possibility of considering such extensions in the

      IARP -- Inter-Autonomous System Routing Protocol

            This is the protocol used between boundary gateways for
            the purpose of routing between autonomous systems.

      IGP -- Interior Gateway Protocol

            This is the protocol used within an autonomous system for
            routing within that autonomous system.

A.2  Architectural Issues

   The architecture of an inter-autonomous system routing environment is
   mutually dependent with the notion of an Autonomous System. In
   general, the architecture should maximize independence of the

Little                                                         [Page 12]

RFC 1126            Inter-Autonomous System Routing         October 1989

   internals of an AS from the internals of other AS's, as well as from
   the inter-autonomous system routing protocols (IARP). This
   independence should allow technological and administrative
   differences among AS's as well as protection against propagation of
   misbehavior.  The following issues address ways to achieve
   interoperation and protection, and to meet certain performance
   criteria. We also put forth a set of minimal constraints to be
   imposed among Autonomous Systems, and between inter- and intra-AS

A.2.1  IGP Behavior

   The IARP should be capable of tolerating an Autonomous System in
   which its IGP is unable to route packets, provides incorrect
   information, and exhibits unstable behavior.  Interfacing to such an
   ill-behaved AS should not produce global instabilities within the
   IARP and the IARP should localize any effects.  On the other hand,
   the IGP should provide a routing environment where the information
   and connectivity provided to the IARP from the IGP does not exhibit
   rapid and continual changes.  An Autonomous System therefore should
   appear as a relatively stable environment.

A.2.2  Independence of Autonomous Systems

   The IARP should not constrain any AS to require the use any one
   specific IGP.  This applies both to IGPs and potentially to any other
   internal protocols.  The architecture should also allow intra-AS
   routing and organizational structures to be hidden from inter-AS use.
   An Autonomous System should not be required to use any one specific
   type of linkage between boundary gateways within the AS.  However,
   there are some minimal constraints that gateways and the associated
   interior routing protocol within an AS must meet in order to be able
   to route Inter-AS traffic, as discussed in Section A.2.6.

A.2.3  General Topology

   The routing architecture should provide significant flexibility
   regarding the interconnection of AS's.  The specification of IARP
   should impose no inherent restriction on either interconnection
   configuration or information passing among autonomous systems. There
   may be administrative and policy limitations on the interconnection
   of AS's, and on the extent to which routing information and data
   traffic may be passed between AS's. However, there should be no
   inherent restrictions imposed by limitations in the design of the
   routing architecture.  The architecture should allow arbitrary
   topological interconnection of Autonomous Systems.  Propagation of
   routing information should not be restricted by the specification of
   the IARP.  For example, the restrictions imposed by the "core model"

Little                                                         [Page 13]

RFC 1126            Inter-Autonomous System Routing         October 1989

   used by EGP are not acceptable.

A.2.4  Routing Firewalls

   We expect AS's to have a certain amount of insulation from other
   AS's.  This protection should apply to both the adequacy and
   stability of routes produced by the routing scheme, and also to the
   amount of overhead traffic and other costs necessary to run the
   routing scheme.  There are several forms which these "routing
   firewalls" may take:

      -  An AS must be able to successfully route its own internal
         traffic in the face of arbitrary failures of other IGPs and the
         IARP.  In other words, the AS should be able to effectively
         shutout the rest of the world.

      -  The IARP should be able to operate correctly in the face of IGP
         failures.  In this case, correct operation is defined as
         recognizing that an AS has failed, and routing around it if
         possible (traffic to or from that AS may of course fail).

      -  In addition, problems in Inter-AS Routing should, as much as
         possible, be limited in the extent of their effect.

   Routing firewalls may be explicit, or may be inherent in the design
   of the algorithms.  We expect that both explicit and inherent
   firewalls will be utilized.  Examples of firewalls include:

      -  Separating Intra- and Inter-AS Routing to some extent
         isolates each of these from problems with the other.  Clearly
         defined interfaces between different modules/protocols provides
         some degree of protection.

      -  Access control restrictions may provide some degree of
         firewalls.  For example, some AS's may be non-transit (won't
         forward transit traffic).  Failures within such AS's may be
         prevented from affecting traffic not associated with that AS.

      -  Protocol design can help.  For example, with link state routing
         you can require that both ends must report a link before is may
         be regarded as up, thereby eliminating the possibility of a
         single node causing fictitious links.

      -  Finally, explicit firewalls may be employed using explicit
         configuration information.

Little                                                         [Page 14]

RFC 1126            Inter-Autonomous System Routing         October 1989

A.2.5  Boundary Gateways

   Boundary gateways will exchange Inter-AS Routing information with
   other boundary gateways using the IARP.  Each AS which is to take
   part in Inter-AS Routing will have one or more boundary gateways, of
   which one or more of these boundary gateways exchanges information
   with peer boundary gateways in other AS's.

   Information related to Inter-AS Routing may be passed between
   connected boundary gateways in different AS's.  Specific designated
   boundary gateways will therefore be required to understand the IARP.
   The external link between the boundary gateways may be accomplished
   by any kind of connectivity that can be modeled as a direct link
   between two gateways -- a LAN, an ARPANET, a satellite link, a
   dedicated line, and so on.

A.2.6  Minimal Constraints on the Autonomous System

   The architectural issues discussed here for inter-AS routing imply
   certain minimal functional constraints that an AS must satisfy in
   order to take part in the Inter-AS Routing scheme.  These minimal
   requirements are described in greater detail in this section. This
   list of functional constraints is not necessarily complete.

A.2.6.1  Internal Links between Boundary Gateways

   In those cases where an AS may act as a transit AS (i.e., may pass
   traffic for which neither the source nor the destination is in that
   AS), the gateways internal to that AS will need to know which
   boundary gateway is to serve as the exit gateway from that AS. There
   are several ways in which this may be accomplished:

      1. Boundary gateways are directly connected

      2. "Tunneling" (i) using source routing (ii) using encapsulation

      3. Interior gateways participate (i) limited participation (ii)
         fully general participation

   With solution (1), the boundary gateways in an AS are directly
   connected.  This eliminates the need for other gateways in the AS to
   have any knowledge of Inter-AS Routing.  Transit traffic is passed
   directly among the boundary gateways of the AS.

   With solution (2), transit traffic may traverse interior gateways,
   but these interior gateways are protected from any need to have
   knowledge about Inter-AS routes by means such as source routing or
   encapsulation.  The boundary gateway by which the packet enters an AS

Little                                                         [Page 15]

RFC 1126            Inter-Autonomous System Routing         October 1989

   determines the boundary gateway which will serve as the exit gateway.
   This may require that the entrance boundary gateway add a source
   route to the packet, or encapsulate the packet in another level of IP
   or gateway-to-gateway header.  This allows boundary gateways to
   forward data traffic using the appropriate tunnelling technique.

   Finally, with solution (3), the interior gateways have some knowledge
   of Inter-AS Routing.  At a minimum, the interior gateways would need
   to know the identity of each boundary gateway, the address(es) that
   can be reached by that gateway, and the Inter-AS metric associated
   with the route to that address(es).  If the IARP allows for separate
   routing for multiple TOS classes, then the information that the
   interior gateways need to know includes a separate Inter-AS metric
   for each TOS class.  The Inter-AS metrics are necessary to allow
   gateways to choose among multiple possible exit boundary gateways.
   In general, it is not necessary for the Inter-AS metrics to have any
   relationship with the metric used within an AS for interior routing.
   The interior gateways do not need to know how to interpret the
   exterior metrics, except to know that each metric is to be
   interpreted as an unsigned integer and a lesser value is preferable
   to a greater value.  It would be possible, but not necessary, for the
   interior gateways to have full knowledge of the IARP.

   It is not necessary for the Inter-AS Routing architecture to specify
   which of these solutions are to be used for any particular AS.
   Rather, it is possible for individual AS's to choose which scheme or
   combination of schemes to use.  Independence of the IARP from the
   internal operation of each AS implies that this decision be left up
   to the internal protocols used in each AS.  The IARP must be able to
   operate as if the boundary gateways were directly connected.

A.2.6.2  Forwarding of Data from the AS

   The scheme used for forwarding transit traffic across an AS also has
   implications for the forwarding of traffic which originates within an
   AS, but whose destination is reachable only from other AS's.  If
   either of solutions (1) or (2) in Section A.2.6.1 is followed, then
   it will be sufficient for an interior gateway to forward such traffic
   to any boundary gateway.  Greater efficiency may optionally be
   achieved in some cases by providing interior gateways with additional
   information which will allow them to choose the "best" boundary
   gateway in some sense.  If solution (3) is followed, then the
   information passed to interior gateways to allow them to forward
   transit traffic will also be sufficient to forward traffic
   originating within that AS.

Little                                                         [Page 16]

RFC 1126            Inter-Autonomous System Routing         October 1989

A.2.6.3  Forwarding of Data to a Destination in the AS

   If a packet whose destination is reachable from an AS arrives at that
   AS, then it is desired that the interior routing protocol used in
   that AS be able to successfully deliver the packet without reliance
   on Inter-AS Routing.  This does not preclude that the Inter-AS
   Routing protocol will deal with partitioned AS's.

   An AS may have access control, security, and policy restrictions that
   restrict which data packets may enter or leave the AS. However, for
   any data packet which is allowed access to the AS, the AS is expected
   to deliver the packet to its destination without further restrictions
   between parts of the AS.  In this sense, the internal structure of
   the AS should not be visible to the IARP.

A.3  Policy Issues

   The interconnection of multiple heterogeneous networks and multiple
   heterogeneous autonomous systems owned and operated by multiple
   administrations into a single combined internet is a very complex
   task.  It is expected that the administrations associated with such
   an internet will wish to impose a variety of constraints on the
   operation of the internet.  Specifically, externally imposed routing
   constraints may include a variety of transit access control,
   administrative policy, and security constraints.

   Transit access control refers to those access control restrictions
   which an AS may impose to restrict which traffic the AS is willing to
   forward.  There are a large number of access control restrictions
   which one could envision being used.  For example, some AS's may wish
   to operate only as "non-transit" AS's, that is, they will only
   forward data traffic which either originates or terminates within
   that AS.  Other AS's may restrict transit traffic to datagrams which
   originate within a specified set of source hosts. Restrictions may
   require that datagrams be associated with specific applications (such
   as electronic mail traffic only), or that datagrams be associated
   with specific classes of users.

   Policy restrictions may allow either the source of datagrams, or the
   organization that is paying for transmission of a datagram, to limit
   which AS's the datagrams may transit.  For example, an organization
   may wish to specify that certain traffic originating within their AS
   should not transit any AS administered by its main competitor.

   Security restrictions on traffic relates to the official security
   classification level of traffic.  As an example, an AS may specify
   that all classified traffic is not allowed to leave its AS.

Little                                                         [Page 17]

RFC 1126            Inter-Autonomous System Routing         October 1989

   The main problem with producing a routing scheme which is sensitive
   to transit access control, administrative policy, and security
   constraints is the associated potential for exponential growth of
   routes.  For example, suppose that there are 20 packets arriving at a
   particular gateway, each for the same destination, but subject to a
   different combination of access control, policy, and security
   constraints.  It is possible that all 20 packets would need to follow
   different routes to the destination.

   This explosive growth of routes leads to the question: "Is it
   practically feasible to deal with complete general external routing
   constraints?" One approach would allow only a smaller subset of
   constraints, chosen to provide some useful level of control while
   allowing minimal impact on the routing protocol.  Further work is
   needed to determine the feasibility of this approach.

   There is another problem with dealing with transit access control,
   policy, and security restrictions in a fully general way.
   Specifically, it is not clear just what the total set of possible
   restrictions should be.  Efforts to study this issue are currently
   underway, but are not expected to produce definitive results within a
   short to medium time frame.  It is therefore not practical to wait
   for this effort to be finished before defining the next generation of
   Inter-AS Routing.

A.4  Service Features

   The following paragraphs discuss issues concerning the services an
   Inter-AS Routing Protocol may provide.  This is not a complete list
   of service issues but does address many of those services which are
   of concern to a significant portion of the community.

A.4.1  Cost on Toll Networks

   Consideration must be given to the use of routing protocols with toll
   (i.e., charge) networks.  Although uncommon in the Internet at the
   moment, they will become more common in the future, and thought needs
   to be given to allowing their inclusion in an efficient and effective

   There are two areas in which concerns of cost intrude.  First,
   provision must be made to include in the routing information
   distributed throughout the system the information that certain links
   cost money, so that traffic patterns may account for the cost.
   Second, the actual operation of the algorithm, in terms of the
   messages that must be exchanged to operate the algorithm, must
   recognize that fact that on certain links, the exchange may have an
   associated cost which must be taken into account.  These areas often

Little                                                         [Page 18]

RFC 1126            Inter-Autonomous System Routing         October 1989

   involve policy questions on the part of the user.  It is a
   requirement of the algorithm that facilities be available to allow
   different groups to answer these questions in different ways.  The
   first area is related to type-of-service routing, and is discussed in
   Section A.4.2.  The second area is discussed below.

   Previous attempts at providing these sorts of controls were
   incomplete because they were not thought through fully; a new effort
   must avoid these pitfalls.  For instance, even though the Hello rate
   in EGP may be adjusted, turning the rate down too low (to control the
   costs) could cause the route to be dropped from databases through

   In a large internet, changes will be occurring constantly; a
   simplistic mechanism might mean that a site which is only connected
   by toll networks has to either accept having an old picture of the
   network, or spend more to keep a more current picture of things.
   However, that is not necessarily the case if incomplete knowledge and
   cache-based techniques are used more. For instance, if a site
   connected only by toll links keeps an incomplete or less up-to-date
   map of the situation, an agreement with a neighbor which does not
   labor under these restrictions might allow it to get up-to-date
   information only when needed.

A.4.2  Type-of-Service Routing

   The need for type-of-service (TOS) has been increasing as networks
   become more heterogeneous in physical channel components, high-level
   applications, and administrative management.  For instance, some
   recently installed fiber cables provide abundant communication
   bandwidths, while old narrow-band channels will still be with us for
   a long time period.  Electronic mail traffic tolerates delivery
   delays and low throughput.  New image transmissions are coming up;
   these require high bandwidths but are not effected by a few bit
   errors.  Furthermore, some networks may soon install accounting
   functions to charge users, while others may still provide free

   Considering the long life span of a new routing architecture, it is
   mandatory that it be built with mechanisms to provide TOS routing.
   Unfortunately, we have had very little experience with TOS routing,
   even with a single network.  No TOS routing system has ever been
   field-tested on a large-scale basis.

   We foresee the need for TOS routing and recognize the possible
   complexities and difficulties in design and implementation.  We also
   consider that new applications coming along may require novel
   services that are unforeseeable today.  We feel a practical approach

Little                                                         [Page 19]

RFC 1126            Inter-Autonomous System Routing         October 1989

   is to provide a small set of TOS routing functions as a first step
   while the design of the architecture should be such that new classes
   of TOS can be easily added later and incrementally deployed.  The
   Inter-AS Routing Architecture should allow both globally and locally
   defined TOS classes.

   We intend to address TOS routing based on the following metrics:

      -  Delay

      -  Throughput

      -  Cost

   Delay and throughput are the main network performance concerns.
   Considering that some networks may soon start charging users for the
   transmission services provided, the cost should also be added as a
   factor in route selection.

   Reliability is not included in the above list.  Different
   applications with different reliability requirements will differ in
   terms of what Transport Protocol they use.  However, IP offers only a
   "moderate" level of reliability, suitable to applications such as
   voice, or possibly even less than that required by voice. The level
   of reliability offered by IP will not differ substantially based on
   the application.  Thus the requested level of reliability will not
   affect Inter-AS Routing.

   Delay and throughput will be measured from the physical
   characteristics of communication channels, without considering
   instantaneous load.  This is necessary in order to provide stable
   routes, and to minimize the overhead caused by the Inter-AS Routing

   A number of TOS service classes may be defined according to these
   metrics.  Each class will present defined requirements for each of
   the TOS metrics.  For example, one class may require low delay,
   require only low throughput, and require low cost.

A.4.3  Multipath Routing

   There are two types of multipath routing which are useful for Inter-
   AS Routing: (1) there may be multiple gateways between any two
   neighboring AS's; (2) there may be multiple AS-to-AS paths between
   any pair of source and destination AS's.  Both forms of multipath are
   useful in order to allow for loadsplitting.  Provision for multipath
   routing in the IARP is desirable, but not an absolute requirement.

Little                                                         [Page 20]

RFC 1126            Inter-Autonomous System Routing         October 1989

A.5  Performance Issues

   The following paragraphs discuss issues related to the performance of
   an Inter-AS Routing Protocol.  This discussion addresses size as well
   as efficiency considerations.

A.5.1  Adaptive Routing

   It is necessary that the Inter-AS Routing scheme respond in a timely
   fashion to major network problems, such as the failure of specific
   network resources.  In this sense, Inter-AS Routing needs to use
   adaptive routing mechanisms similar to those commonly used within
   individual networks and AS's.  It is important that the adaptive
   routing mechanism chosen for Inter-AS Routing achieve rapid
   convergence following internet topological changes, with little or
   none of the serious convergence problems (such as "counting to
   infinity") that have been experienced in some existing dynamic
   routing protocols.

   The Inter-AS Routing scheme must provide stability of routes.  It is
   totally unacceptable for routes to vary on a frequent basis.  This
   requirement is not meant to limit the ability of the routing
   algorithm to react rapidly to major topological changes, such as the
   loss of connectivity between two AS's.  The need for adaptive routing
   does not imply any desire for load-based routing.

A.5.2  Large Internets

   One key question in terms of the targets is the maximum size of the
   Internet this algorithm is supposed to support.  To some degree, this
   is tied to the timeline for which this protocol is expected to be
   active.  However, it is necessary to have some size targets.
   Techniques that work at one order of size may be impractical in a
   system ten or twenty times larger.

   Over the past five years there has been an approximate doubling of
   the Internet each year.  In January 1988, there were more than 330
   operational networks and more than 700 network assigned numbers.
   Exact figures for the future growth rate of the Internet are
   difficult to predict accurately.  However, if this doubling trend
   continues, we would reach 10,000 nets sometime near January 1993.

   Taking a projection purely on the number of networks (the top level
   routing entity) may be overly conservative since the introduction and
   wide use of subnets has absorbed some growth, but will not continue
   to be able to do so.  In addition, there are plans being discussed
   that will continue or accelerate the current rate of growth.
   Nonetheless, the number of networks is very important because

Little                                                         [Page 21]

RFC 1126            Inter-Autonomous System Routing         October 1989

   networks constitute the "top level entities" in the current
   addressing structure.

   The implications of the size parameter are fairly serious.  The
   current system has only one level of addressing above subnets. While
   it is possible to adjust certain parameters (for example, the
   unsolicited or unnecessary retransmission rate) to produce a larger
   but less robust system, other parameters (for example, the rate of
   change in the system) cannot be so adjusted.  This will provide
   eventual limits on the size of a system that can be dealt with in a
   flat address space.

   The exact size that necessitates moving from the current single-
   level system to a multi-level system is not clear.  Among the
   parameters which affect it are the assumed minimum speed of links in
   the system (faster links can allocate more bandwidth to routing
   traffic before it becomes obtrusive), speed and memory capacity of
   routing nodes (needed to store and process routing data), rate at
   which topology changes, etc.  The maximum size which can be handled
   in a single layer is generally thought to be somewhere on the order
   of 10 thousand objects.  The IARP must be designed to deal with
   internets bigger than this.

A.5.3  Addressing Implications

   Given the current rate of growth of the Internet, we can expect that
   the current addressing structure will become unworkable early within
   the lifetime of the new IARP.  It is therefore essential that any new
   IARP be able to use a new addressing format which allows for
   addressing hierarchies beyond the network level.  Any new IARP should
   allow for graceful migration from the current routing protocols, and
   should also allow for graceful migration from a routing scheme based
   on the current addressing, to a scheme based on a new multi-level
   addressing format such as that described by OSI 8473.

A.5.4  Memory, CPU, and Bandwidth Costs

   Routing costs can be measured in terms of the memory needed to store
   routing information, the CPU costs of calculating routes and
   forwarding packets, and the bandwidth costs of exchanging routing
   information and of forwarding packets.  These significant factors
   should provide the basis for comparison between competing proposals
   in IARP design.

Little                                                         [Page 22]

RFC 1126            Inter-Autonomous System Routing         October 1989

   The routing architecture will be driven by the expected size of the
   Internet, the expected memory capacity of the gateways, capacity of
   the Inter-AS links, and the computing speed of the gateways. Given
   our experience with the current Internet, it is clearly necessary for
   the scheme to function adequately even if the Internet grows more
   quickly than we predict and its capacity grows more slowly.  Memory,
   CPU, and bandwidth costs should be in line with what is economically
   practical at any point in time given the size of the Internet at that

A.6  Other Issues

   The following are issues of a general nature and includes discussion
   of items which have been considered to be best left for future

A.6.1  Implementation

   The specification of IARP should allow interoperation among multi-
   vendor implementations.  This requires that multiple vendors be able
   to implement the same protocol, and that equipment from multiple
   vendors be able to interoperate successfully.

   There are potential practical difficulties of realizing multi-vendor
   interoperation.  Any such difficulty should not be inherent to the
   protocol specifications.  Towards this end, we should produce a
   protocol specification that is precise and unambiguous.  This implies
   that the specification should include a detailed specification using
   Pseudo-Code or a Formal Description Technique.

A.6.2  Configuration

   It is expected that any IARP will require a certain amount of
   configuration information to be maintained by gateways.  However, in
   practice it is often difficult to maintain configuration information
   in a fully correct and up-to-date form.  Problems in configuration
   have been known to cause significant problems in existing operational
   networks and internets.  The design of an Inter-AS Routing
   architecture must therefore simplify the maintenance of configuration
   information, consistent with other requirements. Simplification of
   configuration information may require minimizing the amount of
   configuration information, and using automated or semi-automated
   configuration mechanisms.

A.6.3  Migration

   In any event, whether the address format changes or not, a viable
   transition plan which allows for interoperability must be arranged.

Little                                                         [Page 23]

RFC 1126            Inter-Autonomous System Routing         October 1989

   In a system of this magnitude, which is in operational use, a
   coordinated change is not possible.  Where possible, changes should
   not affect the hosts, since deploying such a change is probably
   several orders of magnitude more difficult than changing only the
   gateways, due to the larger number of host implementations as well as
   hosts.  There are two important questions that need to be addressed:
   (1) migration from the existing EGP to a new IARP; (2) migration from
   the current DD IP to future protocols (including the ISO IP, and
   other future protocols).

A.6.4  Load-Based Routing

   Some existing networks are able to route packets based on current
   load in the network.  For example, one approach to congestion
   involves adjusting the routes in real time to send as much traffic as
   possible on lightly loaded network links.

   This sort of load-based routing is a relatively delicate procedure
   which is prone to instability.  It is particularly difficult to
   achieve stability in multi-vendor environments, in large internets,
   and in environments characterized by a large variation in network
   characteristics.  For these reasons, we believe that it would be a
   mistake to attempt to achieve effective load-based routing in an
   Inter-AS Routing scheme.

A.6.5  Non-Interference Policies

   There are policies which are in effect, or desired to be in effect,
   which are based upon the concept of non-interference.  These policies
   state that the utilization of a given resource is permissible by one
   party as long as that utilization does not disrupt the current or
   future utilization of another party.  These policies are often of the
   kind "you may use the excess capacity of my link" without
   guaranteeing any capacity will be available.  The expectation is to
   be able to utilize the link as needed, perhaps to the exclusion of
   the other party.  The problem with supporting such a policy is the
   need to be cognizant of highly dynamic state information and the
   implicit requirement to adapt to these changes. Rapid, persistent,
   and non-deterministic state changes would lead to routing
   oscillations and looping.  We do not believe it is feasible to
   support policies based on these considerations in a large
   internetworking environment based on the current design of IP.

Security Considerations

   Security issues are not addressed in this memo.

Little                                                         [Page 24]

RFC 1126            Inter-Autonomous System Routing         October 1989

Author's Address

   Mike Little
   Science Applications International Corporation (SAIC)
   8619 Westwood Center Drive
   Vienna, Virginia  22182

   Phone: 703-734-9000

   EMail: little@SAIC.COM

Little                                                         [Page 25]