Network Working Group                                          D. Estrin
Request for Comments:  1322                                          USC
                                                             Y. Rekhter
                                                                    IBM
                                                                S. Hotz
                                                                    USC
                                                               May 1992


              A Unified Approach to Inter-Domain Routing

Status of this Memo

  This memo provides information for the Internet community.  It does
  not specify an Internet standard.  Distribution of this memo is
  unlimited.

Abstract

  This memo is an informational RFC which outlines one potential
  approach for inter-domain routing in future global internets.  The
  focus is on scalability to very large networks and functionality, as
  well as scalability, to support routing in an environment of
  heterogeneous services, requirements, and route selection criteria.

  Note: The work of D. Estrin and S. Hotz was supported by the National
  Science Foundation under contract number NCR-9011279, with matching
  funds from GTE Laboratories.  The work of Y. Rekhter was supported by
  the Defense Advanced Research Projects Agency, under contract
  DABT63-91-C-0019.  Views and conclusions expressed in this paper are
  not necessarily those of the Defense Advanced Research Projects
  Agency and National Science Foundation.

1.0 Motivation

  The global internet can be modeled as a collection of hosts
  interconnected via transmission and switching facilities.  Control
  over the collection of hosts and the transmission and switching
  facilities that compose the networking resources of the global
  internet is not homogeneous, but is distributed among multiple
  administrative authorities.  Resources under control of a single
  administration form a domain.  In order to support each domain's
  autonomy and heterogeneity, routing consists of two distinct
  components: intra-domain (interior) routing, and inter-domain
  (exterior) routing.  Intra-domain routing provides support for data
  communication between hosts where data traverses transmission and
  switching facilities within a single domain.  Inter-domain routing
  provides support for data communication between hosts where data



Estrin, Rekhter & Hotz                                          [Page 1]

RFC 1322       A Unified Approach to Inter-Domain Routing       May 1992


  traverses transmission and switching facilities spanning multiple
  domains.  The entities that forward packets across domain boundaries
  are called border routers (BRs).  The entities responsible for
  exchanging inter-domain routing information are called route servers
  (RSs).  RSs and BRs may be colocated.

  As the global internet grows, both in size and in the diversity of
  routing requirements, providing inter-domain routing that can
  accommodate both of these factors becomes more and more crucial.  The
  number and diversity of routing requirements is increasing due to:
  (a) transit restrictions imposed by source, destination, and transit
  networks, (b) different types of services offered and required, and
  (c) the presence of multiple carriers with different charging
  schemes.  The combinatorial explosion of mixing and matching these
  different criteria weighs heavily on the mechanisms provided by
  conventional hop-by-hop routing architectures ([ISIS10589, OSPF,
  Hedrick88, EGP]).

  Current work on inter-domain routing within the Internet community
  has diverged in two directions: one is best represented by the Border
  Gateway Protocol (BGP)/Inter-Domain Routeing Protocol (IDRP)
  architectures ([BGP91, Honig90, IDRP91]), and another is best
  represented by the Inter-Domain Policy Routing (IDPR) architecture
  ([IDPR90, Clark90]).  In this paper we suggest that the two
  architectures are quite complementary and should not be considered
  mutually exclusive.

  We expect that over the next 5 to 10 years, the types of services
  available will continue to evolve and that specialized facilities
  will be employed to provide new services.  While the number and
  variety of routes provided by hop-by-hop routing architectures with
  type of service (TOS) support (i.e., multiple, tagged routes) may be
  sufficient for a large percentage of traffic, it is important that
  mechanisms be in place to support efficient routing of specialized
  traffic types via special routes.  Examples of special routes are:
  (1) a route that travels through one or more transit domains that
  discriminate according to the source domain, (2) a route that travels
  through transit domains that support a service that is not widely or
  regularly used.  We refer to all other routes as generic.

  Our desire to support special routes efficiently led us to
  investigate the dynamic installation of routes ([Breslau-Estrin91,
  Clark90, IDPR90]).  In a previous paper ([Breslau-Estrin91]), we
  evaluated the algorithmic design choices for inter-domain policy
  routing with specific attention to accommodating source-specific and
  other "special" routes.  The conclusion was that special routes are
  best supported with source-routing and extended link-state
  algorithms; we refer to this approach as source-demand routing



Estrin, Rekhter & Hotz                                          [Page 2]

RFC 1322       A Unified Approach to Inter-Domain Routing       May 1992


  [Footnote:  The Inter-Domain Policy Routing (IDPR) architecture uses
  these techniques.].  However, a source-demand routing architecture,
  used as the only means of inter-domain routing, has scaling problems
  because it does not lend itself to general hierarchical clustering
  and aggregation of routing and forwarding information.  For example,
  even if a particular route from an intermediate transit domain X, to
  a destination domain Y is shared by 1,000 source-domains, IDPR
  requires that state for each of the 1,000 routes be setup and
  maintained in the transit border routers between X and Y.  In
  contrast, an alternative approach to inter-domain routing, based on
  hop-by-hop routing and a distributed route-computation algorithm
  (described later), provides extensive support for aggregation and
  abstraction of reachability, topology, and forwarding information.
  The Border Gateway Protocol (BGP) and Inter-Domain Routeing Protocol
  (IDRP) use these techniques ([BGP91, IDRP91]).  While the BGP/IDRP
  architecture is capable of accommodating very large numbers of
  datagram networks, it does not provide support for specialized
  routing requirements as flexibly and efficiently as IDPR-style
  routing.

1.1 Overview of the Unified Architecture

  We want to support special routes and we want to exploit aggregation
  when a special route is not needed.  Therefore, our scalable inter-
  domain routing architecture consists of two major components:
  source-demand routing (SDR), and node routing (NR).  The NR component
  computes and installs routes that are shared by a significant number
  of sources.  These generic routes are commonly used and warrant wide
  propagation, consequently, aggregation of routing information is
  critical.  The SDR component computes and installs specialized routes
  that are not shared by enough sources to justify computation by NR
  [Footnote: Routes that are only needed sporadically (i.e., the demand
  for them is not continuous or otherwise predictable) are also
  candidates for SDR.].  The potentially large number of different
  specialized routes, combined with their sparse utilization, make them
  too costly to support with the NR mechanism.

  A useful analogy to this approach is the manufacturing of consumer
  products.  When predictable patterns of demand exist, firms produce
  objects and sell them as "off the shelf" consumer goods.  In our
  architecture NR provides off-the-shelf routes.  If demand is not
  predictable, then firms accept special orders and produce what is
  demanded at the time it is needed.  In addition, if a part is so
  specialized that only a single or small number of consumers need it,
  the  consumer may repeatedly special order the part, even if it is
  needed in a predictable manner, because the consumer does not
  represent a big enough market for the producer to bother managing the
  item as part of its regular production.  SDR provides such special



Estrin, Rekhter & Hotz                                          [Page 3]

RFC 1322       A Unified Approach to Inter-Domain Routing       May 1992


  order, on-demand routes.

  By combining NR and SDR routing we propose to support inter-domain
  routing in internets of practically-unlimited size, while at the same
  time providing efficient support for specialized routing
  requirements.

  The development of this architecture does assume that routing
  requirements will be diverse and that special routes will be needed.
  On the other hand, the architecture does not depend on assumptions
  about the particular types of routes demanded or on the distribution
  of that demand.  Routing will adapt naturally over time to changing
  traffic patterns and new services by shifting computation and
  installation of particular types of routes between the two components
  of the hybrid architecture [Footnote: Before continuing with our
  explanation of this architecture, we wish to state up front that
  supporting highly specialized routes for all source-destination pairs
  in an internet, or even anything close to that number, is not
  feasible in any routing architecture that we can foresee.  In other
  words, we do not believe that any foreseeable routing architecture
  can support unconstrained proliferation of user requirements and
  network services.  At the same time, this is not necessarily a
  problem.  The capabilities of the architecture may in fact exceed the
  requirements of the users.  Moreover, some of the requirements that
  we regard as infeasible from the inter-domain routing point of view,
  may be supported by means completely outside of routing.
  Nevertheless, the caveat is stated here to preempt unrealistic
  expectations.].

  While the packet forwarding functions of the NR and SDR components
  have little or no coupling with each other, the connectivity
  information exchange mechanism of the SDR component relies on
  services provided by the NR component.

1.2 Outline

  The remainder of this report is organized as follows.  Section 2
  outlines the requirements and priorities that guide the design of the
  NR and SDR components.  Sections 3 and 4 describe the NR and SDR
  design choices, respectively, in light of these requirements.
  Section 5 describes protocol support for the unified architecture and
  briefly discusses transition issues.  We conclude with a brief
  summary.








Estrin, Rekhter & Hotz                                          [Page 4]

RFC 1322       A Unified Approach to Inter-Domain Routing       May 1992


2.0 Architectural Requirements and Priorities

  In order to justify our design choices for a scalable inter-domain
  routing architecture, we must articulate our evaluation criteria and
  priorities.  This section defines complexity, abstraction, policy,
  and type of service requirements.

2.1 Complexity

  Inter-domain routing complexity must be evaluated on the basis of the
  following performance metrics: (1) storage overhead, (2)
  computational overhead, and (3) message overhead.  This evaluation is
  essential to determining the scalability of any architecture.

2.1.1 Storage Overhead

  The storage overhead of an entity that participates in inter-domain
  routing comes from two sources: Routing Information Base (RIB), and
  Forwarding Information Base (FIB) overhead.  The RIB contains the
  routing information that entities exchange via the inter-domain
  routing protocol; the RIB is the input to the route computation.  The
  FIB contains the information that the entities use to forward the
  inter-domain traffic; the FIB is the output of the route computation.
  For an acceptable level of storage overhead, the amount of
  information in both FIBs and RIBs should grow significantly slower
  than linearly (e.g., close to logarithmically) with the total number
  of domains in an internet.  To satisfy this requirement with respect
  to the RIB, the architecture must provide mechanisms for either
  aggregation and abstraction of routing and forwarding information, or
  retrieval of a subset of this information on demand.  To satisfy this
  requirement with respect to the FIB, the architecture must provide
  mechanisms for either aggregation of the forwarding information (for
  the NR computed routes), or dynamic installation/tear down of this
  information (for the SDR computed routes).

  Besides being an intrinsically important evaluation metric, storage
  overhead has a direct impact on computational and bandwidth
  complexity.  Unless the computational complexity is fixed (and
  independent of the total number of domains), the storage overhead has
  direct impact on the computational complexity of the architecture
  since the routing information is used as an input to route
  computation. Moreover, unless the architecture employs incremental
  updates, where only changes to the routing information are
  propagated, the storage overhead has direct impact on the bandwidth
  overhead of the architecture since the exchange of routing
  information constitutes most of the bandwidth overhead.





Estrin, Rekhter & Hotz                                          [Page 5]

RFC 1322       A Unified Approach to Inter-Domain Routing       May 1992


2.1.2 Computational Overhead

  The NR component will rely primarily on precomputation of routes.  If
  inter-domain routing is ubiquitous, then the precomputed routes
  include all reachable destinations.  Even if policy constraints make
  fully ubiquitous routing impossible, the precomputed routes are
  likely to cover a very large percentage of all reachable
  destinations.  Therefore the complexity of this computation must be
  as small as possible.  Specifically, it is highly desirable that the
  architecture would employ some form of partial computation, where
  changes in topology would require less than complete recomputation.
  Even if complete recomputation is necessary, its complexity should be
  less than linear with the total number of domains.

  The SDR component will use on-demand computation and caching.
  Therefore the complexity of this computation can be somewhat higher.
  Another reason for relaxed complexity requirements for SDR is that
  SDR is expected to compute routes to a smaller number of destinations
  than is NR (although SDR route computation may be invoked more
  frequently).

  Under no circumstances is computational complexity allowed to become
  exponential (for either the NR or SDR component).

2.1.3 Bandwidth Overhead

  The bandwidth consumed by routing information distribution should be
  limited.  However, the possible use of data compression techniques
  and the increasing speed of network links make this less important
  than route computation and storage overhead.  Bandwidth overhead may
  be further contained by using incremental (rather than complete)
  exchange of routing information.

  While storage and bandwidth overhead may be interrelated, if
  incremental updates are used then bandwidth overhead is negligible in
  the steady state (no changes in topology), and is independent of the
  storage overhead.  In other words, use of incremental updates
  constrains the bandwidth overhead to the dynamics of the internet.
  Therefore, improvements in stability of the physical links, combined
  with techniques to dampen the effect of topological instabilities,
  will make the bandwidth overhead even less important.

2.2 Aggregation

  Aggregation and abstraction of routing and forwarding information
  provides a very powerful mechanism for satisfying storage,
  computational, and bandwidth constraints.  The ability to aggregate,
  and subsequently abstract, routing and forwarding information is



Estrin, Rekhter & Hotz                                          [Page 6]

RFC 1322       A Unified Approach to Inter-Domain Routing       May 1992


  essential to the scaling of the architecture [Footnote: While we can
  not prove that there are no other ways to achieve scaling, we are not
  aware of any mechanism other than clustering that allows information
  aggregation/abstraction.  Therefore, the rest of the paper assumes
  that clustering is used for information aggregation/abstraction.].
  This is especially true with respect to the NR component, since the
  NR component must be capable of providing routes to all or almost all
  reachable destinations.

  At the same time, since preserving each domain's independence and
  autonomy is one of the crucial requirements of inter-domain routing,
  the architecture must strive for the maximum flexibility of its
  aggregation scheme, i.e., impose as few preconditions, and as little
  global coordination, as possible on the participating domains.

  The Routing Information Base (RIB) carries three types of
  information: (1) topology (i.e., the interconnections between domains
  or groups of domains), (2) network layer reachability, and (3)
  transit constraint.  Aggregation of routing information should
  provide a reduction of all three components.  Aggregation of
  forwarding information will follow from reachability information
  aggregation.

  Clustering (by forming routing domain confederations) serves the
  following aggregation functions: (1) to hide parts of the actual
  physical topology, thus abstracting topological information, (2) to
  combine a set of reachable destination entities into a single entity
  and reduce storage overhead, and (3) to express transit constraints
  in terms of clusters, rather than individual domains.

  As argued in [Breslau-Estrin91], the architecture must allow
  confederations to be formed and changed without extensive
  configuration and coordination; in particular, forming a
  confederation should not require global coordination (such as that
  required in ECMA ([ECMA89]).  In addition, aggregation should not
  require explicit designation of the relative placement of each domain
  relative to another; i.e., domains or confederations of domains
  should not be required to agree on a partial ordering (i.e., who is
  above whom, etc.).












Estrin, Rekhter & Hotz                                          [Page 7]

RFC 1322       A Unified Approach to Inter-Domain Routing       May 1992


  The architecture should allow different domains to use different
  methods of aggregation and abstraction.  For example, a research
  collaborator at IBM might route to USC as a domain-level entity in
  order to take advantage of some special TOS connectivity to, or even
  through, USC.  Whereas, someone else at Digital Equipment Corporation
  might see information at the level of the California Educational
  Institutions Confederation, and know only that USC is a member.
  Alternatively, USC might see part of the internal structure within
  the IBM Confederation (at the domain's level), whereas UCLA may route
  based on the confederation of IBM domains as a whole.

  Support for confederations should be flexible.  Specifically, the
  architecture should allow confederations to overlap without being
  nested, i.e., a single domain, or a group of domains may be part of
  more than one confederation.  For example, USC may be part of the
  California Educational Institutions Confederation and part of the US
  R&D Institutions Confederation; one is not a subset of the other.
  Another example: T.J.  Watson Research Center might be part of
  NYSERNET Confederation and part of IBM-R&D-US Confederation.  While
  the above examples describe cases where overlap consists of a single
  domain, there may be other cases where multiple domains overlap.  As
  an example consider the set of domains that form the IBM
  Confederation, and another set of domains that form the DEC
  Confederation.  Within IBM there is a domain IBM-Research, and
  similarly within DEC there is a domain DEC-Research.  Both of these
  domains could be involved in some collaborative effort, and thus have
  established direct links.  The architecture should allow restricted
  use of these direct links, so that other domains within the IBM
  Confederation would not be able to use it to talk to other domains
  within the DEC Confederation.  A similar example exists when a
  multinational corporation forms a confederation, and the individual
  branches within each country also belong to their respective country
  confederations.  The corporation may need to protect itself from
  being used as an inter-country transit domain (due to internal, or
  international, policy).  All of the above examples illustrate a
  situation where confederations overlap, and it is necessary to
  control the traffic traversing the overlapping resources.

  While flexible aggregation should be accommodated in any inter-domain
  architecture, the extent to which this feature is exploited will have
  direct a effect on the scalability associated with aggregation.  At
  the same time, the exploitation of this feature depends on the way
  addresses are assigned.  Specifically, scaling associated with
  forwarding information depends heavily on the assumption that there
  will be general correspondence between the hierarchy of address
  registration authorities, and the way routing domains and routing
  domain confederations are organized (see Section 2.6).




Estrin, Rekhter & Hotz                                          [Page 8]

RFC 1322       A Unified Approach to Inter-Domain Routing       May 1992


2.3 Routing Policies

  Routing policies that the architecture must support may be broadly
  classified into transit policies and route selection policies
  [Breslau-Estrin 91, Estrin89].

  Restrictions imposed via transit policies may be based on a variety
  of factors.  The architecture should be able to support restrictions
  based on the source, destination, type of services (TOS), user class
  identification (UCI), charging, and path [Estrin89 , Little89].  The
  architecture must allow expression of transit policies on all routes,
  both NR and SDR.  Even if NR routes are widely used and have fewer
  source or destination restrictions, NR routes may have some transit
  qualifiers (e.g., TOS, charging, or user-class restriction).  If the
  most widely-usable route to a destination has policy qualifiers, it
  should be advertiseable by NR and the transit constraints should be
  explicit.

  Route selection policies enable each domain to select a particular
  route among multiple routes to the same destination.  To maintain
  maximum autonomy and independence between domains, the architecture
  must support heterogeneous route selection policies, where each
  domain can establish its own criteria for selecting routes.  This
  argument was made earlier with respect to source route selection
  ([IDPR90, Clark90, Breslau-Estrin91]).  In addition, each
  intermediate transit domain must have the flexibility to apply its
  own selection criteria to the routes made available to it by its
  neighbors.  This is really just a corollary to the above; i.e., if we
  allow route selection policy to be expressed for NR routes, we can
  not assume all domains will apply the same policy.  The support for
  dissimilar route selection policies is one of the key factors that
  distinguish inter-domain routing from intra-domain routing ([ECMA89,
  Estrin89]).  However, it is a non-goal of the architecture to support
  all possible route selection policies.  For more on unsupported route
  selection policies see Section 2.3.2 below.

2.3.1 Routing Information Hiding

  The architecture should not require all domains within an internet to
  reveal their connectivity and transit constraints to each other.
  Domains should be able to enforce their transit policies while
  limiting the advertisement of their policy and connectivity
  information as much as possible; such advertisement will be driven by
  a "need to know" criteria.  Moreover, advertising a transit policy to
  domains that can not use this policy will increase the amount of
  routing information that must be stored, processed, and propagated.
  Not only may it be impractical for a router to maintain full inter-
  domain topology and policy information, it may not be permitted to



Estrin, Rekhter & Hotz                                          [Page 9]

RFC 1322       A Unified Approach to Inter-Domain Routing       May 1992


  obtain this information.

2.3.2 Policies Not Supported

  In this and previous papers we have argued that a global inter-domain
  routing architecture should support a wide range of policies.  In
  this section we identify several types of policy that we explicitly
  do not propose to support.  In general our reasoning is pragmatic; we
  think such policy types are either very expensive in terms of
  overhead, or may lead to routing instabilities.

  1. Route selection policies contingent on external behavior.
     The route selection process may be modeled by a function that
     assigns a non-negative integer to a route, denoting the degree
     of preference.  Route selection applies this function to all
     feasible routes to a given destination, and selects the route
     with the highest value.  To provide a stable environment, the
     preference function should not use as an input the status and
     attributes of other routes (either to the same or to a
     different destination).

  2. Transit policies contingent on external behavior.
     To provide a stable environment, the domain's transit policies
     can not be automatically affected by any information external
     to the domain.  Specifically, these policies can not be modified,
     automatically, in response to information about other domains'
     transit policies, or routes selected by local or other domains.
     Similarly, transit policies can not be automatically modified
     in response to information about performance characteristics of
     either local or external domains.

  3. Policies contingent on external state (e.g., load).
     It is a non-goal of the architecture to support load-sensitive
     routing for generic routes.  However, it is possible that some
     types of service may employ load information to select among
     alternate SDR routes.

  4. Very large number of simultaneous SDR routes.
     It is a non-goal of the architecture to support a very large
     number of simultaneous SDR routes through any single router.
     Specifically, the FIB storage overhead associated with SDR
     routes must be comparable with that of NR routes, and should
     always be bound by the complexity requirements outlined earlier
     [Foonote: As discussed earlier, theoretically the state overhead
     could grow O(N^2) with the number of domains in an internet.
     However, there is no evidence or intuition to suggest that
     this will be a limiting factor on the wide utilization of SDR,
     provided that NR is available to handle generic routes.].



Estrin, Rekhter & Hotz                                         [Page 10]

RFC 1322       A Unified Approach to Inter-Domain Routing       May 1992


2.4 Support for TOS Routing

  Throughout this document we refer to support for type of service
  (TOS) routing.  There is a great deal of research and development
  activity currently underway to explore network architectures and
  protocols for high-bandwidth, multimedia traffic.  Some of this
  traffic is delay sensitive, while some requires high throughput.  It
  is unrealistic to assume that a single communication fabric will be
  deployed homogeneously across the internet (including all
  metropolitan, regional, and backbone networks) that will support all
  types of traffic uniformly.  To support diverse traffic requirements
  in a heterogeneous environment, various resource management
  mechanisms will be used in different parts of the global internet
  (e.g., resource reservation of various kinds) [ST2-90, Zhang91].

  In this context, it is the job of routing protocols to locate routes
  that can potentially support the particular TOS requested.  It is
  explicitly not the job of the general routing protocols to locate
  routes that are guaranteed to have resources available at the
  particular time of the route request.  In other words, it is not
  practical to assume that instantaneous resource availability will be
  known at all remote points in the global internet.  Rather, once a
  TOS route has been identified, an application requiring particular
  service guarantees will attempt to use the route (e.g., using an
  explicit setup message if so required by the underlying networks).
  In Section 4 we describe additional services that may be provided to
  support more adaptive route selection for special TOS [Footnote:
  Adaptive route selection implies adaptability only during the route
  selection process.  Once a route is selected, it is not going to be a
  subject to any adaptations, except when it becomes infeasible.].

2.5 Commonality between Routing Components

  While it is acceptable for the NR and SDR components to be
  dissimilar, we do recognize that such a solution is less desirable --
  all other things being equal.  In theory, there are advantages in
  having the NR and SDR components use similar algorithms and
  mechanisms.  Code and databases could be shared and the architecture
  would be more manageable and comprehensible.  On the other hand,
  there may be some benefit (e.g., robustness) if the two parts of the
  architecture are heterogeneous, and not completely inter-dependent.
  In Section 5 we list several areas in which opportunities for
  increased commonality (unification) will be exploited.

2.6 Interaction with Addressing

  The proposed architecture should be applicable to various addressing
  schemes.  There are no specific assumptions about a particular



Estrin, Rekhter & Hotz                                         [Page 11]

RFC 1322       A Unified Approach to Inter-Domain Routing       May 1992


  address structure, except that this structure should facilitate
  information aggregation, without forcing rigid hierarchical routing.

  Beyond this requirement, most of the proposals for extending the IP
  address space, for example, can be used in conjunction with our
  architecture.  But our architecture itself does not provide (or
  impose) a particular solution to the addressing problem.












































Estrin, Rekhter & Hotz                                         [Page 12]

RFC 1322       A Unified Approach to Inter-Domain Routing       May 1992


3.0 Design Choices for Node Routing (NR)

  This section addresses the design choices made for the NR component
  in light of the above architectural requirements and priorities.  All
  of our discussion of NR assumes hop-by-hop routing.  Source routing
  is subject to different constraints and is used for the complementary
  SDR component.

3.1 Overview of NR

  The NR component is designed and optimized for an environment where a
  large percentage of packets will travel over routes that can be
  shared by multiple sources and that have predictable traffic
  patterns.  The efficiency of the NR component improves when a number
  of source domains share a particular route from some intermediate
  point to a destination.  Moreover, NR is best suited to provide
  routing for inter-domain data traffic that is either steady enough to
  justify the existence of a route, or predictable, so that a route may
  be installed when needed (based on the prediction, rather than on the
  actual traffic).  Such routes lend themselves to distributed route
  computation and installation procedures.

  Routes that are installed in routers, and information that was used
  by the routers to compute these routes, reflect the known operational
  state of the routing facilities (as well as the policy constraints)
  at the time of route computation.  Route computation is driven by
  changes in either operational status of routing facilities or policy
  constraints.  The NR component supports route computation that is
  dynamically adaptable to both changes in topology and policy.  The NR
  component allows time-dependent selection or deletion of routes.
  However, time dependency must be predictable (e.g., advertising a
  certain route only after business hours) and routes should be used
  widely enough to warrant inclusion in NR.

  The proposed architecture assumes that most of the inter-domain
  conversations will be forwarded via routes computed and installed by
  the NR component.

  Moreover, the exchange of routing information necessary for the SDR
  component depends on facilities provided by the NR component; i.e.,
  NR policies must allow SDR reachability information to travel.
  Therefore, the architecture requires that all domains in an internet
  implement and participate in NR.  Since scalability (with respect to
  the size of the internet) is one of the fundamental requirements for
  the NR component, it must provide multiple mechanisms with various
  degrees of sophistication for information aggregation and
  abstraction.




Estrin, Rekhter & Hotz                                         [Page 13]

RFC 1322       A Unified Approach to Inter-Domain Routing       May 1992


  The potential reduction of routing and forwarding information depends
  very heavily on the way addresses are assigned and how domains and
  their confederations are structured.  "If there is no correspondence
  between the address registration hierarchy and the organisation of
  routeing domains, then ... each and every routeing domain in the OSIE
  would need a table entry potentially at every boundary IS of every
  other routeing domain" ([Oran89]).  Indeed, scaling in the NR
  component is almost entirely predicated on the assumption that there
  will be general correspondence between the hierarchy of address
  assignment authorities and the way routing domains are organised, so
  that the efficient and frequent aggregation of routing and forwarding
  information will be possible.  Therefore, given the nature of inter-
  domain routing in general, and the NR component in particular,
  scalability of the architecture depends very heavily on the
  flexibility of the scheme for information aggregation and
  abstraction, and on the preconditions that such a scheme imposes.
  Moreover, given a flexible architecture, the operational efficiency
  (scalability) of the global internet, or portions thereof, will
  depend on tradeoffs made between flexibility and aggregation.

  While the NR component is optimized to satisfy the common case
  routing requirements for an extremely large population of users, this
  does not imply that routes produced by the NR component would not be
  used for different types of service (TOS).  To the contrary, if a TOS
  becomes sufficiently widely used (i.e., by multiple domains and
  predictably over time), then it may warrant being computed by the NR
  component.

  To summarize, the NR component is best suited to provide routes that
  are used by more than a single domain.  That is, the efficiency of
  the NR component improves when a number of source domains share a
  particular route from some intermediate point to a destination.
  Moreover, NR is best suited to provide routing for inter-domain data
  traffic that is either steady enough to justify the existence of a
  route, or predictable, so that a route may be installed when needed,
  (based on the prediction, rather than on the actual traffic).

3.2 Routing Algorithm Choices for NR

  Given that a NR component based on hop-by-hop routing is needed to
  provide scalable, efficient inter-domain routing, the remainder of
  this section considers the fundamental design choices for the NR
  routing algorithm.

  Typically the debate surrounding routing algorithms focuses on link
  state and distance vector protocols.  However, simple distance vector
  protocols (i.e., Routing Information Protocol [Hedrick88]), do not
  scale because of convergence problems.  Improved distance vector



Estrin, Rekhter & Hotz                                         [Page 14]

RFC 1322       A Unified Approach to Inter-Domain Routing       May 1992


  protocols, such as those discussed in [Jaffee82, Zaumen91, Shin87],
  have been developed to address this issue using synchronization
  mechanisms or additional path information.  In the case of inter-
  domain routing, having additional path information is essential to
  supporting policy.  Therefore, the algorithms we consider for NR are
  link state and one we call path vector (PV).  Whereas the
  characteristics of link state algorithms are generally understood
  (for example, [Zaumen 91]), we must digress from our evaluation
  discussion to describe briefly the newer concept of the PV algorithm
  [Footnote: We assume that some form of SPF algorithm will be used to
  compute paths over the topology database when LS algorithms are used
  [Dijkstra59, OSPF]].

3.2.1 Path Vector Protocol Overview

  The Border Gateway Protocol (BGP) (see [BGP91]) and the Inter Domain
  Routing Protocol (IDRP) (see [IDRP91]) are examples of path vector
  (PV) protocols [Footnote: BGP is an inter-autonomous system routing
  protocol for TCP/IP internets.  IDRP is an OSI inter-domain routing
  protocol that is being progressed toward standardization within ISO.
  Since in terms of functionality BGP represents a proper subset of
  IDRP, for the rest of the paper we will only consider IDRP.].

  The routing algorithm employed by PV bears a certain resemblance to
  the traditional Bellman-Ford routing algorithm in the sense that each
  border router advertises the destinations it can reach to its
  neighboring BRs.  However,the PV routing algorithm augments the
  advertisement of reachable destinations with information that
  describes various properties of the paths to these destinations.

  This information is expressed in terms of path attributes.  To
  emphasize the tight coupling between the reachable destinations and
  properties of the paths to these destinations, PV defines a route as
  a pairing between a destination and the attributes of the path to
  that destination.  Thus the name, path-vector protocol, where a BR
  receives from its neighboring BR a vector that contains paths to a
  set of destinations [Footnote: The term "path-vector protocol" bears
  an intentional similarity to the term "distance-vector protocol",
  where a BR receives from each of its neighbors a vector that contains
  distances to a set of destinations.].  The path, expressed in terms
  of the domains (or confederations) traversed so far, is carried in a
  special path attribute which records the sequence of routing domains
  through which the reachability information has passed.  Suppression
  of routing loops is implemented via this special path attribute, in
  contrast to LS and distance vector which use a globally-defined
  monotonically-increasing metric for route selection [Shin87].

  Because PV does not require all routing domains to have homogeneous



Estrin, Rekhter & Hotz                                         [Page 15]

RFC 1322       A Unified Approach to Inter-Domain Routing       May 1992


  criteria (policies) for route selection, route selection policies
  used by one routing domain are not necessarily known to other routing
  domains.  To maintain the maximum degree of autonomy and independence
  between routing domains, each domain which participates in PV may
  have its own view of what constitutes an optimal route.  This view is
  based solely on local route selection policies and the information
  carried in the path attributes of a route.  PV standardizes only the
  results of the route selection procedure, while allowing the
  selection policies that affect the route selection to be non-standard
  [Footnote: This succinct observation is attributed to Ross Callon
  (Digital Equipment Corporation).].

3.3 Complexity

  Given the above description of PV algorithms, we can compare them to
  LS algorithms in terms of the three complexity parameters defined
  earlier.

3.3.1 Storage Overhead

  Without any aggregation of routing information, and ignoring storage
  overhead associated with transit constraints, it is possible to show
  that under some rather general assumptions the average case RIB
  storage overhead of the PV scheme for a single TOS ranges between
  O(N) and O(Nlog(N)), where N is the total number of routing domains
  ([Rekhter91]).  The LS RIB, with no aggregation of routing
  information, no transit constraints, a single homogeneous route
  selection policy across all the domains, and a single TOS, requires a
  complete domain-level topology map whose size is O(N).

  Supporting heterogeneous route selection and transit policies with
  hop-by-hop forwarding and LS requires each domain to know all other
  domains route selection and transit policies.  This may significantly
  increase the amount of routing information that must be stored by
  each domain.  If the number of policies advertised grows with the
  number of domains, then the storage could become unsupportable.  In
  contrast, support for heterogeneous route selection policies has no
  effect on the storage complexity of the PV scheme (aside from simply
  storing the local policy information).  The presence of transit
  constraints in PV results in a restricted distribution of routing
  information, thus further reducing storage overhead.  In contrast,
  with LS no such reduction is possible since each domain must know
  every other domain's transit policies.  Finally, some of the transit
  constraints (e.g., path sensitive constraints) can be expressed in a
  more concise form in PV (see aggregation discussion below).

  The ability to further restrict storage overhead is facilitated by
  the PV routing algorithm, where route computation precedes routing



Estrin, Rekhter & Hotz                                         [Page 16]

RFC 1322       A Unified Approach to Inter-Domain Routing       May 1992


  information dissemination, and only routing information associated
  with the routes selected by a domain is distributed to adjacent
  domains.  In contrast, route selection with LS is decoupled from the
  distribution of routing information, and has no effect on such
  distribution.

  While theoretically routing information aggregation can be used to
  reduce storage complexity in both LS and PV, only aggregation of
  topological information would yield the same gain for both.
  Aggregating transit constraints with LS may result in either reduced
  connectivity or less information reduction, as compared with PV.
  Aggregating heterogeneous route selection policies in LS is highly
  problematic, at best.  In PV, route selection policies are not
  distributed, thus making aggregation of route selection policies a
  non-issue [Footnote: Although a domain's selection policies are not
  explicitly distributed, they have an impact on the routes available
  to other domains.  A route that may be preferred by a particular
  domain, and not prohibited by transit restrictions, may still be
  unavailable due to the selection policies of some intermediate
  domain.  The ability to compute and install alternative routes that
  may be lost using hop-by-hop routing (either LS of PV) is the
  critical functionality provided by SDR.].

  Support for multiple TOSs has the same impact on storage overhead for
  both LS and for PV.

  Potentially the LS FIB may be smaller if routes are computed at each
  node on demand.  However, the gain of such a scheme depends heavily
  on the traffic patterns, memory size, and caching strategy.  If there
  is not a high degree of locality, severely degraded performance can
  result due to excessive overall computation time and excessive
  computation delay when forwarding packets to a new destination.  If
  demand driven route computation is not used for LS, then the size of
  the FIB would be the same for both LS and PV.

















Estrin, Rekhter & Hotz                                         [Page 17]

RFC 1322       A Unified Approach to Inter-Domain Routing       May 1992


3.3.2 Route Computation Complexity

  Even if all domains employ exactly the same route selection policy,
  computational complexity of PV is smaller than that of LS.  The PV
  computation consists of evaluating a newly arrived route and
  comparing it with the existing one [Footnote: Some additional checks
  are required when an update is received to insure that a routing loop
  has not been created.].  Whereas, conventional LS computation
  requires execution of an SPF algorithm such as Dijkstra's.

  With PV, topology changes only result in the recomputation of routes
  affected by these changes, which is more efficient than complete
  recomputation.  However, because of the inclusion of full path
  information with each distance vector, the effect of a topology
  change may propagate farther than in traditional distance vector
  algorithms.  On the other hand, the number of affected domains will
  still be smaller with PV than with conventional LS hop-by-hop
  routing.  With PV, only those domains whose routes are affected by
  the changes have to recompute, while with conventional LS hop-by-hop
  routing all domains must recompute.  While it is also possible to
  employ partial recomputation with LS (i.e., when topology changes,
  only the affected routes are recomputed), "studies suggest that with
  a very small number of link changes (perhaps 2) the expected
  computational complexity of the incremental update exceeds the
  complete recalculation" ([ANSI87-150R]).  However these checks are
  much simpler than executing a full SPF algorithm.

  Support for heterogeneous route selection policies has serious
  implications for the computational complexity.  The major problem
  with supporting heterogeneous route selection policies with LS is the
  computational complexity of the route selection itself.
  Specifically, we are not aware of any algorithm with less than
  exponential time complexity that would be capable of computing routes
  to all destinations, with LS hop-by-hop routing and heterogeneous
  route selection policies.  In contrast, PV allows each domain to make
  its route selection autonomously, based only on local policies.
  Therefore support for dissimilar route selection policies has no
  negative implications for the complexity of route computation in PV.
  Similarly, providing support for path-sensitive transit policies in
  LS implies exponential computation, while in PV such support has no
  impact on the complexity of route computation.

  In summary, because NR will rely primarily on precomputation of
  routes, aggregation is essential to the long-term scalability of the
  architecture.  To the extent aggregation is facilitated with PV, so
  is reduced computational complexity.  While similar arguments may be
  made for LS, the aggregation capabilities that can be achieved with
  LS are more problematic because of LS' reliance on consistent



Estrin, Rekhter & Hotz                                         [Page 18]

RFC 1322       A Unified Approach to Inter-Domain Routing       May 1992


  topology maps at each node.  It is also not clear what additional
  computational complexity will be associated with aggregation of
  transit constraints and heterogeneous route selection policies in LS.

3.3.3 Bandwidth Overhead

  PV routing updates include fully-expanded information.  A complete
  route for each supported TOS is advertised.  In LS, TOS only
  contributes a factor increase per link advertised, which is much less
  than the number of complete routes.  Although TOSs may be encoded
  more efficiently with LS than with PV, link state information is
  flooded to all domains, while with PV, routing updates are
  distributed only to the domains that actually use them.  Therefore,
  it is difficult to make a general statement about which scheme
  imposes more bandwidth overhead, all other factors being equal.

  Moreover, this is perhaps really not an important difference, since
  we are more concerned with the number of messages than with the
  number of bits (because of compression and greater link bandwidth, as
  well as the increased physical stability of links).

3.4 Aggregation

  Forming confederations of domains, for the purpose of consistent,
  hop-by-hop, LS route computation, requires that domains within a
  confederation have consistent policies.  In addition, LS
  confederation requires that any lower level entity be a member of
  only one higher level entity.  In general, no intra-confederation
  information can be made visible outside of a confederation, or else
  routing loops may occur as a result of using an inconsistent map of
  the network at different domains.  Therefore, the use of
  confederations with hop-by-hop LS is limited because each domain (or
  confederation) can only be a part of one higher level confederation
  and only export policies consistent with that confederation (see
  examples in Section 2.2).  These restrictions are likely to impact
  the scaling capabilities of the architecture quite severely.

  In comparison, PV can accommodate different confederation definitions
  because looping is avoided by the use of full path information.
  Consistent network maps are not needed at each route server, since
  route computation precedes routing information dissemination.
  Consequently, PV confederations can be nested, disjoint, or
  overlapping.  A domain (or confederation) can export different
  policies or TOS as part of different confederations, thus providing
  the extreme flexibility that is crucial for the overall scaling and
  extensibility of the architecture.

  In summary, aggregation is essential to achieve acceptable complexity



Estrin, Rekhter & Hotz                                         [Page 19]

RFC 1322       A Unified Approach to Inter-Domain Routing       May 1992


  bounds, and flexibility is essential to achieve acceptable
  aggregation across the global, decentralized internet.  PV's
  strongest advantage stems from its flexibility.

3.5 Policy

  The need to allow expression of transit policy constraints on any
  route (i.e., NR routes as well as SDR routes), by itself, can be
  supported by either LS or PV.  However, the associated complexities
  of supporting transit policy constraints are noticeably higher for LS
  than for PV.  This is due to the need to flood all transit policies
  with LS, where with PV transit policies are controlled via restricted
  distribution of routing information.  The latter always imposes less
  overhead than flooding.

  While all of the transit constraints that can be supported with LS
  can be supported with PV, the reverse is not true.  In other words,
  there are certain transit constraints (e.g., path-sensitive transit
  constraints) that are easily supported with PV, and are prohibitively
  expensive (in terms of complexity) to support in LS.  For example, it
  is not clear how NR with LS could support something like ECMA-style
  policies that are based on hierarchical relations between domains,
  while support for such policies is straightforward with PV.

  As indicated above, support for heterogeneous route selection
  policies, in view of its computational and storage complexity, is
  impractical with LS hop-by-hop routing.  In contrast, PV can
  accommodate heterogeneous route selection with little additional
  overhead.

3.6 Information Hiding

  PV has a clear advantage with respect to selective information
  hiding.  LS with hop-by-hop routing hinges on the ability of all
  domains to have exactly the same information; this clearly
  contradicts the notion of selective information hiding.  That is, the
  requirement to perform selective information hiding is unsatisfiable
  with LS hop-by-hop routing.

3.7 Commonality between NR and SDR Components

  In [Breslau-Estrin91] we argued for the use of LS in conjunction with
  SDR.  Therefore there is some preference for using LS with NR.
  However, there are several reasons why NR and SDR would not use
  exactly the same routing information, even if they did use the same
  algorithm.  Moreover, there are several opportunities for unifying
  the management (distribution and storage) of routing and forwarding
  information, even if dissimilar algorithms are used.



Estrin, Rekhter & Hotz                                         [Page 20]

RFC 1322       A Unified Approach to Inter-Domain Routing       May 1992


  In considering the differences between NR and SDR we must address
  several areas:

    1. Routing information and distribution protocol: LS for SDR is
       quite different from the LS in NR.  For example, SDR LS need
       not aggregate domains; to the contrary SDR LS  requires detailed
       information to generate special routes.

       In addition, consistency requirements (essential for NR) are
       unnecessary for the SDR component.  Therefore LS information for
       the SDR component can be retrieved on-demand, while the NR
       component must use flooding of topology information.

    2. Route computation algorithm: It is not clear whether route
       computation algorithm(s)  can be shared between the SDR and NR
       components, given the difficulty of supporting  heterogeneous
       route selection policies in NR.

    3. Forwarding information: The use of dissimilar route computation
       algorithms does not preclude common handling of packet
       forwarding.  Even if LS were used for NR, the requirement would
       be the same, i.e., that the forwarding agent can determine
       whether to use a NR  precomputed route or an SDR installed route
       to forward a particular data packet.

  In conclusion, using similar algorithms and mechanisms for SDR and NR
  components would have benefits.  However, these benefits do not
  dominate the other factors as discussed before.























Estrin, Rekhter & Hotz                                         [Page 21]

RFC 1322       A Unified Approach to Inter-Domain Routing       May 1992


3.8 Summary

  Given the performance complexity issues associated with global
  routing, aggregation of routing information is essential; at the same
  time we have argued that such aggregation must be flexible.  Given
  the difficulties of supporting LS hop-by-hop routing in the presence
  of (a) flexible aggregation, (b) heterogeneous route selection
  policies, and (c) incomplete or inconsistent routing information, we
  see no alternative but to employ PV for the NR component of our
  architecture.

  Based on the above tradeoffs, our NR component employs a PV
  architecture, where route computation and installation is done in a
  distributed fashion by the routers participating in the NR component
  [Footnote: Packet forwarding along routes produced by the NR
  component can be accomplished by either source routing or hop-by-hop
  routing.  The latter is the primary choice because it reduces the
  amount of state in routers (if route setup is employed), or avoids
  encoding an explicit source route in network layer packets.  However,
  the architecture does not preclude the use of source routing (or
  route setup) along the routes computed, but not installed, by the NR
  component.].

  The distributed algorithm combines some of the features of link state
  with those of distance vector algorithms; in addition to next hop
  information, the NR component maintains path attributes for each
  route (e.g., the list of domains or routing domain confederations
  that the routing information has traversed so far).  The path
  attributes that are carried along with a route express a variety of
  routing policies, and make explicit the entire route to the
  destination.  With aggregation, this is a superset of the domains
  that form the path to the destination.



















Estrin, Rekhter & Hotz                                         [Page 22]

RFC 1322       A Unified Approach to Inter-Domain Routing       May 1992


4.0 Source-demand routing (SDR)

  Inter-domain routers participating in the SDR component forward
  packets according to routing information computed and installed by
  the domain that originates the traffic (source routing domain).

  It is important to realize that requiring route installation by the
  source routing domain is not a matter of choice, but rather a
  necessity.  If a particular route is used by a small number of
  domains (perhaps only one) then it is more appropriate to have the
  source compute and install the special route instead of burdening the
  intermediate nodes with the task of looking for and selecting a route
  with the specialized requirements.  In addition, if the demand for
  the route is unpredictable, and thus can be determined only by the
  source, it should be up to the source to install the route.

  In general, information that is used by source routing domains for
  computing source-demand routes reflects administrative (but not
  operational) status of the routing facilities (i.e., configured
  topology and policy) [Footnote: If SDR uses NR information then
  operational status could be considered in some route selection.].
  Consequently, it is possible for a source routing domain to compute a
  route that is not operational at route installation time.  The SDR
  component attempts to notify the source domain of failures when route
  installation is attempted.  Similarly, the SDR component provides
  mechanisms for the source routing domain to be notified of failures
  along previously-installed active routes.  In other words, the SDR
  component performs routing that is adaptive to topological changes;
  however, the adaptability is achieved as a consequence of the route
  installation and route management mechanisms.  This is different from
  the NR component, where status changes are propagated and
  incorporated by nodes as soon as possible.  Therefore, to allow
  faster adaptation to changes in the operational status of routing
  facilities, the SDR component allows the source domain to switch to a
  route computed by the NR component, if failure along the source-
  demand route is detected (either during the route installation phase,
  or after the route is installed), and if policy permits use of the NR
  route.

  The NR component will group domains into confederations to achieve
  its scaling goals (see [IDRP91]).  In contrast, SDR will allow an
  AD-level route to be used by an individual domain without allowing
  use by the entire confederation to which the domain belongs.
  Similarly, a single transit domain may support a policy or special
  TOS that is not supported by other domains in its confederation(s).
  In other words, the architecture uses SDR to support non-
  hierarchical, non-aggregated policies where and when needed.
  Consequently, SDR by itself does not have the scaling properties of



Estrin, Rekhter & Hotz                                         [Page 23]

RFC 1322       A Unified Approach to Inter-Domain Routing       May 1992


  NR.  In compensation, SDR does not require a complete, global domain
  map if portions of the world are never traversed or communicated
  with.  As a result of the looser routing structure, SDR does not
  guarantee that a participating source routing domain will always have
  sufficient information to compute a route to a destination.  In
  addition, if the domain does have sufficient information, it is
  possible that the quantity may be large enough to preclude storage
  and/or route computation in a timely fashion.  However, despite the
  lack of guarantees, it is a goal of the architecture to provide
  efficient methods whereby sources can obtain the information needed
  to compute desired routes [Footnote: The primary goal of policy or
  TOS routing is to compute a route that satisfies a set of specialized
  requirements, and these requirements take precedence over optimality.
  In other words, even if a routing domain that participates in SDR or
  NR has sufficient information to compute a route, given a particular
  set of requirements, the architecture does not guarantee that the
  computed route is optimal.].

  Essential to SDR is the assumption that the routes installed on
  demand will be used sparingly.  The architecture assumes that at any
  given moment the set of all source-demand routes installed in an
  internet forms a small fraction of the total number of source-demand
  routes that can potentially be installed by all the routing domains.
  It is an assumption of the architecture that the number of routes
  installed in a BR by the SDR component should be on the order of log
  N (where N is the total number of routing domains in the Internet),
  so that the scaling properties of the SDR component are comparable
  with the scaling properties of the NR component.  The NR component
  achieves this property as a result of hierarchy.

  Note that the above requirement does not imply that only a few
  domains can participate in SDR, or that routes installed by the SDR
  component must have short life times.  What the requirement does
  imply, is that the product of the number of routes specified by
  domains that participate in SDR, times the average SDR-route life
  time, is bounded.  For example, the architecture allows either a
  small number of SDR routes with relatively long average life times,
  or a large number of SDR routes with relatively short average life
  times.  But the architecture clearly prohibits a large number of SDR
  routes with relatively long average life times.  The number of SDR
  routes is a function of the number of domains using SDR routes and
  the number of routes used per source domain.

  In summary, SDR is well suited for traffic that (1) is not widely-
  used enough (or is not sufficiently predictable or steady) to justify
  computation and maintenance by the NR component, and (2) whose
  duration is significantly longer than the time it takes to perform
  the route installation procedure.



Estrin, Rekhter & Hotz                                         [Page 24]

RFC 1322       A Unified Approach to Inter-Domain Routing       May 1992


  The architecture does not require all domains in the Internet to
  participate in SDR.  Therefore, issues of scalability (with respect
  to the size of the Internet) become less crucial (though still
  important) to the SDR component.  Instead, the primary focus of the
  SDR component is shifted towards the ability to compute routes that
  satisfy specialized requirements, where we assume that the total
  number of domains requiring special routes simultaneously through the
  same part of the network is small relative to the total population.

4.1 Path Vector vs. Link State for SDR

  It is feasible to use either a distance vector or link state method
  of route computation along with source routing.  One could imagine,
  for instance, a protocol like BGP in which the source uses the full
  AD path information it receives in routing updates to create a source
  route. Such a protocol could address some of the deficiencies
  identified with distance vector, hop-by-hop designs.  However, we opt
  against further discussion of such a protocol because there is less
  to gain by using source routing without also using a link state
  scheme.  The power of source routing, in the context of inter-AD
  policy routing, is in giving the source control over the entire
  route.  This goal cannot be realized fully when intermediate nodes
  control which legal routes are advertised to neighbors, and therefore
  to a source.

  In other words, intermediate nodes should be able to preclude the use
  of a route by expressing a transit policy, but if a route is not
  precluded (i.e.,  is legal according to all ADs in the route), the
  route should be made available to the source independent of an
  intermediate domain's preferences for how its own traffic flows.

  Therefore, the SDR component employs an IDPR-like architecture in
  which link-state style updates are distributed with explicit policy
  terms included in each update along with the advertising node's
  connectivity.

4.2 Distribution of Routing Information

  By using a hop-by-hop NR component based on PV to complement the
  source-routing SDR component, we have alleviated the pressure to
  aggregate SDR forwarding information; the large percentage of inter-
  domain traffic carried, simultaneously, by any particular border
  router will be forwarded using aggregated NR forwarding information.
  However, the use of NR does not address the other major scaling
  problem associated with SDR: that of distributing, storing, and
  computing over a complete domain-level topology map.  In this section
  we describe promising opportunities for improving the scaling
  properties of SDR routing information distribution, storage, and



Estrin, Rekhter & Hotz                                         [Page 25]

RFC 1322       A Unified Approach to Inter-Domain Routing       May 1992


  computation.

  Note that we do not propose to solve this problem in the same way
  that we solve it for NR.  A priori abstraction will not be employed
  since different domains may require different methods of abstracting
  the same routing information.  For example, if we aggregate routing
  information of domains that do not share the same policy and TOS
  characteristics (i.e., services), then outside of the aggregate, only
  those services that are offered by all domains in the aggregate will
  be advertised.  In order to locate special routes, SDR only uses
  aggregates when the component domains (and in turn the aggregate)
  advertise the required TOS and policy descriptions.  When the
  required TOS or policy characteristics are not offered by an
  aggregate, full information about the component domains is used to
  construct a route through those domains that do support the
  particular characteristics.  Consequently, we need some other, more
  flexible, means of reducing the amount of information distributed and
  held.  We address two issues in turn: distribution of configured
  topology and policy information, and distribution of dynamic status
  information.

4.2.1 Configured Information

  Information about the existence of inter-domain links, and policies
  maintained by domains, changes slowly over time.  This is referred to
  as configured information.  In the current IDPR specification
  complete, global, configuration information is kept by a route server
  in each domain.  Route servers (RS) are the entities that compute
  source routes.  On startup a RS can download the connectivity
  database from a neighbor RS; as domains, inter-domain links, or
  policies change, the changes are flooded to a RS in each domain.

  We have not yet specified the exact mechanisms for distributing
  configured connectivity information for SDR.  However, unlike the
  current IDPR specification, the SDR component will not flood all
  configured information globally.  Several alternate methods for
  organizing and distributing information are under investigation.

  Configured information may be regularly distributed via an out-of-
  band channel, e.g., CD/ROM.  In a similar vein, this information
  could be posted in several well-known locations for retrieval, e.g.,
  via FTP.  Between these "major" updates, aggregated collections of
  changes may be flooded globally.  Moreover, limited flooding (e.g.,
  by hop-count) could be used as appropriate to the "importance" of the
  change; while a policy change in a major backbone may still be
  flooded globally, a new inter-regional link may be flooded only
  within those regions, and information about an additional link to a
  non-transit domain may not be available until the next regularly-



Estrin, Rekhter & Hotz                                         [Page 26]

RFC 1322       A Unified Approach to Inter-Domain Routing       May 1992


  scheduled "major" distribution.

  Changes that are not distributed as they occur will not necessarily
  be discovered.  However, a route server may learn pertinent
  information by direct query of remote servers, or through error
  messages resulting from traffic sent along failed routes.  Complete
  global flooding may be avoided by using some combination of these
  mechanisms.

  Even if an initial implementation uses a simple global flood, we must
  study the problem of structuring connectivity information such that
  it can be retrieved or distributed in a more selective manner, while
  still allowing sources to discover desired routes.  For example, we
  imagine RSs requesting filtered information from each other.  How the
  RSs should define filters that will get enough information to find
  special routes, while also effectively limiting the information, is
  an open question.  Again, the question is how to effectively
  anticipate and describe what information is needed in advance of
  computing the route.

  The essential dilemma is that networks are not organized in a nicely
  geographical or topologically consistent manner (e.g., it is not
  effective to ask for all networks going east-west that are within a
  certain north-south region of the target), hence a source domain does
  not know what information it needs (or should ask for) until it
  searches for, and discovers, the actual path.  Even with a central
  database, techniques are needed to structure configuration
  information so that the potential paths that are most likely to be
  useful are explored first, thereby reducing the time required for
  route computation.

  One promising approach organizes information using route fragments
  (partial paths) [Footnote: Route fragments were first suggested by
  Dave Clark and Noel Chiappa.].  Although the number of route
  fragments grows faster than the number of domains (at least O(N^2)),
  we can selectively choose those that will be useful to compute
  routes.  In particular, for each stub domain, fragments would be
  constructed to several well-known backbones [Footnote: Route
  fragments may be computed by a destination's route server and either
  made available via information service queries or global flooding.
  In addition, NR computed routes may be used as SDR route fragments.].
  Among its benefits, this approach aggregates domain information in a
  manner useful for computing source-routes, and provides an index,
  namely the destination, which facilitates on-demand reference and
  retrieval of information pertinent to a particular route computation.
  At this point, it is not clear how route fragments will affect SDR's
  ability to discover non-hierarchical routes.




Estrin, Rekhter & Hotz                                         [Page 27]

RFC 1322       A Unified Approach to Inter-Domain Routing       May 1992


4.2.2 Dynamic Status Information

  Assuming a technique for global or partial distribution of configured
  information, a second issue is whether, and how, to distribute
  dynamic status information (i.e., whether an inter-domain connection
  is up or down).

  In the current version of IDPR, dynamic status information is flooded
  globally in addition to configuration information.  We propose to
  distribute status information based strictly on locality.  First,
  dynamic information will be advertised within a small hop-count
  radius.  This simple and low-overhead mechanism exploits topological
  locality.  In addition to flooding status updates to nearby nodes, we
  also want to provide more accurate route information for long
  distance communications that entails more than a few network hops.
  Reverse path update (RPU) is a mechanism for sending dynamic status
  information to nodes that are outside the k-hop radius used for
  updates, but that nevertheless would obtain better service (fewer
  failed setups) by having access to the dynamic information [Estrin-
  etal91].

  RPU uses the existing active routes (represented by installed setup
  state or by a cache of the most recent source routes sent via the
  node in question) as a hint for distribution of event notifications.
  Instead of reporting only the status of the route being used, RPU
  reports the status of the domain's other inter-domain connections.
  If source routing exhibits route locality, the source is more likely
  to use other routes going through the node in question; in any case
  the overhead of the information about other links will be minimal.

  In this way, sources will receive status information from regions of
  the network through which they maintain active routes, even if those
  regions are more than k hops away.  Using such a scheme, k could be
  small to maximize efficiency, and RPU could be used to reduce the
  incidence of failed routes resulting from inaccurate status
  information.  This will be useful if long-path communication exhibits
  route locality with respect to regions that are closer to the
  destination (and therefore outside the k hop radius of flooded
  information).  In such situations, flooding information to the source
  of the long route would be inefficient because k would have to be
  equal to the length of the route, and in almost all cases, the
  percentage of nodes that would use the information decreases
  significantly with larger k.

4.3 Source-Demand Route Management

  SDR may be built either on top of the network layer supported by the
  NR component, or in parallel with it.  SDR forwarding will be



Estrin, Rekhter & Hotz                                         [Page 28]

RFC 1322       A Unified Approach to Inter-Domain Routing       May 1992


  supported via two techniques: loose source-routing and route setup.

  The first technique, loose source-routing, would allow the originator
  of a packet to specify a sequence of domains that the packet should
  traverse on its path to a destination.  Forwarding such a packet
  within a domain, or even between domains within a confederation,
  would be left to intra-domain routing.  This avoids per-connection
  state and supports transaction traffic.

  The second technique, route setup, will be based on mechanisms
  developed for IDPR and described in [IDPR90].  It is well suited to
  conversations that persist significantly longer than a round-trip-
  time.  The setup protocol defines packet formats and the processing
  of route installation request packets (i.e, setup packets).  When a
  source generates a setup packet, the first border router along the
  specified source route checks the setup request, and if accepted,
  installs routing information; this information includes a path ID,
  the previous and next hops, and whatever other accounting-related
  information the particular domain requires.  The setup packet is
  passed on to the next BR in the domain-level source route, and the
  same procedure is carried out [Footnote: The setup packet may be
  forwarded optimistically, i.e., before checks are completed, to
  reduce latency.].  When the setup packet reaches the destination, an
  accept message is propagated back hop by hop, and each BR en route
  activates its routing information.  Subsequent data packets traveling
  along the same path to the destination include a path ID in the
  packet.  That path ID is used to locate the appropriate next-hop
  information for each packet.

  Border routers that support both the NR and the SDR components, must
  be able to determine what forwarding mechanism to use.  That is, when
  presented with a network layer PDU, such a BR should be able to make
  an unambiguous decision about whether forwarding of that PDU should
  be handled by the NR or the SDR component.  Discrimination mechanisms
  are dependent on whether the new network layer introduced by the SDR
  component is built on top of, or in parallel with, the network layers
  supported by the NR component.  Once the discrimination is made,
  packets that have to be forwarded via routes installed by the SDR
  component are forwarded to the exit port associated with the
  particular Path ID in the packet header.  Packets that have to be
  forwarded via routes installed by the NR component are forwarded to
  the exit port associated with the particular destination and Type of
  Service parameters (if present) in their packet headers.

  Next, we describe the primary differences between the IDPR setup
  procedure previously specified, and the procedure we propose to
  develop for this hybrid architecture.




Estrin, Rekhter & Hotz                                         [Page 29]

RFC 1322       A Unified Approach to Inter-Domain Routing       May 1992


  During route installation, if a BR on the path finds that the
  remainder of the indicated route from the BR to the destination is
  identical to the NR route from the BR to the destination, then the BR
  can turn off the SDR route at that point and map it onto the NR
  route.  For this to occur, the specifications of the SDR route must
  completely match those of the NR route.  In addition, the entire
  forward route must be equivalent (i.e., the remaining hops to the
  destination).

  Moreover, if the NR route changes during the course of an active SDR
  route, and the new NR route does not match the SDR route, then the
  SDR route must be installed for the remainder of the way to the
  destination.  Consequently, when an SDR route is mapped onto an NR
  route, the original setup packet must be saved.  A packet traveling
  from a source to destination may therefore traverse both an SDR and
  an NR route segment; however, a packet will not traverse another SDR
  segment after traveling over an NR segment.  However, during
  transient periods packets could traverse the wrong route and
  therefore this must be an optional and controllable feature.

  A source can also request notification if a previously-down link or
  node returns to operation some time after a requested route setup
  fails.  If a BR on the route discovers that the requested next-hop BR
  is not available, the BR can add the source to a notification list
  and when the next-hop BR becomes reachable, a notification can be
  sent back to the source.  This provides a means of flushing out bad
  news when it is no longer true.  For example, a domain might decide
  to route through a secondary route when its preferred route fails;
  the notification mechanism would inform the source in a timely manner
  when its preferred route is available again.

  A third option addresses adaptation after route installation.  During
  packet forwarding along an active SDR route, if a BR finds that the
  SDR route has failed, it may redirect the traffic along an existing
  NR route to the destination.  This adaptation is allowed only if use
  of the NR route does not violate policy; for example, it may provide
  a less desirable type of service.  This is done only if the source
  selects the option at route setup time.  It is also up to the source
  whether it is to be notified of such actions.

  When a SDR route does fail, the detecting BR sends notification to
  the source(s) of the active routes that are affected.  Optionally,
  the detecting BR may include additional information about the state
  of other BRs in the same domain.  In particular, the BR can include
  its domain's most recent "update" indicating that domain's inter-
  domain links and policy.  This can be helpful to the extent there is
  communication locality; i.e., if alternative routes might be used
  that traverse the domain in question, but avoid the failed BR.



Estrin, Rekhter & Hotz                                         [Page 30]

RFC 1322       A Unified Approach to Inter-Domain Routing       May 1992


  In summary, when a route is first installed, the source has several
  options (which are represented by flags in the route setup packet):

    1. If an NR route is available that satisfies all local policy
       and TOS, then use it.  Otherwise...

    2. Indicate whether the source wants to allow the setup to
       default to a NR route if the SDR route setup fails.

    3. Request notification of mapping to a NR route.

    4. Request additional configured information on failure.

    5. Request addition to a notification list for resource
       re-availability.

    6. Allow data packets to be rerouted to a NR route when failure
       happens after setup (so long  as no policy is violated).

    7. Request notification of a reroute of data packets.

    8. Request additional configured information on failure notice
       when the route is active.

    9. Request addition to a notification list if an active route
       fails.

























Estrin, Rekhter & Hotz                                         [Page 31]

RFC 1322       A Unified Approach to Inter-Domain Routing       May 1992


5.0 The Unified Architecture

  In addition to further evaluation and implementation of the proposed
  architecture, future research must investigate opportunities for
  increased unification of the two components of our architecture.  We
  are investigating several opportunities for additional commonality:

    1. Routing Information Base:
       Perhaps a single RIB could be shared by both NR and SDR.
       NR routes can be represented as a directed graph labeled
       with flags (on the nodes or links) corresponding to the
       generic transit constraints.  SDR requires that this graph
       be augmented by links with non-generic policies that have
       been discovered and maintained for computing special routes;
       in addition, special policy flags may be added to links
       already maintained by the NR component.

    2. Information Distribution:
       The NR path vectors could include address(es) of repositories
       for SDR-update information for each AD (or confederation) to
       assist the SDR component in retrieving selective information
       on demand.  For domains with minimal policies, where the space
       required for policy information is smaller than the space
       required for a repository address (e.g., if the policies for
       the domain listed are all wildcard), the NR path vectors could
       include a flag to that effect.

    3. Packet Forwarding:
       We should consider replacing the current IDPR-style network
       layer (which contains a global path identifier used in
       forwarding data packets to the next policy gateway on an
       IDPR route)  with a standard header (e.g., IP or CLNP),
       augmented with some option fields.  This would  unify the
       packet header parsing and forwarding functions for SDR and NR,
       and possibly eliminate some encapsulation overhead.

    4. Reachability Information:
       Currently IDRP distributes network reachability information
       within updates, whereas IDPR only distributes domain
       reachability information.  IDPR uses a domain name service
       function to map network numbers to domain numbers; the latter
       is needed to make the routing decision.   We should consider
       obtaining the network reachability and domain information in
       a unified manner.







Estrin, Rekhter & Hotz                                         [Page 32]

RFC 1322       A Unified Approach to Inter-Domain Routing       May 1992


5.1 Applicability to Various Network Layer Protocols

  The proposed architecture is designed to accommodate such existing
  network layer protocols as IP ([Postel81]), CLNP ([ISO-473-88]), and
  ST-II ([ST2-90]).  In addition, we intend for this architecture to
  support future network layer mechanisms, e.g., Clark and Jacobson's
  proposal or Braden and Casner's Integrated Services IP.  However on
  principal we can not make sweeping guarantees in advance of the
  mechanisms themselves.  In any case, not all of the mentioned
  protocols will be able to utilize all of the capabilities provided by
  the architecture.  For instance, unless the increase in the number of
  different types of services offered is matched by the ability of a
  particular network layer protocol to unambiguously express requests
  for such different types of services, the capability of the
  architecture to support routing in the presence of a large number of
  different types of service is largely academic.  That is, not all
  components of the architecture will have equal importance for
  different network layer protocols.  On the other hand, this
  architecture is designed to serve the future global internetworking
  environment.  The extensive research and development currently
  underway to implement and evaluate network mechanisms for different
  types of service suggests that future networks will offer such
  services.

  One of the fundamental issues in the proposed architecture is the
  issue of single versus multiple protocols.  The architecture does not
  make any assumptions about whether each network layer is going to
  have its own inter-domain routing protocol, or a single inter-domain
  routing protocol will be able to cover multiple network layers
  [Footnote: Similar issue already arose with respect to the intra-
  domain routing protocol, which generated sufficient amount of
  controversy within the Internet community.  It is our opinion, that
  the issue of single versus multiple protocols is more complex for the
  inter-domain routing than for the intra-domain routing.].  That is,
  the proposed architecture can be realized either by a single inter-
  domain routing protocol covering multiple network layers, or by
  multiple inter-domain routing protocols (with the same architecture)
  tailored to a specific network layer [Footnote: If the single
  protocol strategy is adopted, then it is likely that IDRP will be
  used as a base for the NR component.  Since presently IDRP is
  targeted towards CLNP, further work is needed to augment it to
  support IP and ST-II.  If the multiple protocol strategy is adopted,
  then it is likely that BGP will be used as a base for the NR
  component for IP, and IDRP will be used as a base for the NR
  component for CLNP.  Further work is needed to specify protocol in
  support for the NR component for ST-II.  Additional work may be
  needed to specify new features that may be added to BGP.].




Estrin, Rekhter & Hotz                                         [Page 33]

RFC 1322       A Unified Approach to Inter-Domain Routing       May 1992


5.2 Transition

  The proposed architecture is not intended for full deployment in the
  short term future.  We are proposing this architecture as a goal
  towards which we can begin guiding our operational and research
  investment over the next 5 years.

  At the same time, the architecture does not require wholesale
  overhaul of the existing Internet.  The NR component may be phased in
  gradually.  For example, the NR component for IP may be phased in by
  replacing existing EGP-2 routing with BGP routing.  Once the NR
  component is in place, it can be augmented by the facilities provided
  by the SDR component.

  The most critical components of the architecture needed to support
  SDR include route installation and packet forwarding in the routers
  that support SDR.  Participation as a transit routing domain requires
  that the domain can distribute local configuration information (LCI)
  and that some of its routers implement the route installation and
  route management protocols.  Participation as a source requires that
  the domain have access to a RS to compute routes, and that the source
  domain has a router that implements the route installation and route
  management protocols.  In addition, a network management entity must
  describe local configuration information and send it to the central
  repository(ies).  A collection and distribution mechanism must be put
  in place, even if it is centralized.

























Estrin, Rekhter & Hotz                                         [Page 34]

RFC 1322       A Unified Approach to Inter-Domain Routing       May 1992


6.0 Conclusions and Future Work

  In summary, the proposed architecture combines hop-by-hop path-
  vector, and source-routed link-state, protocols, and uses each for
  that which it is best suited: NR uses PV and multiple, flexible,
  levels of confederations to support efficient routing of generic
  packets over generic routes; SDR uses LS computation over a database
  of configured and dynamic information to route special traffic over
  special routes.  In the past, the community has viewed these two as
  mutually exclusive; to the contrary, they are quite complementary and
  it is fortunate that we, as a community, have pursued both paths in
  parallel.  Together these two approaches will flexibly and
  efficiently support TOS and policy routing in very large global
  internets.

  It is now time to consider the issues associated with combining and
  integrating the two.  We must go back and look at both architectures
  and their constituent protocols, eliminate redundancies, fill in new
  holes, and provide seamless integration.

7.0 Acknowledgments

  We would like to thank Hans-Werner Braun (San Diego Supercomputer
  Center), Lee Breslau (USC), Scott Brim (Cornell University), Tony Li
  (cisco Systems), Doug Montgomery (NIST), Tassos Nakassis (NIST),
  Martha Steenstrup (BBN), and Daniel Zappala (USC) for their comments
  on a previous draft.
























Estrin, Rekhter & Hotz                                         [Page 35]

RFC 1322       A Unified Approach to Inter-Domain Routing       May 1992


8.0 References

  [ANSI 87-150R]  "Intermediate System to Intermediate System Intra-
  Domain Routing Exchange Protocol", ANSI X3S3.3/87-150R.

  [BGP 91]  Lougheed, K., and Y. Rekhter, "A Border Gateway Protocol 3
  (BGP-3)", RFC 1267, cisco Systems, T.J. Watson Research Center, IBM
  Corp., October 1991.

  [Breslau-Estrin 91]  Breslau, L., and D. Estrin, "Design and
  Evaluation of Inter-Domain Policy Routing Protocols", To appear in
  Journal  of Internetworking Research and Experience, 1991.  (Earlier
  version appeared in ACM Sigcomm 1990.)

  [Clark 90]  Clark, D., "Policy Routing in Internetworks", Journal of
  Internetworking Research and Experience, Vol.  1, pp. 35-52, 1990.

  [Dijkstra 59]  Dijkstra, E., "A Note on Two Problems in Connection
  with Graphs", Numer. Math., Vol.  1, 1959, pp. 269-271.

  [ECMA89]  "Inter-Domain Intermediate Systems Routing", Draft
  Technical Report ECMA TR/ISR, ECMA/TC32-TG 10/89/56, May 1989.

  [EGP]  Rosen, E., "Exterior Gateway Protocol (EGP)", RFC 827, BBN,
  October 1982.

  [Estrin 89]  Estrin, D., "Policy Requirements for Inter
  Administrative Domain Routing", RFC 1125, USC Computer Science
  Department, November 1989.

  [Estrin-etal91]  Estrin, D., Breslau, L., and L. Zhang, "Protocol
  Mechanisms for Adaptive Routing in Global Multimedia Internets",
  University of Southern California, Computer Science Department
  Technical Report, CS-SYS-91-04, November 1991.

  [Hedrick 88]  Hedrick, C., "Routing Information Protocol", RFC 1058,
  Rutgers University, June 1988.

  [Honig 90]  Honig, J., Katz, D., Mathis, M., Rekhter, Y., and J. Yu,
  "Application of the Border Gateway Protocol in the Internet", RFC
  1164, Cornell Univ. Theory Center, Merit/NSFNET, Pittsburgh
  Supercomputing Center, T.J. Watson Research Center, IBM Corp., June
  1990.

  [IDPR90]  Steenstrup, M., "Inter-Domain Policy Routing Protocol
  Specification and Usage: Version 1", Work in Progress, February 1991.





Estrin, Rekhter & Hotz                                         [Page 36]

RFC 1322       A Unified Approach to Inter-Domain Routing       May 1992


  [IDRP91]  "Intermediate System to Intermediate System Inter-domain
  Routeing Exchange Protocol", ISO/IEC/ JTC1/SC6 CD10747.

  [ISIS10589]  "Information Processing Systems - Telecommunications and
  Information Exchange between Systems - Intermediate System to
  Intermediate System Intra-Domain Routing Exchange Protocol for use in
  Conjunction with the protocol for providing the Connectionless-mode
  Network Service (ISO 8473)", ISO/IEC 10589.

  [ISO-473 88]  "Protocol for providing the connectionless-mode network
  service", ISO 8473, 1988.

  [Jaffee 82]  Jaffee, J., and F. Moss, "A Responsive Distributed
  Routing Algorithm for Computer Networks", IEEE Transactions on
  Communications, July 1982.

  [Little 89]  Little, M., "Goals and Functional Requirements for
  Inter-Autonomous System Routing", RFC 1126, SAIC, October 1989.

  [Oran 89]  Oran, D., "Expert's Paper: The Relationship between
  Addressing and Routeing", ISO/JTC1/SC6/WG2, 1989.

  [OSPF]  Moy, J., "The Open Shortest Path First (OSPF) Specification",
  RFC 1131, Proteon, October 1989.

  [Postel 81]  Postel, J., "Internet Protocol", RFC 791, DARPA,
  September 1981.

  [Rekhter 91]  Rekhter, Y., "IDRP protocol analysis: storage
  complexity", IBM Research Report RC17298(#76515), October 1991.

  [Shin87] Shin, K., and M. Chen, "Performance Analysis of Distributed
  Routing Strategies Free of Ping-Pong-Type Looping", IEEE Transactions
  on Computers, February 1987.

  [ST2-90]  Topolcic, C., "Experimental Internet Stream Protocol,
  version 2 (ST II)", RFC 1190, CIP Working Group, October 1990.

  [Zaumen 91] Zaumen, W., and J. Garcia-Luna-Aceves, "Dynamics of Link
  State and Loop-free Distance-Vector Routing Algorithms", ACM Sigcomm
  '91, Zurich, Switzerland, September 1991.

  [Zhang 91] Zhang, L., "Virtual Clock: A New Traffic Control Algorithm
  for Packet Switching Networks".







Estrin, Rekhter & Hotz                                         [Page 37]

RFC 1322       A Unified Approach to Inter-Domain Routing       May 1992


Security Considerations

  Security issues are not discussed in this memo.

Authors' Addresses

  Deborah Estrin
  University of Southern California
  Computer Science Department, MC 0782
  Los Angeles, California 90089-0782

  Phone: (310) 740-4524
  EMail: [email protected]


  Yakov Rekhter
  IBM T.J. Watson Research Center
  P.O. Box 218
  Yorktown Heights, New York 10598

  Phone: (914) 945-3896
  EMail: [email protected]


  Steven Hotz
  University of Southern California
  Computer Science Department, MC 0782
  Los Angeles, California 90089-0782

  Phone: (310) 822-1511
  EMail: [email protected]




















Estrin, Rekhter & Hotz                                         [Page 38]


You are viewing proxied material from gopher.rbfh.de. The copyright of proxied material belongs to its original authors. Any comments or complaints in relation to proxied material should be directed to the original authors of the content concerned. Please see the disclaimer for more details.