RFC 5773






Internet Research Task Force (IRTF)                            E. Davies
Request for Comments: 5773                              Folly Consulting
Category: Historic                                              A. Doria
ISSN: 2070-1721                                                      LTU
                                                           February 2010


       Analysis of Inter-Domain Routing Requirements and History

Abstract



   This document analyzes the state of the Internet domain-based routing
   system, concentrating on Inter-Domain Routing (IDR) and also
   considering the relationship between inter-domain and intra-domain
   routing.  The analysis is carried out with respect to RFC 1126 and
   other IDR requirements and design efforts looking at the routing
   system as it appeared to be in 2001 with editorial additions
   reflecting developments up to 2006.  It is the companion document to
   "A Set of Possible Requirements for a Future Routing Architecture"
   (RFC 5772), which is a discussion of requirements for the future
   routing architecture, addressing systems developments and future
   routing protocols.  This document summarizes discussions held several
   years ago by members of the IRTF Routing Research Group (IRTF RRG)
   and other interested parties.  The document is published with the
   support of the IRTF RRG as a record of the work completed at that
   time, but with the understanding that it does not necessarily
   represent either the latest technical understanding or the technical
   consensus of the research group at the date of publication.

Status of This Memo



   This document is not an Internet Standards Track specification; it is
   published for the historical record.

   This document defines a Historic Document for the Internet community.
   This document is a product of the Internet Research Task Force
   (IRTF).  The IRTF publishes the results of Internet-related research
   and development activities.  These results might not be suitable for
   deployment.  This RFC represents the individual opinion(s) of one or
   more members of the Routing Research Group of the Internet Research
   Task Force (IRTF).  Documents approved for publication by the IRSG
   are not a candidate for any level of Internet Standard; see Section 2
   of RFC 5741.

   Information about the current status of this document, any errata,
   and how to provide feedback on it may be obtained at
   http://www.rfc-editor.org/info/rfc5773.




Davies & Doria                  Historic                        [Page 1]

RFC 5773                       IDR History                 February 2010


Copyright Notice



   Copyright (c) 2010 IETF Trust and the persons identified as the
   document authors.  All rights reserved.

   This document is subject to BCP 78 and the IETF Trust's Legal
   Provisions Relating to IETF Documents
   (http://trustee.ietf.org/license-info) in effect on the date of
   publication of this document.  Please review these documents
   carefully, as they describe your rights and restrictions with respect
   to this document.








































Davies & Doria                  Historic                        [Page 2]

RFC 5773                       IDR History                 February 2010


Table of Contents



   1. Provenance of This Document .....................................4
   2. Introduction ....................................................5
      2.1. Background .................................................7
   3. Historical Perspective ..........................................7
      3.1. The Legacy of RFC 1126 .....................................7
           3.1.1. General Requirements ................................8
           3.1.2. "Functional Requirements" ..........................13
           3.1.3. "Non-Goals" ........................................21
      3.2. ISO OSI IDRP, BGP, and the Development of Policy Routing ..25
      3.3. Nimrod Requirements .......................................30
      3.4. PNNI ......................................................32
   4. Recent Research Work ...........................................33
      4.1. Developments in Internet Connectivity .....................33
      4.2. DARPA NewArch Project .....................................34
           4.2.1. Defending the End-to-End Principle .................35
   5. Existing Problems of BGP and the Current Inter-/Intra-Domain
      Architecture ...................................................35
      5.1. BGP and Auto-Aggregation ..................................36
      5.2. Convergence and Recovery Issues ...........................36
      5.3. Non-Locality of Effects of Instability and
           Misconfiguration ..........................................37
      5.4. Multi-Homing Issues .......................................37
      5.5. AS Number Exhaustion ......................................38
      5.6. Partitioned ASs ...........................................39
      5.7. Load Sharing ..............................................40
      5.8. Hold-Down Issues ..........................................40
      5.9. Interaction between Inter-Domain Routing and
           Intra-Domain Routing ......................................40
      5.10. Policy Issues ............................................42
      5.11. Security Issues ..........................................42
      5.12. Support of MPLS and VPNS .................................43
      5.13. IPv4/IPv6 Ships in the Night .............................43
      5.14. Existing Tools to Support Effective Deployment of
            Inter-Domain Routing .....................................44
           5.14.1. Routing Policy Specification Language RPSL
                   (RFC 2622 and RFC 2650) and RIPE NCC Database
                   (RIPE 157) ........................................44
   6. Security Considerations ........................................45
   7. Acknowledgments ................................................45
   8. Informative References .........................................46









Davies & Doria                  Historic                        [Page 3]

RFC 5773                       IDR History                 February 2010


1.  Provenance of This Document



   In 2001, the IRTF Routing Research Group (IRTF RRG) chairs, Abha
   Ahuja and Sean Doran, decided to establish a sub-group to look at
   requirements for inter-domain routing (IDR).  A group of well-known
   routing experts was assembled to develop requirements for a new
   routing architecture.  Their mandate was to approach the problem
   starting from a blank slate.  This group was free to take any
   approach, including a revolutionary approach, in developing
   requirements for solving the problems they saw in inter-domain
   routing.  Their eventual approach documented requirements for a
   complete future routing and addressing architecture rather than just
   the requirements for IDR.

   Simultaneously, an independent effort was started in Sweden with a
   similar goal.  A team, calling itself Babylon, with participation
   from vendors, service providers, and academia, assembled to
   understand the history of inter-domain routing, to research the
   problems seen by the service providers, and to develop a proposal of
   requirements for a follow-on to the current routing architecture.
   This group's approach required an evolutionary approach starting from
   current routing architecture and practice.  In other words, the group
   limited itself to developing an evolutionary strategy and
   consequently assumed that the architecture would probably remain
   domain-based.  The Babylon group was later folded into the IRTF RRG
   as Sub-Group B to distinguish it from the original RRG Sub-Group A.

   This document, which was a part of Sub-Group B's output, provides a
   snapshot of the state of Inter-Domain Routing (IDR) at the time of
   original writing (2001) with some minor updates to take into account
   developments since that date, bringing it up to date in 2006.  The
   development of the new requirements set was then motivated by an
   analysis of the problems that IDR has been encountering in the recent
   past.  This document is intended as a counterpart to the Routing
   Requirements document ("A Set of Possible Requirements for a Future
   Routing Architecture"), which documents the requirements for future
   routing systems as captured separately by the IRTF RRG Sub-Groups A
   and B [RFC5772].

   The IRTF RRG supported publication of this document as a historical
   record of the work completed on the understanding that it does not
   necessarily represent either the latest technical understanding or
   the technical consensus of the research group at the time of
   publication.  The document has had substantial review by members of
   the Babylon team, members of the IRTF RRG, and others over the years.






Davies & Doria                  Historic                        [Page 4]

RFC 5773                       IDR History                 February 2010


2.  Introduction



   For the greater part of its existence, the Internet has used a
   domain-oriented routing system whereby the routers and other nodes
   making up the infrastructure are partitioned into a set of
   administrative domains, primarily along ownership lines.  Individual
   routing domains (also known as Autonomous Systems (ASs)), which maybe
   a subset of an administrative domain, are made up of a finite,
   connected set of nodes (at least in normal operation).  Each routing
   domain is subject to a coherent set of routing and other policies
   managed by a single administrative authority.  The domains are
   interlinked to form the greater Internet, producing a very large
   network: in practice, we have to treat this network as if it were
   infinite in extent as there is no central knowledge about the whole
   network of domains.  An early presentation of the concept of routing
   domains can be found in Paul Francis' OSI routing architecture paper
   from 1987 [Tsuchiya87] (Paul Francis was formerly known as Paul
   Tsuchiya).

   The domain concept and domain-oriented routing has become so
   fundamental to Internet-routing thinking that it is generally taken
   as an axiom these days and not even defined again (cf., [NewArch03]).
   The issues discussed in the present document notwithstanding, it has
   proved to be a robust and successful architectural concept that
   brings with it the possibility of using different routing mechanisms
   and protocols within the domains (intra-domain) and between the
   domains (inter-domain).  This is an attractive division, because
   intra-domain protocols can exploit the well-known finite scope of the
   domain and the mutual trust engendered by shared ownership to give a
   high degree of control to the domain administrators, whereas inter-
   domain routing lives in an essentially infinite region featuring a
   climate of distrust built on a multitude of competitive commercial
   agreements and driven by less-than-fully-public policies from each
   component domain.  Of course, like any other assumption that has been
   around for a very long time, the domain concept should be reevaluated
   to make sure that it is still helping!

   It is generally accepted that there are major shortcomings in the
   inter-domain routing of the Internet today and that these may result
   in severe routing problems within an unspecified period of time.
   Remedying these shortcomings will require extensive research to tie
   down the exact failure modes that lead to these shortcomings and
   identify the best techniques to remedy the situation.  Comparatively,
   intra-domain routing works satisfactorily, and issues with intra-
   domain routing are mainly associated with the interface between
   intra- and inter-domain routing.





Davies & Doria                  Historic                        [Page 5]

RFC 5773                       IDR History                 February 2010


      Reviewer's Note: Even in 2001, there was a wide difference of
      opinion across the community regarding the shortcomings of inter-
      domain routing.  In the years between writing and publication,
      further analysis, changes in operational practice, alterations to
      the demands made on inter-domain routing, modifications made to
      BGP and a recognition of the difficulty of finding a replacement
      may have altered the views of some members of the community.

   Changes in the nature and quality of the services that users want
   from the Internet are difficult to provide within the current
   framework, as they impose requirements never foreseen by the original
   architects of the Internet routing system.

   The kind of radical changes that have to be accommodated are
   epitomized by the advent of IPv6 and the application of IP mechanisms
   to private commercial networks that offer specific service guarantees
   beyond the best-effort services of the public Internet.  Major
   changes to the inter-domain routing system are inevitable to provide
   an efficient underpinning for the radically changed and increasingly
   commercially-based networks that rely on the IP protocol suite.

   Current practice stresses the need to separate the concerns of the
   control plane and the forwarding plane in a router: this document
   will follow this practice, but we still use the term "routing" as a
   global portmanteau to cover all aspects of the system.

   This document provides a historical perspective on the current state
   of inter-domain routing and its relationship to intra-domain routing
   in Section 3 by revisiting the previous IETF requirements document
   intended to steer the development of a future routing system.  These
   requirements, which informed the design of the Border Gateway
   Protocol (BGP) in 1989, are contained in RFC 1126 -- "Goals and
   Functional Requirements for Inter-Autonomous System Routing"
   [RFC1126].

   Section 3 also looks at some other work on requirements for domain-
   based routing that was carried out before and after RFC 1126 was
   published.  This work fleshes out the historical perspective and
   provides some additional insights into alternative approaches that
   may be instructive when building a new set of requirements.

   The motivation for change and the inspiration for some of the
   requirements for new routing architectures derive from the problems
   attributable to the current domain-based routing system that are
   being experienced in the Internet today.  These will be discussed in
   Section 5.





Davies & Doria                  Historic                        [Page 6]

RFC 5773                       IDR History                 February 2010


2.1.  Background



   Today's Internet uses an addressing and routing structure that has
   developed in an ad hoc, more or less upwards-compatible fashion.  The
   structure has progressed from supporting a non-commercial Internet
   with a single administrative domain to a solution that is able to
   control today's multi-domain, federated Internet, carrying traffic
   between the networks of commercial, governmental, and not-for-profit
   participants.  This is not achieved without a great deal of 24/7
   vigilance and operational activity by network operators: Internet
   routing often appears to be running close to the limits of stability.
   As well as directing traffic to its intended endpoint, inter-domain
   routing mechanisms are expected to implement a host of domain-
   specific routing policies for competing, communicating domains.  The
   result is not ideal, particularly as regards inter-domain routing
   mechanisms, but it does a pretty fair job at its primary goal of
   providing any-to-any connectivity to many millions of computers.

   Based on a large body of anecdotal evidence, but also on a growing
   body of experimental evidence [Labovitz02] and analytic work on the
   stability of BGP under certain policy specifications [Griffin99], the
   main Internet inter-domain routing protocol, BGP version 4 (BGP-4),
   appears to have a number of problems.  These problems are discussed
   in more detail in Section 5.  Additionally, the hierarchical nature
   of the inter-domain routing problem appears to be changing as the
   connectivity between domains becomes increasingly meshed [RFC3221],
   which alters some of the scaling and structuring assumptions on which
   BGP-4 is built.  Patches and fix-ups may relieve some of these
   problems, but others may require a new architecture and new
   protocols.

3.  Historical Perspective



3.1.  The Legacy of RFC 1126



   RFC 1126 [RFC1126] outlined a set of requirements that were intended
   to guide the development of BGP.

      Editors' Note: When this document was reviewed by Yakov Rekhter,
      one of the designers of BGP, his view was that "While some people
      expected a set of requirements outlined in RFC 1126 to guide the
      development of BGP, in reality the development of BGP happened
      completely independently of RFC 1126.  In other words, from the
      point of view of the development of BGP, RFC 1126 turned out to be
      totally irrelevant".  On the other hand, it appears that BGP, as
      currently implemented, has met a large proportion of these
      requirements, especially for unicast traffic.




Davies & Doria                  Historic                        [Page 7]

RFC 5773                       IDR History                 February 2010


   While the network is demonstrably different from what it was in 1989,
   having:

   o  moved from single to multiple administrative control,

   o  increased in size by several orders of magnitude, and

   o  migrated from a fairly tree-like connectivity graph to a meshier
      style

   many of the same requirements remain.  As a first step in setting
   requirements for the future, we need to understand the requirements
   that were originally set for the current protocols.  In charting a
   future architecture, we must first be sure to do no harm.  This means
   a future domain-based routing system has to support, as its base
   requirement, the level of function that is available today.

   The following sections each relate to a requirement, or non-
   requirement listed in RFC 1126.  In fact, the section names are
   direct quotes from the document.  The discussion of these
   requirements covers the following areas:

   Explanation:       Optional interpretation for today's audience of
                      the original intent of the requirement.

   Relevance:         Is the requirement of RFC 1126 still relevant, and
                      to what degree?  Should it be understood
                      differently in today's environment?

   Current practice:  How well is the requirement met by current
                      protocols and practice?

3.1.1.  General Requirements



3.1.1.1.  "Route to Destination"



   Timely routing to all reachable destinations, including multi-homing
   and multicast.

   Relevance:         Valid, but requirements for multi-homing need
                      further discussion and elucidation.  The
                      requirement should include multiple-source
                      multicast routing.








Davies & Doria                  Historic                        [Page 8]

RFC 5773                       IDR History                 February 2010


   Current practice:  Multi-homing is not efficient, and the proposed
                      inter-domain multicast protocol Border Gateway
                      Multicast Protocol (BGMP) [RFC3913] is an add-on
                      to BGP following many of the same strategies but
                      not integrated into the BGP framework.

                         Editors' Note: Multicast routing has moved on
                         again since this was originally written.  By
                         2006, BGMP had been effectively superseded.
                         Multicast routing now uses Multi-protocol BGP
                         [RFC4760], the Multicast Source Discovery
                         Protocol (MSDP) [RFC3618], and Protocol
                         Independent Multicast - Sparse Mode (PIM-SM)
                         [RFC2362], [RFC4601], especially the Source
                         Specific Multicast (SSM) subset.

3.1.1.2.  "Routing is Assured"



   This requires that a user be notified within a reasonable time period
   after persistent attempts, about inability to provide a service.

   Relevance:         Valid.

   Current practice:  There are ICMP messages for this; but, in many
                      cases, they are not used, either because of fears
                      about creating message storms or uncertainty about
                      whether the end system can do anything useful with
                      the resulting information.  IPv6 implementations
                      may be able to make better use of the information
                      as they may have alternative addresses that could
                      be used to exploit an alternative routing.

3.1.1.3.  "Large System"



   The architecture was designed to accommodate the growth of the
   Internet.

   Relevance:         Valid.  Properties of Internet topology might be
                      an issue for future scalability (topology varies
                      from very sparse to quite dense at present).
                      Instead of setting out to accommodate growth in a
                      specific time period, indefinite growth should be
                      accommodated.  On the other hand, such growth has
                      to be accommodated without making the protocols
                      too expensive -- trade-offs may be necessary.






Davies & Doria                  Historic                        [Page 9]

RFC 5773                       IDR History                 February 2010


   Current practice:  Scalability of the current protocols will not be
                      sufficient under the current rate of growth.
                      There are problems with BGP convergence for large
                      dense topologies, problems with the slow speed of
                      routing information propagation between routers in
                      transit domains through the intra-domain protocol,
                      for example, when a failure requires traffic to be
                      redirected to an alternative exit point from the
                      domain (see Section 5.9), limited support for
                      hierarchy, etc.

3.1.1.4.  "Autonomous Operation"



   This requirement encapsulates the need for administrative domains
   ("Autonomous Systems" - AS) to be able to operate autonomously as
   regards setting routing policy:

   Relevance:         Valid.  There may need to be additional
                      requirements for adjusting policy decisions to the
                      global functionality and for avoiding
                      contradictory policies.  This would decrease the
                      possibility of unstable routing behavior.

                      There is a need for handling various degrees of
                      trust in autonomous operations, ranging from no
                      trust (e.g., between separate ISPs) to very high
                      trust where the domains have a common goal of
                      optimizing their mutual policies.

                      Policies for intra-domain operations should, in
                      some cases, be revealed, using suitable
                      abstractions.

   Current practice:  Policy management is in the control of network
                      managers, as required, but there is little support
                      for handling policies at an abstract level for a
                      domain.

                      Cooperating administrative entities decide about
                      the extent of cooperation independently.  This can
                      lead to inconsistent, and potentially incompatible
                      routing policies being applied in notionally
                      cooperating domains.  As discussed in Sections
                      5.2, 5.3, and 5.10, lack of coordination combined
                      with global range of effects of BGP policies
                      results in occasional disruption of Internet
                      routing over an area far wider than the domains
                      that are not cooperating effectively.



Davies & Doria                  Historic                       [Page 10]

RFC 5773                       IDR History                 February 2010


3.1.1.5.  "Distributed System"



   The routing environment is a distributed system.  The distributed
   routing environment supports redundancy and diversity of nodes and
   links.  Both the controlling rule sets, which implement the routing
   policies, and the places where operational control is applied,
   through decisions on path selection, are distributed (primarily in
   the routers).

   Relevance:         Valid.  RFC 1126 is very clear that we should not
                      be using centralized solutions, but maybe we need
                      a discussion on trade-offs between common
                      knowledge and distribution (i.e., to allow for
                      uniform policy routing, e.g., Global System for
                      Mobile Communications (GSM) systems are in a sense
                      centralized, but with hierarchies).

   Current practice:  Routing is very distributed, but lacking the
                      ability to consider optimization over several hops
                      or domains.

                         Editors' Note: Also, coordinating the
                         implementation of a set of routing policies
                         across a large domain with many routers running
                         BGP is difficult.  The policies have to be
                         turned into BGP rules and applied individually
                         to each router, giving opportunities for
                         mismatch and error.

3.1.1.6.  "Provide A Credible Environment"



   The routing environment and services should be based upon mechanisms
   and information that exhibit both integrity and security.  That is,
   the routers should always be working with credible data derived
   through the reliable operation of protocols.  Security from unwanted
   modification and influence is required.

   Relevance:         Valid.

   Current practice:  BGP provides a limited mechanism for
                      authentication and security of peering sessions,
                      but this does not guarantee the authenticity or
                      validity of the routing information that is
                      exchanged.







Davies & Doria                  Historic                       [Page 11]

RFC 5773                       IDR History                 February 2010


                      There are certainly security problems with the
                      current practice.  The Routing Protocol Security
                      Requirements (rpsec) working group has been
                      struggling to agree on a set of requirements for
                      BGP security since early 2002.

                         Editors' Note: Proposals for authenticating BGP
                         routing information using certificates were
                         under development by the Secure Inter-Domain
                         Routing (sidr) working group from 2006 through
                         2008.

3.1.1.7.  "Be A Managed Entity"



   This requires that the routing system provides adequate information
   on the state of the network to allow resource, problem, and fault
   management to be carried out effectively and expeditiously.  The
   system must also provide controls that allow managers to use this
   information to make informed decisions and use it to control the
   operation of the routing system.

   Relevance:         The requirement is reasonable, but we might need
                      to be more specific on what information should be
                      available, e.g., to prevent routing oscillations.

   Current practice:  All policies are determined locally, where they
                      may appear reasonable but there is limited global
                      coordination through the routing policy databases
                      operated by the Internet registries (AfriNIC,
                      APNIC, ARIN, LACNIC, RIPE, etc.).

                      Operators are not required to register their
                      policies; even when policies are registered, it is
                      difficult to check that the actual policies in use
                      in other domains match the declared policies.
                      Therefore, a manager cannot guarantee to design
                      and implement policies that will interoperate with
                      those of other domains to provide stable routing.

                         Editors' Note: Operators report that management
                         of BGP-based routing remains a function that
                         needs highly-skilled operators and continual
                         attention.








Davies & Doria                  Historic                       [Page 12]

RFC 5773                       IDR History                 February 2010


3.1.1.8.  "Minimize Required Resources"



   Relevance:         Valid.  However, the paragraph states that
                      assumptions on significant upgrades shouldn't be
                      made.  Although this is reasonable, a new
                      architecture should perhaps be prepared to use
                      upgrades when they occur.

   Current practice:  Most bandwidth is consumed by the exchange of the
                      Network Layer Reachability Information (NLRI).
                      Usage of processing cycles ("Central Processor
                      Usage" - CPU) depends on the stability of the
                      Internet.  Both phenomena have a local nature, so
                      there are not scaling problems with bandwidth and
                      CPU usage.  Instability of routing increases the
                      consumption of resources in any case.  The number
                      of networks in the Internet dominates memory
                      requirements -- this is a scaling problem.

3.1.2.  "Functional Requirements"



3.1.2.1.  "Route Synthesis Requirements"



3.1.2.1.1.  "Route around failures dynamically"


   Relevance:         Valid.  Should perhaps be stronger.  Only
                      providing a best-effort attempt may not be enough
                      if real-time services are to be provided for.
                      Detection of failures may need to be faster than
                      100 ms to avoid being noticed by end-users.

   Current practice:  Latency of fail-over is too high; sometimes
                      minutes or longer.

3.1.2.1.2.  "Provide loop free paths"


   Relevance:         Valid.  Loops should occur only with negligible
                      probability and duration.

   Current practice:  Both link-state intra-domain routing and BGP
                      inter-domain routing (if correctly configured) are
                      forwarding-loop-free after having converged.
                      However, convergence time for BGP can be very
                      long, and poorly designed routing policies may
                      result in a number of BGP speakers engaging in a
                      cyclic pattern of advertisements and withdrawals
                      that never converges to a stable result [RFC3345].
                      Part of the reason for long convergence times is



Davies & Doria                  Historic                       [Page 13]

RFC 5773                       IDR History                 February 2010


                      the non-locality of the effects of changes in BGP
                      advertisements (see Section 5.3).  Modifying the
                      inter-domain routing protocol to make the effects
                      of changes less global, and convergence a more
                      local condition, might improve performance,
                      assuming a suitable modification could be
                      developed.

3.1.2.1.3.  "Know when a path or destination is unavailable"


   Relevance:         Valid to some extent, but there is a trade-off
                      between aggregation and immediate knowledge of
                      reachability.  It requires that routing tables
                      contain enough information to determine that the
                      destination is unknown or a path cannot be
                      constructed to reach it.

   Current practice:  Knowledge about lost reachability propagates
                      slowly through the networks due to slow
                      convergence for route withdrawals.

3.1.2.1.4.  "Provide paths sensitive to administrative policies"


   Relevance:         Valid.  Policy control of routing has become
                      increasingly important as the Internet has turned
                      into a business.

   Current practice:  Supported to some extent.  Policies can only be
                      applied locally in an AS and not globally.  Policy
                      information supplied has a very small probability
                      of affecting policies in other ASs.  Furthermore,
                      only static policies are supported; between static
                      policies and policies dependent upon volatile
                      events of great celerity, there should exist
                      events of which routing should be aware.  Lastly,
                      there is no support for policies other than route-
                      properties (such as AS-origin, AS-path,
                      destination prefix, Multi-Exit Discriminator-
                      values (MED-values), etc).

                         Editors' Note: Subsequent to the original issue
                         of this document, mechanisms that acknowledge
                         the business relationships of operators have
                         been developed such as the NOPEER community
                         attribute [RFC3765].  However, the level of
                         usage of this attribute is apparently not very
                         great.




Davies & Doria                  Historic                       [Page 14]

RFC 5773                       IDR History                 February 2010


3.1.2.1.5.  "Provide paths sensitive to user policies"


   Relevance:         Valid to some extent, as they may conflict with
                      the policies of the network administrator.  It is
                      likely that this requirement will be met by means
                      of different bit-transport services offered by an
                      operator, but at the cost of adequate
                      provisioning, authentication, and policing when
                      utilizing the service.

   Current practice:  Not supported in normal routing.  Can be
                      accomplished to some extent with loose source
                      routing, resulting in inefficient forwarding in
                      the routers.  The various attempts to introduce
                      Quality of Service (QoS -- e.g., Integrated
                      Services and Differentiated Services (Diffserv))
                      can also be seen as means to support this
                      requirement, but they have met with limited
                      success in terms of providing alternate routes as
                      opposed to providing improved service on the
                      standard route.

                         Editors' Note: From the standpoint of a later
                         time, it would probably be more appropriate to
                         say "total failure" rather than "limited
                         success".

3.1.2.1.6.  "Provide paths which characterize user quality-of-service
            requirements"


   Relevance:         Valid to some extent, as they may conflict with
                      the policies of the operator.  It is likely that
                      this requirement will be met by means of different
                      bit-transport services offered by an operator, but
                      at the cost of adequate provisioning,
                      authentication, and policing when utilizing the
                      service.  It has become clear that offering to
                      provide a particular QoS to any arbitrary
                      destination from a particular source is generally
                      impossible: QoS, except in very "soft" forms such
                      as overall long-term average packet delay, is
                      generally associated with connection-oriented
                      routing.

   Current practice:  Creating routes with specified QoS is not
                      generally possible at present.





Davies & Doria                  Historic                       [Page 15]

RFC 5773                       IDR History                 February 2010


3.1.2.1.7.  "Provide autonomy between inter- and intra-autonomous system
            route synthesis"


   Relevance:         Inter- and intra-domain routing should stay
                      independent, but one should notice that this, to
                      some extent, contradicts the previous three
                      requirements.  There is a trade-off between
                      abstraction and optimality.

   Current practice:  Inter-domain routing is performed independently of
                      intra-domain routing.  Intra-domain routing is
                      however, especially in transit domains, very
                      interrelated with inter-domain routing.

3.1.2.2.  "Forwarding Requirements"



3.1.2.2.1.  "Decouple inter- and intra-autonomous system forwarding
            decisions"


   Relevance:         Valid.

   Current practice:  As explained in Section 3.1.2.1.7, intra-domain
                      forwarding in transit domains is dependent on
                      inter-domain forwarding decisions.

3.1.2.2.2.  "Do not forward datagrams deemed administratively
            inappropriate"


   Relevance:         Valid, and increasingly important in the context
                      of enforcing policies correctly expressed through
                      routing advertisements but flouted by rogue peers
                      that send traffic for which a route has not been
                      advertised.  On the other hand, packets that have
                      been misrouted due to transient routing problems
                      perhaps should be forwarded to reach the
                      destination, although along an unexpected path.

   Current practice:  At stub domains (i.e., domains that do not provide
                      any transit service for any other domains but that
                      connect directly to one or more transit domains),
                      there is packet filtering, e.g., to catch source
                      address spoofing on outgoing traffic or to filter
                      out unwanted incoming traffic.  Filtering can in
                      particular reject traffic (such as unauthorized
                      transit traffic) that has been sent to a domain
                      even when it has not advertised a route for such
                      traffic on a given interface.  The growing class
                      of "middleboxes" (midboxes, e.g., Network Address



Davies & Doria                  Historic                       [Page 16]

RFC 5773                       IDR History                 February 2010


                      Translators -- NATs) is quite likely to apply
                      administrative rules that will prevent the
                      forwarding of packets.  Note that security
                      policies may deliberately hide administrative
                      denials.  In the backbone, intentional packet
                      dropping based on policies is not common.

3.1.2.2.3.  "Do not forward datagrams to failed resources"


   Relevance:         Unclear, although it is clearly desirable to
                      minimize the waste of forwarding resources by
                      discarding datagrams that cannot be delivered at
                      the earliest opportunity.  There is a trade-off
                      between scalability and keeping track of
                      unreachable resources.  The requirement
                      effectively imposes a requirement on adjacent
                      nodes to monitor for failures and take steps to
                      cause rerouting at the earliest opportunity, if a
                      failure is detected.  However, packets that are
                      already in-flight or queued on a failed link
                      cannot generally be rescued.

   Current practice:  Routing protocols use both internal adjacency
                      management sub-protocols (e.g., "hello" protocols)
                      and information from equipment and lower-layer
                      link watchdogs to keep track of failures in
                      routers and connecting links.  Failures will
                      eventually result in the routing protocol
                      reconfiguring the routing to avoid (if possible) a
                      failed resource, but this is generally very slow
                      (30s or more).  In the meantime, datagrams may
                      well be forwarded to failed resources.  In general
                      terms, end hosts and some non-router middleboxes
                      do not participate in these notifications, and
                      failures of such boxes will not affect the routing
                      system.

3.1.2.2.4.  "Forward datagram according to its characteristics"


   Relevance:         Valid.  This is necessary in enabling
                      differentiation in the network, based on QoS,
                      precedence, policy or security.

   Current practice:  Ingress and egress filtering can be done based on
                      policy.  Some networks discriminate on the basis
                      of requested QoS.





Davies & Doria                  Historic                       [Page 17]

RFC 5773                       IDR History                 February 2010


3.1.2.3.  "Information Requirements"



3.1.2.3.1.  "Provide a distributed and descriptive information base"


   Relevance:         Valid.  However, an alternative arrangement of
                      information bases, possibly with an element of
                      centralization for the domain (as mentioned in
                      Section 3.1.1.5) might offer some advantages
                      through the ability to optimize across the domain
                      and respond more quickly to changes and failures.

   Current practice:  The information base is distributed, but it is
                      unclear whether it supports all necessary routing
                      functionality.

3.1.2.3.2.  "Determine resource availability"


   Relevance:         Valid.  It should be possible to determine the
                      availability and levels of availability of any
                      resource (such as bandwidth) needed to carry out
                      routing.  This prevents needing to discover
                      unavailability through failure.  Resource location
                      and discovery is arguably a separate concern that
                      could be addressed outside the core routing
                      requirements.

   Current practice:  Resource availability is predominantly handled
                      outside of the routing system.

3.1.2.3.3.  "Restrain transmission utilization"


   Relevance:         Valid.  However, certain requirements in the
                      control plane, such as fast detection of faults
                      may be worth consumption of more resources.
                      Similarly, simplicity of implementation may make
                      it cheaper to "back haul" traffic to central
                      locations to minimize the cost of routing if
                      bandwidth is cheaper than processing.

   Current practice:  BGP messages probably do not ordinarily consume
                      excessive resources, but might during erroneous
                      conditions.  In the data plane, the nearly
                      universal adoption of shortest-path protocols
                      could be considered to result in minimization of
                      transmission utilization.






Davies & Doria                  Historic                       [Page 18]

RFC 5773                       IDR History                 February 2010


3.1.2.3.4.  "Allow limited information exchange"


   Relevance:         Valid.  But perhaps routing could be improved if
                      certain information (especially policies) could be
                      available either globally or at least for a wider-
                      defined locality.

                         Editors' Note: Limited information exchange
                         would be potentially compatible with a more
                         local form of convergence than BGP tries to
                         achieve today.  Limited information exchange is
                         potentially incompatible with global
                         convergence.

   Current practice:  Policies are used to determine which reachability
                      information is exported, but neighbors receiving
                      the information are not generally aware of the
                      policies that resulted in this export.

3.1.2.4.  "Environmental Requirements"



3.1.2.4.1.  "Support a packet-switching environment"


   Relevance:         Valid, but routing system should, perhaps, not be
                      limited to this exclusively.

   Current practice:  Supported.

3.1.2.4.2.  "Accommodate a connection-less oriented user transport
            service"


   Relevance:         Valid, but routing system should, perhaps, not be
                      limited to this exclusively.

   Current practice:  Accommodated.

3.1.2.4.3.  "Accommodate 10K autonomous systems and 100K networks"


   Relevance:         No longer valid.  Needs to be increased --
                      potentially indefinitely.  It is extremely
                      difficult to foresee the future size expansion of
                      the Internet, so the Utopian solution would be to
                      achieve an Internet whose architecture is scale
                      invariant.  Regrettably, this may not be
                      achievable without introducing undesirable
                      complexity and a suitable trade-off between
                      complexity and scalability is likely to be
                      necessary.



Davies & Doria                  Historic                       [Page 19]

RFC 5773                       IDR History                 February 2010


   Current Practice:  Supported, but perhaps reaching its limit.  Since
                      the original version of this document was written
                      in 2001, the number of ASs advertised has grown
                      from around 8000 to 20000, and almost 35000 AS
                      numbers have been allocated by the regional
                      registries [Huston05].  If this growth continues,
                      the original 16-bit AS space in BGP-4 will be
                      exhausted in less than 5 years.  Planning for an
                      extended AS space is now an urgent requirement.

                         Editors' Note: At the time of publication, 32-
                         bit AS numbers have been introduced and are
                         being deployed.

3.1.2.4.4.  "Allow for arbitrary interconnection of autonomous systems"


   Relevance:         Valid.  However, perhaps not all interconnections
                      should be accessible globally.

   Current practice:  BGP-4 allows for arbitrary interconnections.

3.1.2.5.  "General Objectives"



3.1.2.5.1.  "Provide routing services in a timely manner"


   Relevance:         Valid, as stated before.  It might be acceptable
                      for a more complex service to take longer to
                      deliver, but it still has to meet the
                      application's requirements -- routing has to be at
                      the service of the end-to-end principle.

                         Editors' Note: Delays in setting up connections
                         due to network functions such as NAT boxes are
                         becoming increasingly problematic.  The routing
                         system should try to keep any routing delay to
                         a minimum.

   Current practice:  More or less, with the exception of convergence
                      and fault robustness.

3.1.2.5.2.  "Minimize constraints on systems with limited resources"


   Relevance:         Valid.

   Current practice:  Systems with limited resources are typically stub
                      domains that advertise very little information.





Davies & Doria                  Historic                       [Page 20]

RFC 5773                       IDR History                 February 2010


3.1.2.5.3.  "Minimize impact of dissimilarities between autonomous
            systems"


   Relevance:         Important.  This requirement is critical to a
                      future architecture.  In a domain-based routing
                      environment where the internal properties of
                      domains may differ radically, it will be important
                      to be sure that these dissimilarities are
                      minimized at the borders.

   Current: practice: For the most part, this capability is not really
                      required in today's networks since the intra-
                      domain attributes are broadly similar across
                      domains.

3.1.2.5.4.  "Accommodate the addressing schemes and protocol mechanisms
            of the autonomous systems"


   Relevance:         Important, probably more so than when RFC 1126 was
                      originally developed because of the potential
                      deployment of IPv6, wider usage of MPLS, and the
                      increasing usage of VPNs.

   Current practice:  Only one global addressing scheme is supported in
                      most autonomous systems, but the availability of
                      IPv6 services is steadily increasing.  Some global
                      backbones support IPv6 routing and forwarding.

3.1.2.5.5.  "Must be implementable by network vendors"


   Relevance:         Valid, but note that what can be implemented today
                      is different from what was possible when RFC 1126
                      was written: a future domain-based routing
                      architecture should not be unreasonably
                      constrained by past limitations.

   Current practice:  BGP was implemented and meets a large proportion
                      of the original requirements.

3.1.3.  "Non-Goals"



   RFC 1126 also included a section discussing non-goals.  This section
   discusses the extent to which these are still non-goals.  It also
   considers whether the fact that they were non-goals adversely affects
   today's IDR system.






Davies & Doria                  Historic                       [Page 21]

RFC 5773                       IDR History                 February 2010


3.1.3.1.  "Ubiquity"



   The authors of RFC 1126 were explicitly saying that IP and its inter-
   domain routing system need not be deployed in every AS, and a
   participant should not necessarily expect to be able to reach a given
   AS, possibly because of routing policies.  In a sense, this "non-
   goal" has effectively been achieved by the Internet and IP protocols.
   This requirement reflects a different worldview where there was
   serious competition for network protocols, which is really no longer
   the case.  Ubiquitous deployment of inter-domain routing in
   particular has been achieved and must not be undone by any proposed
   future domain-based routing architecture.  On the other hand:

   o  ubiquitous connectivity cannot be reached in a policy-sensitive
      environment and should not be an aim.

         Editors' Note: It has been pointed out that this statement
         could be interpreted as being contrary to the Internet mission
         of providing universal connectivity.  The fact that limits to
         connectivity will be added as operational requirements in a
         policy-sensitive environment should not imply that a future
         domain-based routing architecture contains intrinsic limits on
         connectivity.

   o  it must not be required that the same routing mechanisms are used
      throughout, provided that they can interoperate appropriately.

   o  the information needed to control routing in a part of the network
      should not necessarily be ubiquitously available, and it must be
      possible for an operator to hide commercially sensitive
      information that is not needed outside a domain.

   o  the introduction of IPv6 reintroduces an element of diversity into
      the world of network protocols, but the similarities of IPv4 and
      IPv6 as regards routing and forwarding make this event less likely
      to drive an immediate diversification in routing systems.  The
      potential for further growth in the size of the network enabled by
      IPv6 is very likely to require changes in the future: whether this
      results in the replacement of one de facto ubiquitous system with
      another remains to be seen but cannot be a requirement -- it will
      have to interoperate with BGP during the transition.

   Relevance:         De facto essential for a future domain-based
                      routing architecture, but what is required is
                      ubiquity of the routing system rather than
                      ubiquity of connectivity and it must be capable of
                      a gradual takeover through interoperation with the
                      existing system.



Davies & Doria                  Historic                       [Page 22]

RFC 5773                       IDR History                 February 2010


   Current practice:  De facto ubiquity achieved.

3.1.3.2.  "Congestion control"



   Relevance:         It is not clear if this non-goal was to be applied
                      to routing or forwarding.  It is definitely a non-
                      goal to adapt the choice of route when there is
                      transient congestion.  However, to add support for
                      congestion avoidance (e.g., Explicit Congestion
                      Notification (ECN) and ICMP messages) in the
                      forwarding process would be a useful addition.
                      There is also extensive work going on in traffic
                      engineering that should result in congestion
                      avoidance through routing as well as in
                      forwarding.

   Current practice:  Some ICMP messages (e.g., source quench) exist to
                      deal with congestion control, but these are not
                      generally used as they either make the problem
                      worse or there is no mechanism to reflect the
                      message into the application that is providing the
                      source.

3.1.3.3.  "Load splitting"



   Relevance:         This should neither be a non-goal nor an explicit
                      goal.  It might be desirable in some cases and
                      should be considered as an optional architectural
                      feature.

   Current practice:  Can be implemented by exporting different prefixes
                      on different links, but this requires manual
                      configuration and does not consider actual load.

                         Editors' Note: This configuration is carried
                         out extensively as of 2006 and has been a
                         significant factor in routing table bloat.  If
                         this need is a real operational requirement, as
                         it seems to be for multi-homed or otherwise
                         richly connected sites, it will be necessary to
                         reclassify this as a real and important goal.










Davies & Doria                  Historic                       [Page 23]

RFC 5773                       IDR History                 February 2010


3.1.3.4.  "Maximizing the utilization of resources"



   Relevance:         Valid.  Cost-efficiency should be striven for; we
                      note that maximizing resource utilization does not
                      always lead to the greatest cost-efficiency.

   Current practice:  Not currently part of the system, though often a
                      "hacked in" feature done with manual
                      configuration.

3.1.3.5.  "Schedule to deadline service"



   This non-goal was put in place to ensure that the IDR did not have to
   meet real-time deadline goals such as might apply to Constant Bit
   Rate (CBR) real-time services in ATM.

   Relevance:         The hard form of deadline services is still a non-
                      goal for the future domain-based routing
                      architecture, but overall delay bounds are much
                      more of the essence than was the case when RFC
                      1126 was written.

   Current practice:  Service providers are now offering overall
                      probabilistic delay bounds on traffic contracts.
                      To implement these contracts, there is a
                      requirement for a rather looser form of delay
                      sensitive routing.

3.1.3.6.  "Non-interference policies of resource utilization"



   The requirement in RFC 1126 is somewhat opaque, but appears to imply
   that what we would today call QoS routing is a non-goal and that
   routing would not seek to control the elastic characteristics of
   Internet traffic whereby a TCP connection can seek to utilize all the
   spare bandwidth on a route, possibly to the detriment of other
   connections sharing the route or crossing it.

   Relevance:         Open Issue.  It is not clear whether dynamic QoS
                      routing can or should be implemented.  Such a
                      system would seek to control the admission and
                      routing of traffic depending on current or recent
                      resource utilization.  This would be particularly
                      problematic where traffic crosses an ownership
                      boundary because of the need for potentially
                      commercially sensitive information to be made
                      available outside the ownership boundary.





Davies & Doria                  Historic                       [Page 24]

RFC 5773                       IDR History                 February 2010


   Current practice:  Routing does not consider dynamic resource
                      availability.  Forwarding can support service
                      differentiation.

3.2.  ISO OSI IDRP, BGP, and the Development of Policy Routing



   During the decade before the widespread success of the World Wide
   Web, ISO was developing the communications architecture and protocol
   suite Open Systems Interconnection (OSI).  For a considerable part of
   this time, OSI was seen as a possible competitor for and even a
   replacement for the IP suite as this basis for the Internet.  The
   technical developments of the two protocols were quite heavily
   interrelated with each providing ideas and even components that were
   adapted into the other suite.

   During the early stages of the development of OSI, the IP suite was
   still mainly in use on the ARPANET and the relatively small scale
   first phase NSFNET.  This was effectively a single administrative
   domain with a simple tree-structured network in a three-level
   hierarchy connected to a single logical exchange point (the NSFNET
   backbone).  In the second half of the 1980s, the NSFNET was starting
   on the growth and transformation that would lead to today's Internet.
   It was becoming clear that the backbone routing protocol, the
   Exterior Gateway Protocol (EGP) [RFC0904], was not going to cope even
   with the limited expansion being planned.  EGP is an "all informed"
   protocol that needed to know the identities of all gateways, and this
   was no longer reasonable.  With the increasing complexity of the
   NSFNET and the linkage of the NSFNET network to other networks, there
   was a desire for policy-based routing that would allow administrators
   to manage the flow of packets between networks.  The first version of
   the Border Gateway Protocol (BGP-1) [RFC1105] was developed as a
   replacement for EGP with policy capabilities -- a stopgap EGP version
   3 had been created as an interim measure while BGP was developed.
   BGP was designed to work on a hierarchically structured network, such
   as the original NSFNET, but could also work on networks that were at
   least partially non-hierarchical where there were links between ASs
   at the same level in the hierarchy (we would now call these "peering
   arrangements") although the protocol made a distinction between
   different kinds of links (links are classified as upwards, downwards,
   or sideways).  ASs themselves were a "fix" for the complexity that
   developed in the three-tier structure of the NSFNET.

   Meanwhile, the OSI architects, led by Lyman Chapin, were developing a
   much more general architecture for large-scale networks.  They had
   recognized that no one node, especially an end-system (host), could
   or should attempt to remember routes from "here" to "anywhere" --
   this sounds obvious today, but was not so obvious 20 years ago.  They
   were also considering hierarchical networks with independently



Davies & Doria                  Historic                       [Page 25]

RFC 5773                       IDR History                 February 2010


   administered domains -- a model already well entrenched in the
   public-switched telephone network.  This led to a vision of a network
   with multiple independent administrative domains with an arbitrary
   interconnection graph and a hierarchy of routing functionality.  This
   architecture was fairly well established by 1987 [Tsuchiya87].  The
   architecture initially envisaged a three-level routing functionality
   hierarchy in which each layer had significantly different
   characteristics:

   1.  *End-system to intermediate system (IS) routing (host to
       router)*, in which the principal functions are discovery and
       redirection.

   2.  *Intra-domain IS-IS routing (router to router)*, in which "best"
       routes between end-systems in a single administrative domain are
       computed and used.  A single algorithm and routing protocol would
       be used throughout any one domain.

   3.  *Inter-domain IS-IS routing (router to router)*, in which routes
       between routing domains within administrative domains are
       computed (routing is considered separately between administrative
       domains and routing domains).

   Level 3 of this hierarchy was still somewhat fuzzy.  Tsuchiya says:

      The last two components, Inter-Domain and Inter-Administration
      routing, are less clear-cut.  It is not obvious what should be
      standardized with respect to these two components of routing.  For
      example, for Inter-Domain routing, what can be expected from the
      Domains?  By asking Domains to provide some kind of external
      behavior, we limit their autonomy.  If we expect nothing of their
      external behavior, then routing functionality will be minimal.

      Across administrations, it is not known how much trust there will
      be.  In fact, the definition of trust itself can only be
      determined by the two or more administrations involved.

      Fundamentally, the problem with Inter-Domain and Inter-
      Administration routing is that autonomy and mistrust are both
      antithetical to routing.  Accomplishing either will involve a
      number of tradeoffs which will require more knowledge about the
      environments within which they will operate.

   Further refinement of the model occurred over the next couple of
   years and a more fully formed view is given by Huitema and Dabbous in
   1989 [Huitema90].  By this stage, work on the original IS-IS link-
   state protocol, originated by the Digital Equipment Corporation
   (DEC), was fairly advanced and was close to becoming a Draft



Davies & Doria                  Historic                       [Page 26]

RFC 5773                       IDR History                 February 2010


   International Standard.  IS-IS is of course a major component of
   intra-domain routing today and inspired the development of the Open
   Shortest Path First (OSPF) family.  However, Huitema and Dabbous were
   not able to give any indication of protocol work for Level 3.  There
   are hints of possible use of centralized route servers.

   In the meantime, the NSFNET consortium and the IETF had been
   struggling with the rapid growth of the NSFNET.  It had been clear
   since fairly early on that EGP was not suitable for handling the
   expanding network and the race was on to find a replacement.  There
   had been some intent to include a metric in EGP to facilitate routing
   decisions, but no agreement could be reached on how to define the
   metric.  The lack of trust was seen as one of the main reasons that
   EGP could not establish a globally acceptable routing metric: again
   this seems to be a clearly futile aim from this distance in time!
   Consequently, EGP became effectively a rudimentary path-vector
   protocol that linked gateways with Autonomous Systems.  It was
   totally reliant on the tree-structured network to avoid routing
   loops, and the all-informed nature of EGP meant that update packets
   became very large.  BGP version 1 [RFC1105] was standardized in 1989,
   but it had been in development for some time before this and had
   already seen action in production networks prior to standardization.
   BGP was the first real path-vector routing protocol and was intended
   to relieve some of the scaling problems as well as providing policy-
   based routing.  Routes were described as paths along a "vector" of
   ASs without any associated cost metric.  This way of describing
   routes was explicitly intended to allow detection of routing loops.
   It was assumed that the intra-domain routing system was loop-free
   with the implication that the total routing system would be loop-free
   if there were no loops in the AS path.  Note that there were no
   theoretical underpinnings for this work, and it traded freedom from
   routing loops for guaranteed convergence.

   Also, the NSFNET was a government-funded research and education
   network.  Commercial companies that were partners in some of the
   projects were using the NSFNET for their research activities, but it
   was becoming clear that these companies also needed networks for
   commercial traffic.  NSFNET had put in place "acceptable use"
   policies that were intended to limit the use of the network.
   However, there was little or no technology to support the legal
   framework.

   Practical experience, IETF IAB discussion (centered in the Internet
   Architecture Task Force) and the OSI theoretical work were by now
   coming to the same conclusions:

   o  Networks were going to be composed out of multiple administrative
      domains (the federated network),



Davies & Doria                  Historic                       [Page 27]

RFC 5773                       IDR History                 February 2010


   o  The connections between these domains would be an arbitrary graph
      and certainly not a tree,

   o  The administrative domains would wish to establish distinctive,
      independent routing policies through the graph of Autonomous
      Systems, and

   o  Administrative domains would have a degree of distrust of each
      other that would mean that policies would remain opaque.

   These views were reflected by Susan Hares' (working for Merit
   Networks at that time) contribution to the Internet Architecture
   (INARC) workshop in 1989, summarized in the report of the workshop
   [INARC89]:

      The rich interconnectivity within the Internet causes routing
      problems today.  However, the presenter believes the problem is
      not the high degree of interconnection, but the routing protocols
      and models upon which these protocols are based.  Rich
      interconnectivity can provide redundancy which can help packets
      moving even through periods of outages.  Our model of interdomain
      routing needs to change.  The model of autonomous confederations
      and autonomous systems [RFC0975] no longer fits the reality of
      many regional networks.  The ISO models of administrative domain
      and routing domains better fit the current Internet's routing
      structure.

      With the first NSFNET backbone, NSF assumed that the Internet
      would be used as a production network for research traffic.  We
      cannot stop these networks for a month and install all new routing
      protocols.  The Internet will need to evolve its changes to
      networking protocols while still continuing to serve its users.
      This reality colors how plans are made to change routing
      protocols.

   It is also interesting to note that the difficulties of organizing a
   transition were recognized at this stage and have not been seriously
   explored or resolved since.

   Policies would primarily be interested in controlling which traffic
   should be allowed to transit a domain (to satisfy commercial
   constraints or acceptable use policies), thereby controlling which
   traffic uses the resources of the domain.  The solution adopted by
   both the IETF and OSI was a form of distance vector hop-by-hop
   routing with explicit policy terms.  The reasoning for this choice
   can be found in Breslau and Estrin's 1990 paper [Breslau90]
   (implicitly -- because some other alternatives are given such as a
   link state with policy suggestion, which, with hindsight, would have



Davies & Doria                  Historic                       [Page 28]

RFC 5773                       IDR History                 February 2010


   even greater problems than BGP on a global scale network).
   Traditional distance-vector protocols exchanged routing information
   in the form of a destination and a metric.  The new protocols
   explicitly associated policy expressions with the route by including
   either a list of the source ASs that are permitted to use the route
   described in the routing update, and/or a list of all ASs traversed
   along the advertised route.

   Parallel protocol developments were already in progress by the time
   this paper was published: BGP version 2 [RFC1163] in the IETF and the
   Inter-Domain Routing Protocol (IDRP) [ISO10747], which would be the
   Level 3 routing protocol for the OSI architecture.  IDRP was
   developed under the aegis of the ANSI XS3.3 working group led by
   Lyman Chapin and Charles Kunzinger.  The two protocols were very
   similar in basic design, but IDRP has some extra features, some of
   which have been incorporated into later versions of BGP; others may
   yet be so, and still others may be seen to be inappropriate.  Breslau
   and Estrin summarize the design of IDRP as follows:

      IDRP attempts to solve the looping and convergence problems
      inherent in distance vector routing by including full AD
      (Administrative Domain -- essentially the equivalent of what are
      now called ASs) path information in routing updates.  Each routing
      update includes the set of ADs that must be traversed in order to
      reach the specified destination.  In this way, routes that contain
      AD loops can be avoided.

      IDRP updates also contain additional information relevant to
      policy constraints.  For instance, these updates can specify what
      other ADs are allowed to receive the information described in the
      update.  In this way, IDRP is able to express source specific
      policies.  The IDRP protocol also provides the structure for the
      addition of other types of policy related information in routing
      updates.  For example, User Class Identifiers (UCI) could also be
      included as policy attributes in routing updates.

      Using the policy route attributes IDRP provides the framework for
      expressing more fine grained policy in routing decisions.
      However, because it uses hop-by-hop distance vector routing, it
      only allows a single route to each destination per-QOS to be
      advertised.  As the policy attributes associated with routes
      become more fine grained, advertised routes will be applicable to
      fewer sources.  This implies a need for multiple routes to be
      advertised for each destination in order to increase the
      probability that sources have acceptable routes available to them.
      This effectively replicates the routing table per forwarding
      entity for each QoS, UCI, source combination that might appear in




Davies & Doria                  Historic                       [Page 29]

RFC 5773                       IDR History                 February 2010


      a packet.  Consequently, we claim that this approach does not
      scale well as policies become more fine grained, i.e., source or
      UCI specific policies.

   Over the next three or four years, successive versions of BGP (BGP-2
   [RFC1163], BGP-3 [RFC1267], and BGP-4 [RFC1771]) were deployed to
   cope with the growing and by now commercialized Internet.  From BGP-2
   onwards, BGP made no assumptions about an overall structure of
   interconnections allowing it to cope with today's dense web of
   interconnections between ASs.  BGP version 4 was developed to handle
   the change from classful to classless addressing.  For most of this
   time, IDRP was being developed in parallel, and both protocols were
   implemented in the Merit gatedaemon routing protocol suite.  During
   this time, there was a movement within the IETF that saw BGP as a
   stopgap measure to be used until the more sophisticated IDRP could be
   adapted to run over IP instead of the OSI connectionless protocol
   Connectionless Network Protocol (CLNP).  However, unlike its intra-
   domain counterpart IS-IS, which has stood the test of time, and
   indeed proved to be more flexible than OSPF, IDRP was ultimately not
   adopted by the market.  By the time the NSFNET backbone was
   decommissioned in 1995, BGP-4 was the inter-domain routing protocol
   of choice and OSI's star was already beginning to wane.  IDRP is now
   little remembered.

   A more complete account of the capabilities of IDRP can be found in
   Chapter 14 of David Piscitello and Lyman Chapin's book "Open Systems
   Networking: TCP/IP and OSI", which is now readable on the Internet
   [Chapin94].

   IDRP also contained quite extensive means for securing routing
   exchanges, much of it based on X.509 certificates for each router and
   public-/private-key encryption of routing updates.

   Some of the capabilities of IDRP that might yet appear in a future
   version of BGP include the ability to manage routes with explicit QoS
   classes and the concept of domain confederations (somewhat different
   from the confederation mechanism in today's BGP) as an extra level in
   the hierarchy of routing.

3.3.  Nimrod Requirements



   Nimrod as expressed by Noel Chiappa in his early document, "A New IP
   Routing and Addressing Architecture" [Chiappa91] and later in the
   NIMROD working group documents [RFC1753] and [RFC1992] established a
   number of requirements that need to be considered by any new routing
   architecture.  The Nimrod requirements took RFC 1126 as a starting
   point and went further.




Davies & Doria                  Historic                       [Page 30]

RFC 5773                       IDR History                 February 2010


   The three goals of Nimrod, quoted from [RFC1992], were as follows:

   1.  To support a dynamic internetwork of _arbitrary size_ (our
       emphasis) by providing mechanisms to control the amount of
       routing information that must be known throughout an
       internetwork.

   2.  To provide service-specific routing in the presence of multiple
       constraints imposed by service providers and users.

   3.  To admit incremental deployment throughout an internetwork.

   It is certain that these goals should be considered requirements for
   any new domain-based routing architecture.

   o  As discussed in other sections of this document, the rate of
      growth of the amount of information needed to maintain the routing
      system is such that the system may not be able to scale up as the
      Internet expands as foreseen.  And yet, as the services and
      constraints upon those services grow, there is a need for more
      information to be maintained by the routing system.  One of the
      key terms in the first requirements is "control".  While
      increasing amounts of information need to be known and maintained
      in the Internet, the amounts and kinds of information that are
      distributed can be controlled.  This goal should be reflected in
      the requirements for the future domain-based architecture.

   o  If anything, the demand for specific services in the Internet has
      grown since 1996 when the Nimrod architecture was published.
      Additionally, the kinds of constraints that service providers need
      to impose upon their networks and that services need to impose
      upon the routing have also increased.  Any changes made to the
      network in the last half-decade have not significantly improved
      this situation.

   o  The ability to incrementally deploy any new routing architecture
      within the Internet is still an absolute necessity.  It is
      impossible to imagine that a new routing architecture could
      supplant the current architecture on a flag day.

   At one point in time, Nimrod, with its addressing and routing
   architectures, was seen as a candidate for IPng.  History shows that
   it was not accepted as the IPng, having been ruled out of the
   selection process by the IESG in 1994 on the grounds that it was "too
   much of a research effort" [RFC1752], although input for the
   requirements of IPng was explicitly solicited from Chiappa [RFC1753].
   Instead, IPv6 has been put forth as the IPng.  Without entering a
   discussion of the relative merits of IPv6 versus Nimrod, it is



Davies & Doria                  Historic                       [Page 31]

RFC 5773                       IDR History                 February 2010


   apparent that IPv6, while it may solve many problems, does not solve
   the critical routing problems in the Internet today.  In fact, in
   some sense, it exacerbates them by adding a requirement for support
   of two Internet protocols and their respective addressing methods.
   In many ways, the addition of IPv6 to the mix of methods in today's
   Internet only points to the fact that the goals, as set forth by the
   Nimrod team, remain as necessary goals.

   There is another sense in which the study of Nimrod and its
   architecture may be important to deriving a future domain-based
   routing architecture.  Nimrod can be said to have two derivatives:

   o  Multi-Protocol Label Switching (MPLS), in that it took the notion
      of forwarding along well-known paths.

   o  Private Network-Node Interface (PNNI), in that it took the notion
      of abstracting topological information and using that information
      to create connections for traffic.

   It is important to note, that whilst MPLS and PNNI borrowed ideas
   from Nimrod, neither of them can be said to be an implementation of
   this architecture.

3.4.  PNNI



   The Private Network-Node Interface (PNNI) routing protocol was
   developed under the ATM Forum's auspices as a hierarchical route
   determination protocol for ATM, a connection-oriented architecture.
   It is reputed to have developed several of its methods from a study
   of the Nimrod architecture.  What can be gained from an analysis of
   what did and did not succeed in PNNI?

   The PNNI protocol includes the assumption that all peer groups are
   willing to cooperate, and that the entire network is under the same
   top administration.  Are there limitations that stem from this "world
   node" presupposition?  As discussed in [RFC3221], the Internet is no
   longer a clean hierarchy, and there is a lot of resistance to having
   any sort of "ultimate authority" controlling or even brokering
   communication.

   PNNI is the first deployed example of a routing protocol that uses
   abstract map exchange (as opposed to distance-vector or link-state
   mechanisms) for inter-domain routing information exchange.  One
   consequence of this is that domains need not all use the same
   mechanism for map creation.  What were the results of this
   abstraction and source-based route calculation mechanism?





Davies & Doria                  Historic                       [Page 32]

RFC 5773                       IDR History                 February 2010


   Since the authors of this document do not have experience running a
   PNNI network, the comments above are from a theoretical perspective.
   Further research on these issues based on operational experience is
   required.

4.  Recent Research Work



4.1.  Developments in Internet Connectivity



   The work commissioned from Geoff Huston by the Internet Architecture
   Board [RFC3221] draws a number of conclusions from the analysis of
   BGP routing tables and routing registry databases:

   o  The connectivity between provider ASs is becoming more like a
      dense mesh than the tree structure that was commonly assumed to be
      commonplace a couple of years ago.  This has been driven by the
      increasing amounts charged for peering and transit traffic by
      global service providers.  Local direct peering and Internet
      exchanges are becoming steadily more common as the cost of local
      fibre connections drops.

   o  End-user sites are increasingly resorting to multi-homing onto two
      or more service providers as a way of improving resiliency.  This
      has a knock-on effect of spectacularly fast depletion of the
      available pool of AS numbers as end-user sites require public AS
      numbers to become multi-homed and corresponding increase in the
      number of prefixes advertised in BGP.

   o  Multi-homed sites are using advertisement of longer prefixes in
      BGP as a means of traffic engineering to load spread across their
      multiple external connections with further impact on the size of
      the BGP tables.

   o  Operational practices are not uniform, and in some cases lack of
      knowledge or training is leading to instability and/or excessive
      advertisement of routes by incorrectly configured BGP speakers.

   o  All these factors are quickly negating the advantages in limiting
      the expansion of BGP routing tables that were gained by the
      introduction of Classless Inter-Domain Routing (CIDR) and
      consequent prefix aggregation in BGP.  It is also now impossible
      for IPv6 to realize the worldview in which the default-free zone
      would be limited to perhaps 10,000 prefixes.

   o  The typical "width" of the Internet in AS hops is now around five,
      and much less in many cases.





Davies & Doria                  Historic                       [Page 33]

RFC 5773                       IDR History                 February 2010


   These conclusions have a considerable impact on the requirements for
   the future domain-based routing architecture:

   o  Topological hierarchy (e.g., mandating a tree-structured
      connectivity) cannot be relied upon to deliver scalability of a
      large Internet routing system.

   o  Aggregation cannot be relied upon to constrain the size of routing
      tables for an all-informed routing system.

4.2.  DARPA NewArch Project



   DARPA funded a project to think about a new architecture for future
   generation Internet, called NewArch (see
   http://www.isi.edu/newarch/).  Work started in the first half of 2000
   and the main project finished in 2003 [NewArch03].

   The main development is to conclude that as the Internet becomes
   mainstream infrastructure, fewer and fewer of the requirements are
   truly global but may apply with different force or not at all in
   certain parts of the network.  This (it is claimed) makes the
   compilation of a single, ordered list of requirements deeply
   problematic.  Instead, we may have to produce multiple requirement
   sets with support for differing requirement importance at different
   times and in different places.  This "meta-requirement" significantly
   impacts architectural design.

   Potential new technical requirements identified so far include:

   o  Commercial environment concerns such as richer inter-provider
      policy controls and support for a variety of payment models

   o  Trustworthiness

   o  Ubiquitous mobility

   o  Policy driven self-organization ("deep auto-configuration")

   o  Extreme short-timescale resource variability

   o  Capacity allocation mechanisms

   o  Speed, propagation delay, and delay/bandwidth product issues

   Non-technical or political "requirements" include:

   o  Legal and Policy drivers such as




Davies & Doria                  Historic                       [Page 34]

RFC 5773                       IDR History                 February 2010


      *  Privacy and free/anonymous speech

      *  Intellectual property concerns

      *  Encryption export controls

      *  Law enforcement surveillance regulations

      *  Charging and taxation issues

   o  Reconciling national variations and consistent operation in a
      worldwide infrastructure

   The conclusions of the work are now summarized in the final report
   [NewArch03].

4.2.1.  Defending the End-to-End Principle



   One of the participants in DARPA NewArch work (Dave Clark) with one
   of his associates has also published a very interesting paper
   analyzing the impact of some of the new requirements identified in
   NewArch (see Section 4.2) on the end-to-end principle that has guided
   the development of the Internet to date [Clark00].  Their primary
   conclusion is that the loss of trust between the users at the ends of
   end-to-end has the most fundamental effect on the Internet.  This is
   clear in the context of the routing system, where operators are
   unwilling to reveal the inner workings of their networks for
   commercial reasons.  Similarly, trusted third parties and their
   avatars (mainly midboxes of one sort or another) have a major impact
   on the end-to-end principles and the routing mechanisms that went
   with them.  Overall, the end-to-end principles should be defended so
   far as is possible -- some changes are already too deeply embedded to
   make it possible to go back to full trust and openness -- at least
   partly as a means of staving off the day when the network will ossify
   into an unchangeable form and function (much as the telephone network
   has done).  The hope is that by that time, a new Internet will appear
   to offer a context for unfettered innovation.

5.  Existing Problems of BGP and the Current Inter-/Intra-Domain
    Architecture



   Although most of the people who have to work with BGP today believe
   it to be a useful, working protocol, discussions have brought to
   light a number of areas where BGP or the relationship between BGP and
   the intra-domain routing protocols in use today could be improved.
   BGP-4 has been and continues to be extended since it was originally
   introduced in [RFC1771] and the protocol as deployed has been
   documented in [RFC4271].  This section is, to a large extent, a wish



Davies & Doria                  Historic                       [Page 35]

RFC 5773                       IDR History                 February 2010


   list for the future domain-based routing architecture based on those
   areas where BGP is seen to be lacking, rather than simply a list of
   problems with BGP.  The shortcomings of today's inter-domain routing
   system have also been extensively surveyed in "Architectural
   Requirements for Inter-Domain Routing in the Internet" [RFC3221],
   particularly with respect to its stability and the problems produced
   by explosions in the size of the Internet.

5.1.  BGP and Auto-Aggregation



   The initial stability followed by linear growth rates of the number
   of routing objects (prefixes) that was achieved by the introduction
   of CIDR around 1994, has now been once again been replaced by near-
   exponential growth of number of routing objects.  The granularity of
   many of the objects advertised in the default-free zone is very small
   (prefix length of 22 or longer): this granularity appears to be a by-
   product of attempts to perform precision traffic engineering related
   to increasing levels of multi-homing.  At present, there is no
   mechanism in BGP that would allow an AS to aggregate such prefixes
   without advance knowledge of their existence, even if it was possible
   to deduce automatically that they could be aggregated.  Achieving
   satisfactory auto-aggregation would also significantly reduce the
   non-locality problems associated with instability in peripheral ASs.

   On the other hand, it may be that alterations to the connectivity of
   the net as described in [RFC3221] and Section 2.5.1 may limit the
   usefulness of auto-aggregation.

5.2.  Convergence and Recovery Issues



   BGP today is a stable protocol under most circumstances, but this has
   been achieved at the expense of making the convergence time of the
   inter-domain routing system very slow under some conditions.  This
   has a detrimental effect on the recovery of the network from
   failures.

   The timers that control the behavior of BGP are typically set to
   values in the region of several tens of seconds to a few minutes,
   which constrains the responsiveness of BGP to failure conditions.

   In the early days of deployment of BGP, poor network stability and
   router software problems lead to storms of withdrawals closely
   followed by re-advertisements of many prefixes.  To control the load
   on routing software imposed by these "route flaps", route-flap
   damping was introduced into BGP.  Most operators have now implemented
   a degree of route-flap damping in their deployments of BGP.  This
   restricts the number of times that the routing tables will be
   rebuilt, even if a route is going up and down very frequently.



Davies & Doria                  Historic                       [Page 36]

RFC 5773                       IDR History                 February 2010


   Unfortunately, route-flap damping responds to multiple flaps by
   increasing the route suppression time exponentially, which can result
   in some parts of the Internet being unreachable for hours at a time.

   There is evidence ([RFC3221] and measurements by some of the Sub-
   Group B members [Jiang02]) that in today's network, route flap is
   disproportionately associated with the fine-grained prefixes (length
   22 or longer) associated with traffic engineering at the periphery of
   the network.  Auto-aggregation, as previously discussed, would tend
   to mask such instability and prevent it being propagated across the
   whole network.  Another question that needs to be studied is the
   continuing need for an architecture that requires global convergence.
   Some of our studies (unpublished) show that, in some localities at
   least, the network never actually reaches stability; i.e., it never
   really globally converges.  Can a global, and beyond, network be
   designed with the requirement of global convergence?

5.3.  Non-Locality of Effects of Instability and Misconfiguration



   There have been a number of instances, some of which are well
   documented, of a mistake in BGP configuration in a single peripheral
   AS propagating across the whole Internet and resulting in misrouting
   of most of the traffic in the Internet.

   Similarly, a single route flap in a single peripheral AS can require
   route table recalculation across the entire Internet.

   This non-locality of effects is highly undesirable, and it would be a
   considerable improvement if such effects were naturally limited to a
   small area of the network around the problem.  This is another
   argument for an architecture that does not require global
   convergence.

5.4.  Multi-Homing Issues



   As discussed previously, the increasing use of multi-homing as a
   robustness technique by peripheral networks requires that multiple
   routes have to be advertised for such domains.  These routes must not
   be aggregated close in to the multi-homed domain as this would defeat
   the traffic engineering implied by multi-homing and currently cannot
   be aggregated further away from the multi-homed domain due to the
   lack of auto-aggregation capabilities.  Consequentially, the default-
   free zone routing table is growing exponentially, as it was before
   CIDR.

   The longest prefix match routing technique introduced by CIDR, and
   implemented in BGP-4, when combined with provider address allocation
   is an obstacle to effective multi-homing if load sharing across the



Davies & Doria                  Historic                       [Page 37]

RFC 5773                       IDR History                 February 2010


   multiple links is required.  If an AS has been allocated, its
   addresses from an upstream provider, the upstream provider can
   aggregate those addresses with those of other customers and need only
   advertise a single prefix for a range of customers.  But, if the
   customer AS is also connected to another provider, the second
   provider is not able to aggregate the customer addresses because they
   are not taken from his allocation, and will therefore have to
   announce a more specific route to the customer AS.  The longest match
   rule will then direct all traffic through the second provider, which
   is not as required.

   Example:


                                  \       /
                                 AS1     AS2
                                    \   /
                                     AS3


                       Figure 1: Address Aggregation

   In Figure 1, AS3 has received its addresses from AS1, which means AS1
   can aggregate.  But if AS3 wants its traffic to be seen equally both
   ways, AS3 is forced to announce both the aggregate and the more
   specific route to AS2.

   This problem has induced many ASs to apply for their own address
   allocation even though they could have been allocated from an
   upstream provider further exacerbating the default-free zone route
   table size explosion.  This problem also interferes with the desire
   of many providers in the default-free zone to route only prefixes
   that are equal to or shorter than 20 or 19 bits.

   Note that some problems that are referred to as multi-homing issues
   are not, and should not be, solvable through the routing system
   (e.g., where a TCP load distributor is needed), and multi-homing is
   not a panacea for the general problem of robustness in a routing
   system [Berkowitz01].

      Editors' Note: A more recent analysis of multi-homing can be found
      in [RFC4116].

5.5.  AS Number Exhaustion



   The domain identifier or AS number is a 16-bit number.  When this
   paper was originally written in 2001, allocation of AS numbers was
   increasing 51% a year [RFC3221] and exhaustion by 2005 was predicted.



Davies & Doria                  Historic                       [Page 38]

RFC 5773                       IDR History                 February 2010


   According to some recent work again by Huston [Huston05], the rate of
   increase dropped off after the business downturn, but as of July
   2005, well over half the available AS numbers (39000 out of 64510)
   had been allocated by IANA and around 20000 were visible in the
   global BGP routing tables.  A year later, these figures had grown to
   42000 (April 2006) and 23000 (August 2006), respectively, and the
   rate of allocation is currently about 3500 per year.  Depending on
   the curve-fitting model used to predict when exhaustion will occur,
   the pool will run out somewhere between 2010 and 2013.  There appear
   to be other factors at work in this rate of increase beyond an
   increase in the number of ISPs in business, although there is a fair
   degree of correlation between these numbers.  AS numbers are now used
   for a number of purposes beyond that of identifying large routing
   domains: multi-homed sites acquire an AS number in order to express
   routing preferences to their various providers and AS numbers are
   used part of the addressing mechanism for MPLS/BGP-based virtual
   private networks (VPNs) [RFC4364].  The IETF has had a proposal under
   development for over four years to increase the available range of AS
   numbers to 32 bits [RFC4893].  Much of the slowness in development is
   due to the deployment challenge during transition.  Because of the
   difficulties of transition, deployment needs to start well in advance
   of actual exhaustion so that the network as a whole is ready for the
   new capability when it is needed.  This implies that standardization
   needs to be complete and implementations available at least well in
   advance of expected exhaustion so that deployment of upgrades that
   can handle the longer AS numbers, should be starting around 2008, to
   give a reasonable expectation that the change has been rolled out
   across a large fraction of the Internet by the time exhaustion
   occurs.

      Editors' Note: The Regional Internet Registries (RIRs) are
      planning to move to assignment of the longer AS numbers by default
      on 1 January 2009, but there are concerns that significant numbers
      of routers will not have been upgraded by then.

5.6.  Partitioned ASs



   Tricks with discontinuous ASs are used by operators, for example, to
   implement anycast.  Discontinuous ASs may also come into being by
   chance if a multi-homed domain becomes partitioned as a result of a
   fault and part of the domain can access the Internet through each
   connection.  It may be desirable to make support for this kind of
   situation more transparent than it is at present.








Davies & Doria                  Historic                       [Page 39]

RFC 5773                       IDR History                 February 2010


5.7.  Load Sharing



   Load splitting or sharing was not a goal of the original designers of
   BGP and it is now a problem for today's network designers and
   managers.  Trying to fool BGP into load sharing between several links
   is a constantly recurring exercise for most operators today.

5.8.  Hold-Down Issues



   As with the interval between "hello" messages in OSPF, the typical
   size and defined granularity (seconds to tens of seconds) of the
   "keepalive" time negotiated at start-up for each BGP connection
   constrains the responsiveness of BGP to link failures.

   The recommended values and the available lower limit for this timer
   were set to limit the overhead caused by keepalive messages when link
   bandwidths were typically much lower than today.  Analysis and
   experiment ([Alaettinoglu00], [Sandiick00] and [RFC4204]) indicate
   that faster links could sustain a much higher rate of keepalive
   messages without significantly impacting normal data traffic.  This
   would improve responsiveness to link and node failures but with a
   corresponding increase in the risk of instability, if the error
   characteristics of the link are not taken properly into account when
   setting the keepalive interval.

      Editors' Note: A "fast" liveness protocol has been specified in
      [Katz10].

   An additional problem with the hold-down mechanism in BGP is the
   amount of information that has to be exchanged to re-establish the
   database of route advertisements on each side of the link when it is
   re-established after a failure.  Currently any failure, however brief
   forces a full exchange that could perhaps be constrained by retaining
   some state across limited time failures and using revision control,
   transaction and replication techniques to resynchronize the
   databases.  Various techniques have been implemented to try to reduce
   this problem, but they have not yet been standardized.

5.9.  Interaction between Inter-Domain Routing and Intra-Domain Routing



   Today, many operators' backbone routers run both I-BGP and an intra-
   domain protocol to maintain the routes that reach between the borders
   of the domain.  Exporting routes from BGP into the intra-domain
   protocol in use and bringing them back up to BGP is not recommended
   [RFC2791], but it is still necessary for all backbone routers to run
   both protocols.  BGP is used to find the egress point and intra-





Davies & Doria                  Historic                       [Page 40]

RFC 5773                       IDR History                 February 2010


   domain protocol to find the path (next-hop router) to the egress
   point across the domain.  This is not only a management problem but
   may also create other problems:

   o  BGP is a path-vector protocol (i.e., a protocol that uses distance
      metrics possibly overridden by policy metrics), whereas most
      intra-domain protocols are link-state protocols.  As such, BGP is
      not optimized for convergence speed although distance-vector
      algorithms generally require less processing power.  Incidentally,
      more efficient distance-vector algorithms are available such as
      [Xu97].

   o  The metrics used in BGP and the intra-domain protocol are rarely
      comparable or combinable.  Whilst there are arguments that the
      optimizations inside a domain may be different from those for end-
      to-end paths, there are occasions, such as calculating the
      "topologically nearest" server when computable or combinable
      metrics would be of assistance.

   o  The policies that can be implemented using BGP are designed for
      control of traffic exchange between operators, not for controlling
      paths within a domain.  Policies for BGP are most conveniently
      expressed in Routing Policy Support Language (RPSL) [RFC2622] and
      this could be extended if thought desirable to include additional
      policy information.

   o  If the NEXT HOP destination for a set of BGP routes becomes
      inaccessible because of intra-domain protocol problems, the routes
      using the vanished next hop have to be invalidated at the next
      available UPDATE.  Subsequently, if the next-hop route reappears,
      this would normally lead to the BGP speaker requesting a full
      table from its neighbor(s).  Current implementations may attempt
      to circumvent the effects of intra-domain protocol route flap by
      caching the invalid routes for a period in case the next hop is
      restored through the "graceful restart" mechanism.

         Editors' Note: This was standardized as [RFC4724].

   o  Synchronization between intra-domain and inter-domain routing
      information is a problem as long as we use different protocols for
      intra-domain and inter-domain routing, which will most probably be
      the case even in the future because of the differing requirements
      in the two situations.  Some sort of synchronization between those
      two protocols would be useful.  In the RFC "IS-IS Transient
      Blackhole Avoidance" [RFC3277], the intra-domain protocol side of
      the story is covered (there is an equivalent discussion for OSPF).





Davies & Doria                  Historic                       [Page 41]

RFC 5773                       IDR History                 February 2010


   o  Synchronizing in BGP means waiting for the intra-domain protocol
      to know about the same networks as the inter-domain protocol,
      which can take a significant period of time and slows down the
      convergence of BGP by adding the intra-domain protocol convergence
      time into each cycle.  In general, operators no longer attempt
      full synchronization in order to avoid this problem (in general,
      redistributing the entire BGP routing feed into the local intra-
      domain protocol is unnecessary and undesirable but where a domain
      has multiple exits to peers and other non-customer networks,
      changes in BGP routing that affect the exit taken by traffic
      require corresponding re-routing in the intra-domain routing).

5.10.  Policy Issues



   There are several classes of issues with current BGP policy:

   o  Policy is installed in an ad hoc manner in each autonomous system.
      There isn't a method for ensuring that the policy installed in one
      router is coherent with policies installed in other routers.

   o  As described in Griffin [Griffin99] and in McPherson [RFC3345], it
      is possible to create policies for ASs, and instantiate them in
      routers, that will cause BGP to fail to converge in certain types
      of topology

   o  There is no available network model for describing policy in a
      coherent manner.

   Policy management is extremely complex and mostly done without the
   aid of any automated procedures.  The extreme complexity means that a
   highly-qualified specialist is required for policy management of
   border routers.  The training of these specialists is quite lengthy
   and needs to involve long periods of hands-on experience.  There is,
   therefore, a shortage of qualified staff for installing and
   maintaining the routing policies.  Because of the overall complexity
   of BGP, policy management tends to be only a relatively small topic
   within a complete BGP training course and specialized policy
   management training courses are not generally available.

5.11.  Security Issues



   While many of the issues with BGP security have been traced either to
   implementation issues or to operational issues, BGP is vulnerable to
   Distributed Denial of Service (DDoS) attacks.  Additionally, routers
   can be used as unwitting forwarders in DDoS attacks on other systems.






Davies & Doria                  Historic                       [Page 42]

RFC 5773                       IDR History                 February 2010


   Though DDoS attacks can be fought in a variety of ways, mostly using
   filtering methods, it takes constant vigilance.  There is nothing in
   the current architecture or in the protocols that serves to protect
   the forwarders from these attacks.

      Editors' Note: Since the original document was written, the issue
      of inter-domain routing security has been studied in much greater
      depth.  The rpsec working group has gone into the security issues
      in great detail [RFC4593] and readers should refer to that work to
      understand the security issues.

5.12.  Support of MPLS and VPNS



   Recently, BGP has been modified to function as a signaling protocol
   for MPLS and for VPNs [RFC4364].  Some people see this overloading of
   the BGP protocol as a boon whilst others see it as a problem.  While
   it was certainly convenient as a vehicle for vendors to deliver extra
   functionality to their products, it has exacerbated some of the
   performance and complexity issues of BGP.  Two important problems are
   that, the additional state that must be retained and refreshed to
   support VPN (Virtual Private Network) tunnels and that BGP does not
   provide end-to-end notification making it difficult to confirm that
   all necessary state has been installed or updated.

   It is an open question whether VPN signaling protocols should remain
   separate from the route determination protocols.

5.13.  IPv4/IPv6 Ships in the Night



   The fact that service providers need to maintain two completely
   separate networks, one for IPv4 and one for IPv6, has been a real
   hindrance to the introduction of IPv6.  When IPv6 does get widely
   deployed, it will do so without causing the disappearance of IPv4.
   This means that unless something is done, service providers would
   need to maintain the two networks in perpetuity (at least on the
   foreshortened timescale which the Internet world uses).

   It is possible to use a single set of BGP speakers with multi-
   protocol extensions [RFC4760] to exchange information about both IPv4
   and IPv6 routes between domains, but the use of TCP as the transport
   protocol for the information exchange results in an asymmetry when
   choosing to use one of TCP over IPv4 or TCP over IPv6.  Successful
   information exchange confirms one of IPv4 or IPv6 reachability
   between the speakers but not the other, making it possible that
   reachability is being advertised for a protocol for which it is not
   present.





Davies & Doria                  Historic                       [Page 43]

RFC 5773                       IDR History                 February 2010


   Also, current implementations do not allow a route to be advertised
   for both IPv4 and IPv6 in the same UPDATE message, because it is not
   possible to explicitly link the reachability information for an
   address family to the corresponding next-hop information.  This could
   be improved, but currently results in independent UPDATEs being
   exchanged for each address family.

5.14.  Existing Tools to Support Effective Deployment of Inter-Domain
       Routing



   The tools available to network operators to assist in configuring and
   maintaining effective inter-domain routing in line with their defined
   policies are limited, and almost entirely passive.

   o  There are no tools to facilitate the planning of the routing of a
      domain (either intra- or inter-domain); there are a limited number
      of display tools that will visualize the routing once it has been
      configured.

   o  There are no tools to assist in converting business policy
      specifications into the Routing Policy Specification Language
      (RPSL) language (see Section 5.14.1); there are limited tools to
      convert the RPSL into BGP commands and to check, post-facto, that
      the proposed policies are consistent with the policies in adjacent
      domains (always provided that these have been revealed and
      accurately documented).

   o  There are no tools to monitor BGP route changes in real-time and
      warn the operator about policy inconsistencies and/or
      instabilities.

   The following section summarizes the tools that are available to
   assist with the use of RPSL.  Note they are all batch mode tools used
   off-line from a real network.  These tools will provide checks for
   skilled inter-domain routing configurers but limited assistance for
   the novice.

5.14.1.  Routing Policy Specification Language RPSL (RFC 2622 and RFC
         2650) and RIPE NCC Database (RIPE 157)



   Routing Policy Specification Language (RPSL) [RFC2622] enables a
   network operator to describe routes, routers, and Autonomous Systems
   (ASs) that are connected to the local AS.

   Using the RPSL language (see [RFC2650]) a distributed database is
   created to describe routing policies in the Internet as described by
   each AS independently.  The database can be used to check the
   consistency of routing policies stored in the database.



Davies & Doria                  Historic                       [Page 44]

RFC 5773                       IDR History                 February 2010


   Tools exist [IRRToolSet] that can use the database to (among other
   things):

   o  Flag when two neighboring network operators specify conflicting or
      inconsistent routing information exchanges with each other and
      also detect global inconsistencies where possible;

   o  Extract all AS-paths between two networks that are allowed by
      routing policy from the routing policy database; display the
      connectivity a given network has according to current policies.

   The database queries enable a partial-static solution to the
   convergence problem.  They analyze routing policies of a very limited
   part of Internet and verify that they do not contain conflicts that
   could lead to protocol divergence.  The static analysis of
   convergence of the entire system has exponential time complexity, so
   approximation algorithms would have to be used.

   The toolset also allows router configurations to be generated from
   RPSL specifications.

      Editors' Note: The "Internet Routing Registry Toolset" was
      originally developed by the University of Southern California's
      Information Sciences Institute (ISI) between 1997 and 2001 as the
      "Routing Arbiter ToolSet" (RAToolSet) project.  The toolset is no
      longer developed by ISI but is used worldwide, so after a period
      of improvement by RIPE NCC, it has now been transferred to the
      Internet Systems Consortium (ISC) for ongoing maintenance as a
      public resource.

6.  Security Considerations



   As this is an informational document on the history of requirements
   in IDR and on the problems facing the current Internet IDR
   architecture, it does not as such create any security problems.  On
   the other hand, some of the problems with today's Internet routing
   architecture do create security problems, and these have been
   discussed in the text above.

7.  Acknowledgments



   The document is derived from work originally produced by Babylon.
   Babylon was a loose association of individuals from academia, service
   providers, and vendors whose goal was to discuss issues in Internet
   routing with the intention of finding solutions for those problems.






Davies & Doria                  Historic                       [Page 45]

RFC 5773                       IDR History                 February 2010


   The individual members who contributed materially to this document
   are: Anders Bergsten, Howard Berkowitz, Malin Carlzon, Lenka Carr
   Motyckova, Elwyn Davies, Avri Doria, Pierre Fransson, Yong Jiang,
   Dmitri Krioukov, Tove Madsen, Olle Pers, and Olov Schelen.

   Thanks also go to the members of Babylon and others who did
   substantial reviews of this material.  Specifically, we would like to
   acknowledge the helpful comments and suggestions of the following
   individuals: Loa Andersson, Tomas Ahlstrom, Erik Aman, Thomas
   Eriksson, Niklas Borg, Nigel Bragg, Thomas Chmara, Krister Edlund,
   Owe Grafford, Susan Hares, Torbjorn Lundberg, David McGrew, Jasminko
   Mulahusic, Florian-Daniel Otel, Bernhard Stockman, Tom Worster, and
   Roberto Zamparo.

   In addition, the authors are indebted to the folks who wrote all the
   references we have consulted in putting this paper together.  This
   includes not only the references explicitly listed below, but also
   those who contributed to the mailing lists we have been participating
   in for years.

   The editors thank Lixia Zhang, as IRSG document shepherd, for her
   help and her perseverance, without which this document would never
   have been published.

   Finally, it is the editors who are responsible for any lack of
   clarity, any errors, glaring omissions or misunderstandings.

8.  Informative References



   [Alaettinoglu00]
              Alaettinoglu, C., Jacobson, V., and H. Yu, "Towards Milli-
              Second IGP Convergence", Work in Progress, November 2000.

   [Berkowitz01]
              Berkowitz, H. and D. Krioukov, "To Be Multihomed:
              Requirements and Definitions", Work in Progress,
              July 2001.

   [Breslau90]
              Breslau, L. and D. Estrin, "An Architecture for Network-
              Layer Routing in OSI", Proceedings of the ACM symposium on
              Communications architectures & protocols , 1990.

   [Chapin94]
              Piscitello, D. and A. Chapin, "Open Systems Networking:
              TCP/IP & OSI", Addison-Wesley Copyright assigned to
              authors, 1994, <http://www.interisle.net/OSN/OSN.html>.




Davies & Doria                  Historic                       [Page 46]

RFC 5773                       IDR History                 February 2010


   [Chiappa91]
              Chiappa, J., "A New IP Routing and Addressing
              Architecture", Work in Progress, 1991.

   [Clark00]  Clark, D. and M. Blumenthal, "Rethinking the design of the
              Internet: The end to end arguments vs. the brave new
              world", August 2000,
              <http://dspace.mit.edu/handle/1721.1/1519>.

   [Griffin99]
              Griffin, T. and G. Wilfong, "An Analysis of BGP
              Convergence Properties", Association for Computing
              Machinery Proceedings of SIGCOMM '99, 1999.

   [Huitema90]
              Huitema, C. and W. Dabbous, "Routeing protocols
              development in the OSI architecture",  Proceedings of
              ISCIS V Turkey, 1990.

   [Huston05]
              Huston, G., "Exploring Autonomous System Numbers", The ISP
              Column , August 2005,
              <http://www.potaroo.net/ispcol/2005-08/as.html>.

   [INARC89]  Mills, D., Ed. and M. Davis, Ed., "Internet Architecture
              Workshop: Future of the Internet System Architecture and
              TCP/IP Protocols - Report", Internet Architecture Task
              Force INARC, 1990, <http://www.eecis.udel.edu/~mills/
              database/papers/inarc.pdf>.

   [IRRToolSet]
              Internet Systems Consortium, "Internet Routing Registry
              Toolset Project", IRR Tool Set Website, 2006,
              <http://www.isc.org/index.pl?/sw/IRRToolSet/>.

   [ISO10747]
              ISO/IEC, "Protocol for Exchange of Inter-Domain Routeing
              Information among Intermediate Systems to support
              Forwarding of ISO 8473 PDUs", International Standard
              10747 , 1993.

   [Jiang02]  Jiang, Y., Doria, A., Olsson, D., and F. Pettersson,
              "Inter-domain Routing Stability Measurement", 2002,
              <http://psg.com/~avri/papers/paper-yong-
              hpsr2002-final.pdf>.

   [Katz10]   Katz, D. and D. Ward, "Bidirectional Forwarding
              Detection", Work in Progress, January 2010.



Davies & Doria                  Historic                       [Page 47]

RFC 5773                       IDR History                 February 2010


   [Labovitz02]
              Labovitz, C., Ahuja, A., Farnam, J., and A. Bose,
              "Experimental Measurement of Delayed Convergence", NANOG ,
              2002.

   [NewArch03]
              Clark, D., Sollins, K., Wroclawski, J., Katabi, D., Kulik,
              J., Yang, X., Braden, R., Faber, T., Falk, A., Pingali,
              V., Handley, M., and N. Chiappa, "New Arch: Future
              Generation Internet Architecture", December 2003,
              <http://www.isi.edu/newarch/iDOCS/final.finalreport.pdf>.

   [RFC0904]  Mills, D., "Exterior Gateway Protocol formal
              specification", RFC 904, April 1984.

   [RFC0975]  Mills, D., "Autonomous confederations", RFC 975,
              February 1986.

   [RFC1105]  Lougheed, K. and J. Rekhter, "Border Gateway Protocol
              (BGP)", RFC 1105, June 1989.

   [RFC1126]  Little, M., "Goals and functional requirements for inter-
              autonomous system routing", RFC 1126, October 1989.

   [RFC1163]  Lougheed, K. and Y. Rekhter, "Border Gateway Protocol
              (BGP)", RFC 1163, June 1990.

   [RFC1267]  Lougheed, K. and Y. Rekhter, "Border Gateway Protocol 3
              (BGP-3)", RFC 1267, October 1991.

   [RFC1752]  Bradner, S. and A. Mankin, "The Recommendation for the IP
              Next Generation Protocol", RFC 1752, January 1995.

   [RFC1753]  Chiappa, J., "IPng Technical Requirements Of the Nimrod
              Routing and Addressing Architecture", RFC 1753,
              December 1994.

   [RFC1771]  Rekhter, Y. and T. Li, "A Border Gateway Protocol 4
              (BGP-4)", RFC 1771, March 1995.

   [RFC1992]  Castineyra, I., Chiappa, N., and M. Steenstrup, "The
              Nimrod Routing Architecture", RFC 1992, August 1996.

   [RFC2362]  Estrin, D., Farinacci, D., Helmy, A., Thaler, D., Deering,
              S., Handley, M., and V. Jacobson, "Protocol Independent
              Multicast-Sparse Mode (PIM-SM): Protocol Specification",
              RFC 2362, June 1998.




Davies & Doria                  Historic                       [Page 48]

RFC 5773                       IDR History                 February 2010


   [RFC2622]  Alaettinoglu, C., Villamizar, C., Gerich, E., Kessens, D.,
              Meyer, D., Bates, T., Karrenberg, D., and M. Terpstra,
              "Routing Policy Specification Language (RPSL)", RFC 2622,
              June 1999.

   [RFC2650]  Meyer, D., Schmitz, J., Orange, C., Prior, M., and C.
              Alaettinoglu, "Using RPSL in Practice", RFC 2650,
              August 1999.

   [RFC2791]  Yu, J., "Scalable Routing Design Principles", RFC 2791,
              July 2000.

   [RFC3221]  Huston, G., "Commentary on Inter-Domain Routing in the
              Internet", RFC 3221, December 2001.

   [RFC3277]  McPherson, D., "Intermediate System to Intermediate System
              (IS-IS) Transient Blackhole Avoidance", RFC 3277,
              April 2002.

   [RFC3345]  McPherson, D., Gill, V., Walton, D., and A. Retana,
              "Border Gateway Protocol (BGP) Persistent Route
              Oscillation Condition", RFC 3345, August 2002.

   [RFC3618]  Fenner, B. and D. Meyer, "Multicast Source Discovery
              Protocol (MSDP)", RFC 3618, October 2003.

   [RFC3765]  Huston, G., "NOPEER Community for Border Gateway Protocol
              (BGP) Route Scope Control", RFC 3765, April 2004.

   [RFC3913]  Thaler, D., "Border Gateway Multicast Protocol (BGMP):
              Protocol Specification", RFC 3913, September 2004.

   [RFC4116]  Abley, J., Lindqvist, K., Davies, E., Black, B., and V.
              Gill, "IPv4 Multihoming Practices and Limitations",
              RFC 4116, July 2005.

   [RFC4204]  Lang, J., "Link Management Protocol (LMP)", RFC 4204,
              October 2005.

   [RFC4271]  Rekhter, Y., Li, T., and S. Hares, "A Border Gateway
              Protocol 4 (BGP-4)", RFC 4271, January 2006.

   [RFC4364]  Rosen, E. and Y. Rekhter, "BGP/MPLS IP Virtual Private
              Networks (VPNs)", RFC 4364, February 2006.

   [RFC4593]  Barbir, A., Murphy, S., and Y. Yang, "Generic Threats to
              Routing Protocols", RFC 4593, October 2006.




Davies & Doria                  Historic                       [Page 49]

RFC 5773                       IDR History                 February 2010


   [RFC4601]  Fenner, B., Handley, M., Holbrook, H., and I. Kouvelas,
              "Protocol Independent Multicast - Sparse Mode (PIM-SM):
              Protocol Specification (Revised)", RFC 4601, August 2006.

   [RFC4724]  Sangli, S., Chen, E., Fernando, R., Scudder, J., and Y.
              Rekhter, "Graceful Restart Mechanism for BGP", RFC 4724,
              January 2007.

   [RFC4760]  Bates, T., Chandra, R., Katz, D., and Y. Rekhter,
              "Multiprotocol Extensions for BGP-4", RFC 4760,
              January 2007.

   [RFC4893]  Vohra, Q. and E. Chen, "BGP Support for Four-octet AS
              Number Space", RFC 4893, May 2007.

   [RFC5772]  Doria, A., Davies, E., and F. Kastenholz, "A Set of
              Possible Requirements for a Future Routing Architecture",
              RFC 5772, February 2010.

   [Sandiick00]
              Sandick, H., Squire, M., Cain, B., Duncan, I., and B.
              Haberman, "Fast LIveness Protocol (FLIP)", Work
              in Progress, February 2000.

   [Tsuchiya87]
              Tsuchiya, P., "An Architecture for Network-Layer Routing
              in OSI", Proceedings of the ACM workshop on Frontiers in
              computer communications technology , 1987.

   [Xu97]     Xu, Z., Dai, S., and J. Garcia-Luna-Aceves, "A More
              Efficient Distance Vector Routing Algorithm", Proc IEEE
              MILCOM 97, Monterey, California, Nov 1997, <http://
              www.cse.ucsc.edu/research/ccrg/publications/
              zhengyu.milcom97.pdf>.

















Davies & Doria                  Historic                       [Page 50]

RFC 5773                       IDR History                 February 2010


Authors' Addresses



   Elwyn B. Davies
   Folly Consulting
   Soham, Cambs
   UK

   Phone: +44 7889 488 335
   EMail: elwynd@dial.pipex.com


   Avri Doria
   LTU
   Lulea,   971 87
   Sweden

   Phone: +1 401 663 5024
   EMail: avri@acm.org

































Davies & Doria                  Historic                       [Page 51]