RFC 9049

Internet Research Task Force (IRTF)                      S. Dawkins, Ed.
Request for Comments: 9049                               Tencent America
Category: Informational                                        June 2021
ISSN: 2070-1721

             Path Aware Networking: Obstacles to Deployment
                    (A Bestiary of Roads Not Taken)


   This document is a product of the Path Aware Networking Research
   Group (PANRG).  At the first meeting of the PANRG, the Research Group
   agreed to catalog and analyze past efforts to develop and deploy Path
   Aware techniques, most of which were unsuccessful or at most
   partially successful, in order to extract insights and lessons for
   Path Aware networking researchers.

   This document contains that catalog and analysis.

Status of This Memo

   This document is not an Internet Standards Track specification; it is
   published for informational purposes.

   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 consensus of the Path Aware
   Networking Research Group of the Internet Research Task Force (IRTF).
   Documents approved for publication by the IRSG are not candidates for
   any level of Internet Standard; see Section 2 of RFC 7841.

   Information about the current status of this document, any errata,
   and how to provide feedback on it may be obtained at

Copyright Notice

   Copyright (c) 2021 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
   (https://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.

Table of Contents

   1.  Introduction
     1.1.  What Do "Path" and "Path Awareness" Mean in This Document?
   2.  A Perspective on This Document
     2.1.  Notes for the Reader
     2.2.  A Note about Path Aware Techniques Included in This
     2.3.  Architectural Guidance
     2.4.  Terminology Used in This Document
     2.5.  Methodology for Contributions
   3.  Applying the Lessons We've Learned
   4.  Summary of Lessons Learned
     4.1.  Justifying Deployment
     4.2.  Providing Benefits for Early Adopters
     4.3.  Providing Benefits during Partial Deployment
     4.4.  Outperforming End-to-End Protocol Mechanisms
     4.5.  Paying for Path Aware Techniques
     4.6.  Impact on Operational Practices
     4.7.  Per-Connection State
     4.8.  Keeping Traffic on Fast Paths
     4.9.  Endpoints Trusting Intermediate Nodes
     4.10. Intermediate Nodes Trusting Endpoints
     4.11. Reacting to Distant Signals
     4.12. Support in Endpoint Protocol Stacks
     4.13. Planning for Failure
   5.  Future Work
   6.  Contributions
     6.1.  Stream Transport (ST, ST2, ST2+)
       6.1.1.  Reasons for Non-deployment
       6.1.2.  Lessons Learned
     6.2.  Integrated Services (IntServ)
       6.2.1.  Reasons for Non-deployment
       6.2.2.  Lessons Learned
     6.3.  Quick-Start TCP
       6.3.1.  Reasons for Non-deployment
       6.3.2.  Lessons Learned
     6.4.  ICMP Source Quench
       6.4.1.  Reasons for Non-deployment
       6.4.2.  Lessons Learned
     6.5.  Triggers for Transport (TRIGTRAN)
       6.5.1.  Reasons for Non-deployment
       6.5.2.  Lessons Learned
     6.6.  Shim6
       6.6.1.  Reasons for Non-deployment
       6.6.2.  Lessons Learned
       6.6.3.  Addendum on Multipath TCP
     6.7.  Next Steps in Signaling (NSIS)
       6.7.1.  Reasons for Non-deployment
       6.7.2.  Lessons Learned
     6.8.  IPv6 Flow Labels
       6.8.1.  Reasons for Non-deployment
       6.8.2.  Lessons Learned
     6.9.  Explicit Congestion Notification (ECN)
       6.9.1.  Reasons for Non-deployment
       6.9.2.  Lessons Learned
   7.  Security Considerations
   8.  IANA Considerations
   9.  Informative References

   Author's Address

1.  Introduction

   This document describes the lessons that IETF participants have
   learned (and learned the hard way) about Path Aware networking over a
   period of several decades.  It also provides an analysis of reasons
   why various Path Aware techniques have seen limited or no deployment.

   This document represents the consensus of the Path Aware Networking
   Research Group (PANRG).

1.1.  What Do "Path" and "Path Awareness" Mean in This Document?

   One of the first questions reviewers of this document have asked is
   "What's the definition of a Path, and what's the definition of Path
   Awareness?"  That is not an easy question to answer for this

   These terms have definitions in other PANRG documents [PANRG] and are
   still the subject of some discussion in the Research Group, as of the
   date of this document.  But because this document reflects work
   performed over several decades, the technologies described in
   Section 6 significantly predate the current definitions of "Path" and
   "Path Aware" in use in the Path Aware Networking Research Group, and
   it is unlikely that all the contributors to Section 6 would have had
   the same understanding of these terms.  Those technologies were
   considered "Path Aware" in early PANRG discussions and so are
   included in this retrospective document.

   It is worth noting that the definitions of "Path" and "Path Aware" in
   [PANRG-PATH-PROPERTIES] would apply to Path Aware techniques at a
   number of levels of the Internet protocol architecture ([RFC1122],
   plus several decades of refinements), but the contributions received
   for this document tended to target the transport layer and to treat a
   "Path" constructed by routers as opaque.  It would be useful to
   consider how applicable the Lessons Learned cataloged in this
   document are, at other layers, and that would be a fine topic for
   follow-on research.

   The current definition of "Path" in the Path Aware Networking
   Research Group appears in Section 2 ("Terminology") in
   [PANRG-PATH-PROPERTIES].  That definition is included here as a
   convenience to the reader.

   |  Path: A sequence of adjacent path elements over which a packet can
   |  be transmitted, starting and ending with a node.  A path is
   |  unidirectional.  Paths are time-dependent, i.e., the sequence of
   |  path elements over which packets are sent from one node to another
   |  may change.  A path is defined between two nodes.  For multicast
   |  or broadcast, a packet may be sent by one node and received by
   |  multiple nodes.  In this case, the packet is sent over multiple
   |  paths at once, one path for each combination of sending and
   |  receiving node; these paths do not have to be disjoint.  Note that
   |  an entity may have only partial visibility of the path elements
   |  that comprise a path and visibility may change over time.
   |  Different entities may have different visibility of a path and/or
   |  treat path elements at different levels of abstraction.

   The current definition of Path Awareness, used by the Path Aware
   Networking Research Group, appears in Section 1.1 ("Definition") in
   [PANRG-QUESTIONS].  That definition is included here as a convenience
   to the reader.

   |  For purposes of this document, "path aware networking" describes
   |  endpoint discovery of the properties of paths they use for
   |  communication across an internetwork, and endpoint reaction to
   |  these properties that affects routing and/or data transfer.  Note
   |  that this can and already does happen to some extent in the
   |  current Internet architecture; this definition expands current
   |  techniques of path discovery and manipulation to cross
   |  administrative domain boundaries and up to the transport and
   |  application layers at the endpoints.
   |  Expanding on this definition, a "path aware internetwork" is one
   |  in which endpoint discovery of path properties and endpoint
   |  selection of paths used by traffic exchanged by the endpoint are
   |  explicitly supported, regardless of the specific design of the
   |  protocol features which enable this discovery and selection.

2.  A Perspective on This Document

   At the first meeting of the Path Aware Networking Research Group
   [PANRG], at IETF 99 [PANRG-99], Olivier Bonaventure led a discussion
   of "A Decade of Path Awareness" [PATH-Decade], on attempts, which
   were mostly unsuccessful for a variety of reasons, to exploit Path
   Aware techniques and achieve a variety of goals over the past decade.
   At the end of that discussion, two things were abundantly clear.

   *  The Internet community has accumulated considerable experience
      with many Path Aware techniques over a long period of time, and

   *  Although some Path Aware techniques have been deployed (for
      example, Differentiated Services, or Diffserv [RFC2475]), most of
      these techniques haven't seen widespread adoption and deployment.
      Even "successful" techniques like Diffserv can face obstacles that
      prevent wider usage.  The reasons for non-adoption and limited
      adoption and deployment are many and are worthy of study.

   The meta-lessons from that experience were as follows:

   *  Path Aware networking has been more Research than Engineering, so
      establishing an IRTF Research Group for Path Aware networking was
      the right thing to do [RFC7418].

   *  Analyzing a catalog of past experience to learn the reasons for
      non-adoption would be a great first step for the Research Group.

   Allison Mankin, as IRTF Chair, officially chartered the Path Aware
   Networking Research Group in July 2018.

   This document contains the analysis performed by that Research Group
   (Section 4), based on that catalog (Section 6).

2.1.  Notes for the Reader

   This Informational document discusses Path Aware protocol mechanisms
   considered, and in some cases standardized, by the Internet
   Engineering Task Force (IETF), and it considers Lessons Learned from
   those mechanisms.  The intention is to inform the work of protocol
   designers, whether in the IRTF, the IETF, or elsewhere in the
   Internet ecosystem.

   As an Informational document published in the IRTF Stream, this
   document has no authority beyond the quality of the analysis it

2.2.  A Note about Path Aware Techniques Included in This Document

   This document does not catalog every proposed Path Aware technique
   that was not adopted and deployed.  Instead, we limited our focus to
   technologies that passed through the IETF community and still
   identified enough techniques to provide background for the lessons
   included in Section 4 to inform researchers and protocol engineers in
   their work.

   No shame is intended for the techniques included in this document.
   As shown in Section 4, the quality of specific techniques had little
   to do with whether they were deployed or not.  Based on the
   techniques cataloged in this document, it is likely that when these
   techniques were put forward, the proponents were trying to engineer
   something that could not be engineered without first carrying out
   research.  Actual shame would be failing to learn from experience and
   failing to share that experience with other networking researchers
   and engineers.

2.3.  Architectural Guidance

   As background for understanding the Lessons Learned contained in this
   document, the reader is encouraged to become familiar with the
   Internet Architecture Board's documents on "What Makes for a
   Successful Protocol?"  [RFC5218] and "Planning for Protocol Adoption
   and Subsequent Transitions" [RFC8170].

   Although these two documents do not specifically target Path Aware
   networking protocols, they are helpful resources for readers seeking
   to improve their understanding of considerations for successful
   adoption and deployment of any protocol.  For example, the basic
   success factors described in Section 2.1 of [RFC5218] are helpful for
   readers of this document.

   Because there is an economic aspect to decisions about deployment,
   the IAB Workshop on Internet Technology Adoption and Transition
   [ITAT] report [RFC7305] also provides food for thought.

   Several of the Lessons Learned in Section 4 reflect considerations
   described in [RFC5218], [RFC7305], and [RFC8170].

2.4.  Terminology Used in This Document

   The terms "node" and "element" in this document have the meaning
   defined in [PANRG-PATH-PROPERTIES].

2.5.  Methodology for Contributions

   This document grew out of contributions by various IETF participants
   with experience with one or more Path Aware techniques.

   There are many things that could be said about the Path Aware
   techniques that have been developed.  For the purposes of this
   document, contributors were requested to provide

   *  the name of a technique, including an abbreviation if one was

   *  if available, a long-term pointer to the best reference describing
      the technique.

   *  a short description of the problem the technique was intended to

   *  a short description of the reasons why the technique wasn't

   *  a short statement of the lessons that researchers can learn from
      our experience with this technique.

3.  Applying the Lessons We've Learned

   The initial scope for this document was roughly "What mistakes have
   we made in the decade prior to [PANRG-99], that we shouldn't make
   again?"  Some of the contributions in Section 6 predate the initial
   scope.  The earliest Path Aware technique referred to in Section 6 is
   [IEN-119], which was published in the late 1970s; see Section 6.1.
   Given that the networking ecosystem has evolved continuously, it
   seems reasonable to consider how to apply these lessons.

   The PANRG reviewed the Lessons Learned (Section 4) contained in the
   May 23, 2019 draft version of this document at IETF 105
   [PANRG-105-Min] and carried out additional discussion at IETF 106
   [PANRG-106-Min].  Table 1 provides the "sense of the room" about each
   lesson after those discussions.  The intention was to capture whether
   a specific lesson seems to be

   *  "Invariant" - well-understood and is likely to be applicable for
      any proposed Path Aware networking solution.

   *  "Variable" - has impeded deployment in the past but might not be
      applicable in a specific technique.  Engineering analysis to
      understand whether the lesson is applicable is prudent.

   *  "Not Now" - a characteristic that tends to turn up a minefield
      full of dragons.  Prudent network engineers will wish to avoid
      gambling on a technique that relies on this, until something
      significant changes.

   Section 6.9 on Explicit Congestion Notification (ECN) was added
   during the review and approval process, based on a question from
   Martin Duke.  Section 6.9, as contained in the March 8, 2021 draft
   version of this document, was discussed at [PANRG-110] and is
   summarized in Section 4.13, describing a new Lesson Learned.

    | Lesson                                              | Category  |
    | Justifying Deployment (Section 4.1)                 | Invariant |
    | Providing Benefits for Early Adopters (Section 4.2) | Invariant |
    | Providing Benefits during Partial Deployment        | Invariant |
    | (Section 4.3)                                       |           |
    | Outperforming End-to-End Protocol Mechanisms        | Variable  |
    | (Section 4.4)                                       |           |
    | Paying for Path Aware Techniques (Section 4.5)      | Invariant |
    | Impact on Operational Practices (Section 4.6)       | Invariant |
    | Per-Connection State (Section 4.7)                  | Variable  |
    | Keeping Traffic on Fast Paths (Section 4.8)         | Variable  |
    | Endpoints Trusting Intermediate Nodes (Section 4.9) | Not Now   |
    | Intermediate Nodes Trusting Endpoints               | Not Now   |
    | (Section 4.10)                                      |           |
    | Reacting to Distant Signals (Section 4.11)          | Variable  |
    | Support in Endpoint Protocol Stacks (Section 4.12)  | Variable  |
    | Planning for Failure (Section 4.13)                 | Invariant |

                                  Table 1

   "Justifying Deployment", "Providing Benefits for Early Adopters",
   "Paying for Path Aware Techniques", "Impact on Operational
   Practices", and "Planning for Failure" were considered to be
   Invariant -- the sense of the room was that these would always be
   considerations for any proposed Path Aware technique.

   "Providing Benefits during Partial Deployment" was added after IETF
   105, during Research Group Last Call, and is also considered to be

   For "Outperforming End-to-End Protocol Mechanisms", there is a trade-
   off between improved performance from Path Aware techniques and
   additional complexity required by some Path Aware techniques.

   *  For example, if you can obtain the same understanding of path
      characteristics from measurements obtained over a few more round
      trips, endpoint implementers are unlikely to be eager to add
      complexity, and many attributes can be measured from an endpoint,
      without assistance from intermediate nodes.

   For "Per-Connection State", the key questions discussed in the
   Research Group were "how much state" and "where state is maintained".

   *  Integrated Services (IntServ) (Section 6.2) required state at
      every participating intermediate node for every connection between
      two endpoints.  As the Internet ecosystem has evolved, carrying
      many connections in a tunnel that appears to intermediate nodes as
      a single connection has become more common, so that additional
      end-to-end connections don't add additional state to intermediate
      nodes between tunnel endpoints.  If these tunnels are encrypted,
      intermediate nodes between tunnel endpoints can't distinguish
      between connections, even if that were desirable.

   For "Keeping Traffic on Fast Paths", we noted that this was true for
   many platforms, but not for all.

   *  For backbone routers, this is likely an Invariant, but for
      platforms that rely more on general-purpose computers to make
      forwarding decisions, this may not be a fatal flaw for Path Aware

   For "Endpoints Trusting Intermediate Nodes" and "Intermediate Nodes
   Trusting Endpoints", these lessons point to the broader need to
   revisit the Internet Threat Model.

   *  We noted with relief that discussions about this were already
      underway in the IETF community at IETF 105 (see the Security Area
      Open Meeting minutes [SAAG-105-Min] for discussion of
      [INTERNET-THREAT-MODEL] and [FARRELL-ETM]), and the Internet
      Architecture Board has created a mailing list for continued
      discussions [model-t], but we recognize that there are Path Aware
      networking aspects of this effort, requiring research.

   For "Reacting to Distant Signals", we noted that not all attributes
   are equal.

   *  If an attribute is stable over an extended period of time, is
      difficult to observe via end-to-end mechanisms, and is valuable,
      Path Aware techniques that rely on that attribute to provide a
      significant benefit become more attractive.

   *  Analysis to help identify attributes that are useful enough to
      justify deployment of Path Aware techniques that make use of those
      attributes would be helpful.

   For "Support in Endpoint Protocol Stacks", we noted that Path Aware
   applications must be able to identify and communicate requirements
   about path characteristics.

   *  The de facto sockets API has no way of signaling application
      expectations for the network path to the protocol stack.

4.  Summary of Lessons Learned

   This section summarizes the Lessons Learned from the contributed
   subsections in Section 6.

   Each Lesson Learned is tagged with one or more contributions that
   encountered this obstacle as a significant impediment to deployment.
   Other contributed techniques may have also encountered this obstacle,
   but this obstacle may not have been the biggest impediment to
   deployment for those techniques.

   It is useful to notice that sometimes an obstacle might impede
   deployment, while at other times, the same obstacle might prevent
   adoption and deployment entirely.  The Research Group discussed
   distinguishing between obstacles that impede and obstacles that
   prevent, but it appears that the boundary between "impede" and
   "prevent" can shift over time -- some of the Lessons Learned are
   based on both a) Path Aware techniques that were not deployed and b)
   Path Aware techniques that were deployed but were not deployed widely
   or quickly.  See Sections 6.6 and 6.6.3 for examples of this shifting

4.1.  Justifying Deployment

   The benefit of Path Awareness must be great enough to justify making
   changes in an operational network.  The colloquial U.S. American
   English expression, "If it ain't broke, don't fix it" is a "best
   current practice" on today's Internet.  (See Sections 6.3, 6.4, 6.5,
   and 6.9, in addition to [RFC5218].)

4.2.  Providing Benefits for Early Adopters

   Providing benefits for early adopters can be key -- if everyone must
   deploy a technique in order for the technique to provide benefits, or
   even to work at all, the technique is unlikely to be adopted widely
   or quickly.  (See Sections 6.2 and 6.3, in addition to [RFC5218].)

4.3.  Providing Benefits during Partial Deployment

   Some proposals require that all path elements along the full length
   of the path must be upgraded to support a new technique, before any
   benefits can be seen.  This is likely to require coordination between
   operators who control a subset of path elements, and between
   operators and end users if endpoint upgrades are required.  If a
   technique provides benefits when only a part of the path has been
   upgraded, this is likely to encourage adoption and deployment.  (See
   Sections 6.2, 6.3, and 6.9, in addition to [RFC5218].)

4.4.  Outperforming End-to-End Protocol Mechanisms

   Adaptive end-to-end protocol mechanisms may respond to feedback
   quickly enough that the additional realizable benefit from a new Path
   Aware mechanism that tries to manipulate nodes along a path, or
   observe the attributes of nodes along a path, may be much smaller
   than anticipated.  (See Sections 6.3 and 6.5.)

4.5.  Paying for Path Aware Techniques

   "Follow the money."  If operators can't charge for a Path Aware
   technique to recover the costs of deploying it, the benefits to the
   operator must be really significant.  Corollary: if operators charge
   for a Path Aware technique, the benefits to users of that Path Aware
   technique must be significant enough to justify the cost.  (See
   Sections 6.1, 6.2, 6.5, and 6.9.)

4.6.  Impact on Operational Practices

   The impact of a Path Aware technique requiring changes to operational
   practices can affect how quickly or widely a promising technique is
   deployed.  The impacts of these changes may make deployment more
   likely, but they often discourage deployment.  (See Section 6.6,
   including Section 6.6.3.)

4.7.  Per-Connection State

   Per-connection state in intermediate nodes has been an impediment to
   adoption and deployment in the past, because of added cost and
   complexity.  Often, similar benefits can be achieved with much less
   finely grained state.  This is especially true as we move from the
   edge of the network, further into the routing core.  (See
   Sections 6.1 and 6.2.)

4.8.  Keeping Traffic on Fast Paths

   Many modern platforms, especially high-end routers, have been
   designed with hardware that can make simple per-packet forwarding
   decisions ("fast paths") but have not been designed to make heavy use
   of in-band mechanisms such as IPv4 and IPv6 Router Alert Options
   (RAOs) that require more processing to make forwarding decisions.
   Packets carrying in-band mechanisms are diverted to other processors
   in the router with much lower packet-processing rates.  Operators can
   be reluctant to deploy techniques that rely heavily on in-band
   mechanisms because they may significantly reduce packet throughput.
   (See Section 6.7.)

4.9.  Endpoints Trusting Intermediate Nodes

   If intermediate nodes along the path can't be trusted, it's unlikely
   that endpoints will rely on signals from intermediate nodes to drive
   changes to endpoint behaviors.  We note that "trust" is not binary --
   one low level of trust applies when a node receiving a message can
   confirm that the sender of the message has visibility of the packets
   on the path it is seeking to control [RFC8085] (e.g., an ICMP
   Destination Unreachable message [RFC0792] that includes the Internet
   Header + 64 bits of Original Data Datagram payload from the source).
   A higher level of trust can arise when an endpoint has established a
   short-term, or even long-term, trust relationship with network nodes.
   (See Sections 6.4 and 6.5.)

4.10.  Intermediate Nodes Trusting Endpoints

   If the endpoints do not have any trust relationship with the
   intermediate nodes along a path, operators have been reluctant to
   deploy techniques that rely on endpoints sending unauthenticated
   control signals to routers.  (See Sections 6.2 and 6.7.)  (We also
   note that this still remains a factor hindering deployment of

4.11.  Reacting to Distant Signals

   Because the Internet is a distributed system, if the distance that
   information from distant path elements travels to a Path Aware host
   is sufficiently large, the information may no longer accurately
   represent the state and situation at the distant host or elements
   along the path when it is received locally.  In this case, the
   benefit that a Path Aware technique provides will be inconsistent and
   may not always be beneficial.  (See Section 6.3.)

4.12.  Support in Endpoint Protocol Stacks

   Just because a protocol stack provides a new feature/signal does not
   mean that applications will use the feature/signal.  Protocol stacks
   may not know how to effectively utilize Path Aware techniques,
   because the protocol stack may require information from applications
   to permit the technique to work effectively, but applications may not
   a priori know that information.  Even if the application does know
   that information, the de facto sockets API has no way of signaling
   application expectations for the network path to the protocol stack.
   In order for applications to provide these expectations to protocol
   stacks, we need an API that signals more than the packets to be sent.
   (See Sections 6.1 and 6.2.)

4.13.  Planning for Failure

   If early implementers discover severe problems with a new feature,
   that feature is likely to be disabled, and convincing implementers to
   re-enable that feature can be very difficult and can require years or
   decades.  In addition to testing, partial deployment for a subset of
   users, implementing instrumentation that will detect degraded user
   experience, and even "failback" to a previous version or "failover"
   to an entirely different implementation are likely to be helpful.
   (See Section 6.9.)

5.  Future Work

   By its nature, this document has been retrospective.  In addition to
   considering how the Lessons Learned to date apply to current and
   future Path Aware networking proposals, it's also worth considering
   whether there is deeper investigation left to do.

   *  We note that this work was based on contributions from experts on
      various Path Aware techniques, and all of the contributed
      techniques involved unicast protocols.  We didn't consider how
      these lessons might apply to multicast, and, given anecdotal
      reports at the IETF 109 Media Operations (MOPS) Working Group
      meeting of IP multicast offerings within data centers at one or
      more cloud providers [MOPS-109-Min], it might be useful to think
      about Path Awareness in multicast, before we have a history of
      unsuccessful deployments to document.

   *  The question of whether a mechanism supports admission control,
      based on either endpoints or applications, is associated with Path
      Awareness.  One of the motivations of IntServ and a number of
      other architectures (e.g., Deterministic Networking [RFC8655]) is
      the ability to "say no" to an application based on resource
      availability on a path, before the application tries to inject
      traffic onto that path and discovers the path does not have the
      capacity to sustain enough utility to meet the application's
      minimum needs.  The question of whether admission control is
      needed comes up repeatedly, but we have learned a few useful
      lessons that, while covered implicitly in some of the Lessons
      Learned provided in this document, might be explained explicitly:

      -  We have gained a lot of experience with application-based
         adaptation since the days where applications just injected
         traffic inelastically into the network.  Such adaptations seem
         to work well enough that admission control is of less value to
         these applications.

      -  There are end-to-end measurement techniques that can steer
         traffic at the application layer (Content Delivery Networks
         (CDNs), multi-CDNs like Conviva [Conviva], etc.).

      -  We noted in Section 4.12 that applications often don't know how
         to utilize Path Aware techniques.  This includes not knowing
         enough about their admission control threshold to be able to
         ask accurately for the resources they need, whether this is
         because the application itself doesn't know or because the
         application has no way to signal its expectations to the
         underlying protocol stack.  To date, attempts to help them
         haven't gotten anywhere (e.g., the multiple-TSPEC (Traffic
         Specification) additions to RSVP to attempt to mirror codec
         selection by applications [INTSERV-MULTIPLE-TSPEC] expired in

   *  We note that this work took the then-current IP network
      architecture as given, at least at the time each technique was
      proposed.  It might be useful to consider aspects of the now-
      current IP network architecture that ease, or impede, Path Aware
      techniques.  For example, there is limited ability in IP to
      constrain bidirectional paths to be symmetric, and information-
      centric networking protocols such as Named Data Networking (NDN)
      and Content-Centric Networking (CCNx) [RFC8793] must force
      bidirectional path symmetry using protocol-specific mechanisms.

6.  Contributions

   Contributions on these Path Aware techniques were analyzed to arrive
   at the Lessons Learned captured in Section 4.

   Our expectation is that most readers will not need to read through
   this section carefully, but we wanted to record these hard-fought
   lessons as a service to others who may revisit this document, so
   they'll have the details close at hand.

6.1.  Stream Transport (ST, ST2, ST2+)

   The suggested references for Stream Transport are:

   *  "ST - A Proposed Internet Stream Protocol" [IEN-119]

   *  "Experimental Internet Stream Protocol: Version 2 (ST-II)"

   *  "Internet Stream Protocol Version 2 (ST2) Protocol Specification -
      Version ST2+" [RFC1819]

   The first version of Stream Transport, ST [IEN-119], was published in
   the late 1970s and was implemented and deployed on the ARPANET at
   small scale.  It was used throughout the 1980s for experimental
   transmission of voice, video, and distributed simulation.

   The second version of the ST specification (ST2) [RFC1190] [RFC1819]
   was an experimental connection-oriented internetworking protocol that
   operated at the same layer as connectionless IP.  ST2 packets could
   be distinguished by their IP header version numbers (IP, at that
   time, used version number 4, while ST2 used version number 5).

   ST2 used a control plane layered over IP to select routes and reserve
   capacity for real-time streams across a network path, based on a flow
   specification communicated by a separate protocol.  The flow
   specification could be associated with QoS state in routers,
   producing an experimental resource reservation protocol.  This
   allowed ST2 routers along a path to offer end-to-end guarantees,
   primarily to satisfy the QoS requirements for real-time services over
   the Internet.

6.1.1.  Reasons for Non-deployment

   Although implemented in a range of equipment, ST2 was not widely used
   after completion of the experiments.  It did not offer the
   scalability and fate-sharing properties that have come to be desired
   by the Internet community.

   The ST2 protocol is no longer in use.

6.1.2.  Lessons Learned

   As time passed, the trade-off between router processing and link
   capacity changed.  Links became faster, and the cost of router
   processing became comparatively more expensive.

   The ST2 control protocol used "hard state" -- once a route was
   established, and resources were reserved, routes and resources
   existed until they were explicitly released via signaling.  A soft-
   state approach was thought superior to this hard-state approach and
   led to development of the IntServ model described in Section 6.2.

6.2.  Integrated Services (IntServ)

   The suggested references for IntServ are:

   *  "Integrated Services in the Internet Architecture: an Overview"

   *  "Specification of the Controlled-Load Network Element Service"

   *  "Specification of Guaranteed Quality of Service" [RFC2212]

   *  "General Characterization Parameters for Integrated Service
      Network Elements" [RFC2215]

   *  "Resource ReSerVation Protocol (RSVP) -- Version 1 Functional
      Specification" [RFC2205]

   In 1994, when the IntServ architecture document [RFC1633] was
   published, real-time traffic was first appearing on the Internet.  At
   that time, bandwidth was still a scarce commodity.  Internet Service
   Providers built networks over DS3 (45 Mbps) infrastructure, and sub-
   rate (< 1 Mbps) access was common.  Therefore, the IETF anticipated a
   need for a fine-grained QoS mechanism.

   In the IntServ architecture, some applications can require service
   guarantees.  Therefore, those applications use RSVP [RFC2205] to
   signal QoS reservations across network paths.  Every router in the
   network that participates in IntServ maintains per-flow soft state to
   a) perform call admission control and b) deliver guaranteed service.

   Applications use Flow Specifications (Flow Specs, or FLOWSPECs)
   [RFC2210] to describe the traffic that they emit.  RSVP reserves
   capacity for traffic on a per-Flow-Spec basis.

6.2.1.  Reasons for Non-deployment

   Although IntServ has been used in enterprise and government networks,
   IntServ was never widely deployed on the Internet because of its
   cost.  The following factors contributed to operational cost:

   *  IntServ must be deployed on every router that is on a path where
      IntServ is to be used.  Although it is possible to include a
      router that does not participate in IntServ along the path being
      controlled, if that router is likely to become a bottleneck,
      IntServ cannot be used to avoid that bottleneck along the path.

   *  IntServ maintained per-flow state.

   As IntServ was being discussed, the following occurred:

   *  For many expected uses, it became more cost effective to solve the
      QoS problem by adding bandwidth.  Between 1994 and 2000, Internet
      Service Providers upgraded their infrastructures from DS3 (45
      Mbps) to OC-48 (2.4 Gbps).  This meant that even if an endpoint
      was using IntServ in an IntServ-enabled network, its requests
      would rarely, if ever, be denied, so endpoints and Internet
      Service Providers had little reason to enable IntServ.

   *  Diffserv [RFC2475] offered a more cost-effective, albeit less
      fine-grained, solution to the QoS problem.

6.2.2.  Lessons Learned

   The following lessons were learned:

   *  Any mechanism that requires every participating on-path router to
      maintain per-flow state is not likely to succeed, unless the
      additional cost for offering the feature can be recovered from the

   *  Any mechanism that requires an operator to upgrade all of its
      routers is not likely to succeed, unless the additional cost for
      offering the feature can be recovered from the user.

   In environments where IntServ has been deployed, trust relationships
   with endpoints are very different from trust relationships on the
   Internet itself.  There are often clearly defined hierarchies in
   Service Level Agreements (SLAs) governing well-defined transport
   flows operating with predetermined capacity and latency requirements
   over paths where capacity or other attributes are constrained.

   IntServ was never widely deployed to manage capacity across the
   Internet.  However, the technique that it produced was deployed for
   reasons other than bandwidth management.  RSVP is widely deployed as
   an MPLS signaling mechanism.  BGP reuses the RSVP concept of Filter
   Specs to distribute firewall filters, although they are called "Flow
   Spec Component Types" in BGP [RFC5575].

6.3.  Quick-Start TCP

   The suggested references for Quick-Start TCP are:

   *  "Quick-Start for TCP and IP" [RFC4782]

   *  "Determining an appropriate sending rate over an underutilized
      network path" [SAF07]

   *  "Fast Startup Internet Congestion Control for Broadband
      Interactive Applications" [Sch11]

   *  "Using Quick-Start to enhance TCP-friendly rate control
      performance in bidirectional satellite networks" [QS-SAT]

   Quick-Start is defined in an Experimental RFC [RFC4782] and is a TCP
   extension that leverages support from the routers on the path to
   determine an allowed initial sending rate for a path through the
   Internet, either at the start of data transfers or after idle
   periods.  Without information about the path, a sender cannot easily
   determine an appropriate initial sending rate.  The default TCP
   congestion control therefore uses the safe but time-consuming slow-
   start algorithm [RFC5681].  With Quick-Start, connections are allowed
   to use higher initial sending rates if there is significant unused
   bandwidth along the path and if the sender and all of the routers
   along the path approve the request.

   By examining the Time To Live (TTL) field in Quick-Start packets, a
   sender can determine if routers on the path have approved the Quick-
   Start request.  However, this method is unable to take into account
   the routers hidden by tunnels or other network nodes invisible at the
   IP layer.

   The protocol also includes a nonce that provides protection against
   cheating routers and receivers.  If the Quick-Start request is
   explicitly approved by all routers along the path, the TCP host can
   send at up to the approved rate; otherwise, TCP would use the default
   congestion control.  Quick-Start requires modifications in the
   involved end systems as well as in routers.  Due to the resulting
   deployment challenges, Quick-Start was only proposed in [RFC4782] for
   controlled environments.

   The Quick-Start mechanism is a lightweight, coarse-grained, in-band,
   network-assisted fast startup mechanism.  The benefits are studied by
   simulation in a research paper [SAF07] that complements the protocol
   specification.  The study confirms that Quick-Start can significantly
   speed up mid-sized data transfers.  That paper also presents router
   algorithms that do not require keeping per-flow state.  Later studies
   [Sch11] comprehensively analyze Quick-Start with a full Linux
   implementation and with a router fast-path prototype using a network
   processor.  In both cases, Quick-Start could be implemented with
   limited additional complexity.

6.3.1.  Reasons for Non-deployment

   However, experiments with Quick-Start in [Sch11] revealed several

   *  Having information from the routers along the path can reduce the
      risk of congestion but cannot avoid it entirely.  Determining
      whether there is unused capacity is not trivial in actual router
      and host implementations.  Data about available capacity visible
      at the IP layer may be imprecise, and due to the propagation
      delay, information can already be outdated when it reaches a
      sender.  There is a trade-off between the speedup of data
      transfers and the risk of congestion even with Quick-Start.  This
      could be mitigated by only allowing Quick-Start to access a
      proportion of the unused capacity along a path.

   *  For scalable router fast-path implementations, it is important to
      enable parallel processing of packets, as this is a widely used
      method, e.g., in network processors.  One challenge is
      synchronization of information between packets that are processed
      in parallel, which should be avoided as much as possible.

   *  Only some types of application traffic can benefit from Quick-
      Start.  Capacity needs to be requested and discovered.  The
      discovered capacity needs to be utilized by the flow, or it
      implicitly becomes available for other flows.  Failing to use the
      requested capacity may have already reduced the pool of Quick-
      Start capacity that was made available to other competing Quick-
      Start requests.  The benefit is greatest when senders use this
      only for bulk flows and avoid sending unnecessary Quick-Start
      requests, e.g., for flows that only send a small amount of data.
      Choosing an appropriate request size requires application-internal
      knowledge that is not commonly expressed by the transport API.
      How a sender can determine the rate for an initial Quick-Start
      request is still a largely unsolved problem.

   There is no known deployment of Quick-Start for TCP or other IETF

6.3.2.  Lessons Learned

   Some lessons can be learned from Quick-Start.  Despite being a very
   lightweight protocol, Quick-Start suffers from poor incremental
   deployment properties regarding both a) the required modifications in
   network infrastructure and b) its interactions with applications.
   Except for corner cases, congestion control can be quite efficiently
   performed end to end in the Internet, and in modern stacks there is
   not much room for significant improvement by additional network

   After publication of the Quick-Start specification, there have been
   large-scale experiments with an initial window of up to 10 segments
   [RFC6928].  This alternative "IW10" approach can also ramp up data
   transfers faster than the standard congestion control, but it only
   requires sender-side modifications.  As a result, this approach can
   be easier and incrementally deployed in the Internet.  While
   theoretically Quick-Start can outperform "IW10", the improvement in
   completion time for data transfer times can, in many cases, be small.
   After publication of [RFC6928], most modern TCP stacks have increased
   their default initial window.

6.4.  ICMP Source Quench

   The suggested reference for ICMP Source Quench is:

   *  "Internet Control Message Protocol" [RFC0792]

   The ICMP Source Quench message [RFC0792] allowed an on-path router to
   request the source of a flow to reduce its sending rate.  This method
   allowed a router to provide an early indication of impending
   congestion on a path to the sources that contribute to that

6.4.1.  Reasons for Non-deployment

   This method was deployed in Internet routers over a period of time;
   the reaction of endpoints to receiving this signal has varied.  For
   low-speed links, with low multiplexing of flows the method could be
   used to regulate (momentarily reduce) the transmission rate.
   However, the simple signal does not scale with link speed or with the
   number of flows sharing a link.

   The approach was overtaken by the evolution of congestion control
   methods in TCP [RFC2001], and later also by other IETF transports.
   Because these methods were based upon measurement of the end-to-end
   path and an algorithm in the endpoint, they were able to evolve and
   mature more rapidly than methods relying on interactions between
   operational routers and endpoint stacks.

   After ICMP Source Quench was specified, the IETF began to recommend
   that transports provide end-to-end congestion control [RFC2001].  The
   Source Quench method has been obsoleted by the IETF [RFC6633], and
   both hosts and routers must now silently discard this message.

6.4.2.  Lessons Learned

   This method had several problems.

   First, [RFC0792] did not sufficiently specify how the sender would
   react to the ICMP Source Quench signal from the path (e.g.,
   [RFC1016]).  There was ambiguity in how the sender should utilize
   this additional information.  This could lead to unfairness in the
   way that receivers (or routers) responded to this message.

   Second, while the message did provide additional information, the
   Explicit Congestion Notification (ECN) mechanism [RFC3168] provided a
   more robust and informative signal for network nodes to provide early
   indication that a path has become congested.

   The mechanism originated at a time when the Internet trust model was
   very different.  Most endpoint implementations did not attempt to
   verify that the message originated from an on-path node before they
   utilized the message.  This made it vulnerable to Denial-of-Service
   (DoS) attacks.  In theory, routers might have chosen to use the
   quoted packet contained in the ICMP payload to validate that the
   message originated from an on-path node, but this would have
   increased per-packet processing overhead for each router along the
   path and would have required transport functionality in the router to
   verify whether the quoted packet header corresponded to a packet the
   router had sent.  In addition, Section 5.2 of [RFC4443] noted
   ICMPv6-based attacks on hosts that would also have threatened routers
   processing ICMPv6 Source Quench payloads.  As time passed, it became
   increasingly obvious that the lack of validation of the messages
   exposed receivers to a security vulnerability where the messages
   could be forged to create a tangible DoS opportunity.

6.5.  Triggers for Transport (TRIGTRAN)

   The suggested references for TRIGTRAN are:



   TCP [RFC0793] has a well-known weakness -- the end-to-end flow
   control mechanism has only a single signal, the loss of a segment,
   detected when no acknowledgment for the lost segment is received at
   the sender.  There are multiple reasons why the sender might not have
   received an acknowledgment for the segment.  To name several, the
   segment could have been trapped in a routing loop, damaged in
   transmission and failed checksum verification at the receiver, or
   lost because some intermediate device discarded the packet, or any of
   a variety of other things could have happened to the acknowledgment
   on the way back from the receiver to the sender.  TCP implementations
   since the late 1980s have made the "safe" decision and have
   interpreted the loss of a segment as evidence that the path between
   two endpoints may have become congested enough to exhaust buffers on
   intermediate hops, so that the TCP sender should "back off" -- reduce
   its sending rate until it knows that its segments are now being
   delivered without loss [RFC5681].

   The thinking behind TRIGTRAN was that if a path completely stopped
   working because a link along the path was "down", somehow something
   along the path could signal TCP when that link returned to service,
   and the sending TCP could retry immediately, without waiting for a
   full retransmission timeout (RTO) period.

6.5.1.  Reasons for Non-deployment

   The early dreams for TRIGTRAN were dashed because of an assumption
   that TRIGTRAN triggers would be unauthenticated.  This meant that any
   "safe" TRIGTRAN mechanism would have relied on a mechanism such as
   setting the IPv4 TTL or IPv6 Hop Count to 255 at a sender and testing
   that it was 254 upon receipt, so that a receiver could verify that a
   signal was generated by an adjacent sender known to be on the path
   being used and not some unknown sender that might not even be on the
   path (e.g., "The Generalized TTL Security Mechanism (GTSM)"
   [RFC5082]).  This situation is very similar to the case for ICMP
   Source Quench messages as described in Section 6.4, which were also
   unauthenticated and could be sent by an off-path attacker, resulting
   in deprecation of ICMP Source Quench message processing [RFC6633].

   TRIGTRAN's scope shrunk from "the path is down" to "the first-hop
   link is down."

   But things got worse.

   Because TRIGTRAN triggers would only be provided when the first-hop
   link was "down", TRIGTRAN triggers couldn't replace normal TCP
   retransmission behavior if the path failed because some link further
   along the network path was "down".  So TRIGTRAN triggers added
   complexity to an already-complex TCP state machine and did not allow
   any existing complexity to be removed.

   There was also an issue that the TRIGTRAN signal was not sent in
   response to a specific host that had been sending packets and was
   instead a signal that stimulated a response by any sender on the
   link.  This needs to scale when there are multiple flows trying to
   use the same resource, yet the sender of a trigger has no
   understanding of how many of the potential traffic sources will
   respond by sending packets -- if recipients of the signal "back off"
   their responses to a trigger to improve scaling, then that
   immediately mitigates the benefit of the signal.

   Finally, intermediate forwarding nodes required modification to
   provide TRIGTRAN triggers, but operators couldn't charge for TRIGTRAN
   triggers, so there was no way to recover the cost of modifying,
   testing, and deploying updated intermediate nodes.

   Two TRIGTRAN BOFs were held, at IETF 55 [TRIGTRAN-55] and IETF 56
   [TRIGTRAN-56], but this work was not chartered, and there was no
   interest in deploying TRIGTRAN unless it was chartered and
   standardized in the IETF.

6.5.2.  Lessons Learned

   The reasons why this work was not chartered, much less deployed,
   provide several useful lessons for researchers.

   *  TRIGTRAN started with a plausible value proposition, but
      networking realities in the early 2000s forced reductions in scope
      that led directly to reductions in potential benefits but no
      corresponding reductions in costs and complexity.

   *  These reductions in scope were the direct result of an inability
      for hosts to trust or authenticate TRIGTRAN signals they received
      from the network.

   *  Operators did not believe they could charge for TRIGTRAN
      signaling, because first-hop links didn't fail frequently and
      TRIGTRAN provided no reduction in operating expenses, so there was
      little incentive to purchase and deploy TRIGTRAN-capable network

   It is also worth noting that the targeted environment for TRIGTRAN in
   the late 1990s contained links with a relatively small number of
   directly connected hosts -- for instance, cellular or satellite
   links.  The transport community was well aware of the dangers of
   sender synchronization based on multiple senders receiving the same
   stimulus at the same time, but the working assumption for TRIGTRAN
   was that there wouldn't be enough senders for this to be a meaningful
   problem.  In the 2010s, it was common for a single "link" to support
   many senders and receivers, likely requiring TRIGTRAN senders to wait
   some random amount of time before sending after receiving a TRIGTRAN
   signal, which would have reduced the benefits of TRIGTRAN even more.

6.6.  Shim6

   The suggested reference for Shim6 is:

   *  "Shim6: Level 3 Multihoming Shim Protocol for IPv6" [RFC5533]

   The IPv6 routing architecture [RFC1887] assumed that most sites on
   the Internet would be identified by Provider Assigned IPv6 prefixes,
   so that Default-Free Zone routers only contained routes to other
   providers, resulting in a very small IPv6 global routing table.

   For a single-homed site, this could work well.  A multihomed site
   with only one upstream provider could also work well, although BGP
   multihoming from a single upstream provider was often a premium
   service (costing more than twice as much as two single-homed sites),
   and if the single upstream provider went out of service, all of the
   multihomed paths could fail simultaneously.

   IPv4 sites often multihomed by obtaining Provider Independent
   prefixes and advertising these prefixes through multiple upstream
   providers.  With the assumption that any multihomed IPv4 site would
   also multihome in IPv6, it seemed likely that IPv6 routing would be
   subject to the same pressures to announce Provider Independent
   prefixes, resulting in an IPv6 global routing table that exhibited
   the same explosive growth as the IPv4 global routing table.  During
   the early 2000s, work began on a protocol that would provide
   multihoming for IPv6 sites without requiring sites to advertise
   Provider Independent prefixes into the IPv6 global routing table.

   This protocol, called "Shim6", allowed two endpoints to exchange
   multiple addresses ("Locators") that all mapped to the same endpoint
   ("Identity").  After an endpoint learned multiple Locators for the
   other endpoint, it could send to any of those Locators with the
   expectation that those packets would all be delivered to the endpoint
   with the same Identity.  Shim6 was an example of an "Identity/Locator
   Split" protocol.

   Shim6, as defined in [RFC5533] and related RFCs, provided a workable
   solution for IPv6 multihoming using Provider Assigned prefixes,
   including capability discovery and negotiation, and allowing end-to-
   end application communication to continue even in the face of path
   failure, because applications don't see Locator failures and continue
   to communicate with the same Identity using a different Locator.

6.6.1.  Reasons for Non-deployment

   Note that the problem being addressed was "site multihoming", but
   Shim6 was providing "host multihoming".  That meant that the decision
   about what path would be used was under host control, not under edge
   router control.

   Although more work could have been done to provide a better technical
   solution, the biggest impediments to Shim6 deployment were
   operational and business considerations.  These impediments were
   discussed at multiple network operator group meetings, including
   [Shim6-35] at [NANOG-35].

   The technical issues centered around concerns that Shim6 relied on
   the host to track all the connections, while also tracking Identity/
   Locator mappings in the kernel and tracking failures to recognize
   that an available path has failed.

   The operational issues centered around concerns that operators were
   performing traffic engineering on traffic aggregates.  With Shim6,
   these operator traffic engineering policies must be pushed down to
   individual hosts.

   In addition, operators would have no visibility or control over the
   decision of hosts choosing to switch to another path.  They expressed
   concerns that relying on hosts to steer traffic exposed operator
   networks to oscillation based on feedback loops, if hosts moved from
   path to path frequently.  Given that Shim6 was intended to support
   multihoming across operators, operators providing only one of the
   paths would have even less visibility as traffic suddenly appeared
   and disappeared on their networks.

   In addition, firewalls that expected to find a TCP or UDP transport-
   level protocol header in the IP payload would see a Shim6 Identity
   header instead, and they would not perform transport-protocol-based
   firewalling functions because the firewall's normal processing logic
   would not look past the Identity header.  The firewall would perform
   its default action, which would most likely be to drop packets that
   don't match any processing rule.

   The business issues centered on reducing or removing the ability to
   sell BGP multihoming service to their own customers, which is often
   more expensive than two single-homed connectivity services.

6.6.2.  Lessons Learned

   It is extremely important to take operational concerns into account
   when a Path Aware protocol is making decisions about path selection
   that may conflict with existing operational practices and business

6.6.3.  Addendum on Multipath TCP

   During discussions in the PANRG session at IETF 103 [PANRG-103-Min],
   Lars Eggert, past Transport Area Director, pointed out that during
   charter discussions for the Multipath TCP Working Group [MP-TCP],
   operators expressed concerns that customers could use Multipath TCP
   to load-share TCP connections across operators simultaneously and
   compare passive performance measurements across network paths in real
   time, changing the balance of power in those business relationships.
   Although the Multipath TCP Working Group was chartered, this concern
   could have acted as an obstacle to deployment.

   Operator objections to Shim6 were focused on technical concerns, but
   this concern could have also been an obstacle to Shim6 deployment if
   the technical concerns had been overcome.

6.7.  Next Steps in Signaling (NSIS)

   The suggested references for Next Steps in Signaling (NSIS) are:

   *  the concluded working group charter [NSIS-CHARTER-2001]

   *  "GIST: General Internet Signalling Transport" [RFC5971]

   *  "NAT/Firewall NSIS Signaling Layer Protocol (NSLP)" [RFC5973]

   *  "NSIS Signaling Layer Protocol (NSLP) for Quality-of-Service
      Signaling" [RFC5974]

   *  "Authorization for NSIS Signaling Layer Protocols" [RFC5981]

   The NSIS Working Group worked on signaling techniques for network-
   layer resources (e.g., QoS resource reservations, Firewall and NAT

   When RSVP [RFC2205] was used in deployments, a number of questions
   came up about its perceived limitations and potential missing
   features.  The issues noted in the NSIS Working Group charter
   [NSIS-CHARTER-2001] include interworking between domains with
   different QoS architectures, mobility and roaming for IP interfaces,
   and complexity.  Later, the lack of security in RSVP was also
   recognized [RFC4094].

   The NSIS Working Group was chartered to tackle those issues and
   initially focused on QoS signaling as its primary use case.  However,
   over time a new approach evolved that introduced a modular
   architecture using two application-specific signaling protocols: a)
   the NSIS Signaling Layer Protocol (NSLP) on top of b) a generic
   signaling transport protocol (the NSIS Transport Layer Protocol

   NTLP is defined in [RFC5971].  Two types of NSLPs are defined: an
   NSLP for QoS signaling [RFC5974] and an NSLP for NATs/firewalls

6.7.1.  Reasons for Non-deployment

   The obstacles for deployment can be grouped into implementation-
   related aspects and operational aspects.

   *  Implementation-related aspects:

      Although NSIS provides benefits with respect to flexibility,
      mobility, and security compared to other network signaling
      techniques, hardware vendors were reluctant to deploy this
      solution, because it would require additional implementation
      effort and would result in additional complexity for router

      NTLP mainly operates as a path-coupled signaling protocol, i.e.,
      its messages are processed at the control plane of each
      intermediate node that is also forwarding the data flows.  This
      requires a mechanism to intercept signaling packets while they are
      forwarded in the same manner (especially along the same path) as
      data packets.  NSIS uses the IPv4 and IPv6 Router Alert Option
      (RAO) to allow for interception of those path-coupled signaling
      messages, and this technique requires router implementations to
      correctly understand and implement the handling of RAOs, e.g., to
      only process packets with RAOs of interest and to leave packets
      with irrelevant RAOs in the fast forwarding processing path (a
      comprehensive discussion of these issues can be found in
      [RFC6398]).  The latter was an issue with some router
      implementations at the time of standardization.

      Another reason is that path-coupled signaling protocols that
      interact with routers and request manipulation of state at these
      routers (or any other network element in general) are under
      scrutiny: a packet (or sequence of packets) out of the mainly
      untrusted data path is requesting creation and manipulation of
      network state.  This is seen as potentially dangerous (e.g., opens
      up a DoS threat to a router's control plane) and difficult for an
      operator to control.  Path-coupled signaling approaches were
      considered problematic (see also Section 3 of [RFC6398]).  There
      are recommendations on how to secure NSIS nodes and deployments
      (e.g., [RFC5981]).

   *  Operational Aspects:

      NSIS not only required trust between customers and their provider,
      but also among different providers.  In particular, QoS signaling
      techniques would require some kind of dynamic SLA support that
      would imply (potentially quite complex) bilateral negotiations
      between different Internet Service Providers.  This complexity was
      not considered to be justified, and increasing the bandwidth (and
      thus avoiding bottlenecks) was cheaper than actively managing
      network resource bottlenecks by using path-coupled QoS signaling
      techniques.  Furthermore, an end-to-end path typically involves
      several provider domains, and these providers need to closely
      cooperate in cases of failures.

6.7.2.  Lessons Learned

   One goal of NSIS was to decrease the complexity of the signaling
   protocol, but a path-coupled signaling protocol comes with the
   intrinsic complexity of IP-based networks, beyond the complexity of
   the signaling protocol itself.  Sources of intrinsic complexity

   *  the presence of asymmetric routes between endpoints and routers.

   *  the lack of security and trust at large in the Internet

   *  the presence of different trust boundaries.

   *  the effects of best-effort networks (e.g., robustness to packet

   *  divergence from the fate-sharing principle (e.g., state within the

   Any path-coupled signaling protocol has to deal with these realities.

   Operators view the use of IPv4 and IPv6 Router Alert Options (RAOs)
   to signal routers along the path from end systems with suspicion,
   because these end systems are usually not authenticated and heavy use
   of RAOs can easily increase the CPU load on routers that are designed
   to process most packets using a hardware "fast path" and diverting
   packets containing RAOs to a slower, more capable processor.

6.8.  IPv6 Flow Labels

   The suggested reference for IPv6 Flow Labels is:

   *  "IPv6 Flow Label Specification" [RFC6437]

   IPv6 specifies a 20-bit Flow Label field [RFC6437], included in the
   fixed part of the IPv6 header and hence present in every IPv6 packet.
   An endpoint sets the value in this field to one of a set of
   pseudorandomly assigned values.  If a packet is not part of any flow,
   the flow label value is set to zero [RFC3697].  A number of Standards
   Track and Best Current Practice RFCs (e.g., [RFC8085], [RFC6437],
   [RFC6438]) encourage IPv6 endpoints to set a non-zero value in this
   field.  A multiplexing transport could choose to use multiple flow
   labels to allow the network to either independently forward its
   subflows or use one common value for the traffic aggregate.  The flow
   label is present in all fragments.  IPsec was originally put forward
   as one important use case for this mechanism and does encrypt the
   field [RFC6438].

   Once set, the flow label can provide information that can help inform
   network nodes about subflows present at the transport layer, without
   needing to interpret the setting of upper-layer protocol fields
   [RFC6294].  This information can also be used to coordinate how
   aggregates of transport subflows are grouped when queued in the
   network and to select appropriate per-flow forwarding when choosing
   between alternate paths [RFC6438] (e.g., for Equal-Cost Multipath
   (ECMP) routing and Link Aggregation Groups (LAGs)).

6.8.1.  Reasons for Non-deployment

   Despite the field being present in every IPv6 packet, the mechanism
   did not receive as much use as originally envisioned.  One reason is
   that to be useful it requires engagement by two different

   *  Endpoint Implementation:

      For network nodes along a path to utilize the flow label, there
      needs to be a non-zero value inserted in the field [RFC6437] at
      the sending endpoint.  There needs to be an incentive for an
      endpoint to set an appropriate non-zero value.  The value should
      appropriately reflect the level of aggregation the traffic expects
      to be provided by the network.  However, this requires the stack
      to know granularity at which flows should be identified (or,
      conversely, which flows should receive aggregated treatment),
      i.e., which packets carry the same flow label.  Therefore, setting
      a non-zero value may result in additional choices that need to be
      made by an application developer.

      Although the original flow label standard [RFC3697] forbids any
      encoding of meaning into the flow label value, the opportunity to
      use the flow label as a covert channel or to signal other meta-
      information may have raised concerns about setting a non-zero
      value [RFC6437].

      Before methods are widely deployed to use this method, there could
      be no incentive for an endpoint to set the field.

   *  Operational support in network nodes:

      A benefit can only be realized when a network node along the path
      also uses this information to inform its decisions.  Network
      equipment (routers and/or middleboxes) need to include appropriate
      support in order to utilize the field when making decisions about
      how to classify flows or forward packets.  The use of any optional
      feature in a network node also requires corresponding updates to
      operational procedures and therefore is normally only introduced
      when the cost can be justified.

      A benefit from utilizing the flow label is expected to be
      increased quality of experience for applications -- but this comes
      at some operational cost to an operator and requires endpoints to
      set the field.

6.8.2.  Lessons Learned

   The flow label is a general-purpose header field for use by the path.
   Multiple uses have been proposed.  One candidate use was to reduce
   the complexity of forwarding decisions.  However, modern routers can
   use a "fast path", often taking advantage of hardware to accelerate
   processing.  The method can assist in more complex forwarding, such
   as ECMP routing and load balancing.

   Although [RFC6437] recommended that endpoints should by default
   choose uniformly distributed labels for their traffic, the
   specification permitted an endpoint to choose to set a zero value.
   This ability of endpoints to choose to set a flow label of zero has
   had consequences on deployability:

   *  Before wide-scale support by endpoints, it would be impossible to
      rely on a non-zero flow label being set.  Network nodes therefore
      would need to also employ other techniques to realize equivalent
      functions.  An example of a method is one assuming semantics of
      the source port field to provide entropy input to a network-layer
      hash.  This use of a 5-tuple to classify a packet represents a
      layering violation [RFC6294].  When other methods have been
      deployed, they increase the cost of deploying standards-based
      methods, even though they may offer less control to endpoints and
      result in potential interaction with other uses/interpretation of
      the field.

   *  Even though the flow label is specified as an end-to-end field,
      some network paths have been observed to not transparently forward
      the flow label.  This could result from non-conformant equipment
      or could indicate that some operational networks have chosen to
      reuse the protocol field for other (e.g., internal) purposes.
      This results in lack of transparency, and a deployment hurdle to
      endpoints expecting that they can set a flow label that is
      utilized by the network.  The more recent practice of "greasing"
      [GREASE] would suggest that a different outcome could have been
      achieved if endpoints were always required to set a non-zero

   *  [RFC1809] noted that setting the choice of the flow label value
      can depend on the expectations of the traffic generated by an
      application, which suggests that an API should be presented to
      control the setting or policy that is used.  However, many
      currently available APIs do not have this support.

   A growth in the use of encrypted transports (e.g., QUIC [RFC9000])
   seems likely to raise issues similar to those discussed above and
   could motivate renewed interest in utilizing the flow label.

6.9.  Explicit Congestion Notification (ECN)

   The suggested references for Explicit Congestion Notification (ECN)

   *  "Recommendations on Queue Management and Congestion Avoidance in
      the Internet" [RFC2309]

   *  "A Proposal to add Explicit Congestion Notification (ECN) to IP"

   *  "The Addition of Explicit Congestion Notification (ECN) to IP"

   *  "Implementation Report on Experiences with Various TCP RFCs"
      [vista-impl], slides 6 and 7

   *  "Implementation and Deployment of ECN" (at [SallyFloyd])

   In the early 1990s, the large majority of Internet traffic used TCP
   as its transport protocol, but TCP had no way to detect path
   congestion before the path was so congested that packets were being
   dropped.  These congestion events could affect all senders using a
   path, either by "lockout", where long-lived flows monopolized the
   queues along a path, or by "full queues", where queues remain full,
   or almost full, for a long period of time.

   In response to this situation, "Active Queue Management" (AQM) was
   deployed in the network.  A number of AQM disciplines have been
   deployed, but one common approach was that routers dropped packets
   when a threshold buffer length was reached, so that transport
   protocols like TCP that were responsive to loss would detect this
   loss and reduce their sending rates.  Random Early Detection (RED)
   was one such proposal in the IETF.  As the name suggests, a router
   using RED as its AQM discipline that detected time-averaged queue
   lengths passing a threshold would choose incoming packets
   probabilistically to be dropped [RFC2309].

   Researchers suggested providing "explicit congestion notifications"
   to senders when routers along the path detected that their queues
   were building, giving senders an opportunity to "slow down" as if a
   loss had occurred, giving path queues time to drain, while the path
   still had sufficient buffer capacity to accommodate bursty arrivals
   of packets from other senders.  This was proposed as an experiment in
   [RFC2481] and standardized in [RFC3168].

   A key aspect of ECN was the use of IP header fields rather than IP
   options to carry explicit congestion notifications, since the
   proponents recognized that

      Many routers process the "regular" headers in IP packets more
      efficiently than they process the header information in IP

   Unlike most of the Path Aware technologies included in this document,
   the story of ECN continues to the present day and encountered a large
   number of Lessons Learned during that time.  The early history of ECN
   (non-)deployment provides Lessons Learned that were not captured by
   other contributions in Section 6, so that is the emphasis in this
   section of the document.

6.9.1.  Reasons for Non-deployment

   ECN deployment relied on three factors -- support in client
   implementations, support in router implementations, and deployment
   decisions in operational networks.

   The proponents of ECN did so much right, anticipating many of the
   Lessons Learned now recognized in Section 4.  They recognized the
   need to support incremental deployment (Section 4.2).  They
   considered the impact on router throughput (Section 4.8).  They even
   considered trust issues between end nodes and the network, for both
   non-compliant end nodes (Section 4.10) and non-compliant routers
   (Section 4.9).

   They were rewarded with ECN being implemented in major operating
   systems, for both end nodes and routers.  A number of implementations
   are listed under "Implementation and Deployment of ECN" at

   What they did not anticipate was routers that would crash when they
   saw bits 6 and 7 in the IPv4 Type of Service (TOS) octet [RFC0791] /
   IPv6 Traffic Class field [RFC2460], which [RFC2481] redefined to be
   "Currently Unused", being set to a non-zero value.

   As described in [vista-impl] ("IGD" stands for "Intermediate Gateway

   |  IGD problem #1: one of the most popular versions from one of the
   |  most popular vendors.  When a data packet arrives with either
   |  ECT(0) or ECT(1) (indicating successful ECN capability
   |  negotiation) indicated, router crashed.  Cannot be recovered at
   |  TCP layer [sic]

   This implementation, which would be run on a significant percentage
   of Internet end nodes, was shipped with ECN disabled, as was true for
   several of the other implementations listed under "Implementation and
   Deployment of ECN" at [SallyFloyd].  Even if subsequent router
   vendors fixed these implementations, ECN was still disabled on end
   nodes, and given the trade-off between the benefits of enabling ECN
   (somewhat better behavior during congestion) and the risks of
   enabling ECN (possibly crashing a router somewhere along the path),
   ECN tended to stay disabled on implementations that supported ECN for
   decades afterwards.

6.9.2.  Lessons Learned

   Of the contributions included in Section 6, ECN may be unique in
   providing these lessons:

   *  Even if you do everything right, you may trip over implementation
      bugs in devices you know nothing about, that will cause severe
      problems that prevent successful deployment of your Path Aware

   *  After implementations disable your Path Aware technology, it may
      take years, or even decades, to convince implementers to re-enable
      it by default.

   These two lessons, taken together, could be summarized as "you get
   one chance to get it right."

   During discussion of ECN at [PANRG-110], we noted that "you get one
   chance to get it right" isn't quite correct today, because operating
   systems on so many host systems are frequently updated, and transport
   protocols like QUIC [RFC9000] are being implemented in user space and
   can be updated without touching installed operating systems.  Neither
   of these factors were true in the early 2000s.

   We think that these restatements of the ECN Lessons Learned are more
   useful for current implementers:

   *  Even if you do everything right, you may trip over implementation
      bugs in devices you know nothing about, that will cause severe
      problems that prevent successful deployment of your Path Aware
      technology.  Testing before deployment isn't enough to ensure
      successful deployment.  It is also necessary to "deploy gently",
      which often means deploying for a small subset of users to gain
      experience and implementing feedback mechanisms to detect that
      user experience is being degraded.

   *  After implementations disable your Path Aware technology, it may
      take years, or even decades, to convince implementers to re-enable
      it by default.  This might be based on the difficulty of
      distributing implementations that enable it by default, but it is
      just as likely to be based on the "bad taste in the mouth" that
      implementers have after an unsuccessful deployment attempt that
      degraded user experience.

   With these expansions, the two lessons, taken together, could be more
   helpfully summarized as "plan for failure" -- anticipate what your
   next step will be, if initial deployment is unsuccessful.

   ECN deployment was also hindered by non-deployment of AQM in many
   devices, because of operator interest in QoS features provided in the
   network, rather than using the network to assist end systems in
   providing for themselves.  But that's another story, and the AQM
   Lessons Learned are already covered in other contributions in
   Section 6.

7.  Security Considerations

   This document describes Path Aware techniques that were not adopted
   and widely deployed on the Internet, so it doesn't affect the
   security of the Internet.

   If this document meets its goals, we may develop new techniques for
   Path Aware networking that would affect the security of the Internet,
   but security considerations for those techniques will be described in
   the corresponding RFCs that specify them.

8.  IANA Considerations

   This document has no IANA actions.

9.  Informative References

              Wikipedia, "Colossal Cave Adventure", June 2021,

   [Conviva]  "Conviva Precision : Data Sheet", January 2021,

              Farrell, S., "We're gonna need a bigger threat model",
              Work in Progress, Internet-Draft, draft-farrell-etm-03, 6
              July 2019, <https://datatracker.ietf.org/doc/html/draft-

   [GREASE]   Thomson, M., "Long-term Viability of Protocol Extension
              Mechanisms", Work in Progress, Internet-Draft, draft-iab-
              use-it-or-lose-it-00, 7 August 2019,

   [IEN-119]  Forgie, J., "ST - A Proposed Internet Stream Protocol",
              September 1979,

              Arkko, J., "Changes in the Internet Threat Model", Work in
              Progress, Internet-Draft, draft-arkko-arch-internet-
              threat-model-01, 8 July 2019,

              Polk, J. and S. Dhesikan, "Integrated Services (IntServ)
              Extension to Allow Signaling of Multiple Traffic
              Specifications and Multiple Flow Specifications in
              RSVPv1", Work in Progress, Internet-Draft, draft-ietf-
              tsvwg-intserv-multiple-tspec-02, 25 February 2013,

   [ITAT]     "IAB Workshop on Internet Technology Adoption and
              Transition (ITAT) 2013", December 2013,

   [model-t]  "Model-t -- Discussions of changes in Internet deployment
              patterns and their impact on the Internet threat model",
              model-t mailing list,

              "Media Operations Working Group - IETF 109 Minutes",
              November 2020,

   [MP-TCP]   "Multipath TCP Working Group Home Page",

   [NANOG-35] "NANOG 35 Agenda", North American Network Operators' Group
              (NANOG), October 2005,

              "Next Steps In Signaling Working Group Charter", March

   [PANRG]    "Path Aware Networking Research Group Home Page",

              "Path Aware Networking Research Group - IETF 103 Minutes",
              November 2018,

              "Path Aware Networking Research Group - IETF 105 Minutes",
              July 2019,

              "Path Aware Networking Research Group - IETF 106 Minutes",
              November 2019,

              "Path Aware Networking Research Group - IETF 110", March

   [PANRG-99] "Path Aware Networking Research Group - IETF 99", July

              Enghardt, T. and C. Krähenbühl, "A Vocabulary of Path
              Properties", Work in Progress, Internet-Draft, draft-irtf-
              panrg-path-properties-02, 22 February 2021,

              Trammell, B., "Current Open Questions in Path Aware
              Networking", Work in Progress, Internet-Draft, draft-irtf-
              panrg-questions-09, 16 April 2021,

              Bonaventure, O., "A Decade of Path Awareness", July 2017,

   [QS-SAT]   Secchi, R., Sathiaseelan, A., Potortì, F., Gotta, A., and
              G. Fairhurst, "Using Quick-Start to enhance TCP-friendly
              rate control performance in bidirectional satellite
              networks", DOI 10.1002/sat.929, May 2009,

   [RFC0791]  Postel, J., "Internet Protocol", STD 5, RFC 791,
              DOI 10.17487/RFC0791, September 1981,

   [RFC0792]  Postel, J., "Internet Control Message Protocol", STD 5,
              RFC 792, DOI 10.17487/RFC0792, September 1981,

   [RFC0793]  Postel, J., "Transmission Control Protocol", STD 7,
              RFC 793, DOI 10.17487/RFC0793, September 1981,

   [RFC1016]  Prue, W. and J. Postel, "Something a Host Could Do with
              Source Quench: The Source Quench Introduced Delay
              (SQuID)", RFC 1016, DOI 10.17487/RFC1016, July 1987,

   [RFC1122]  Braden, R., Ed., "Requirements for Internet Hosts -
              Communication Layers", STD 3, RFC 1122,
              DOI 10.17487/RFC1122, October 1989,

   [RFC1190]  Topolcic, C., "Experimental Internet Stream Protocol:
              Version 2 (ST-II)", RFC 1190, DOI 10.17487/RFC1190,
              October 1990, <https://www.rfc-editor.org/info/rfc1190>.

   [RFC1633]  Braden, R., Clark, D., and S. Shenker, "Integrated
              Services in the Internet Architecture: an Overview",
              RFC 1633, DOI 10.17487/RFC1633, June 1994,

   [RFC1809]  Partridge, C., "Using the Flow Label Field in IPv6",
              RFC 1809, DOI 10.17487/RFC1809, June 1995,

   [RFC1819]  Delgrossi, L., Ed. and L. Berger, Ed., "Internet Stream
              Protocol Version 2 (ST2) Protocol Specification - Version
              ST2+", RFC 1819, DOI 10.17487/RFC1819, August 1995,

   [RFC1887]  Rekhter, Y., Ed. and T. Li, Ed., "An Architecture for IPv6
              Unicast Address Allocation", RFC 1887,
              DOI 10.17487/RFC1887, December 1995,

   [RFC2001]  Stevens, W., "TCP Slow Start, Congestion Avoidance, Fast
              Retransmit, and Fast Recovery Algorithms", RFC 2001,
              DOI 10.17487/RFC2001, January 1997,

   [RFC2205]  Braden, R., Ed., Zhang, L., Berson, S., Herzog, S., and S.
              Jamin, "Resource ReSerVation Protocol (RSVP) -- Version 1
              Functional Specification", RFC 2205, DOI 10.17487/RFC2205,
              September 1997, <https://www.rfc-editor.org/info/rfc2205>.

   [RFC2210]  Wroclawski, J., "The Use of RSVP with IETF Integrated
              Services", RFC 2210, DOI 10.17487/RFC2210, September 1997,

   [RFC2211]  Wroclawski, J., "Specification of the Controlled-Load
              Network Element Service", RFC 2211, DOI 10.17487/RFC2211,
              September 1997, <https://www.rfc-editor.org/info/rfc2211>.

   [RFC2212]  Shenker, S., Partridge, C., and R. Guerin, "Specification
              of Guaranteed Quality of Service", RFC 2212,
              DOI 10.17487/RFC2212, September 1997,

   [RFC2215]  Shenker, S. and J. Wroclawski, "General Characterization
              Parameters for Integrated Service Network Elements",
              RFC 2215, DOI 10.17487/RFC2215, September 1997,

   [RFC2309]  Braden, B., Clark, D., Crowcroft, J., Davie, B., Deering,
              S., Estrin, D., Floyd, S., Jacobson, V., Minshall, G.,
              Partridge, C., Peterson, L., Ramakrishnan, K., Shenker,
              S., Wroclawski, J., and L. Zhang, "Recommendations on
              Queue Management and Congestion Avoidance in the
              Internet", RFC 2309, DOI 10.17487/RFC2309, April 1998,

   [RFC2460]  Deering, S. and R. Hinden, "Internet Protocol, Version 6
              (IPv6) Specification", RFC 2460, DOI 10.17487/RFC2460,
              December 1998, <https://www.rfc-editor.org/info/rfc2460>.

   [RFC2475]  Blake, S., Black, D., Carlson, M., Davies, E., Wang, Z.,
              and W. Weiss, "An Architecture for Differentiated
              Services", RFC 2475, DOI 10.17487/RFC2475, December 1998,

   [RFC2481]  Ramakrishnan, K. and S. Floyd, "A Proposal to add Explicit
              Congestion Notification (ECN) to IP", RFC 2481,
              DOI 10.17487/RFC2481, January 1999,

   [RFC3168]  Ramakrishnan, K., Floyd, S., and D. Black, "The Addition
              of Explicit Congestion Notification (ECN) to IP",
              RFC 3168, DOI 10.17487/RFC3168, September 2001,

   [RFC3697]  Rajahalme, J., Conta, A., Carpenter, B., and S. Deering,
              "IPv6 Flow Label Specification", RFC 3697,
              DOI 10.17487/RFC3697, March 2004,

   [RFC4094]  Manner, J. and X. Fu, "Analysis of Existing Quality-of-
              Service Signaling Protocols", RFC 4094,
              DOI 10.17487/RFC4094, May 2005,

   [RFC4443]  Conta, A., Deering, S., and M. Gupta, Ed., "Internet
              Control Message Protocol (ICMPv6) for the Internet
              Protocol Version 6 (IPv6) Specification", STD 89,
              RFC 4443, DOI 10.17487/RFC4443, March 2006,

   [RFC4782]  Floyd, S., Allman, M., Jain, A., and P. Sarolahti, "Quick-
              Start for TCP and IP", RFC 4782, DOI 10.17487/RFC4782,
              January 2007, <https://www.rfc-editor.org/info/rfc4782>.

   [RFC5082]  Gill, V., Heasley, J., Meyer, D., Savola, P., Ed., and C.
              Pignataro, "The Generalized TTL Security Mechanism
              (GTSM)", RFC 5082, DOI 10.17487/RFC5082, October 2007,

   [RFC5218]  Thaler, D. and B. Aboba, "What Makes for a Successful
              Protocol?", RFC 5218, DOI 10.17487/RFC5218, July 2008,

   [RFC5533]  Nordmark, E. and M. Bagnulo, "Shim6: Level 3 Multihoming
              Shim Protocol for IPv6", RFC 5533, DOI 10.17487/RFC5533,
              June 2009, <https://www.rfc-editor.org/info/rfc5533>.

   [RFC5575]  Marques, P., Sheth, N., Raszuk, R., Greene, B., Mauch, J.,
              and D. McPherson, "Dissemination of Flow Specification
              Rules", RFC 5575, DOI 10.17487/RFC5575, August 2009,

   [RFC5681]  Allman, M., Paxson, V., and E. Blanton, "TCP Congestion
              Control", RFC 5681, DOI 10.17487/RFC5681, September 2009,

   [RFC5971]  Schulzrinne, H. and R. Hancock, "GIST: General Internet
              Signalling Transport", RFC 5971, DOI 10.17487/RFC5971,
              October 2010, <https://www.rfc-editor.org/info/rfc5971>.

   [RFC5973]  Stiemerling, M., Tschofenig, H., Aoun, C., and E. Davies,
              "NAT/Firewall NSIS Signaling Layer Protocol (NSLP)",
              RFC 5973, DOI 10.17487/RFC5973, October 2010,

   [RFC5974]  Manner, J., Karagiannis, G., and A. McDonald, "NSIS
              Signaling Layer Protocol (NSLP) for Quality-of-Service
              Signaling", RFC 5974, DOI 10.17487/RFC5974, October 2010,

   [RFC5981]  Manner, J., Stiemerling, M., Tschofenig, H., and R. Bless,
              Ed., "Authorization for NSIS Signaling Layer Protocols",
              RFC 5981, DOI 10.17487/RFC5981, February 2011,

   [RFC6294]  Hu, Q. and B. Carpenter, "Survey of Proposed Use Cases for
              the IPv6 Flow Label", RFC 6294, DOI 10.17487/RFC6294, June
              2011, <https://www.rfc-editor.org/info/rfc6294>.

   [RFC6398]  Le Faucheur, F., Ed., "IP Router Alert Considerations and
              Usage", BCP 168, RFC 6398, DOI 10.17487/RFC6398, October
              2011, <https://www.rfc-editor.org/info/rfc6398>.

   [RFC6437]  Amante, S., Carpenter, B., Jiang, S., and J. Rajahalme,
              "IPv6 Flow Label Specification", RFC 6437,
              DOI 10.17487/RFC6437, November 2011,

   [RFC6438]  Carpenter, B. and S. Amante, "Using the IPv6 Flow Label
              for Equal Cost Multipath Routing and Link Aggregation in
              Tunnels", RFC 6438, DOI 10.17487/RFC6438, November 2011,

   [RFC6633]  Gont, F., "Deprecation of ICMP Source Quench Messages",
              RFC 6633, DOI 10.17487/RFC6633, May 2012,

   [RFC6928]  Chu, J., Dukkipati, N., Cheng, Y., and M. Mathis,
              "Increasing TCP's Initial Window", RFC 6928,
              DOI 10.17487/RFC6928, April 2013,

   [RFC7305]  Lear, E., Ed., "Report from the IAB Workshop on Internet
              Technology Adoption and Transition (ITAT)", RFC 7305,
              DOI 10.17487/RFC7305, July 2014,

   [RFC7418]  Dawkins, S., Ed., "An IRTF Primer for IETF Participants",
              RFC 7418, DOI 10.17487/RFC7418, December 2014,

   [RFC8085]  Eggert, L., Fairhurst, G., and G. Shepherd, "UDP Usage
              Guidelines", BCP 145, RFC 8085, DOI 10.17487/RFC8085,
              March 2017, <https://www.rfc-editor.org/info/rfc8085>.

   [RFC8170]  Thaler, D., Ed., "Planning for Protocol Adoption and
              Subsequent Transitions", RFC 8170, DOI 10.17487/RFC8170,
              May 2017, <https://www.rfc-editor.org/info/rfc8170>.

   [RFC8655]  Finn, N., Thubert, P., Varga, B., and J. Farkas,
              "Deterministic Networking Architecture", RFC 8655,
              DOI 10.17487/RFC8655, October 2019,

   [RFC8793]  Wissingh, B., Wood, C., Afanasyev, A., Zhang, L., Oran,
              D., and C. Tschudin, "Information-Centric Networking
              (ICN): Content-Centric Networking (CCNx) and Named Data
              Networking (NDN) Terminology", RFC 8793,
              DOI 10.17487/RFC8793, June 2020,

   [RFC9000]  Iyengar, J., Ed. and M. Thomson, Ed., "QUIC: A UDP-Based
              Multiplexed and Secure Transport", RFC 9000,
              DOI 10.17487/RFC9000, May 2021,

              "Security Area Open Meeting - IETF 105 Minutes", July
              2019, <https://datatracker.ietf.org/meeting/105/materials/

   [SAF07]    Sarolahti, P., Allman, M., and S. Floyd, "Determining an
              appropriate sending rate over an underutilized network
              path", Computer Networks: The International Journal of
              Computer and Telecommunications Networking, Volume 51,
              Number 7, DOI 10.1016/j.comnet.2006.11.006, May 2007,

              Floyd, S., "ECN (Explicit Congestion Notification) in TCP/
              IP", June 2009, <https://www.icir.org/floyd/ecn.html>.

   [Sch11]    Scharf, M., "Fast Startup Internet Congestion Control for
              Broadband Interactive Applications", Ph.D. Thesis,
              University of Stuttgart, April 2011.

   [Shim6-35] Meyer, D., Huston, G., Schiller, J., and V. Gill, "IAB
              IPv6 Multihoming Panel at NANOG 35", North American
              Network Operators' Group (NANOG), October 2005,

              "Triggers for Transport BOF at IETF 55", November 2002,

              "Triggers for Transport BOF at IETF 56", March 2003,

              Sridharan, M., Bansal, D., and D. Thaler, "Implementation
              Report on Experiences with Various TCP RFCs", March 2007,


   Initial material for Section 6.1 on ST2 was provided by Gorry

   Initial material for Section 6.2 on IntServ was provided by Ron

   Initial material for Section 6.3 on Quick-Start TCP was provided by
   Michael Scharf, who also provided suggestions to improve this section
   after it was edited.

   Initial material for Section 6.4 on ICMP Source Quench was provided
   by Gorry Fairhurst.

   Initial material for Section 6.5 on Triggers for Transport (TRIGTRAN)
   was provided by Spencer Dawkins.

   Section 6.6 on Shim6 builds on initial material describing obstacles,
   which was provided by Erik Nordmark, with background added by Spencer

   Initial material for Section 6.7 on Next Steps in Signaling (NSIS)
   was provided by Roland Bless and Martin Stiemerling.

   Initial material for Section 6.8 on IPv6 Flow Labels was provided by
   Gorry Fairhurst.

   Initial material for Section 6.9 on Explicit Congestion Notification
   was provided by Spencer Dawkins.

   Our thanks to Adrian Farrel, Bob Briscoe, C.M. Heard, David Black,
   Eric Kinnear, Erik Auerswald, Gorry Fairhurst, Jake Holland, Joe
   Touch, Joeri de Ruiter, Kireeti Kompella, Mohamed Boucadair, Randy
   Presuhn, Roland Bless, Ruediger Geib, Theresa Enghardt, and Wes Eddy,
   who provided review comments on this document as a "work in process".

   Mallory Knodel reviewed this document for the Internet Research
   Steering Group and provided many helpful suggestions.

   David Oran also provided helpful comments and text suggestions on
   this document during Internet Research Steering Group balloting.  In
   particular, Section 5 reflects his review.

   Benjamin Kaduk, Martin Duke, and Rob Wilton provided helpful comments
   during Internet Engineering Steering Group conflict review.

   Special thanks to Adrian Farrel for helping Spencer navigate the
   twisty little passages of Flow Specs and Filter Specs in IntServ,
   RSVP, MPLS, and BGP.  They are all alike, except when they are
   different [Colossal-Cave].

Author's Address

   Spencer Dawkins (editor)
   Tencent America
   United States of America

   Email: spencerdawkins.ietf@gmail.com