| RFC 0000 | What Not To Do | May 2021 |
| Dawkins | Informational | [Page] |
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.¶
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 a candidate 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 https://www.rfc-editor.org/info/rfc0000.¶
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.¶
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 Networking techniques have seen limited or no deployment.¶
This document represents the consensus of the Path Aware Networking Research Group (PANRG).¶
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 document.¶
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 networking 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 a "black box". 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, and endpoint reaction to these properties that affects routing and/or transmission; note that this can and already does happen to some extent in the current Internet architecture. 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.¶
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 meta-lessons from that experience were¶
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).¶
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 contains.¶
This document does not catalog every proposed path aware networking 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.¶
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].¶
LB: Also adding a test for an RFC whose title ends in an exclamation point: "I'm Being Attacked by PRISONER.IANA.ORG!" [RFC6305].¶
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 Setion 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].¶
The terms "node" and "element" in this document have the meaning defined in [PANRG-PATH-PROPERTIES].¶
This document grew out of contributions by various IETF participants with experience with one or more Path Aware Networking techniques.¶
There are many things that could be said about the Path Aware networking techniques that have been developed. For the purposes of this document, contributors were requested to provide¶
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 Networking technique of those discussed 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 Research Group reviewed the Lessons Learned (Section 4) contained in the May 23, 2019 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¶
Section 6.9 on ECN was added during the review and approval process, based on a question from Martin Duke. That section, along with its Lessons Learned and place in the "Invariant"/"Variable"/"Not Now" taxonomy, as contained in the March 8, 2021 version of this document, was discussed at [PANRG-110].¶
| Lesson | Category |
|---|---|
| Justifying Deployment (Section 4.1) | Invariant |
| Providing Benefits for Early Adopters (Section 4.2) | Invariant |
| Providing Benefits during Partial Deployment (Section 4.3) | Invariant |
| Outperforming End-to-End Protocol Mechanisms (Section 4.4) | Variable |
| 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 (Section 4.10) | Not Now |
| Reacting to Distant Signals (Section 4.11) | Variable |
| Support in Endpoint Protocol Stacks (Section 4.12) | Variable |
| Planning for Failure (Section 4.13) | Invariant |
"Justifying Deployment", "Providing Benefits for Early Adopters", "Paying for Path Aware Techniques", "Impact on Operational Practice", 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 invariant.¶
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 "Per-Connection State", the key questions discussed in the research group were "how much state" and "where state is maintained".¶
For "Keeping Traffic on Fast-Paths", we noted that this was true for many platforms, but not for all.¶
For "Endpoints Trusting Intermediate Nodes" and "Intermediate Nodes Trusting Endpoints", these lessons point to the broader need to revisit the Internet Threat Model.¶
For "Reacting to Distant Signals", we noted that not all attributes are equal.¶
For "Support in Endpoint Protocol Stacks", we noted that Path Aware applications must be able to identify and communicate requirements about path characteristics.¶
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 Path Aware techniques that were not deployed, and Path Aware techniques that were deployed, but were not deployed widely or quickly. See Sections 6.6 and 6.6.3 as one example of this shifting boundary.¶
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].)¶
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].)¶
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].)¶
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.)¶
"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.)¶
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 often discourage deployment. (See Section 6.6, including Section 6.6.3.)¶
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.)¶
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.)¶
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 issuing a message can confirm that it has visibility of the packets on the path it is seeking to control [RFC8085] (e.g., an ICMP message included a quoted packet 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.)¶
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 Diffserv.)¶
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.)¶
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.)¶
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.)¶
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.¶
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 of the document, might be explained explicitly:¶
Contributions on these Path Aware networking 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.¶
The suggested references for Stream Transport are:¶
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 realtime services over the Internet.¶
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.¶
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 existing 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.¶
The suggested references for IntServ are:¶
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 Mpbs) 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) [RFC2210] to describe the traffic that they emit. RSVP reserves capacity for traffic on a per-Flow-Spec basis.¶
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:¶
As IntServ was being discussed, the following occurred:¶
The following lessons were learned:¶
In environments where IntServ has been deployed, trust relationships with endpoints are very different from trust relationships on the Internet itself, and there are often clearly defined hierarchies in Service Level Agreements (SLAs), and well-defined transport flows operating with pre-determined 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].¶
The suggested references for Quick-Start TCP are:¶
Quick-Start [RFC4782] is an Experimental 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 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 analyzes 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.¶
However, experiments with Quick-Start in [Sch11] revealed several challenges:¶
There is no known deployment of Quick-Start for TCP or other IETF transports.¶
Some lessons can be learned from Quick-Start. Despite being a very light-weight protocol, Quick-Start suffers from poor incremental deployment properties, both regarding the required modifications in network infrastructure as well as 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 support.¶
After publication of the Quick-Start specification, there have been large-scale experiments with an initial window of up to 10 MSS [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.¶
The suggested references for ICMP Source Quench are:¶
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 congestion.¶
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 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.¶
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 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, 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 denial of service opportunity.¶
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, and TCP implementations since the late 1980s 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]. More modern TCP stacks have added a growing array of strategies about how to establish the sending rate [RFC5681], but when a path is no longer operational, TCP would continue to retry transmissions, which would fail, again, and double their retransmission timeout (RTO) timers with each failed transmission, with the result that TCP would wait many seconds before retrying a segment, even if the path becomes operational while the sender is waiting for its next retry.¶
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 RTO period.¶
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 which 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 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.¶
The reasons why this work was not chartered, much less deployed, provide several useful lessons for researchers.¶
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 is common for a single "link" to support many senders and receivers on a single link, 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.¶
The suggested references for Shim6 are:¶
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 a global IPv6 routing table that exhibited the same explosive growth as the global IPv4 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.¶
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 would not perform transport-protocol-based firewalling functions because the firewall's normal processing logic would not look past the Identity header.¶
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.¶
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 considerations.¶
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 loadshare 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.¶
The suggested references for Next Steps in Signaling (NSIS) are:¶
The NSIS Working Group worked on signaling techniques for network layer resources (e.g., QoS resource reservations, Firewall and NAT traversal).¶
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 application-specific signaling protocols (the NSIS Signaling Layer Protocol (NSLP)) on top of a generic signaling transport protocol (the NSIS Transport Layer Protocol (NTLP)).¶
The NTLP is defined in [RFC5971]. Two NSLPs are defined: the NSIS Signaling Layer Protocol (NSLP) for Quality-of-Service Signaling [RFC5974] as well as the NAT/Firewall NSIS Signaling Layer Protocol (NSLP) [RFC5973].¶
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 implementations.¶
The NTLP mainly operates as path-coupled signaling protocol, i.e, its messages are processed at the intermediate node's control plane that are 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 packet 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 Denial of Service (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. Especially, QoS signaling techniques would require some kind of dynamic service level agreement 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.¶
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 include:¶
Any path-coupled signaling protocol has to deal with these realities.¶
Operators view the use of IPv4 and IPv6 Router Alert Option (RAO) 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.¶
The suggested references for IPv6 Flow Label are:¶
IPv6 specifies a 20-bit field 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 pseudo-randomly 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 independently forward its subflows, or to 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 Routing (ECMP) and Link Aggregation (LAG)).¶
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 stakeholders:¶
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 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 so they can utilize the field when making decisions about how to classify flows, or to inform forwarding choices. 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.¶
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 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:¶
A growth in the use of encrypted transports, (e.g., QUIC [QUIC-WG]) seems likely to raise similar issues to those discussed above and could motivate renewed interest in utilizing the flow label.¶
The suggested references for Explicit Congestion Notification (ECN) are:¶
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, and 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 that providing "explicit congestion notifications" to senders when routers along the path detected their queues were building, so that some senders would "slow down" as if a loss had occurred, so that the path queues had time to drain, and 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 options.¶
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.¶
There are at least three sub-stories - ECN deployment in clients, ECN deployment in routers, and AQM deployment in operational networks. All three sub-stories mattered.¶
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, both for 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, both for end nodes and for routers. A number of implementations are listed under "Implementation and Deployment of ECN" at [SallyFloyd].¶
What they did not anticipate, was routers that would crash, when they saw bits 6 and 7 in the IPv4 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 Device"),¶
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 tradeoff 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.¶
Of the contributions included in Section 6, ECN may be unique in providing these lessons:¶
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 [RFCYYY1] 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:¶
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.¶
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.¶
This document has no IANA actions.¶
Initial material for Section 6.1 on ST2 was provided by Gorry Fairhurst.¶
Initial material for Section 6.2 on IntServ was provided by Ron Bonica.¶
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 provided by Erik Nordmark, with background added by Spencer Dawkins.¶
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, 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 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].¶