rfc9330.original   rfc9330.txt 
Transport Area Working Group B. Briscoe, Ed. Internet Engineering Task Force (IETF) B. Briscoe, Ed.
Internet-Draft Independent Request for Comments: 9330 Independent
Intended status: Informational K. De Schepper Category: Informational K. De Schepper
Expires: 2 March 2023 Nokia Bell Labs ISSN: 2070-1721 Nokia Bell Labs
M. Bagnulo Braun M. Bagnulo
Universidad Carlos III de Madrid Universidad Carlos III de Madrid
G. White G. White
CableLabs CableLabs
29 August 2022 January 2023
Low Latency, Low Loss, Scalable Throughput (L4S) Internet Service: Low Latency, Low Loss, and Scalable Throughput (L4S) Internet Service:
Architecture Architecture
draft-ietf-tsvwg-l4s-arch-20
Abstract Abstract
This document describes the L4S architecture, which enables Internet This document describes the L4S architecture, which enables Internet
applications to achieve Low queuing Latency, Low Loss, and Scalable applications to achieve low queuing latency, low congestion loss, and
throughput (L4S). L4S is based on the insight that the root cause of scalable throughput control. L4S is based on the insight that the
queuing delay is in the capacity-seeking congestion controllers of root cause of queuing delay is in the capacity-seeking congestion
senders, not in the queue itself. With the L4S architecture all controllers of senders, not in the queue itself. With the L4S
Internet applications could (but do not have to) transition away from architecture, all Internet applications could (but do not have to)
congestion control algorithms that cause substantial queuing delay, transition away from congestion control algorithms that cause
to a new class of congestion controls that can seek capacity with substantial queuing delay and instead adopt a new class of congestion
very little queuing. These are aided by a modified form of explicit controls that can seek capacity with very little queuing. These are
congestion notification (ECN) from the network. With this new aided by a modified form of Explicit Congestion Notification (ECN)
architecture, applications can have both low latency and high from the network. With this new architecture, applications can have
throughput. both low latency and high throughput.
The architecture primarily concerns incremental deployment. It The architecture primarily concerns incremental deployment. It
defines mechanisms that allow the new class of L4S congestion defines mechanisms that allow the new class of L4S congestion
controls to coexist with 'Classic' congestion controls in a shared controls to coexist with 'Classic' congestion controls in a shared
network. The aim is for L4S latency and throughput to be usually network. The aim is for L4S latency and throughput to be usually
much better (and rarely worse), while typically not impacting Classic much better (and rarely worse) while typically not impacting Classic
performance. performance.
Status of This Memo Status of This Memo
This Internet-Draft is submitted in full conformance with the This document is not an Internet Standards Track specification; it is
provisions of BCP 78 and BCP 79. published for informational purposes.
Internet-Drafts are working documents of the Internet Engineering
Task Force (IETF). Note that other groups may also distribute
working documents as Internet-Drafts. The list of current Internet-
Drafts is at https://datatracker.ietf.org/drafts/current/.
Internet-Drafts are draft documents valid for a maximum of six months This document is a product of the Internet Engineering Task Force
and may be updated, replaced, or obsoleted by other documents at any (IETF). It represents the consensus of the IETF community. It has
time. It is inappropriate to use Internet-Drafts as reference received public review and has been approved for publication by the
material or to cite them other than as "work in progress." Internet Engineering Steering Group (IESG). Not all documents
approved by the IESG are candidates for any level of Internet
Standard; see Section 2 of RFC 7841.
This Internet-Draft will expire on 2 March 2023. 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/rfc9330.
Copyright Notice Copyright Notice
Copyright (c) 2022 IETF Trust and the persons identified as the Copyright (c) 2023 IETF Trust and the persons identified as the
document authors. All rights reserved. document authors. All rights reserved.
This document is subject to BCP 78 and the IETF Trust's Legal This document is subject to BCP 78 and the IETF Trust's Legal
Provisions Relating to IETF Documents (https://trustee.ietf.org/ Provisions Relating to IETF Documents
license-info) in effect on the date of publication of this document. (https://trustee.ietf.org/license-info) in effect on the date of
Please review these documents carefully, as they describe your rights publication of this document. Please review these documents
and restrictions with respect to this document. Code Components carefully, as they describe your rights and restrictions with respect
extracted from this document must include Revised BSD License text as to this document. Code Components extracted from this document must
described in Section 4.e of the Trust Legal Provisions and are include Revised BSD License text as described in Section 4.e of the
provided without warranty as described in the Revised BSD License. Trust Legal Provisions and are provided without warranty as described
in the Revised BSD License.
Table of Contents Table of Contents
1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . 3 1. Introduction
1.1. Document Roadmap . . . . . . . . . . . . . . . . . . . . 5 1.1. Document Roadmap
2. L4S Architecture Overview . . . . . . . . . . . . . . . . . . 5 2. L4S Architecture Overview
3. Terminology . . . . . . . . . . . . . . . . . . . . . . . . . 7 3. Terminology
4. L4S Architecture Components . . . . . . . . . . . . . . . . . 9 4. L4S Architecture Components
4.1. Protocol Mechanisms . . . . . . . . . . . . . . . . . . . 9 4.1. Protocol Mechanisms
4.2. Network Components . . . . . . . . . . . . . . . . . . . 10 4.2. Network Components
4.3. Host Mechanisms . . . . . . . . . . . . . . . . . . . . . 13 4.3. Host Mechanisms
5. Rationale . . . . . . . . . . . . . . . . . . . . . . . . . . 15 5. Rationale
5.1. Why These Primary Components? . . . . . . . . . . . . . . 15 5.1. Why These Primary Components?
5.2. What L4S adds to Existing Approaches . . . . . . . . . . 18 5.2. What L4S Adds to Existing Approaches
6. Applicability . . . . . . . . . . . . . . . . . . . . . . . . 21 6. Applicability
6.1. Applications . . . . . . . . . . . . . . . . . . . . . . 21 6.1. Applications
6.2. Use Cases . . . . . . . . . . . . . . . . . . . . . . . . 22 6.2. Use Cases
6.3. Applicability with Specific Link Technologies . . . . . . 24 6.3. Applicability with Specific Link Technologies
6.4. Deployment Considerations . . . . . . . . . . . . . . . . 25 6.4. Deployment Considerations
6.4.1. Deployment Topology . . . . . . . . . . . . . . . . . 25 6.4.1. Deployment Topology
6.4.2. Deployment Sequences . . . . . . . . . . . . . . . . 26 6.4.2. Deployment Sequences
6.4.3. L4S Flow but Non-ECN Bottleneck . . . . . . . . . . . 29 6.4.3. L4S Flow but Non-ECN Bottleneck
6.4.4. L4S Flow but Classic ECN Bottleneck . . . . . . . . . 30 6.4.4. L4S Flow but Classic ECN Bottleneck
6.4.5. L4S AQM Deployment within Tunnels . . . . . . . . . . 30 6.4.5. L4S AQM Deployment within Tunnels
7. IANA Considerations (to be removed by RFC Editor) . . . . . . 30 7. IANA Considerations
8. Security Considerations . . . . . . . . . . . . . . . . . . . 31 8. Security Considerations
8.1. Traffic Rate (Non-)Policing . . . . . . . . . . . . . . . 31 8.1. Traffic Rate (Non-)Policing
8.1.1. (Non-)Policing Rate per Flow . . . . . . . . . . . . 31 8.1.1. (Non-)Policing Rate per Flow
8.1.2. (Non-)Policing L4S Service Rate . . . . . . . . . . . 31 8.1.2. (Non-)Policing L4S Service Rate
8.2. 'Latency Friendliness' . . . . . . . . . . . . . . . . . 32 8.2. 'Latency Friendliness'
8.3. Interaction between Rate Policing and L4S . . . . . . . . 34 8.3. Interaction between Rate Policing and L4S
8.4. ECN Integrity . . . . . . . . . . . . . . . . . . . . . . 35 8.4. ECN Integrity
8.5. Privacy Considerations . . . . . . . . . . . . . . . . . 35 8.5. Privacy Considerations
9. Informative References . . . . . . . . . . . . . . . . . . . 36 9. Informative References
Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . 45 Acknowledgements
Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . 45 Authors' Addresses
1. Introduction 1. Introduction
At any one time, it is increasingly common for all of the traffic in At any one time, it is increasingly common for all of the traffic in
a bottleneck link (e.g. a household's Internet access) to come from a bottleneck link (e.g., a household's Internet access or Wi-Fi) to
applications that prefer low delay: interactive Web, Web services, come from applications that prefer low delay: interactive web, web
voice, conversational video, interactive video, interactive remote services, voice, conversational video, interactive video, interactive
presence, instant messaging, online gaming, remote desktop, cloud- remote presence, instant messaging, online and cloud-rendered gaming,
based applications and video-assisted remote control of machinery and remote desktop, cloud-based applications, cloud-rendered virtual
industrial processes. In the last decade or so, much has been done reality or augmented reality, and video-assisted remote control of
to reduce propagation delay by placing caches or servers closer to machinery and industrial processes. In the last decade or so, much
users. However, queuing remains a major, albeit intermittent, has been done to reduce propagation delay by placing caches or
component of latency. For instance spikes of hundreds of servers closer to users. However, queuing remains a major, albeit
milliseconds are not uncommon, even with state-of-the-art active intermittent, component of latency. For instance, spikes of hundreds
queue management (AQM) [COBALT], [DOCSIS3AQM]. Queuing in access of milliseconds are not uncommon, even with state-of-the-art Active
network bottlenecks is typically configured to cause overall network Queue Management (AQM) [COBALT] [DOCSIS3AQM]. A Classic AQM in an
delay to roughly double during a long-running flow, relative to access network bottleneck is typically configured to buffer the
expected base (unloaded) path delay [BufferSize]. Low loss is also sawteeth of lone flows, which can cause peak overall network delay to
important because, for interactive applications, losses translate roughly double during a long-running flow, relative to expected base
into even longer retransmission delays. (unloaded) path delay [BufferSize]. Low loss is also important
because, for interactive applications, losses translate into even
longer retransmission delays.
It has been demonstrated that, once access network bit rates reach It has been demonstrated that, once access network bit rates reach
levels now common in the developed world, increasing link capacity levels now common in the developed world, increasing link capacity
offers diminishing returns if latency (delay) is not addressed offers diminishing returns if latency (delay) is not addressed
[Dukkipati06], [Rajiullah15]. Therefore, the goal is an Internet [Dukkipati06] [Rajiullah15]. Therefore, the goal is an Internet
service with very Low queueing Latency, very Low Loss and Scalable service with very low queuing latency, very low loss, and scalable
throughput (L4S). Very low queuing latency means less than throughput. Very low queuing latency means less than 1 millisecond
1 millisecond (ms) on average and less than about 2 ms at the 99th (ms) on average and less than about 2 ms at the 99th percentile.
percentile. End-to-end delay above 50 ms [Raaen14] or even above End-to-end delay above 50 ms [Raaen14], or even above 20 ms [NASA04],
20 ms [NASA04] starts to feel unnatural for more demanding starts to feel unnatural for more demanding interactive applications.
interactive applications. So removing unnecessary delay variability Therefore, removing unnecessary delay variability increases the reach
increases the reach of these applications (the distance over which of these applications (the distance over which they are comfortable
they are comfortable to use). This document describes the L4S to use) and/or provides additional latency budget that can be used
for enhanced processing. This document describes the L4S
architecture for achieving these goals. architecture for achieving these goals.
Differentiated services (Diffserv) offers Expedited Forwarding Differentiated services (Diffserv) offers Expedited Forwarding (EF)
(EF [RFC3246]) for some packets at the expense of others, but this [RFC3246] for some packets at the expense of others, but this makes
makes no difference when all (or most) of the traffic at a bottleneck no difference when all (or most) of the traffic at a bottleneck at
at any one time requires low latency. In contrast, L4S still works any one time requires low latency. In contrast, L4S still works well
well when all traffic is L4S - a service that gives without taking when all traffic is L4S -- a service that gives without taking needs
needs none of the configuration or management baggage (traffic none of the configuration or management baggage (traffic policing or
policing, traffic contracts) associated with favouring some traffic traffic contracts) associated with favouring some traffic flows over
flows over others. others.
Queuing delay degrades performance intermittently [Hohlfeld14]. It Queuing delay degrades performance intermittently [Hohlfeld14]. It
occurs when a large enough capacity-seeking (e.g. TCP) flow is occurs i) when a large enough capacity-seeking (e.g., TCP) flow is
running alongside the user's traffic in the bottleneck link, which is running alongside the user's traffic in the bottleneck link, which is
typically in the access network. Or when the low latency application typically in the access network, or ii) when the low latency
is itself a large capacity-seeking or adaptive rate (e.g. interactive application is itself a large capacity-seeking or adaptive rate flow
video) flow. At these times, the performance improvement from L4S (e.g., interactive video). At these times, the performance
must be sufficient that network operators will be motivated to deploy improvement from L4S must be sufficient for network operators to be
it. motivated to deploy it.
Active Queue Management (AQM) is part of the solution to queuing Active Queue Management (AQM) is part of the solution to queuing
under load. AQM improves performance for all traffic, but there is a under load. AQM improves performance for all traffic, but there is a
limit to how much queuing delay can be reduced by solely changing the limit to how much queuing delay can be reduced by solely changing the
network; without addressing the root of the problem. network without addressing the root of the problem.
The root of the problem is the presence of standard congestion The root of the problem is the presence of standard congestion
control (Reno [RFC5681]) or compatible variants control (Reno [RFC5681]) or compatible variants (e.g., CUBIC
(e.g. CUBIC [RFC8312]) that are used in TCP and in other transports [RFC8312]) that are used in TCP and in other transports, such as QUIC
such as QUIC [RFC9000]. We shall use the term 'Classic' for these [RFC9000]. We shall use the term 'Classic' for these Reno-friendly
Reno-friendly congestion controls. Classic congestion controls congestion controls. Classic congestion controls induce relatively
induce relatively large saw-tooth-shaped excursions up the queue and large sawtooth-shaped excursions of queue occupancy. So if a network
down again, which have been growing as flow rate scales [RFC3649]. operator naively attempts to reduce queuing delay by configuring an
So if a network operator naively attempts to reduce queuing delay by AQM to operate at a shallower queue, a Classic congestion control
configuring an AQM to operate at a shallower queue, a Classic will significantly underutilize the link at the bottom of every
congestion control will significantly underutilize the link at the sawtooth. These sawteeth have also been growing in duration as flow
bottom of every saw-tooth. rate scales (see Section 5.1 and [RFC3649]).
It has been demonstrated that if the sending host replaces a Classic It has been demonstrated that, if the sending host replaces a Classic
congestion control with a 'Scalable' alternative, when a suitable AQM congestion control with a 'Scalable' alternative, the performance
is deployed in the network the performance under load of all the under load of all the above interactive applications can be
above interactive applications can be significantly improved. For significantly improved once a suitable AQM is deployed in the
instance, queuing delay under heavy load with the example DCTCP/DualQ network. Taking the example solution cited below that uses Data
solution cited below on a DSL or Ethernet link is roughly 1 to 2 Center TCP (DCTCP) [RFC8257] and a Dual-Queue Coupled AQM [RFC9332]
milliseconds at the 99th percentile without losing link utilization on a DSL or Ethernet link, queuing delay under heavy load is roughly
[DualPI2Linux], [DCttH19] (for other link types, see Section 6.3). 1-2 ms at the 99th percentile without losing link utilization
[L4Seval22] [DualPI2Linux] (for other link types, see Section 6.3).
This compares with 5-20 ms on _average_ with a Classic congestion This compares with 5-20 ms on _average_ with a Classic congestion
control and current state-of-the-art AQMs such as FQ-CoDel [RFC8290], control and current state-of-the-art AQMs, such as Flow Queue CoDel
PIE [RFC8033] or DOCSIS PIE [RFC8034] and about 20-30 ms at the 99th [RFC8290], Proportional Integral controller Enhanced (PIE) [RFC8033],
percentile [DualPI2Linux]. or DOCSIS PIE [RFC8034] and about 20-30 ms at the 99th percentile
[DualPI2Linux].
L4S is designed for incremental deployment. It is possible to deploy L4S is designed for incremental deployment. It is possible to deploy
the L4S service at a bottleneck link alongside the existing best the L4S service at a bottleneck link alongside the existing best
efforts service [DualPI2Linux] so that unmodified applications can efforts service [DualPI2Linux] so that unmodified applications can
start using it as soon as the sender's stack is updated. Access start using it as soon as the sender's stack is updated. Access
networks are typically designed with one link as the bottleneck for networks are typically designed with one link as the bottleneck for
each site (which might be a home, small enterprise or mobile device), each site (which might be a home, small enterprise, or mobile
so deployment at either or both ends of this link should give nearly device), so deployment at either or both ends of this link should
all the benefit in the respective direction. With some transport give nearly all the benefit in the respective direction. With some
protocols, namely TCP and SCTP, the sender has to check that the transport protocols, namely TCP [ACCECN], the sender has to check
receiver has been suitably updated to give more accurate feedback, that the receiver has been suitably updated to give more accurate
whereas with more recent transport protocols such as QUIC and DCCP, feedback, whereas with more recent transport protocols, such as QUIC
[RFC9000] and Datagram Congestion Control Protocol (DCCP) [RFC4340],
all receivers have always been suitable. all receivers have always been suitable.
This document presents the L4S architecture. It consists of three This document presents the L4S architecture. It consists of three
components: network support to isolate L4S traffic from classic components: network support to isolate L4S traffic from Classic
traffic; protocol features that allow network elements to identify traffic; protocol features that allow network elements to identify
L4S traffic; and host support for L4S congestion controls. The L4S traffic; and host support for L4S congestion controls. The
protocol is defined separately [I-D.ietf-tsvwg-ecn-l4s-id] as an protocol is defined separately in [RFC9331] as an experimental change
experimental change to Explicit Congestion Notification (ECN). This to Explicit Congestion Notification (ECN). This document describes
document describes and justifies the component parts and how they and justifies the component parts and how they interact to provide
interact to provide the scalable, low latency, low loss Internet the low latency, low loss, and scalable Internet service. It also
service. It also details the approach to incremental deployment, as details the approach to incremental deployment, as briefly summarized
briefly summarized above. above.
1.1. Document Roadmap 1.1. Document Roadmap
This document describes the L4S architecture in three passes. First This document describes the L4S architecture in three passes. First,
this brief overview gives the very high level idea and states the the brief overview in Section 2 gives the very high-level idea and
main components with minimal rationale. This is only intended to states the main components with minimal rationale. This is only
give some context for the terminology definitions that follow in intended to give some context for the terminology definitions that
Section 3, and to explain the structure of the rest of the document. follow in Section 3 and to explain the structure of the rest of the
Then Section 4 goes into more detail on each component with some document. Then, Section 4 goes into more detail on each component
rationale, but still mostly stating what the architecture is, rather with some rationale but still mostly stating what the architecture
than why. Finally, Section 5 justifies why each element of the is, rather than why. Finally, Section 5 justifies why each element
solution was chosen (Section 5.1) and why these choices were of the solution was chosen (Section 5.1) and why these choices were
different from other solutions (Section 5.2). different from other solutions (Section 5.2).
Having described the architecture, Section 6 clarifies its After the architecture has been described, Section 6 clarifies its
applicability; that is, the applications and use-cases that motivated applicability by describing the applications and use cases that
the design, the challenges applying the architecture to various link motivated the design, the challenges applying the architecture to
technologies, and various incremental deployment models: including various link technologies, and various incremental deployment models
the two main deployment topologies, different sequences for (including the two main deployment topologies, different sequences
incremental deployment and various interactions with pre-existing for incremental deployment, and various interactions with preexisting
approaches. The document ends with the usual tailpieces, including approaches). The document ends with the usual tailpieces, including
extensive discussion of traffic policing and other security extensive discussion of traffic policing and other security
considerations in Section 8. considerations in Section 8.
2. L4S Architecture Overview 2. L4S Architecture Overview
Below we outline the three main components to the L4S architecture; Below, we outline the three main components to the L4S architecture:
1) the scalable congestion control on the sending host; 2) the AQM at 1) the Scalable congestion control on the sending host; 2) the AQM at
the network bottleneck; and 3) the protocol between them. the network bottleneck; and 3) the protocol between them.
But first, the main point to grasp is that low latency is not But first, the main point to grasp is that low latency is not
provided by the network - low latency results from the careful provided by the network; low latency results from the careful
behaviour of the scalable congestion controllers used by L4S senders. behaviour of the Scalable congestion controllers used by L4S senders.
The network does have a role - primarily to isolate the low latency The network does have a role, primarily to isolate the low latency of
of the carefully behaving L4S traffic from the higher queuing delay the carefully behaving L4S traffic from the higher queuing delay
needed by traffic with pre-existing Classic behaviour. The network needed by traffic with preexisting Classic behaviour. The network
also alters the way it signals queue growth to the transport - It also alters the way it signals queue growth to the transport. It
uses the Explicit Congestion Notification (ECN) protocol, but it uses the Explicit Congestion Notification (ECN) protocol, but it
signals the very start of queue growth - immediately without the signals the very start of queue growth immediately, without the
smoothing delay typical of Classic AQMs. Because ECN support is smoothing delay typical of Classic AQMs. Because ECN support is
essential for L4S, senders use the ECN field as the protocol that essential for L4S, senders use the ECN field as the protocol that
allows the network to identify which packets are L4S and which are allows the network to identify which packets are L4S and which are
Classic. Classic.
1) Host: Scalable congestion controls already exist. They solve the 1) Host:
scaling problem with Classic congestion controls, such as Reno or
Cubic. Because flow rate has scaled since TCP congestion control
was first designed in 1988, assuming the flow lasts long enough,
it now takes hundreds of round trips (and growing) to recover
after a congestion signal (whether a loss or an ECN mark) as shown
in the examples in Section 5.1 and [RFC3649]. Therefore, control
of queuing and utilization becomes very slack, and the slightest
disturbances (e.g. from new flows starting) prevent a high rate
from being attained.
With a scalable congestion control, the average time from one Scalable congestion controls already exist. They solve the
congestion signal to the next (the recovery time) remains scaling problem with Classic congestion controls, such as Reno or
invariant as the flow rate scales, all other factors being equal. CUBIC. Because flow rate has scaled since TCP congestion control
This maintains the same degree of control over queueing and was first designed in 1988, assuming the flow lasts long enough,
utilization whatever the flow rate, as well as ensuring that high it now takes hundreds of round trips (and growing) to recover
throughput is more robust to disturbances. The scalable control after a congestion signal (whether a loss or an ECN mark), as
used most widely (in controlled environments) is Data Center TCP shown in the examples in Section 5.1 and [RFC3649]. Therefore,
(DCTCP [RFC8257]), which has been implemented and deployed in control of queuing and utilization becomes very slack, and the
Windows Server Editions (since 2012), in Linux and in FreeBSD. slightest disturbances (e.g., from new flows starting) prevent a
Although DCTCP as-is functions well over wide-area round trip high rate from being attained.
times, most implementations lack certain safety features that
would be necessary for use outside controlled environments like
data centres (see Section 6.4.3). So scalable congestion control
needs to be implemented in TCP and other transport protocols
(QUIC, SCTP, RTP/RTCP, RMCAT, etc.). Indeed, between the present
document being drafted and published, the following scalable
congestion controls were implemented: TCP Prague [PragueLinux],
QUIC Prague, an L4S variant of the RMCAT SCReAM
controller [SCReAM] and the L4S ECN part of BBRv2 [BBRv2] intended
for TCP and QUIC transports.
2) Network: L4S traffic needs to be isolated from the queuing With a Scalable congestion control, the average time from one
latency of Classic traffic. One queue per application flow (FQ) congestion signal to the next (the recovery time) remains
is one way to achieve this, e.g. FQ-CoDel [RFC8290]. However, invariant as flow rate scales, all other factors being equal.
using just two queues is sufficient and does not require This maintains the same degree of control over queuing and
inspection of transport layer headers in the network, which is not utilization, whatever the flow rate, as well as ensuring that
always possible (see Section 5.2). With just two queues, it might high throughput is more robust to disturbances. The Scalable
seem impossible to know how much capacity to schedule for each control used most widely (in controlled environments) is DCTCP
queue without inspecting how many flows at any one time are using [RFC8257], which has been implemented and deployed in Windows
each. And it would be undesirable to arbitrarily divide access Server Editions (since 2012), in Linux, and in FreeBSD. Although
network capacity into two partitions. The Dual Queue Coupled AQM DCTCP as-is functions well over wide-area round-trip times
was developed as a minimal complexity solution to this problem. (RTTs), most implementations lack certain safety features that
It acts like a 'semi-permeable' membrane that partitions latency would be necessary for use outside controlled environments, like
but not bandwidth. As such, the two queues are for transition data centres (see Section 6.4.3). Therefore, Scalable congestion
from Classic to L4S behaviour, not bandwidth prioritization. control needs to be implemented in TCP and other transport
protocols (QUIC, Stream Control Transmission Protocol (SCTP),
RTP/RTCP, RTP Media Congestion Avoidance Techniques (RMCAT),
etc.). Indeed, between the present document being drafted and
published, the following Scalable congestion controls were
implemented: Prague over TCP and QUIC [PRAGUE-CC] [PragueLinux],
an L4S variant of the RMCAT SCReAM controller [SCReAM-L4S], and
the L4S ECN part of Bottleneck Bandwidth and Round-trip
propagation time (BBRv2) [BBRv2] intended for TCP and QUIC
transports.
Section 4 gives a high level explanation of how the per-flow-queue 2) Network:
(FQ) and DualQ variants of L4S work, and
[I-D.ietf-tsvwg-aqm-dualq-coupled] gives a full explanation of the
DualQ Coupled AQM framework. A specific marking algorithm is not
mandated for L4S AQMs. Appendices of
[I-D.ietf-tsvwg-aqm-dualq-coupled] give non-normative examples
that have been implemented and evaluated, and give recommended
default parameter settings. It is expected that L4S experiments
will improve knowledge of parameter settings and whether the set
of marking algorithms needs to be limited.
3) Protocol: A sending host needs to distinguish L4S and Classic L4S traffic needs to be isolated from the queuing latency of
packets with an identifier so that the network can classify them Classic traffic. One queue per application flow (FQ) is one way
into their separate treatments. The L4S identifier to achieve this, e.g., FQ-CoDel [RFC8290]. However, using just
spec. [I-D.ietf-tsvwg-ecn-l4s-id] concludes that all alternatives two queues is sufficient and does not require inspection of
involve compromises, but the ECT(1) and CE codepoints of the ECN transport layer headers in the network, which is not always
field represent a workable solution. As already explained, the possible (see Section 5.2). With just two queues, it might seem
network also uses ECN to immediately signal the very start of impossible to know how much capacity to schedule for each queue
queue growth to the transport. without inspecting how many flows at any one time are using each.
And it would be undesirable to arbitrarily divide access network
capacity into two partitions. The Dual-Queue Coupled AQM was
developed as a minimal complexity solution to this problem. It
acts like a 'semi-permeable' membrane that partitions latency but
not bandwidth. As such, the two queues are for transitioning
from Classic to L4S behaviour, not bandwidth prioritization.
3. Terminology Section 4 gives a high-level explanation of how the per-flow
queue (FQ) and DualQ variants of L4S work, and [RFC9332] gives a
full explanation of the DualQ Coupled AQM framework. A specific
marking algorithm is not mandated for L4S AQMs. Appendices of
[RFC9332] give non-normative examples that have been implemented
and evaluated and give recommended default parameter settings.
It is expected that L4S experiments will improve knowledge of
parameter settings and whether the set of marking algorithms
needs to be limited.
[Note to the RFC Editor (to be removed before publication as an RFC): 3) Protocol:
The following definitions are copied from the L4S ECN
spec [I-D.ietf-tsvwg-ecn-l4s-id] for the reader's convenience. A sending host needs to distinguish L4S and Classic packets with
Except, here, Classic CC and Scalable CC are condensed because they an identifier so that the network can classify them into their
refer to Section 5.1 later. Also the definition of Traffic Policing separate treatments. The L4S identifier spec [RFC9331] concludes
is not needed in [I-D.ietf-tsvwg-ecn-l4s-id].] that all alternatives involve compromises, but the ECT(1) and
Congestion Experienced (CE) codepoints of the ECN field represent
a workable solution. As already explained, the network also uses
ECN to immediately signal the very start of queue growth to the
transport.
3. Terminology
Classic Congestion Control: A congestion control behaviour that can Classic Congestion Control: A congestion control behaviour that can
co-exist with standard Reno [RFC5681] without causing coexist with standard Reno [RFC5681] without causing significantly
significantly negative impact on its flow rate [RFC5033]. The negative impact on its flow rate [RFC5033]. The scaling problem
scaling problem with Classic congestion control is explained, with with Classic congestion control is explained, with examples, in
examples, in Section 5.1 and in [RFC3649]. Section 5.1 and in [RFC3649].
Scalable Congestion Control: A congestion control where the average Scalable Congestion Control: A congestion control where the average
time from one congestion signal to the next (the recovery time) time from one congestion signal to the next (the recovery time)
remains invariant as the flow rate scales, all other factors being remains invariant as flow rate scales, all other factors being
equal. For instance, DCTCP averages 2 congestion signals per equal. For instance, DCTCP averages 2 congestion signals per
round-trip whatever the flow rate, as do other recently developed round trip, whatever the flow rate, as do other recently developed
scalable congestion controls, e.g. Relentless TCP [Mathis09], TCP Scalable congestion controls, e.g., Relentless TCP [RELENTLESS],
Prague [I-D.briscoe-iccrg-prague-congestion-control], Prague for TCP and QUIC [PRAGUE-CC] [PragueLinux], BBRv2 [BBRv2]
[PragueLinux], BBRv2 [BBRv2], [BBR-CC], and the L4S variant of SCReAM for real-time media
[I-D.cardwell-iccrg-bbr-congestion-control] and the L4S variant of [SCReAM-L4S] [RFC8298]. See Section 4.3 of [RFC9331] for more
SCReAM for real-time media [SCReAM], [RFC8298]). See Section 4.3 explanation.
of [I-D.ietf-tsvwg-ecn-l4s-id] for more explanation.
Classic service: The Classic service is intended for all the Classic Service: The Classic service is intended for all the
congestion control behaviours that co-exist with Reno [RFC5681] congestion control behaviours that coexist with Reno [RFC5681]
(e.g. Reno itself, Cubic [RFC8312], (e.g., Reno itself, CUBIC [RFC8312], Compound [CTCP], and TFRC
Compound [I-D.sridharan-tcpm-ctcp], TFRC [RFC5348]). The term [RFC5348]). The term 'Classic queue' means a queue providing the
'Classic queue' means a queue providing the Classic service. Classic service.
Low-Latency, Low-Loss Scalable throughput (L4S) service: The 'L4S' Low Latency, Low Loss, and Scalable throughput (L4S) service: The
service is intended for traffic from scalable congestion control 'L4S' service is intended for traffic from Scalable congestion
algorithms, such as the Prague congestion control algorithms, such as the Prague congestion control
control [I-D.briscoe-iccrg-prague-congestion-control], which was [PRAGUE-CC], which was derived from DCTCP [RFC8257]. The L4S
derived from DCTCP [RFC8257]. The L4S service is for more service is for more general traffic than just Prague -- it allows
general traffic than just Prague -- it allows the set of the set of congestion controls with similar scaling properties to
congestion controls with similar scaling properties to Prague to Prague to evolve, such as the examples listed above (Relentless,
evolve, such as the examples listed above (Relentless, SCReAM). SCReAM, etc.). The term 'L4S queue' means a queue providing the
The term 'L4S queue' means a queue providing the L4S service. L4S service.
The terms Classic or L4S can also qualify other nouns, such as The terms Classic or L4S can also qualify other nouns, such as
'queue', 'codepoint', 'identifier', 'classification', 'packet', 'queue', 'codepoint', 'identifier', 'classification', 'packet',
'flow'. For example: an L4S packet means a packet with an L4S and 'flow'. For example, an L4S packet means a packet with an L4S
identifier sent from an L4S congestion control. identifier sent from an L4S congestion control.
Both Classic and L4S services can cope with a proportion of Both Classic and L4S services can cope with a proportion of
unresponsive or less-responsive traffic as well, but in the L4S unresponsive or less-responsive traffic as well but, in the L4S
case its rate has to be smooth enough or low enough to not build a case, its rate has to be smooth enough or low enough to not build
queue (e.g. DNS, VoIP, game sync datagrams, etc.). a queue (e.g., DNS, Voice over IP (VoIP), game sync datagrams,
etc.).
Reno-friendly: The subset of Classic traffic that is friendly to the Reno-friendly: The subset of Classic traffic that is friendly to the
standard Reno congestion control defined for TCP in [RFC5681]. standard Reno congestion control defined for TCP in [RFC5681].
The TFRC spec. [RFC5348] indirectly implies that 'friendly' is The TFRC spec [RFC5348] indirectly implies that 'friendly' is
defined as "generally within a factor of two of the sending rate defined as "generally within a factor of two of the sending rate
of a TCP flow under the same conditions". Reno-friendly is used of a TCP flow under the same conditions". Reno-friendly is used
here in place of 'TCP-friendly', given the latter has become here in place of 'TCP-friendly', given the latter has become
imprecise, because the TCP protocol is now used with so many imprecise, because the TCP protocol is now used with so many
different congestion control behaviours, and Reno is used in non- different congestion control behaviours, and Reno is used in non-
TCP transports such as QUIC [RFC9000]. TCP transports, such as QUIC [RFC9000].
Classic ECN: The original Explicit Congestion Notification (ECN) Classic ECN: The original Explicit Congestion Notification (ECN)
protocol [RFC3168], which requires ECN signals to be treated as protocol [RFC3168] that requires ECN signals to be treated as
equivalent to drops, both when generated in the network and when equivalent to drops, both when generated in the network and when
responded to by the sender. responded to by the sender.
L4S uses the ECN field as an For L4S, the names used for the four codepoints of the 2-bit IP-
identifier [I-D.ietf-tsvwg-ecn-l4s-id] with the names for the four ECN field are unchanged from those defined in the ECN spec
codepoints of the 2-bit IP-ECN field unchanged from those defined [RFC3168], i.e., Not-ECT, ECT(0), ECT(1), and CE, where ECT stands
in the ECN spec [RFC3168]: Not ECT, ECT(0), ECT(1) and CE, where for ECN-Capable Transport and CE stands for Congestion
ECT stands for ECN-Capable Transport and CE stands for Congestion
Experienced. A packet marked with the CE codepoint is termed Experienced. A packet marked with the CE codepoint is termed
'ECN-marked' or sometimes just 'marked' where the context makes 'ECN-marked' or sometimes just 'marked' where the context makes
ECN obvious. ECN obvious.
Site: A home, mobile device, small enterprise or campus, where the Site: A home, mobile device, small enterprise, or campus where the
network bottleneck is typically the access link to the site. Not network bottleneck is typically the access link to the site. Not
all network arrangements fit this model but it is a useful, widely all network arrangements fit this model, but it is a useful,
applicable generalization. widely applicable generalization.
Traffic policing: Limiting traffic by dropping packets or shifting Traffic Policing: Limiting traffic by dropping packets or shifting
them to lower service class (as opposed to introducing delay, them to a lower service class (as opposed to introducing delay,
which is termed traffic shaping). Policing can involve limiting which is termed 'traffic shaping'). Policing can involve limiting
average rate and/or burst size. Policing focused on limiting the average rate and/or burst size. Policing focused on limiting
queuing but not average flow rate is termed congestion policing, queuing but not the average flow rate is termed 'congestion
latency policing, burst policing or queue protection in this policing', 'latency policing', 'burst policing', or 'queue
document. Otherwise, the term rate policing is used. protection' in this document. Otherwise, the term rate policing
is used.
4. L4S Architecture Components 4. L4S Architecture Components
The L4S architecture is composed of the elements in the following The L4S architecture is composed of the elements in the following
three subsections. three subsections.
4.1. Protocol Mechanisms 4.1. Protocol Mechanisms
The L4S architecture involves: a) unassignment of the previous use of The L4S architecture involves: a) unassignment of the previous use of
the identifier; b) reassignment of the same identifier; and c) the identifier; b) reassignment of the same identifier; and c)
optional further identifiers: optional further identifiers:
a. An essential aspect of a scalable congestion control is the use a. An essential aspect of a Scalable congestion control is the use
of explicit congestion signals. 'Classic' ECN [RFC3168] requires of explicit congestion signals. Classic ECN [RFC3168] requires
an ECN signal to be treated as equivalent to drop, both when it an ECN signal to be treated as equivalent to drop, both when it
is generated in the network and when it is responded to by hosts. is generated in the network and when it is responded to by hosts.
L4S needs networks and hosts to support a more fine-grained L4S needs networks and hosts to support a more fine-grained
meaning for each ECN signal that is less severe than a drop, so meaning for each ECN signal that is less severe than a drop, so
that the L4S signals: that the L4S signals:
* can be much more frequent; * can be much more frequent and
* can be signalled immediately, without the significant delay * can be signalled immediately, without the significant delay
required to smooth out fluctuations in the queue. required to smooth out fluctuations in the queue.
To enable L4S, the standards track Classic ECN spec. [RFC3168] To enable L4S, the Standards Track Classic ECN spec [RFC3168] has
has had to be updated to allow L4S packets to depart from the had to be updated to allow L4S packets to depart from the
'equivalent to drop' constraint. [RFC8311] is a standards track 'equivalent-to-drop' constraint. [RFC8311] is a Standards Track
update to relax specific requirements in RFC 3168 (and certain update to relax specific requirements in [RFC3168] (and certain
other standards track RFCs), which clears the way for the other Standards Track RFCs), which clears the way for the
experimental changes proposed for L4S. Also, the ECT(1) experimental changes proposed for L4S. Also, the ECT(1)
codepoint was previously assigned as the experimental ECN codepoint was previously assigned as the experimental ECN nonce
nonce [RFC3540], which RFC 8311 recategorizes as historic to make [RFC3540], which [RFC8311] recategorizes as historic to make the
the codepoint available again. codepoint available again.
b. [I-D.ietf-tsvwg-ecn-l4s-id] specifies that ECT(1) is used as the b. [RFC9331] specifies that ECT(1) is used as the identifier to
identifier to classify L4S packets into a separate treatment from classify L4S packets into a separate treatment from Classic
Classic packets. This satisfies the requirement for identifying packets. This satisfies the requirement for identifying an
an alternative ECN treatment in [RFC4774]. alternative ECN treatment in [RFC4774].
The CE codepoint is used to indicate Congestion Experienced by The CE codepoint is used to indicate Congestion Experienced by
both L4S and Classic treatments. This raises the concern that a both L4S and Classic treatments. This raises the concern that a
Classic AQM earlier on the path might have marked some ECT(0) Classic AQM earlier on the path might have marked some ECT(0)
packets as CE. Then these packets will be erroneously classified packets as CE. Then, these packets will be erroneously
into the L4S queue. Appendix B of the L4S ECN classified into the L4S queue. Appendix B of [RFC9331] explains
spec [I-D.ietf-tsvwg-ecn-l4s-id] explains why five unlikely why five unlikely eventualities all have to coincide for this to
eventualities all have to coincide for this to have any have any detrimental effect, which even then would only involve a
detrimental effect, which even then would only involve a
vanishingly small likelihood of a spurious retransmission. vanishingly small likelihood of a spurious retransmission.
c. A network operator might wish to include certain unresponsive, c. A network operator might wish to include certain unresponsive,
non-L4S traffic in the L4S queue if it is deemed to be smoothly non-L4S traffic in the L4S queue if it is deemed to be paced
enough paced and low enough rate not to build a queue. For smoothly enough and at a low enough rate not to build a queue,
instance, VoIP, low rate datagrams to sync online games, for instance, VoIP, low rate datagrams to sync online games,
relatively low rate application-limited traffic, DNS, LDAP, etc. relatively low rate application-limited traffic, DNS, Lightweight
This traffic would need to be tagged with specific identifiers, Directory Access Protocol (LDAP), etc. This traffic would need
e.g. a low latency Diffserv Codepoint such as Expedited to be tagged with specific identifiers, e.g., a low-latency
Forwarding (EF [RFC3246]), Non-Queue-Building Diffserv codepoint such as Expedited Forwarding (EF) [RFC3246],
(NQB [I-D.ietf-tsvwg-nqb]), or operator-specific identifiers. Non-Queue-Building (NQB) [NQB-PHB], or operator-specific
identifiers.
4.2. Network Components 4.2. Network Components
The L4S architecture aims to provide low latency without the _need_ The L4S architecture aims to provide low latency without the _need_
for per-flow operations in network components. Nonetheless, the for per-flow operations in network components. Nonetheless, the
architecture does not preclude per-flow solutions. The following architecture does not preclude per-flow solutions. The following
bullets describe the known arrangements: a) the DualQ Coupled AQM bullets describe the known arrangements: a) the DualQ Coupled AQM
with an L4S AQM in one queue coupled from a Classic AQM in the other; with an L4S AQM in one queue coupled from a Classic AQM in the other;
b) Per-Flow Queues with an instance of a Classic and an L4S AQM in b) per-flow queues with an instance of a Classic and an L4S AQM in
each queue; c) Dual queues with per-flow AQMs, but no per-flow each queue; and c) Dual queues with per-flow AQMs but no per-flow
queues: queues:
a. The Dual Queue Coupled AQM (illustrated in Figure 1) achieves the a. The Dual-Queue Coupled AQM (illustrated in Figure 1) achieves the
'semi-permeable' membrane property mentioned earlier as follows: 'semi-permeable' membrane property mentioned earlier as follows:
* Latency isolation: Two separate queues are used to isolate L4S * Latency isolation: Two separate queues are used to isolate L4S
queuing delay from the larger queue that Classic traffic needs queuing delay from the larger queue that Classic traffic needs
to maintain full utilization. to maintain full utilization.
* Bandwidth pooling: The two queues act as if they are a single * Bandwidth pooling: The two queues act as if they are a single
pool of bandwidth in which flows of either type get roughly pool of bandwidth in which flows of either type get roughly
equal throughput without the scheduler needing to identify any equal throughput without the scheduler needing to identify any
flows. This is achieved by having an AQM in each queue, but flows. This is achieved by having an AQM in each queue, but
skipping to change at page 11, line 28 skipping to change at line 510
classes of congestion control. Specifically, the Classic AQM classes of congestion control. Specifically, the Classic AQM
generates a drop/mark probability based on congestion in its generates a drop/mark probability based on congestion in its
own queue, which it uses both to drop/mark packets in its own own queue, which it uses both to drop/mark packets in its own
queue and to affect the marking probability in the L4S queue. queue and to affect the marking probability in the L4S queue.
The strength of the coupling of the congestion signalling The strength of the coupling of the congestion signalling
between the two queues is enough to make the L4S flows slow between the two queues is enough to make the L4S flows slow
down to leave the right amount of capacity for the Classic down to leave the right amount of capacity for the Classic
flows (as they would if they were the same type of traffic flows (as they would if they were the same type of traffic
sharing the same queue). sharing the same queue).
Then the scheduler can serve the L4S queue with priority (denoted Then, the scheduler can serve the L4S queue with priority
by the '1' on the higher priority input), because the L4S traffic (denoted by the '1' on the higher priority input), because the
isn't offering up enough traffic to use all the priority that it L4S traffic isn't offering up enough traffic to use all the
is given. Therefore: priority that it is given. Therefore:
* for latency isolation on short time-scales (sub-round-trip) * for latency isolation on short timescales (sub-round-trip),
the prioritization of the L4S queue protects its low latency the prioritization of the L4S queue protects its low latency
by allowing bursts to dissipate quickly; by allowing bursts to dissipate quickly;
* but for bandwidth pooling on longer time-scales (round-trip * but for bandwidth pooling on longer timescales (round-trip and
and longer) the Classic queue creates an equal and opposite longer), the Classic queue creates an equal and opposite
pressure against the L4S traffic to ensure that neither has pressure against the L4S traffic to ensure that neither has
priority when it comes to bandwidth - the tension between priority when it comes to bandwidth -- the tension between
prioritizing L4S and coupling the marking from the Classic AQM prioritizing L4S and coupling the marking from the Classic AQM
results in approximate per-flow fairness. results in approximate per-flow fairness.
To protect against unresponsive traffic taking advantage of the To protect against the prioritization of persistent L4S traffic
prioritization of the L4S queue and starving the Classic queue, deadlocking the Classic queue for a while in some
it is advisable for the priority to be conditional, not strict implementations, it is advisable for the priority to be
(see Appendix A of the DualQ conditional, not strict (see Appendix A of the DualQ spec
spec [I-D.ietf-tsvwg-aqm-dualq-coupled]). [RFC9332]).
When there is no Classic traffic, the L4S queue's own AQM comes When there is no Classic traffic, the L4S queue's own AQM comes
into play. It starts congestion marking with a very shallow into play. It starts congestion marking with a very shallow
queue, so L4S traffic maintains very low queuing delay. queue, so L4S traffic maintains very low queuing delay.
If either queue becomes persistently overloaded, drop of ECN- If either queue becomes persistently overloaded, drop of some
capable packets is introduced, as recommended in Section 7 of the ECN-capable packets is introduced, as recommended in Section 7 of
ECN spec [RFC3168] and Section 4.2.1 of the AQM the ECN spec [RFC3168] and Section 4.2.1 of the AQM
recommendations [RFC7567]. Then both queues introduce the same recommendations [RFC7567]. The trade-offs with different
level of drop (not shown in the figure). approaches are discussed in Section 4.2.3 of the DualQ spec
[RFC9332] (not shown in the figure here).
The Dual Queue Coupled AQM has been specified as generically as The Dual-Queue Coupled AQM has been specified as generically as
possible [I-D.ietf-tsvwg-aqm-dualq-coupled] without specifying possible [RFC9332] without specifying the particular AQMs to use
the particular AQMs to use in the two queues so that designers in the two queues so that designers are free to implement diverse
are free to implement diverse ideas. Informational appendices in ideas. Informational appendices in that document give pseudocode
that draft give pseudocode examples of two different specific AQM examples of two different specific AQM approaches: one called
approaches: one called DualPI2 (pronounced Dual PI DualPI2 (pronounced Dual PI Squared) [DualPI2Linux] that uses the
Squared) [DualPI2Linux] that uses the PI2 variant of PIE, and a PI2 variant of PIE and a zero-config variant of Random Early
zero-config variant of RED called Curvy RED. A DualQ Coupled AQM Detection (RED) called Curvy RED. A DualQ Coupled AQM based on
based on PIE has also been specified and implemented for Low PIE has also been specified and implemented for Low Latency
Latency DOCSIS [DOCSIS3.1]. DOCSIS [DOCSIS3.1].
(3) (2) (3) (2)
.-------^------..------------^------------------. .-------^------..------------^------------------.
,-(1)-----. _____ ,-(1)-----. _____
; ________ : L4S -------. | | ; ________ : L4S -------. | |
:|Scalable| : _\ ||__\_|mark | :|Scalable| : _\ ||__\_|mark |
:| sender | : __________ / / || / |_____|\ _________ :| sender | : __________ / / || / |_____|\ _________
:|________|\; | |/ -------' ^ \1|condit'nl| :|________|\; | |/ -------' ^ \1|condit'nl|
`---------'\_| IP-ECN | Coupling : \|priority |_\ `---------'\_| IP-ECN | Coupling : \|priority |_\
________ / |Classifier| : /|scheduler| / ________ / |Classifier| : /|scheduler| /
|Classic |/ |__________|\ -------. __:__ / |_________| |Classic |/ |__________|\ -------. __:__ / |_________|
| sender | \_\ || | ||__\_|mark/|/ | sender | \_\ || | ||__\_|mark/|/
|________| / || | || / |drop | |________| / || | || / |drop |
Classic -------' |_____| Classic -------' |_____|
Figure 1: Components of an L4S DualQ Coupled AQM Solution: 1) (1) Scalable sending host
Scalable Sending Host; 2) Isolation in separate network (2) Isolation in separate network queues
queues; and 3) Packet Identification Protocol (3) Packet identification protocol
b. Per-Flow Queues and AQMs: A scheduler with per-flow queues such Figure 1: Components of an L4S DualQ Coupled AQM Solution
as FQ-CoDel or FQ-PIE can be used for L4S. For instance within
b. Per-Flow Queues and AQMs: A scheduler with per-flow queues, such
as FQ-CoDel or FQ-PIE, can be used for L4S. For instance, within
each queue of an FQ-CoDel system, as well as a CoDel AQM, there each queue of an FQ-CoDel system, as well as a CoDel AQM, there
is typically also the option of ECN marking at an immediate is typically also the option of ECN marking at an immediate
(unsmoothed) shallow threshold to support use in data centres (unsmoothed) shallow threshold to support use in data centres
(see Sec.5.2.7 of the FQ-CoDel spec [RFC8290]). In Linux, this (see Section 5.2.7 of the FQ-CoDel spec [RFC8290]). In Linux,
has been modified so that the shallow threshold can be solely this has been modified so that the shallow threshold can be
applied to ECT(1) packets [FQ_CoDel_Thresh]. Then, if there is a solely applied to ECT(1) packets [FQ_CoDel_Thresh]. Then, if
flow of non-ECN or ECT(0) packets in the per-flow-queue, the there is a flow of Not-ECT or ECT(0) packets in the per-flow
Classic AQM (e.g. CoDel) is applied; while if there is a flow of queue, the Classic AQM (e.g., CoDel) is applied; whereas, if
ECT(1) packets in the queue, the shallower (typically sub- there is a flow of ECT(1) packets in the queue, the shallower
millisecond) threshold is applied. In addition, ECT(0) and not- (typically sub-millisecond) threshold is applied. In addition,
ECT packets could potentially be classified into a separate flow- ECT(0) and Not-ECT packets could potentially be classified into a
queue from ECT(1) and CE packets to avoid them mixing if they separate flow queue from ECT(1) and CE packets to avoid them
share a common flow-identifier (e.g. in a VPN). mixing if they share a common flow identifier (e.g., in a VPN).
c. Dual-queues, but per-flow AQMs: It should also be possible to use c. Dual queues but per-flow AQMs: It should also be possible to use
dual queues for isolation, but with per-flow marking to control dual queues for isolation but with per-flow marking to control
flow-rates (instead of the coupled per-queue marking of the Dual flow rates (instead of the coupled per-queue marking of the Dual-
Queue Coupled AQM). One of the two queues would be for isolating Queue Coupled AQM). One of the two queues would be for isolating
L4S packets, which would be classified by the ECN codepoint. L4S packets, which would be classified by the ECN codepoint.
Flow rates could be controlled by flow-specific marking. The Flow rates could be controlled by flow-specific marking. The
policy goal of the marking could be to differentiate flow rates policy goal of the marking could be to differentiate flow rates
(e.g. [Nadas20], which requires additional signalling of a per- (e.g., [Nadas20], which requires additional signalling of a per-
flow 'value'), or to equalize flow-rates (perhaps in a similar flow 'value') or to equalize flow rates (perhaps in a similar way
way to Approx Fair CoDel [AFCD], to Approx Fair CoDel [AFCD] [CODEL-APPROX-FAIR] but with two
[I-D.morton-tsvwg-codel-approx-fair], but with two queues not queues not one).
one).
Note that whenever the term 'DualQ' is used loosely without Note that, whenever the term 'DualQ' is used loosely without
saying whether marking is per-queue or per-flow, it means a dual saying whether marking is per queue or per flow, it means a dual-
queue AQM with per-queue marking. queue AQM with per-queue marking.
4.3. Host Mechanisms 4.3. Host Mechanisms
The L4S architecture includes two main mechanisms in the end host The L4S architecture includes two main mechanisms in the end host
that we enumerate next: that we enumerate next:
a. Scalable Congestion Control at the sender: Section 2 defines a a. Scalable congestion control at the sender: Section 2 defines a
scalable congestion control as one where the average time from Scalable congestion control as one where the average time from
one congestion signal to the next (the recovery time) remains one congestion signal to the next (the recovery time) remains
invariant as the flow rate scales, all other factors being equal. invariant as flow rate scales, all other factors being equal.
Data Center TCP is the most widely used example. It has been DCTCP is the most widely used example. It has been documented as
documented as an informational record of the protocol currently an informational record of the protocol currently in use in
in use in controlled environments [RFC8257]. A draft list of controlled environments [RFC8257]. A list of safety and
safety and performance improvements for a scalable congestion performance improvements for a Scalable congestion control to be
control to be usable on the public Internet has been drawn up usable on the public Internet has been drawn up (see the so-
(the so-called 'Prague L4S requirements' in Appendix A of called 'Prague L4S requirements' in Appendix A of [RFC9331]).
The subset that involve risk of harm to others have been captured
[I-D.ietf-tsvwg-ecn-l4s-id]). The subset that involve risk of as normative requirements in Section 4 of [RFC9331]. TCP Prague
harm to others have been captured as normative requirements in [PRAGUE-CC] has been implemented in Linux as a reference
Section 4 of [I-D.ietf-tsvwg-ecn-l4s-id]. TCP implementation to address these requirements [PragueLinux].
Prague [I-D.briscoe-iccrg-prague-congestion-control] has been
implemented in Linux as a reference implementation to address
these requirements [PragueLinux].
Transport protocols other than TCP use various congestion Transport protocols other than TCP use various congestion
controls that are designed to be friendly with Reno. Before they controls that are designed to be friendly with Reno. Before they
can use the L4S service, they will need to be updated to can use the L4S service, they will need to be updated to
implement a scalable congestion response, which they will have to implement a Scalable congestion response, which they will have to
indicate by using the ECT(1) codepoint. Scalable variants are indicate by using the ECT(1) codepoint. Scalable variants are
under consideration for more recent transport protocols, under consideration for more recent transport protocols (e.g.,
e.g. QUIC, and the L4S ECN part of BBRv2 [BBRv2], QUIC), and the L4S ECN part of BBRv2 [BBRv2] [BBR-CC] is a
[I-D.cardwell-iccrg-bbr-congestion-control] is a scalable Scalable congestion control intended for the TCP and QUIC
congestion control intended for the TCP and QUIC transports, transports, amongst others. Also, an L4S variant of the RMCAT
amongst others. Also, an L4S variant of the RMCAT SCReAM SCReAM controller [RFC8298] has been implemented [SCReAM-L4S] for
controller [RFC8298] has been implemented [SCReAM] for media media transported over RTP.
transported over RTP.
Section 4.3 of the L4S ECN spec [I-D.ietf-tsvwg-ecn-l4s-id] Section 4.3 of the L4S ECN spec [RFC9331] defines Scalable
defines scalable congestion control in more detail, and specifies congestion control in more detail and specifies the requirements
the requirements that an L4S scalable congestion control has to that an L4S Scalable congestion control has to comply with.
comply with.
b. The ECN feedback in some transport protocols is already b. The ECN feedback in some transport protocols is already
sufficiently fine-grained for L4S (specifically DCCP [RFC4340] sufficiently fine-grained for L4S (specifically DCCP [RFC4340]
and QUIC [RFC9000]). But others either require update or are in and QUIC [RFC9000]). But others either require updates or are in
the process of being updated: the process of being updated:
* For the case of TCP, the feedback protocol for ECN embeds the * For the case of TCP, the feedback protocol for ECN embeds the
assumption from Classic ECN [RFC3168] that an ECN mark is assumption from Classic ECN [RFC3168] that an ECN mark is
equivalent to a drop, making it unusable for a scalable TCP. equivalent to a drop, making it unusable for a Scalable TCP.
Therefore, the implementation of TCP receivers will have to be Therefore, the implementation of TCP receivers will have to be
upgraded [RFC7560]. Work to standardize and implement more upgraded [RFC7560]. Work to standardize and implement more
accurate ECN feedback for TCP (AccECN) is in accurate ECN feedback for TCP (AccECN) is in progress [ACCECN]
progress [I-D.ietf-tcpm-accurate-ecn], [PragueLinux]. [PragueLinux].
* ECN feedback was only roughly sketched in an appendix of the * ECN feedback was only roughly sketched in the appendix of the
now obsoleted second specification of SCTP [RFC4960], while a now obsoleted second specification of SCTP [RFC4960], while a
fuller specification was proposed in a long-expired fuller specification was proposed in a long-expired document
draft [I-D.stewart-tsvwg-sctpecn]. A new design would need to [ECN-SCTP]. A new design would need to be implemented and
be implemented and deployed before SCTP could support L4S. deployed before SCTP could support L4S.
* For RTP, sufficient ECN feedback was defined in [RFC6679], but * For RTP, sufficient ECN feedback was defined in [RFC6679], but
[RFC8888] defines the latest standards track improvements. [RFC8888] defines the latest Standards Track improvements.
5. Rationale 5. Rationale
5.1. Why These Primary Components? 5.1. Why These Primary Components?
Explicit congestion signalling (protocol): Explicit congestion Explicit congestion signalling (protocol): Explicit congestion
signalling is a key part of the L4S approach. In contrast, use of signalling is a key part of the L4S approach. In contrast, use of
drop as a congestion signal creates a tension because drop is both drop as a congestion signal creates tension because drop is both
an impairment (less would be better) and a useful signal (more an impairment (less would be better) and a useful signal (more
would be better): would be better):
* Explicit congestion signals can be used many times per round * Explicit congestion signals can be used many times per round
trip, to keep tight control, without any impairment. Under trip to keep tight control without any impairment. Under heavy
heavy load, even more explicit signals can be applied, so that load, even more explicit signals can be applied so that the
the queue can be kept short whatever the load. In contrast, queue can be kept short whatever the load. In contrast,
Classic AQMs have to introduce very high packet drop at high Classic AQMs have to introduce very high packet drop at high
load to keep the queue short. By using ECN, an L4S congestion load to keep the queue short. By using ECN, an L4S congestion
control's sawtooth reduction can be smaller and therefore control's sawtooth reduction can be smaller and therefore
return to the operating point more often, without worrying that return to the operating point more often, without worrying that
more sawteeth will cause more signals. The consequent smaller more sawteeth will cause more signals. The consequent smaller
amplitude sawteeth fit between an empty queue and a very amplitude sawteeth fit between an empty queue and a very
shallow marking threshold (~1 ms in the public Internet), so shallow marking threshold (~1 ms in the public Internet), so
queue delay variation can be very low, without risk of under- queue delay variation can be very low, without risk of
utilization. underutilization.
* Explicit congestion signals can be emitted immediately to track * Explicit congestion signals can be emitted immediately to track
fluctuations of the queue. L4S shifts smoothing from the fluctuations of the queue. L4S shifts smoothing from the
network to the host. The network doesn't know the round trip network to the host. The network doesn't know the round-trip
times of any of the flows. So if the network is responsible times (RTTs) of any of the flows. So if the network is
for smoothing (as in the Classic approach), it has to assume a responsible for smoothing (as in the Classic approach), it has
worst case RTT, otherwise long RTT flows would become unstable. to assume a worst case RTT, otherwise long RTT flows would
This delays Classic congestion signals by 100-200 ms. In become unstable. This delays Classic congestion signals by
contrast, each host knows its own round trip time. So, in the 100-200 ms. In contrast, each host knows its own RTT. So, in
L4S approach, the host can smooth each flow over its own RTT, the L4S approach, the host can smooth each flow over its own
introducing no more smoothing delay than strictly necessary RTT, introducing no more smoothing delay than strictly
(usually only a few milliseconds). A host can also choose not necessary (usually only a few milliseconds). A host can also
to introduce any smoothing delay if appropriate, e.g. during choose not to introduce any smoothing delay if appropriate,
flow start-up. e.g., during flow start-up.
Neither of the above are feasible if explicit congestion Neither of the above are feasible if explicit congestion
signalling has to be considered 'equivalent to drop' (as was signalling has to be considered 'equivalent to drop' (as was
required with Classic ECN [RFC3168]), because drop is an required with Classic ECN [RFC3168]), because drop is an
impairment as well as a signal. So drop cannot be excessively impairment as well as a signal. So drop cannot be excessively
frequent, and drop cannot be immediate, otherwise too many drops frequent, and drop cannot be immediate; otherwise, too many drops
would turn out to have been due to only a transient fluctuation in would turn out to have been due to only a transient fluctuation in
the queue that would not have warranted dropping a packet in the queue that would not have warranted dropping a packet in
hindsight. Therefore, in an L4S AQM, the L4S queue uses a new L4S hindsight. Therefore, in an L4S AQM, the L4S queue uses a new L4S
variant of ECN that is not equivalent to drop (see section 5.2 of variant of ECN that is not equivalent to drop (see Section 5.2 of
the L4S ECN spec [I-D.ietf-tsvwg-ecn-l4s-id]), while the Classic the L4S ECN spec [RFC9331]), while the Classic queue uses either
queue uses either Classic ECN [RFC3168] or drop, which are Classic ECN [RFC3168] or drop, which are still equivalent to each
equivalent to each other. other.
Before Classic ECN was standardized, there were various proposals Before Classic ECN was standardized, there were various proposals
to give an ECN mark a different meaning from drop. However, there to give an ECN mark a different meaning from drop. However, there
was no particular reason to agree on any one of the alternative was no particular reason to agree on any one of the alternative
meanings, so 'equivalent to drop' was the only compromise that meanings, so 'equivalent to drop' was the only compromise that
could be reached. RFC 3168 contains a statement that: could be reached. [RFC3168] contains a statement that:
"An environment where all end nodes were ECN-Capable could An environment where all end nodes were ECN-Capable could
allow new criteria to be developed for setting the CE allow new criteria to be developed for setting the CE
codepoint, and new congestion control mechanisms for end-node codepoint, and new congestion control mechanisms for end-node
reaction to CE packets. However, this is a research issue, and reaction to CE packets. However, this is a research issue,
as such is not addressed in this document." and as such is not addressed in this document.
Latency isolation (network): L4S congestion controls keep queue Latency isolation (network): L4S congestion controls keep queue
delay low whereas Classic congestion controls need a queue of the delay low, whereas Classic congestion controls need a queue of the
order of the RTT to avoid under-utilization. One queue cannot order of the RTT to avoid underutilization. One queue cannot have
have two lengths, therefore L4S traffic needs to be isolated in a two lengths; therefore, L4S traffic needs to be isolated in a
separate queue (e.g. DualQ) or queues (e.g. FQ). separate queue (e.g., DualQ) or queues (e.g., FQ).
Coupled congestion notification: Coupling the congestion Coupled congestion notification: Coupling the congestion
notification between two queues as in the DualQ Coupled AQM is not notification between two queues as in the DualQ Coupled AQM is not
necessarily essential, but it is a simple way to allow senders to necessarily essential, but it is a simple way to allow senders to
determine their rate, packet by packet, rather than be overridden determine their rate packet by packet, rather than be overridden
by a network scheduler. An alternative is for a network scheduler by a network scheduler. An alternative is for a network scheduler
to control the rate of each application flow (see discussion in to control the rate of each application flow (see the discussion
Section 5.2). in Section 5.2).
L4S packet identifier (protocol): Once there are at least two L4S packet identifier (protocol): Once there are at least two
treatments in the network, hosts need an identifier at the IP treatments in the network, hosts need an identifier at the IP
layer to distinguish which treatment they intend to use. layer to distinguish which treatment they intend to use.
Scalable congestion notification: A scalable congestion control in Scalable congestion notification: A Scalable congestion control in
the host keeps the signalling frequency from the network high the host keeps the signalling frequency from the network high,
whatever the flow rate, so that queue delay variations can be whatever the flow rate, so that queue delay variations can be
small when conditions are stable, and rate can track variations in small when conditions are stable, and rate can track variations in
available capacity as rapidly as possible otherwise. available capacity as rapidly as possible otherwise.
Low loss: Latency is not the only concern of L4S. The 'Low Loss' Low loss: Latency is not the only concern of L4S. The 'Low Loss'
part of the name denotes that L4S generally achieves zero part of the name denotes that L4S generally achieves zero
congestion loss due to its use of ECN. Otherwise, loss would congestion loss due to its use of ECN. Otherwise, loss would
itself cause delay, particularly for short flows, due to itself cause delay, particularly for short flows, due to
retransmission delay [RFC2884]. retransmission delay [RFC2884].
Scalable throughput: The "Scalable throughput" part of the name Scalable throughput: The 'Scalable throughput' part of the name
denotes that the per-flow throughput of scalable congestion denotes that the per-flow throughput of Scalable congestion
controls should scale indefinitely, avoiding the imminent scaling controls should scale indefinitely, avoiding the imminent scaling
problems with Reno-friendly congestion control problems with Reno-friendly congestion control algorithms
algorithms [RFC3649]. It was known when TCP congestion avoidance [RFC3649]. It was known when TCP congestion avoidance was first
was first developed in 1988 that it would not scale to high developed in 1988 that it would not scale to high bandwidth-delay
bandwidth-delay products (see footnote 6 in [TCP-CA]). Today, products (see footnote 6 in [TCP-CA]). Today, regular broadband
regular broadband flow rates over WAN distances are already beyond flow rates over WAN distances are already beyond the scaling range
the scaling range of Classic Reno congestion control. So `less of Classic Reno congestion control. So 'less unscalable' CUBIC
unscalable' Cubic [RFC8312] and Compound [I-D.sridharan-tcpm-ctcp] [RFC8312] and Compound [CTCP] variants of TCP have been
variants of TCP have been successfully deployed. However, these successfully deployed. However, these are now approaching their
are now approaching their scaling limits. scaling limits.
For instance, we will consider a scenario with a maximum RTT of For instance, we will consider a scenario with a maximum RTT of 30
30 ms at the peak of each sawtooth. As Reno packet rate scales 8x ms at the peak of each sawtooth. As Reno packet rate scales 8
from 1,250 to 10,000 packet/s (from 15 to 120 Mb/s with 1500 B times from 1,250 to 10,000 packet/s (from 15 to 120 Mb/s with 1500
packets), the time to recover from a congestion event rises B packets), the time to recover from a congestion event rises
proportionately by 8x as well, from 422 ms to 3.38 s. It is proportionately by 8 times as well, from 422 ms to 3.38 s. It is
clearly problematic for a congestion control to take multiple clearly problematic for a congestion control to take multiple
seconds to recover from each congestion event. Cubic [RFC8312] seconds to recover from each congestion event. CUBIC [RFC8312]
was developed to be less unscalable, but it is approaching its was developed to be less unscalable, but it is approaching its
scaling limit; with the same max RTT of 30 ms, at 120 Mb/s Cubic scaling limit; with the same max RTT of 30 ms, at 120 Mb/s, CUBIC
is still fully in its Reno-friendly mode, so it takes about 4.3 s is still fully in its Reno-friendly mode, so it takes about 4.3 s
to recover. However, once the flow rate scales by 8x again to to recover. However, once flow rate scales by 8 times again to
960 Mb/s it enters true Cubic mode, with a recovery time of 960 Mb/s it enters true CUBIC mode, with a recovery time of 12.2
12.2 s. From then on, each further scaling by 8x doubles Cubic's s. From then on, each further scaling by 8 times doubles CUBIC's
recovery time (because the cube root of 8 is 2), e.g. at 7.68 Gb/s recovery time (because the cube root of 8 is 2), e.g., at 7.68 Gb/
the recovery time is 24.3 s. In contrast, a scalable congestion s, the recovery time is 24.3 s. In contrast, a Scalable
control like DCTCP or TCP Prague induces 2 congestion signals per congestion control like DCTCP or Prague induces 2 congestion
round trip on average, which remains invariant for any flow rate, signals per round trip on average, which remains invariant for any
keeping dynamic control very tight. flow rate, keeping dynamic control very tight.
For a feel of where the global average lone-flow download sits on For a feel of where the global average lone-flow download sits on
this scale at the time of writing (2021), according to [BDPdata] this scale at the time of writing (2021), according to [BDPdata],
globally averaged fixed access capacity was 103 Mb/s in 2020 and the global average fixed access capacity was 103 Mb/s in 2020 and
averaged base RTT to a CDN was 25-34ms in 2019. Averaging of per- the average base RTT to a CDN was 25 to 34 ms in 2019. Averaging
country data was weighted by Internet user population (data of per-country data was weighted by Internet user population (data
collected globally is necessarily of variable quality, but the collected globally is necessarily of variable quality, but the
paper does double-check that the outcome compares well against a paper does double-check that the outcome compares well against a
second source). So a lone CUBIC flow would at best take about 200 second source). So a lone CUBIC flow would at best take about 200
round trips (5 s) to recover from each of its sawtooth reductions, round trips (5 s) to recover from each of its sawtooth reductions,
if the flow even lasted that long. This is described as 'at best' if the flow even lasted that long. This is described as 'at best'
because it assumes everyone uses an AQM, whereas in reality most because it assumes everyone uses an AQM, whereas in reality, most
users still have a (probably bloated) tail-drop buffer. In the users still have a (probably bloated) tail-drop buffer. In the
tail-drop case, likely average recovery time would be at least 4x tail-drop case, the likely average recovery time would be at least
5 s, if not more, because RTT under load would be at least double 4 times 5 s, if not more, because RTT under load would be at least
that of an AQM, and recovery time depends on the square of RTT. double that of an AQM, and the recovery time of Reno-friendly
flows depends on the square of RTT.
Although work on scaling congestion controls tends to start with Although work on scaling congestion controls tends to start with
TCP as the transport, the above is not intended to exclude other TCP as the transport, the above is not intended to exclude other
transports (e.g. SCTP, QUIC) or less elastic algorithms transports (e.g., SCTP and QUIC) or less elastic algorithms (e.g.,
(e.g. RMCAT), which all tend to adopt the same or similar RMCAT), which all tend to adopt the same or similar developments.
developments.
5.2. What L4S adds to Existing Approaches 5.2. What L4S Adds to Existing Approaches
All the following approaches address some part of the same problem All the following approaches address some part of the same problem
space as L4S. In each case, it is shown that L4S complements them or space as L4S. In each case, it is shown that L4S complements them or
improves on them, rather than being a mutually exclusive alternative: improves on them, rather than being a mutually exclusive alternative:
Diffserv: Diffserv addresses the problem of bandwidth apportionment Diffserv: Diffserv addresses the problem of bandwidth apportionment
for important traffic as well as queuing latency for delay- for important traffic as well as queuing latency for delay-
sensitive traffic. Of these, L4S solely addresses the problem of sensitive traffic. Of these, L4S solely addresses the problem of
queuing latency. Diffserv will still be necessary where important queuing latency. Diffserv will still be necessary where important
traffic requires priority (e.g. for commercial reasons, or for traffic requires priority (e.g., for commercial reasons or for
protection of critical infrastructure traffic) - see protection of critical infrastructure traffic) -- see
[I-D.briscoe-tsvwg-l4s-diffserv]. Nonetheless, the L4S approach [L4S-DIFFSERV]. Nonetheless, the L4S approach can provide low
can provide low latency for all traffic within each Diffserv class latency for all traffic within each Diffserv class (including the
(including the case where there is only the one default Diffserv case where there is only the one default Diffserv class).
class).
Also, Diffserv can only provide a latency benefit if a small Also, Diffserv can only provide a latency benefit if a small
subset of the traffic on a bottleneck link requests low latency. subset of the traffic on a bottleneck link requests low latency.
As already explained, it has no effect when all the applications As already explained, it has no effect when all the applications
in use at one time at a single site (home, small business or in use at one time at a single site (e.g., a home, small business,
mobile device) require low latency. In contrast, because L4S or mobile device) require low latency. In contrast, because L4S
works for all traffic, it needs none of the management baggage works for all traffic, it needs none of the management baggage
(traffic policing, traffic contracts) associated with favouring (traffic policing or traffic contracts) associated with favouring
some packets over others. This lack of management baggage ought some packets over others. This lack of management baggage ought
to give L4S a better chance of end-to-end deployment. to give L4S a better chance of end-to-end deployment.
In particular, because networks tend not to trust end systems to In particular, if networks do not trust end systems to identify
identify which packets should be favoured over others, where which packets should be favoured, they assign packets to Diffserv
networks assign packets to Diffserv classes they tend to use classes themselves. However, the techniques available to such
packet inspection of application flow identifiers or deeper networks, like inspection of flow identifiers or deeper inspection
inspection of application signatures. Thus, nowadays, Diffserv of application signatures, do not always sit well with encryption
doesn't always sit well with encryption of the layers above IP of the layers above IP [RFC8404]. In these cases, users can have
[RFC8404]. So users have to choose between privacy and QoS. either privacy or Quality of Service (QoS), but not both.
As with Diffserv, the L4S identifier is in the IP header. But, in As with Diffserv, the L4S identifier is in the IP header. But, in
contrast to Diffserv, the L4S identifier does not convey a want or contrast to Diffserv, the L4S identifier does not convey a want or
a need for a certain level of quality. Rather, it promises a a need for a certain level of quality. Rather, it promises a
certain behaviour (scalable congestion response), which networks certain behaviour (Scalable congestion response), which networks
can objectively verify if they need to. This is because low delay can objectively verify if they need to. This is because low delay
depends on collective host behaviour, whereas bandwidth priority depends on collective host behaviour, whereas bandwidth priority
depends on network behaviour. depends on network behaviour.
State-of-the-art AQMs: AQMs such as PIE and FQ-CoDel give a State-of-the-art AQMs: AQMs for Classic traffic, such as PIE and FQ-
significant reduction in queuing delay relative to no AQM at all. CoDel, give a significant reduction in queuing delay relative to
L4S is intended to complement these AQMs, and should not distract no AQM at all. L4S is intended to complement these AQMs and
from the need to deploy them as widely as possible. Nonetheless, should not distract from the need to deploy them as widely as
AQMs alone cannot reduce queuing delay too far without possible. Nonetheless, AQMs alone cannot reduce queuing delay too
significantly reducing link utilization, because the root cause of far without significantly reducing link utilization, because the
the problem is on the host - where Classic congestion controls use root cause of the problem is on the host -- where Classic
large saw-toothing rate variations. The L4S approach resolves congestion controls use large sawtoothing rate variations. The
this tension between delay and utilization by enabling hosts to L4S approach resolves this tension between delay and utilization
minimize the amplitude of their sawteeth. A single-queue Classic by enabling hosts to minimize the amplitude of their sawteeth. A
AQM is not sufficient to allow hosts to use small sawteeth for two single-queue Classic AQM is not sufficient to allow hosts to use
reasons: i) smaller sawteeth would not get lower delay in an AQM small sawteeth for two reasons: i) smaller sawteeth would not get
designed for larger amplitude Classic sawteeth, because a queue lower delay in an AQM designed for larger amplitude Classic
can only have one length at a time; and ii) much smaller sawteeth sawteeth, because a queue can only have one length at a time and
implies much more frequent sawteeth, so L4S flows would drive a ii) much smaller sawteeth implies much more frequent sawteeth, so
Classic AQM into a high level of ECN-marking, which would appear L4S flows would drive a Classic AQM into a high level of ECN-
as heavy congestion to Classic flows, which in turn would greatly marking, which would appear as heavy congestion to Classic flows,
reduce their rate as a result (see Section 6.4.4). which in turn would greatly reduce their rate as a result (see
Section 6.4.4).
Per-flow queuing or marking: Similarly, per-flow approaches such as Per-flow queuing or marking: Similarly, per-flow approaches, such as
FQ-CoDel or Approx Fair CoDel [AFCD] are not incompatible with the FQ-CoDel or Approx Fair CoDel [AFCD], are not incompatible with
L4S approach. However, per-flow queuing alone is not enough - it the L4S approach. However, per-flow queuing alone is not enough
only isolates the queuing of one flow from others; not from -- it only isolates the queuing of one flow from others, not from
itself. Per-flow implementations need to have support for itself. Per-flow implementations need to have support for
scalable congestion control added, which has already been done for Scalable congestion control added, which has already been done for
FQ-CoDel in Linux (see Sec.5.2.7 of [RFC8290] and FQ-CoDel in Linux (see Section 5.2.7 of [RFC8290] and
[FQ_CoDel_Thresh]). Without this simple modification, per-flow [FQ_CoDel_Thresh]). Without this simple modification, per-flow
AQMs like FQ-CoDel would still not be able to support applications AQMs, like FQ-CoDel, would still not be able to support
that need both very low delay and high bandwidth, e.g. video-based applications that need both very low delay and high bandwidth,
control of remote procedures, or interactive cloud-based video e.g., video-based control of remote procedures or interactive
(see Note 1 below). cloud-based video (see Note 1 below).
Although per-flow techniques are not incompatible with L4S, it is Although per-flow techniques are not incompatible with L4S, it is
important to have the DualQ alternative. This is because handling important to have the DualQ alternative. This is because handling
end-to-end (layer 4) flows in the network (layer 3 or 2) precludes end-to-end (layer 4) flows in the network (layer 3 or 2) precludes
some important end-to-end functions. For instance: some important end-to-end functions. For instance:
a. Per-flow forms of L4S like FQ-CoDel are incompatible with full A. Per-flow forms of L4S, like FQ-CoDel, are incompatible with
end-to-end encryption of transport layer identifiers for full end-to-end encryption of transport layer identifiers for
privacy and confidentiality (e.g. IPSec or encrypted VPN privacy and confidentiality (e.g., IPsec or encrypted VPN
tunnels, as opposed to DTLS over UDP), because they require tunnels, as opposed to DTLS over UDP), because they require
packet inspection to access the end-to-end transport flow packet inspection to access the end-to-end transport flow
identifiers. identifiers.
In contrast, the DualQ form of L4S requires no deeper In contrast, the DualQ form of L4S requires no deeper
inspection than the IP layer. So, as long as operators take inspection than the IP layer. So as long as operators take
the DualQ approach, their users can have both very low queuing the DualQ approach, their users can have both very low queuing
delay and full end-to-end encryption [RFC8404]. delay and full end-to-end encryption [RFC8404].
b. With per-flow forms of L4S, the network takes over control of B. With per-flow forms of L4S, the network takes over control of
the relative rates of each application flow. Some see it as the relative rates of each application flow. Some see it as
an advantage that the network will prevent some flows running an advantage that the network will prevent some flows running
faster than others. Others consider it an inherent part of faster than others. Others consider it an inherent part of
the Internet's appeal that applications can control their rate the Internet's appeal that applications can control their rate
while taking account of the needs of others via congestion while taking account of the needs of others via congestion
signals. They maintain that this has allowed applications signals. They maintain that this has allowed applications
with interesting rate behaviours to evolve, for instance, with interesting rate behaviours to evolve, for instance: i) a
variable bit-rate video that varies around an equal share variable bit-rate video that varies around an equal share,
rather than being forced to remain equal at every instant, or rather than being forced to remain equal at every instant or
e2e scavenger behaviours [RFC6817] that use less than an equal ii) end-to-end scavenger behaviours [RFC6817] that use less
share of capacity [LEDBAT_AQM]. than an equal share of capacity [LEDBAT_AQM].
The L4S architecture does not require the IETF to commit to The L4S architecture does not require the IETF to commit to
one approach over the other, because it supports both, so that one approach over the other, because it supports both so that
the 'market' can decide. Nonetheless, in the spirit of 'Do the 'market' can decide. Nonetheless, in the spirit of 'Do
one thing and do it well' [McIlroy78], the DualQ option one thing and do it well' [McIlroy78], the DualQ option
provides low delay without prejudging the issue of flow-rate provides low delay without prejudging the issue of flow-rate
control. Then, flow rate policing can be added separately if control. Then, flow rate policing can be added separately if
desired. This allows application control up to a point, but desired. In contrast to scheduling, a policer would allow
the network can still choose to set the point at which it application control up to a point, but the network would still
intervenes to prevent one flow completely starving another. be able to set the point at which it intervened to prevent one
flow completely starving another.
Note: Note:
1. It might seem that self-inflicted queuing delay within a per- 1. It might seem that self-inflicted queuing delay within a per-
flow queue should not be counted, because if the delay wasn't flow queue should not be counted, because if the delay wasn't
in the network it would just shift to the sender. However, in the network, it would just shift to the sender. However,
modern adaptive applications, e.g. HTTP/2 [RFC9113] or some modern adaptive applications, e.g., HTTP/2 [RFC9113] or some
interactive media applications (see Section 6.1), can keep low interactive media applications (see Section 6.1), can keep low
latency objects at the front of their local send queue by latency objects at the front of their local send queue by
shuffling priorities of other objects dependent on the shuffling priorities of other objects dependent on the
progress of other transfers (for example see [lowat]). They progress of other transfers (for example, see [lowat]). They
cannot shuffle objects once they have released them into the cannot shuffle objects once they have released them into the
network. network.
Alternative Back-off ECN (ABE): Here again, L4S is not an Alternative Back-off ECN (ABE): Here again, L4S is not an
alternative to ABE but a complement that introduces much lower alternative to ABE but a complement that introduces much lower
queuing delay. ABE [RFC8511] alters the host behaviour in queuing delay. ABE [RFC8511] alters the host behaviour in
response to ECN marking to utilize a link better and give ECN response to ECN marking to utilize a link better and give ECN
flows faster throughput. It uses ECT(0) and assumes the network flows faster throughput. It uses ECT(0) and assumes the network
still treats ECN and drop the same. Therefore, ABE exploits any still treats ECN and drop the same. Therefore, ABE exploits any
lower queuing delay that AQMs can provide. But, as explained lower queuing delay that AQMs can provide. But, as explained
above, AQMs still cannot reduce queuing delay too far without above, AQMs still cannot reduce queuing delay too much without
losing link utilization (to allow for other, non-ABE, flows). losing link utilization (to allow for other, non-ABE, flows).
BBR: Bottleneck Bandwidth and Round-trip propagation time BBR: Bottleneck Bandwidth and Round-trip propagation time (BBR)
(BBR [I-D.cardwell-iccrg-bbr-congestion-control]) controls queuing [BBR-CC] controls queuing delay end-to-end without needing any
delay end-to-end without needing any special logic in the network, special logic in the network, such as an AQM. So it works pretty
such as an AQM. So it works pretty-much on any path. BBR keeps much on any path. BBR keeps queuing delay reasonably low, but
queuing delay reasonably low, but perhaps not quite as low as with perhaps not quite as low as with state-of-the-art AQMs, such as
state-of-the-art AQMs such as PIE or FQ-CoDel, and certainly PIE or FQ-CoDel, and certainly nowhere near as low as with L4S.
nowhere near as low as with L4S. Queuing delay is also not Queuing delay is also not consistently low, due to BBR's regular
consistently low, due to BBR's regular bandwidth probing spikes bandwidth probing spikes and its aggressive flow start-up phase.
and its aggressive flow start-up phase.
L4S complements BBR. Indeed, BBRv2 can use L4S ECN where L4S complements BBR. Indeed, BBRv2 can use L4S ECN where
available and a scalable L4S congestion control behaviour in available and a Scalable L4S congestion control behaviour in
response to any ECN signalling from the path [BBRv2]. The L4S ECN response to any ECN signalling from the path [BBRv2]. The L4S ECN
signal complements the delay based congestion control aspects of signal complements the delay-based congestion control aspects of
BBR with an explicit indication that hosts can use, both to BBR with an explicit indication that hosts can use, both to
converge on a fair rate and to keep below a shallow queue target converge on a fair rate and to keep below a shallow queue target
set by the network. Without L4S ECN, both these aspects need to set by the network. Without L4S ECN, both these aspects need to
be assumed or estimated. be assumed or estimated.
6. Applicability 6. Applicability
6.1. Applications 6.1. Applications
A transport layer that solves the current latency issues will provide A transport layer that solves the current latency issues will provide
new service, product and application opportunities. new service, product, and application opportunities.
With the L4S approach, the following existing applications also With the L4S approach, the following existing applications also
experience significantly better quality of experience under load: experience significantly better quality of experience under load:
* Gaming, including cloud based gaming; * gaming, including cloud-based gaming;
* VoIP; * VoIP;
* Video conferencing; * video conferencing;
* Web browsing; * web browsing;
* (Adaptive) video streaming; * (adaptive) video streaming; and
* Instant messaging. * instant messaging.
The significantly lower queuing latency also enables some interactive The significantly lower queuing latency also enables some interactive
application functions to be offloaded to the cloud that would hardly application functions to be offloaded to the cloud that would hardly
even be usable today: even be usable today, including:
* Cloud based interactive video; * cloud-based interactive video and
* Cloud based virtual and augmented reality. * cloud-based virtual and augmented reality.
The above two applications have been successfully demonstrated with The above two applications have been successfully demonstrated with
L4S, both running together over a 40 Mb/s broadband access link L4S, both running together over a 40 Mb/s broadband access link
loaded up with the numerous other latency sensitive applications in loaded up with the numerous other latency-sensitive applications in
the previous list as well as numerous downloads - all sharing the the previous list, as well as numerous downloads, with all sharing
same bottleneck queue simultaneously [L4Sdemo16]. For the former, a the same bottleneck queue simultaneously [L4Sdemo16]
panoramic video of a football stadium could be swiped and pinched so [L4Sdemo16-Video]. For the former, a panoramic video of a football
that, on the fly, a proxy in the cloud could generate a sub-window of stadium could be swiped and pinched so that, on the fly, a proxy in
the match video under the finger-gesture control of each user. For the cloud could generate a sub-window of the match video under the
the latter, a virtual reality headset displayed a viewport taken from finger-gesture control of each user. For the latter, a virtual
a 360-degree camera in a racing car. The user's head movements reality headset displayed a viewport taken from a 360-degree camera
controlled the viewport extracted by a cloud-based proxy. In both in a racing car. The user's head movements controlled the viewport
cases, with 7 ms end-to-end base delay, the additional queuing delay extracted by a cloud-based proxy. In both cases, with a 7 ms end-to-
of roughly 1 ms was so low that it seemed the video was generated end base delay, the additional queuing delay of roughly 1 ms was so
locally. low that it seemed the video was generated locally.
Using a swiping finger gesture or head movement to pan a video are Using a swiping finger gesture or head movement to pan a video are
extremely latency-demanding actions -- far more demanding than VoIP. extremely latency-demanding actions -- far more demanding than VoIP
Because human vision can detect extremely low delays of the order of -- because human vision can detect extremely low delays of the order
single milliseconds when delay is translated into a visual lag of single milliseconds when delay is translated into a visual lag
between a video and a reference point (the finger or the orientation between a video and a reference point (the finger or the orientation
of the head sensed by the balance system in the inner ear -- the of the head sensed by the balance system in the inner ear, i.e., the
vestibular system). With an alternative AQM, the video noticeably vestibular system). With an alternative AQM, the video noticeably
lagged behind the finger gestures and head movements. lagged behind the finger gestures and head movements.
Without the low queuing delay of L4S, cloud-based applications like Without the low queuing delay of L4S, cloud-based applications like
these would not be credible without significantly more access these would not be credible without significantly more access-network
bandwidth (to deliver all possible video that might be viewed) and bandwidth (to deliver all possible areas of the video that might be
more local processing, which would increase the weight and power viewed) and more local processing, which would increase the weight
consumption of head-mounted displays. When all interactive and power consumption of head-mounted displays. When all interactive
processing can be done in the cloud, only the data to be rendered for processing can be done in the cloud, only the data to be rendered for
the end user needs to be sent. the end user needs to be sent.
Other low latency high bandwidth applications such as: Other low latency high bandwidth applications, such as:
* Interactive remote presence; * interactive remote presence and
* Video-assisted remote control of machinery or industrial * video-assisted remote control of machinery or industrial processes
processes.
are not credible at all without very low queuing delay. No amount of are not credible at all without very low queuing delay. No amount of
extra access bandwidth or local processing can make up for lost time. extra access bandwidth or local processing can make up for lost time.
6.2. Use Cases 6.2. Use Cases
The following use-cases for L4S are being considered by various The following use cases for L4S are being considered by various
interested parties: interested parties:
* Where the bottleneck is one of various types of access network: * where the bottleneck is one of various types of access network,
e.g. DSL, Passive Optical Networks (PON), DOCSIS cable, mobile, e.g., DSL, Passive Optical Networks (PONs), DOCSIS cable, mobile,
satellite (see Section 6.3 for some technology-specific details) satellite; or where it's a Wi-Fi link (see Section 6.3 for some
technology-specific details)
* Private networks of heterogeneous data centres, where there is no * private networks of heterogeneous data centres, where there is no
single administrator that can arrange for all the simultaneous single administrator that can arrange for all the simultaneous
changes to senders, receivers and network needed to deploy DCTCP: changes to senders, receivers, and networks needed to deploy
DCTCP:
- a set of private data centres interconnected over a wide area - a set of private data centres interconnected over a wide area
with separate administrations, but within the same company with separate administrations but within the same company
- a set of data centres operated by separate companies - a set of data centres operated by separate companies
interconnected by a community of interest network (e.g. for the interconnected by a community of interest network (e.g., for
finance sector) the finance sector)
- multi-tenant (cloud) data centres where tenants choose their - multi-tenant (cloud) data centres where tenants choose their
operating system stack (Infrastructure as a Service - IaaS) operating system stack (Infrastructure as a Service (IaaS))
* Different types of transport (or application) congestion control: * different types of transport (or application) congestion control:
- elastic (TCP/SCTP); - elastic (TCP/SCTP);
- real-time (RTP, RMCAT); - real-time (RTP, RMCAT); and
- query (DNS/LDAP). - query-response (DNS/LDAP).
* Where low delay quality of service is required, but without * where low delay QoS is required but without inspecting or
inspecting or intervening above the IP layer [RFC8404]: intervening above the IP layer [RFC8404]:
- mobile and other networks have tended to inspect higher layers - Mobile and other networks have tended to inspect higher layers
in order to guess application QoS requirements. However, with in order to guess application QoS requirements. However, with
growing demand for support of privacy and encryption, L4S growing demand for support of privacy and encryption, L4S
offers an alternative. There is no need to select which offers an alternative. There is no need to select which
traffic to favour for queuing, when L4S can give favourable traffic to favour for queuing when L4S can give favourable
queuing to all traffic. queuing to all traffic.
* If queuing delay is minimized, applications with a fixed delay * If queuing delay is minimized, applications with a fixed delay
budget can communicate over longer distances, or via a longer budget can communicate over longer distances or via more
chain of service functions [RFC7665] or onion routers. circuitous paths, e.g., longer chains of service functions
[RFC7665] or of onion routers.
* If delay jitter is minimized, it is possible to reduce the * If delay jitter is minimized, it is possible to reduce the
dejitter buffers on the receive end of video streaming, which dejitter buffers on the receiving end of video streaming, which
should improve the interactive experience should improve the interactive experience.
6.3. Applicability with Specific Link Technologies 6.3. Applicability with Specific Link Technologies
Certain link technologies aggregate data from multiple packets into Certain link technologies aggregate data from multiple packets into
bursts, and buffer incoming packets while building each burst. Wi- bursts and buffer incoming packets while building each burst. Wi-Fi,
Fi, PON and cable all involve such packet aggregation, whereas fixed PON, and cable all involve such packet aggregation, whereas fixed
Ethernet and DSL do not. No sender, whether L4S or not, can do Ethernet and DSL do not. No sender, whether L4S or not, can do
anything to reduce the buffering needed for packet aggregation. So anything to reduce the buffering needed for packet aggregation. So
an AQM should not count this buffering as part of the queue that it an AQM should not count this buffering as part of the queue that it
controls, given no amount of congestion signals will reduce it. controls, given no amount of congestion signals will reduce it.
Certain link technologies also add buffering for other reasons, Certain link technologies also add buffering for other reasons,
specifically: specifically:
* Radio links (cellular, Wi-Fi, satellite) that are distant from the * Radio links (cellular, Wi-Fi, or satellite) that are distant from
source are particularly challenging. The radio link capacity can the source are particularly challenging. The radio link capacity
vary rapidly by orders of magnitude, so it is considered desirable can vary rapidly by orders of magnitude, so it is considered
to hold a standing queue that can utilize sudden increases of desirable to hold a standing queue that can utilize sudden
capacity; increases of capacity.
* Cellular networks are further complicated by a perceived need to * Cellular networks are further complicated by a perceived need to
buffer in order to make hand-overs imperceptible; buffer in order to make hand-overs imperceptible.
L4S cannot remove the need for all these different forms of L4S cannot remove the need for all these different forms of
buffering. However, by removing 'the longest pole in the tent' buffering. However, by removing 'the longest pole in the tent'
(buffering for the large sawteeth of Classic congestion controls), (buffering for the large sawteeth of Classic congestion controls),
L4S exposes all these 'shorter poles' to greater scrutiny. L4S exposes all these 'shorter poles' to greater scrutiny.
Until now, the buffering needed for these additional reasons tended Until now, the buffering needed for these additional reasons tended
to be over-specified - with the excuse that none were 'the longest to be over-specified -- with the excuse that none were 'the longest
pole in the tent'. But having removed the 'longest pole', it becomes pole in the tent'. But having removed the 'longest pole', it becomes
worthwhile to minimize them, for instance reducing packet aggregation worthwhile to minimize them, for instance, reducing packet
burst sizes and MAC scheduling intervals. aggregation burst sizes and MAC scheduling intervals.
Also certain link types, particularly radio-based links, are far more Also, certain link types, particularly radio-based links, are far
prone to transmission losses. Section 6.4.3 explains how an L4S more prone to transmission losses. Section 6.4.3 explains how an L4S
response to loss has to be as drastic as a Classic response. response to loss has to be as drastic as a Classic response.
Nonetheless, research referred to in the same section has Nonetheless, research referred to in the same section has
demonstrated potential for considerably more effective loss repair at demonstrated potential for considerably more effective loss repair at
the link layer, due to the relaxed ordering constraints of L4S the link layer, due to the relaxed ordering constraints of L4S
packets. packets.
6.4. Deployment Considerations 6.4. Deployment Considerations
L4S AQMs, whether DualQ [I-D.ietf-tsvwg-aqm-dualq-coupled] or FQ, L4S AQMs, whether DualQ [RFC9332] or FQ [RFC8290], are in themselves
e.g. [RFC8290] are, in themselves, an incremental deployment an incremental deployment mechanism for L4S -- so that L4S traffic
mechanism for L4S - so that L4S traffic can coexist with existing can coexist with existing Classic (Reno-friendly) traffic.
Classic (Reno-friendly) traffic. Section 6.4.1 explains why only Section 6.4.1 explains why only deploying an L4S AQM in one node at
deploying an L4S AQM in one node at each end of the access link will each end of the access link will realize nearly all the benefit of
realize nearly all the benefit of L4S. L4S.
L4S involves both end systems and the network, so Section 6.4.2 L4S involves both the network and end systems, so Section 6.4.2
suggests some typical sequences to deploy each part, and why there suggests some typical sequences to deploy each part and why there
will be an immediate and significant benefit after deploying just one will be an immediate and significant benefit after deploying just one
part. part.
Section 6.4.3 and Section 6.4.4 describe the converse incremental Sections 6.4.3 and 6.4.4 describe the converse incremental deployment
deployment case where there is no L4S AQM at the network bottleneck, case where there is no L4S AQM at the network bottleneck, so any L4S
so any L4S flow traversing this bottleneck has to take care in case flow traversing this bottleneck has to take care in case it is
it is competing with Classic traffic. competing with Classic traffic.
6.4.1. Deployment Topology 6.4.1. Deployment Topology
L4S AQMs will not have to be deployed throughout the Internet before L4S AQMs will not have to be deployed throughout the Internet before
L4S can benefit anyone. Operators of public Internet access networks L4S can benefit anyone. Operators of public Internet access networks
typically design their networks so that the bottleneck will nearly typically design their networks so that the bottleneck will nearly
always occur at one known (logical) link. This confines the cost of always occur at one known (logical) link. This confines the cost of
queue management technology to one place. queue management technology to one place.
The case of mesh networks is different and will be discussed later in The case of mesh networks is different and will be discussed later in
this section. But the known bottleneck case is generally true for this section. However, the known-bottleneck case is generally true
Internet access to all sorts of different 'sites', where the word for Internet access to all sorts of different 'sites', where the word
'site' includes home networks, small- to medium-sized campus or 'site' includes home networks, small- to medium-sized campus or
enterprise networks and even cellular devices (Figure 2). Also, this enterprise networks and even cellular devices (Figure 2). Also, this
known-bottleneck case tends to be applicable whatever the access link known-bottleneck case tends to be applicable whatever the access link
technology; whether xDSL, cable, PON, cellular, line of sight technology, whether xDSL, cable, PON, cellular, line of sight
wireless or satellite. wireless, or satellite.
Therefore, the full benefit of the L4S service should be available in Therefore, the full benefit of the L4S service should be available in
the downstream direction when an L4S AQM is deployed at the ingress the downstream direction when an L4S AQM is deployed at the ingress
to this bottleneck link. And similarly, the full upstream service to this bottleneck link. And similarly, the full upstream service
will be available once an L4S AQM is deployed at the ingress into the will typically be available once an L4S AQM is deployed at the
upstream link. (Of course, multi-homed sites would only see the full ingress into the upstream link. (Of course, multihomed sites would
benefit once all their access links were covered.) only see the full benefit once all their access links were covered.)
______ ______
( ) ( )
__ __ ( ) __ __ ( )
|DQ\________/DQ|( enterprise ) |DQ\________/DQ|( enterprise )
___ |__/ \__| ( /campus ) ___ |__/ \__| ( /campus )
( ) (______) ( ) (______)
( ) ___||_ ( ) ___||_
+----+ ( ) __ __ / \ +----+ ( ) __ __ / \
| DC |-----( Core )|DQ\_______________/DQ|| home | | DC |-----( Core )|DQ\_______________/DQ|| home |
+----+ ( ) |__/ \__||______| +----+ ( ) |__/ \__||______|
(_____) __ (_____) __
|DQ\__/\ __ ,===. |DQ\__/\ __ ,===.
|__/ \ ____/DQ||| ||mobile |__/ \ ____/DQ||| ||mobile
\/ \__|||_||device \/ \__|||_||device
| o | | o |
`---' `---'
Figure 2: Likely location of DualQ (DQ) Deployments in common Figure 2: Likely Location of DualQ (DQ) Deployments in Common
access topologies Access Topologies
Deployment in mesh topologies depends on how overbooked the core is. Deployment in mesh topologies depends on how overbooked the core is.
If the core is non-blocking, or at least generously provisioned so If the core is non-blocking, or at least generously provisioned so
that the edges are nearly always the bottlenecks, it would only be that the edges are nearly always the bottlenecks, it would only be
necessary to deploy an L4S AQM at the edge bottlenecks. For example, necessary to deploy an L4S AQM at the edge bottlenecks. For example,
some data-centre networks are designed with the bottleneck in the some data-centre networks are designed with the bottleneck in the
hypervisor or host NICs, while others bottleneck at the top-of-rack hypervisor or host Network Interface Controllers (NICs), while others
switch (both the output ports facing hosts and those facing the bottleneck at the top-of-rack switch (both the output ports facing
core). hosts and those facing the core).
An L4S AQM would often next be needed where the Wi-Fi links in a home An L4S AQM would often next be needed where the Wi-Fi links in a home
sometimes become the bottleneck. And an L4S AQM would eventually sometimes become the bottleneck. Also an L4S AQM would eventually
also need to be deployed at any other persistent bottlenecks such as need to be deployed at any other persistent bottlenecks, such as
network interconnections, e.g. some public Internet exchange points network interconnections, e.g., some public Internet exchange points
and the ingress and egress to WAN links interconnecting data-centres. and the ingress and egress to WAN links interconnecting data centres.
6.4.2. Deployment Sequences 6.4.2. Deployment Sequences
For any one L4S flow to provide benefit, it requires three (or For any one L4S flow to provide benefit, it requires three (or
sometimes two) parts to have been deployed: i) the congestion control sometimes two) parts to have been deployed: i) the congestion control
at the sender; ii) the AQM at the bottleneck; and iii) older at the sender; ii) the AQM at the bottleneck; and iii) older
transports (namely TCP) need upgraded receiver feedback too. This transports (namely TCP) need upgraded receiver feedback too. This
was the same deployment problem that ECN faced [RFC8170] so we have was the same deployment problem that ECN faced [RFC8170], so we have
learned from that experience. learned from that experience.
Firstly, L4S deployment exploits the fact that DCTCP already exists Firstly, L4S deployment exploits the fact that DCTCP already exists
on many Internet hosts (Windows, FreeBSD and Linux); both servers and on many Internet hosts (e.g., Windows, FreeBSD, and Linux), both
clients. Therefore, an L4S AQM can be deployed at a network servers and clients. Therefore, an L4S AQM can be deployed at a
bottleneck to immediately give a working deployment of all the L4S network bottleneck to immediately give a working deployment of all
parts for testing, as long as the ECT(0) codepoint is switched to the L4S parts for testing, as long as the ECT(0) codepoint is
ECT(1). DCTCP needs some safety concerns to be fixed for general use switched to ECT(1). DCTCP needs some safety concerns to be fixed for
over the public Internet (see Section 4.3 of the L4S ECN general use over the public Internet (see Section 4.3 of the L4S ECN
spec [I-D.ietf-tsvwg-ecn-l4s-id]), but DCTCP is not on by default, so spec [RFC9331]), but DCTCP is not on by default, so these issues can
these issues can be managed within controlled deployments or be managed within controlled deployments or controlled trials.
controlled trials.
Secondly, the performance improvement with L4S is so significant that Secondly, the performance improvement with L4S is so significant that
it enables new interactive services and products that were not it enables new interactive services and products that were not
previously possible. It is much easier for companies to initiate new previously possible. It is much easier for companies to initiate new
work on deployment if there is budget for a new product trial. If, work on deployment if there is budget for a new product trial. In
in contrast, there were only an incremental performance improvement contrast, if there were only an incremental performance improvement
(as with Classic ECN), spending on deployment tends to be much harder (as with Classic ECN), spending on deployment tends to be much harder
to justify. to justify.
Thirdly, the L4S identifier is defined so that initially network Thirdly, the L4S identifier is defined so that network operators can
operators can enable L4S exclusively for certain customers or certain initially enable L4S exclusively for certain customers or certain
applications. But this is carefully defined so that it does not applications. However, this is carefully defined so that it does not
compromise future evolution towards L4S as an Internet-wide service. compromise future evolution towards L4S as an Internet-wide service.
This is because the L4S identifier is defined not only as the end-to- This is because the L4S identifier is defined not only as the end-to-
end ECN field, but it can also optionally be combined with any other end ECN field, but it can also optionally be combined with any other
packet header or some status of a customer or their access link (see packet header or some status of a customer or their access link (see
section 5.4 of [I-D.ietf-tsvwg-ecn-l4s-id]). Operators could do this Section 5.4 of [RFC9331]). Operators could do this anyway, even if
anyway, even if it were not blessed by the IETF. However, it is best it were not blessed by the IETF. However, it is best for the IETF to
for the IETF to specify that, if they use their own local identifier, specify that, if they use their own local identifier, it must be in
it must be in combination with the IETF's identifier. Then, if an combination with the IETF's identifier, ECT(1). Then, if an operator
operator has opted for an exclusive local-use approach, later they has opted for an exclusive local-use approach, they only have to
only have to remove this extra rule to make the service work remove this extra rule later to make the service work across the
Internet-wide - it will already traverse middleboxes, peerings, etc. Internet -- it will already traverse middleboxes, peerings, etc.
+-+--------------------+----------------------+---------------------+ +-+--------------------+----------------------+---------------------+
| | Servers or proxies | Access link | Clients | | | Servers or proxies | Access link | Clients |
+-+--------------------+----------------------+---------------------+ +-+--------------------+----------------------+---------------------+
|0| DCTCP (existing) | | DCTCP (existing) | |0| DCTCP (existing) | | DCTCP (existing) |
+-+--------------------+----------------------+---------------------+ +-+--------------------+----------------------+---------------------+
|1| |Add L4S AQM downstream| | |1| |Add L4S AQM downstream| |
| | WORKS DOWNSTREAM FOR CONTROLLED DEPLOYMENTS/TRIALS | | | WORKS DOWNSTREAM FOR CONTROLLED DEPLOYMENTS/TRIALS |
+-+--------------------+----------------------+---------------------+ +-+--------------------+----------------------+---------------------+
|2| Upgrade DCTCP to | |Replace DCTCP feedb'k| |2| Upgrade DCTCP to | |Replace DCTCP feedb'k|
skipping to change at page 28, line 4 skipping to change at line 1269
+-+--------------------+----------------------+---------------------+ +-+--------------------+----------------------+---------------------+
|2| Upgrade DCTCP to | |Replace DCTCP feedb'k| |2| Upgrade DCTCP to | |Replace DCTCP feedb'k|
| | TCP Prague | | with AccECN | | | TCP Prague | | with AccECN |
| | FULLY WORKS DOWNSTREAM | | | FULLY WORKS DOWNSTREAM |
+-+--------------------+----------------------+---------------------+ +-+--------------------+----------------------+---------------------+
| | | | Upgrade DCTCP to | | | | | Upgrade DCTCP to |
|3| | Add L4S AQM upstream | TCP Prague | |3| | Add L4S AQM upstream | TCP Prague |
| | | | | | | | | |
| | FULLY WORKS UPSTREAM AND DOWNSTREAM | | | FULLY WORKS UPSTREAM AND DOWNSTREAM |
+-+--------------------+----------------------+---------------------+ +-+--------------------+----------------------+---------------------+
Figure 3: Example L4S Deployment Sequence Figure 3: Example L4S Deployment Sequence
Figure 3 illustrates some example sequences in which the parts of L4S Figure 3 illustrates some example sequences in which the parts of L4S
might be deployed. It consists of the following stages, preceded by might be deployed. It consists of the following stages, preceded by
a presumption that DCTCP is already installed at both ends: a presumption that DCTCP is already installed at both ends:
1. DCTCP is not applicable for use over the public Internet, so it 1. DCTCP is not applicable for use over the public Internet, so it
is emphasized here that any DCTCP flow has to be completely is emphasized here that any DCTCP flow has to be completely
contained within a controlled trial environment. contained within a controlled trial environment.
Within this trial environment, once an L4S AQM has been deployed, Within this trial environment, once an L4S AQM has been deployed,
the trial DCTCP flow will experience immediate benefit, without the trial DCTCP flow will experience immediate benefit, without
any other deployment being needed. In this example downstream any other deployment being needed. In this example, downstream
deployment is first, but in other scenarios the upstream might be deployment is first, but in other scenarios, the upstream might
deployed first. If no AQM at all was previously deployed for the be deployed first. If no AQM at all was previously deployed for
downstream access, an L4S AQM greatly improves the Classic the downstream access, an L4S AQM greatly improves the Classic
service (as well as adding the L4S service). If an AQM was service (as well as adding the L4S service). If an AQM was
already deployed, the Classic service will be unchanged (and L4S already deployed, the Classic service will be unchanged (and L4S
will add an improvement on top). will add an improvement on top).
2. In this stage, the name 'TCP 2. In this stage, the name 'TCP Prague' [PRAGUE-CC] is used to
Prague' [I-D.briscoe-iccrg-prague-congestion-control] is used to
represent a variant of DCTCP that is designed to be used in a represent a variant of DCTCP that is designed to be used in a
production Internet environment (that is, it has to comply with production Internet environment (that is, it has to comply with
all the requirements in Section 4 of the L4S ECN all the requirements in Section 4 of the L4S ECN spec [RFC9331],
spec [I-D.ietf-tsvwg-ecn-l4s-id], which then means it can be used which then means it can be used over the public Internet). If
over the public Internet). If the application is primarily the application is primarily unidirectional, 'TCP Prague' at the
unidirectional, 'TCP Prague' at one end will provide all the sending end will provide all the benefit needed, as long as the
benefit needed. receiving end supports Accurate ECN (AccECN) feedback [ACCECN].
For TCP transports, Accurate ECN feedback For TCP transports, AccECN feedback is needed at the other end,
(AccECN) [I-D.ietf-tcpm-accurate-ecn] is needed at the other end,
but it is a generic ECN feedback facility that is already planned but it is a generic ECN feedback facility that is already planned
to be deployed for other purposes, e.g. DCTCP, BBR. The two ends to be deployed for other purposes, e.g., DCTCP and BBR. The two
can be deployed in either order, because, in TCP, an L4S ends can be deployed in either order because, in TCP, an L4S
congestion control only enables itself if it has negotiated the congestion control only enables itself if it has negotiated the
use of AccECN feedback with the other end during the connection use of AccECN feedback with the other end during the connection
handshake. Thus, deployment of TCP Prague on a server enables handshake. Thus, deployment of TCP Prague on a server enables
L4S trials to move to a production service in one direction, L4S trials to move to a production service in one direction,
wherever AccECN is deployed at the other end. This stage might wherever AccECN is deployed at the other end. This stage might
be further motivated by the performance improvements of TCP be further motivated by the performance improvements of TCP
Prague relative to DCTCP (see Appendix A.2 of the L4S ECN Prague relative to DCTCP (see Appendix A.2 of the L4S ECN spec
spec [I-D.ietf-tsvwg-ecn-l4s-id]). [RFC9331]).
Unlike TCP, from the outset, QUIC ECN feedback [RFC9000] has Unlike TCP, from the outset, QUIC ECN feedback [RFC9000] has
supported L4S. Therefore, if the transport is QUIC, one-ended supported L4S. Therefore, if the transport is QUIC, one-ended
deployment of a Prague congestion control at this stage is simple deployment of a Prague congestion control at this stage is simple
and sufficient. and sufficient.
For QUIC, if a proxy sits in the path between multiple origin For QUIC, if a proxy sits in the path between multiple origin
servers and the access bottlenecks to multiple clients, then servers and the access bottlenecks to multiple clients, then
upgrading the proxy with a Scalable congestion control would upgrading the proxy with a Scalable congestion control would
provide the benefits of L4S over all the clients' downstream provide the benefits of L4S over all the clients' downstream
bottlenecks in one go --- whether or not all the origin servers bottlenecks in one go -- whether or not all the origin servers
were upgraded. Conversely, where a proxy has not been upgraded, were upgraded. Conversely, where a proxy has not been upgraded,
the clients served by it will not benefit from L4S at all in the the clients served by it will not benefit from L4S at all in the
downstream, even when any origin server behind the proxy has been downstream, even when any origin server behind the proxy has been
upgraded to support L4S. upgraded to support L4S.
For TCP, a proxy upgraded to support 'TCP Prague' would provide For TCP, a proxy upgraded to support 'TCP Prague' would provide
the benefits of L4S downstream to all clients that support AccECN the benefits of L4S downstream to all clients that support AccECN
(whether or not they support L4S as well). And in the upstream, (whether or not they support L4S as well). And in the upstream,
the proxy would also support AccECN as a receiver, so that any the proxy would also support AccECN as a receiver, so that any
client deploying its own L4S support would benefit in the client deploying its own L4S support would benefit in the
upstream direction, irrespective of whether any origin server upstream direction, irrespective of whether any origin server
beyond the proxy supported AccECN. beyond the proxy supported AccECN.
3. This is a two-move stage to enable L4S upstream. An L4S AQM or 3. This is a two-move stage to enable L4S upstream. An L4S AQM or
TCP Prague can be deployed in either order as already explained. TCP Prague can be deployed in either order as already explained.
To motivate the first of two independent moves, the deferred To motivate the first of two independent moves, the deferred
benefit of enabling new services after the second move has to be benefit of enabling new services after the second move has to be
worth it to cover the first mover's investment risk. As worth it to cover the first mover's investment risk. As
explained already, the potential for new interactive services explained already, the potential for new interactive services
provides this motivation. An L4S AQM also improves the upstream provides this motivation. An L4S AQM also improves the upstream
Classic service - significantly if no other AQM has already been Classic service significantly if no other AQM has already been
deployed. deployed.
Note that other deployment sequences might occur. For instance: the Note that other deployment sequences might occur. For instance, the
upstream might be deployed first; a non-TCP protocol might be used upstream might be deployed first; a non-TCP protocol might be used
end-to-end, e.g. QUIC, RTP; a body such as the 3GPP might require L4S end to end, e.g., QUIC and RTP; a body, such as the 3GPP, might
to be implemented in 5G user equipment, or other random acts of require L4S to be implemented in 5G user equipment; or other random
kindness. acts of kindness might arise.
6.4.3. L4S Flow but Non-ECN Bottleneck 6.4.3. L4S Flow but Non-ECN Bottleneck
If L4S is enabled between two hosts, the L4S sender is required to If L4S is enabled between two hosts, the L4S sender is required to
coexist safely with Reno in response to any drop (see Section 4.3 of coexist safely with Reno in response to any drop (see Section 4.3 of
the L4S ECN spec [I-D.ietf-tsvwg-ecn-l4s-id]). the L4S ECN spec [RFC9331]).
Unfortunately, as well as protecting Classic traffic, this rule Unfortunately, as well as protecting Classic traffic, this rule
degrades the L4S service whenever there is any loss, even if the degrades the L4S service whenever there is any loss, even if the
cause is not persistent congestion at a bottleneck, e.g.: cause is not persistent congestion at a bottleneck, for example:
* congestion loss at other transient bottlenecks, e.g. due to bursts * congestion loss at other transient bottlenecks, e.g., due to
in shallower queues; bursts in shallower queues;
* transmission errors, e.g., due to electrical interference; and
* transmission errors, e.g. due to electrical interference;
* rate policing. * rate policing.
Three complementary approaches are in progress to address this issue, Three complementary approaches are in progress to address this issue,
but they are all currently research: but they are all currently research:
* In Prague congestion control, ignore certain losses deemed * In Prague congestion control, ignore certain losses deemed
unlikely to be due to congestion (using some ideas from unlikely to be due to congestion (using some ideas from BBR
BBR [I-D.cardwell-iccrg-bbr-congestion-control] regarding isolated [BBR-CC] regarding isolated losses). This could mask any of the
losses). This could mask any of the above types of loss while above types of loss while still coexisting with drop-based
still coexisting with drop-based congestion controls. congestion controls.
* A combination of RACK, L4S and link retransmission without * A combination of Recent Acknowledgement (RACK) [RFC8985], L4S, and
resequencing could repair transmission errors without the head of link retransmission without resequencing could repair transmission
line blocking delay usually associated with link-layer errors without the head of line blocking delay usually associated
retransmission [UnorderedLTE], [I-D.ietf-tsvwg-ecn-l4s-id]; with link-layer retransmission [UnorderedLTE] [RFC9331].
* Hybrid ECN/drop rate policers (see Section 8.3). * Hybrid ECN/drop rate policers (see Section 8.3).
L4S deployment scenarios that minimize these issues (e.g. over L4S deployment scenarios that minimize these issues (e.g., over
wireline networks) can proceed in parallel to this research, in the wireline networks) can proceed in parallel to this research, in the
expectation that research success could continually widen L4S expectation that research success could continually widen L4S
applicability. applicability.
6.4.4. L4S Flow but Classic ECN Bottleneck 6.4.4. L4S Flow but Classic ECN Bottleneck
Classic ECN support is starting to materialize on the Internet as an Classic ECN support is starting to materialize on the Internet as an
increased level of CE marking. It is hard to detect whether this is increased level of CE marking. It is hard to detect whether this is
all due to the addition of support for ECN in implementations of FQ- all due to the addition of support for ECN in implementations of FQ-
CoDel and/or FQ-COBALT, which is not generally problematic, because CoDel and/or FQ-COBALT, which is not generally problematic, because
flow-queue (FQ) scheduling inherently prevents a flow from exceeding flow queue (FQ) scheduling inherently prevents a flow from exceeding
the 'fair' rate irrespective of its aggressiveness. However, some of the 'fair' rate irrespective of its aggressiveness. However, some of
this Classic ECN marking might be due to single-queue ECN deployment. this Classic ECN marking might be due to single-queue ECN deployment.
This case is discussed in Section 4.3 of the L4S ECN This case is discussed in Section 4.3 of the L4S ECN spec [RFC9331].
spec [I-D.ietf-tsvwg-ecn-l4s-id].
6.4.5. L4S AQM Deployment within Tunnels 6.4.5. L4S AQM Deployment within Tunnels
An L4S AQM uses the ECN field to signal congestion. So, in common An L4S AQM uses the ECN field to signal congestion. So in common
with Classic ECN, if the AQM is within a tunnel or at a lower layer, with Classic ECN, if the AQM is within a tunnel or at a lower layer,
correct functioning of ECN signalling requires correct propagation of correct functioning of ECN signalling requires standards-compliant
the ECN field up the layers [RFC6040], propagation of the ECN field up the layers [RFC6040] [ECN-SHIM]
[I-D.ietf-tsvwg-rfc6040update-shim], [ECN-ENCAP].
[I-D.ietf-tsvwg-ecn-encap-guidelines].
7. IANA Considerations (to be removed by RFC Editor) 7. IANA Considerations
This specification contains no IANA considerations. This document has no IANA actions.
8. Security Considerations 8. Security Considerations
8.1. Traffic Rate (Non-)Policing 8.1. Traffic Rate (Non-)Policing
8.1.1. (Non-)Policing Rate per Flow 8.1.1. (Non-)Policing Rate per Flow
In the current Internet, ISPs usually enforce separation between the In the current Internet, ISPs usually enforce separation between the
capacity of shared links assigned to different 'sites' capacity of shared links assigned to different 'sites' (e.g.,
(e.g. households, businesses or mobile users - see terminology in households, businesses, or mobile users -- see terminology in
Section 3) using some form of scheduler [RFC0970]. And they use Section 3) using some form of scheduler [RFC0970]. And they use
various techniques like redirection to traffic scrubbing facilities various techniques, like redirection to traffic scrubbing facilities,
to deal with flooding attacks. However, there has never been a to deal with flooding attacks. However, there has never been a
universal need to police the rate of individual application flows - universal need to police the rate of individual application flows --
the Internet has generally always relied on self-restraint of the Internet has generally always relied on self-restraint of
congestion controls at senders for sharing intra-'site' capacity. congestion controls at senders for sharing intra-'site' capacity.
L4S has been designed not to upset this status quo. If a DualQ is L4S has been designed not to upset this status quo. If a DualQ is
used to provide L4S service, section 4.2 of used to provide L4S service, Section 4.2 of [RFC9332] explains how it
[I-D.ietf-tsvwg-aqm-dualq-coupled] explains how it is designed to is designed to give no more rate advantage to unresponsive flows than
give no more rate advantage to unresponsive flows than a single-queue a single-queue AQM would, whether or not there is traffic overload.
AQM would, whether or not there is traffic overload.
Also, in case per-flow rate policing is ever required, it can be Also, in case per-flow rate policing is ever required, it can be
added because it is orthogonal to the distinction between L4S and added because it is orthogonal to the distinction between L4S and
Classic. As explained in Section 5.2, the DualQ variant of L4S Classic. As explained in Section 5.2, the DualQ variant of L4S
provides low delay without prejudging the issue of flow-rate control. provides low delay without prejudging the issue of flow-rate control.
So, if flow-rate control is needed, per-flow-queuing (FQ) with L4S So if flow-rate control is needed, per-flow queuing (FQ) with L4S
support can be used instead, or flow rate policing can be added as a support can be used instead, or flow rate policing can be added as a
modular addition to a DualQ. However, per-flow rate control is not modular addition to a DualQ. However, per-flow rate control is not
usually deployed as a security mechanism, because an active attacker usually deployed as a security mechanism, because an active attacker
can just shard its traffic over more flow IDs if the rate of each is can just shard its traffic over more flow identifiers if the rate of
restricted. each is restricted.
8.1.2. (Non-)Policing L4S Service Rate 8.1.2. (Non-)Policing L4S Service Rate
Section 5.2 explains how Diffserv only makes a difference if some Section 5.2 explains how Diffserv only makes a difference if some
packets get less favourable treatment than others, which typically packets get less favourable treatment than others, which typically
requires traffic rate policing for a low latency class. In contrast, requires traffic rate policing for a low latency class. In contrast,
it should not be necessary to rate-police access to the L4S service it should not be necessary to rate-police access to the L4S service
to protect the Classic service, because L4S is designed to reduce to protect the Classic service, because L4S is designed to reduce
delay without harming the delay or rate of any Classic traffic. delay without harming the delay or rate of any Classic traffic.
During early deployment (and perhaps always), some networks will not During early deployment (and perhaps always), some networks will not
offer the L4S service. In general, these networks should not need to offer the L4S service. In general, these networks should not need to
police L4S traffic. They are required (by both the ECN police L4S traffic. They are required (by both the ECN spec
spec [RFC3168] and the L4S ECN spec [I-D.ietf-tsvwg-ecn-l4s-id]) not [RFC3168] and the L4S ECN spec [RFC9331]) not to change the L4S
to change the L4S identifier, which would interfere with end-to-end identifier, which would interfere with end-to-end congestion control.
congestion control. If they already treat ECN traffic as Not-ECT, If they already treat ECN traffic as Not-ECT, they can merely treat
they can merely treat L4S traffic as Not-ECT too. At a bottleneck, L4S traffic as Not-ECT too. At a bottleneck, such networks will
such networks will introduce some queuing and dropping. When a introduce some queuing and dropping. When a Scalable congestion
scalable congestion control detects a drop it will have to respond control detects a drop, it will have to respond safely with respect
safely with respect to Classic congestion controls (as required in to Classic congestion controls (as required in Section 4.3 of
Section 4.3 of [I-D.ietf-tsvwg-ecn-l4s-id]). This will degrade the [RFC9331]). This will degrade the L4S service to be no better (but
L4S service to be no better (but never worse) than Classic best never worse) than Classic best efforts whenever a non-ECN bottleneck
efforts, whenever a non-ECN bottleneck is encountered on a path (see is encountered on a path (see Section 6.4.3).
Section 6.4.3).
In cases that are expected to be rare, networks that solely support In cases that are expected to be rare, networks that solely support
Classic ECN [RFC3168] in a single queue bottleneck might opt to Classic ECN [RFC3168] in a single queue bottleneck might opt to
police L4S traffic so as to protect competing Classic ECN traffic police L4S traffic so as to protect competing Classic ECN traffic
(for instance, see Section 6.1.3 of the L4S operational (for instance, see Section 6.1.3 of the L4S operational guidance
guidance [I-D.ietf-tsvwg-l4sops]). However, Section 4.3 of the L4S [L4SOPS]). However, Section 4.3 of the L4S ECN spec [RFC9331]
ECN spec [I-D.ietf-tsvwg-ecn-l4s-id] recommends that the sender recommends that the sender adapts its congestion response to properly
adapts its congestion response to properly coexist with Classic ECN coexist with Classic ECN flows, i.e., reverting to the self-restraint
flows, i.e. reverting to the self-restraint approach. approach.
Certain network operators might choose to restrict access to the L4S Certain network operators might choose to restrict access to the L4S
service, perhaps only to selected premium customers as a value-added service, perhaps only to selected premium customers as a value-added
service. Their packet classifier (item 2 in Figure 1) could identify service. Their packet classifier (item 2 in Figure 1) could identify
such customers against some other field (e.g. source address range) such customers against some other field (e.g., source address range),
as well as classifying on the ECN field. If only the ECN L4S as well as classifying on the ECN field. If only the ECN L4S
identifier matched, but not the source address (say), the classifier identifier matched, but not (say) the source address, the classifier
could direct these packets (from non-premium customers) into the could direct these packets (from non-premium customers) into the
Classic queue. Explaining clearly how operators can use additional Classic queue. Explaining clearly how operators can use additional
local classifiers (see section 5.4 of the L4S ECN local classifiers (see Section 5.4 of [RFC9331]) is intended to
spec [I-D.ietf-tsvwg-ecn-l4s-id]) is intended to remove any remove any motivation to clear the L4S identifier. Then at least the
motivation to clear the L4S identifier. Then at least the L4S ECN L4S ECN identifier will be more likely to survive end to end, even
identifier will be more likely to survive end-to-end even though the though the service may not be supported at every hop. Such local
service may not be supported at every hop. Such local arrangements arrangements would only require simple registered/not-registered
would only require simple registered/not-registered packet packet classification, rather than the managed, application-specific
classification, rather than the managed, application-specific traffic traffic policing against customer-specific traffic contracts that
policing against customer-specific traffic contracts that Diffserv Diffserv uses.
uses.
8.2. 'Latency Friendliness' 8.2. 'Latency Friendliness'
Like the Classic service, the L4S service relies on self-restraint - Like the Classic service, the L4S service relies on self-restraint to
limiting rate in response to congestion. In addition, the L4S limit the rate in response to congestion. In addition, the L4S
service requires self-restraint in terms of limiting latency service requires self-restraint in terms of limiting latency
(burstiness). It is hoped that self-interest and guidance on dynamic (burstiness). It is hoped that self-interest and guidance on dynamic
behaviour (especially flow start-up, which might need to be behaviour (especially flow start-up, which might need to be
standardized) will be sufficient to prevent transports from sending standardized) will be sufficient to prevent transports from sending
excessive bursts of L4S traffic, given the application's own latency excessive bursts of L4S traffic, given the application's own latency
will suffer most from such behaviour. will suffer most from such behaviour.
Because the L4S service can reduce delay without discernibly Because the L4S service can reduce delay without discernibly
increasing the delay of any Classic traffic, it should not be increasing the delay of any Classic traffic, it should not be
necessary to police L4S traffic to protect the delay of Classic. necessary to police L4S traffic to protect the delay of Classic
However, whether burst policing becomes necessary to protect other traffic. However, whether burst policing becomes necessary to
L4S traffic remains to be seen. Without it, there will be potential protect other L4S traffic remains to be seen. Without it, there will
for attacks on the low latency of the L4S service. be potential for attacks on the low latency of the L4S service.
If needed, various arrangements could be used to address this If needed, various arrangements could be used to address this
concern: concern:
Local bottleneck queue protection: A per-flow (5-tuple) queue Local bottleneck queue protection: A per-flow (5-tuple) queue
protection function [I-D.briscoe-docsis-q-protection] has been protection function [DOCSIS-Q-PROT] has been developed for the low
developed for the low latency queue in DOCSIS, which has adopted latency queue in DOCSIS, which has adopted the DualQ L4S
the DualQ L4S architecture. It protects the low latency service architecture. It protects the low latency service from any queue-
from any queue-building flows that accidentally or maliciously building flows that accidentally or maliciously classify
classify themselves into the low latency queue. It is designed to themselves into the low latency queue. It is designed to score
score flows based solely on their contribution to queuing (not flows based solely on their contribution to queuing (not flow rate
flow rate in itself). Then, if the shared low latency queue is at in itself). Then, if the shared low latency queue is at risk of
risk of exceeding a threshold, the function redirects enough exceeding a threshold, the function redirects enough packets of
packets of the highest scoring flow(s) into the Classic queue to the highest scoring flow(s) into the Classic queue to preserve low
preserve low latency. latency.
Distributed traffic scrubbing: Rather than policing locally at each Distributed traffic scrubbing: Rather than policing locally at each
bottleneck, it may only be necessary to address problems bottleneck, it may only be necessary to address problems
reactively, e.g. punitively target any deployments of new bursty reactively, e.g., punitively target any deployments of new bursty
malware, in a similar way to how traffic from flooding attack malware, in a similar way to how traffic from flooding attack
sources is rerouted via scrubbing facilities. sources is rerouted via scrubbing facilities.
Local bottleneck per-flow scheduling: Per-flow scheduling should Local bottleneck per-flow scheduling: Per-flow scheduling should
inherently isolate non-bursty flows from bursty (see Section 5.2 inherently isolate non-bursty flows from bursty flows (see
for discussion of the merits of per-flow scheduling relative to Section 5.2 for discussion of the merits of per-flow scheduling
per-flow policing). relative to per-flow policing).
Distributed access subnet queue protection: Per-flow queue Distributed access subnet queue protection: Per-flow queue
protection could be arranged for a queue structure distributed protection could be arranged for a queue structure distributed
across a subnet intercommunicating using lower layer control across a subnet intercommunicating using lower layer control
messages (see Section 2.1.4 of [QDyn]). For instance, in a radio messages (see Section 2.1.4 of [QDyn]). For instance, in a radio
access network, user equipment already sends regular buffer status access network, user equipment already sends regular buffer status
reports to a radio network controller, which could use this reports to a radio network controller, which could use this
information to remotely police individual flows. information to remotely police individual flows.
Distributed Congestion Exposure to Ingress Policers: The Congestion Distributed Congestion Exposure to ingress policers: The Congestion
Exposure (ConEx) architecture [RFC7713] uses egress audit to Exposure (ConEx) architecture [RFC7713] uses an egress audit to
motivate senders to truthfully signal path congestion in-band motivate senders to truthfully signal path congestion in-band,
where it can be used by ingress policers. An edge-to-edge variant where it can be used by ingress policers. An edge-to-edge variant
of this architecture is also possible. of this architecture is also possible.
Distributed Domain-edge traffic conditioning: An architecture Distributed domain-edge traffic conditioning: An architecture
similar to Diffserv [RFC2475] may be preferred, where traffic is similar to Diffserv [RFC2475] may be preferred, where traffic is
proactively conditioned on entry to a domain, rather than proactively conditioned on entry to a domain, rather than
reactively policed only if it leads to queuing once combined with reactively policed only if it leads to queuing once combined with
other traffic at a bottleneck. other traffic at a bottleneck.
Distributed core network queue protection: The policing function Distributed core network queue protection: The policing function
could be divided between per-flow mechanisms at the network could be divided between per-flow mechanisms at the network
ingress that characterize the burstiness of each flow into a ingress that characterize the burstiness of each flow into a
signal carried with the traffic, and per-class mechanisms at signal carried with the traffic and per-class mechanisms at
bottlenecks that act on these signals if queuing actually occurs bottlenecks that act on these signals if queuing actually occurs
once the traffic converges. This would be somewhat similar to once the traffic converges. This would be somewhat similar to
[Nadas20], which is in turn similar to the idea behind core [Nadas20], which is in turn similar to the idea behind core
stateless fair queuing. stateless fair queuing.
No single one of these possible queue protection capabilities is No single one of these possible queue protection capabilities is
considered an essential part of the L4S architecture, which works considered an essential part of the L4S architecture, which works
without any of them under non-attack conditions (much as the Internet without any of them under non-attack conditions (much as the Internet
normally works without per-flow rate policing). Indeed, even where normally works without per-flow rate policing). Indeed, even where
latency policers are deployed, under normal circumstances they would latency policers are deployed, under normal circumstances, they would
not intervene, and if operators found they were not necessary they not intervene, and if operators found they were not necessary, they
could disable them. Part of the L4S experiment will be to see could disable them. Part of the L4S experiment will be to see
whether such a function is necessary, and which arrangements are most whether such a function is necessary and which arrangements are most
appropriate to the size of the problem. appropriate to the size of the problem.
8.3. Interaction between Rate Policing and L4S 8.3. Interaction between Rate Policing and L4S
As mentioned in Section 5.2, L4S should remove the need for low As mentioned in Section 5.2, L4S should remove the need for low
latency Diffserv classes. However, those Diffserv classes that give latency Diffserv classes. However, those Diffserv classes that give
certain applications or users priority over capacity, would still be certain applications or users priority over capacity would still be
applicable in certain scenarios (e.g. corporate networks). Then, applicable in certain scenarios (e.g., corporate networks). Then,
within such Diffserv classes, L4S would often be applicable to give within such Diffserv classes, L4S would often be applicable to give
traffic low latency and low loss as well. Within such a Diffserv traffic low latency and low loss as well. Within such a Diffserv
class, the bandwidth available to a user or application is often class, the bandwidth available to a user or application is often
limited by a rate policer. Similarly, in the default Diffserv class, limited by a rate policer. Similarly, in the default Diffserv class,
rate policers are sometimes used to partition shared capacity. rate policers are sometimes used to partition shared capacity.
A classic rate policer drops any packets exceeding a set rate, A Classic rate policer drops any packets exceeding a set rate,
usually also giving a burst allowance (variants exist where the usually also giving a burst allowance (variants exist where the
policer re-marks non-compliant traffic to a discard-eligible Diffserv policer re-marks noncompliant traffic to a discard-eligible Diffserv
codepoint, so they can be dropped elsewhere during contention). codepoint, so they can be dropped elsewhere during contention).
Whenever L4S traffic encounters one of these rate policers, it will Whenever L4S traffic encounters one of these rate policers, it will
experience drops and the source will have to fall back to a Classic experience drops and the source will have to fall back to a Classic
congestion control, thus losing the benefits of L4S (Section 6.4.3). congestion control, thus losing the benefits of L4S (Section 6.4.3).
So, in networks that already use rate policers and plan to deploy So in networks that already use rate policers and plan to deploy L4S,
L4S, it will be preferable to redesign these rate policers to be more it will be preferable to redesign these rate policers to be more
friendly to the L4S service. friendly to the L4S service.
L4S-friendly rate policing is currently a research area (note that L4S-friendly rate policing is currently a research area (note that
this is not the same as latency policing). It might be achieved by this is not the same as latency policing). It might be achieved by
setting a threshold where ECN marking is introduced, such that it is setting a threshold where ECN marking is introduced, such that it is
just under the policed rate or just under the burst allowance where just under the policed rate or just under the burst allowance where
drop is introduced. For instance the two-rate three-colour drop is introduced. For instance, the two-rate, three-colour marker
marker [RFC2698] or a PCN threshold and excess-rate marker [RFC5670] [RFC2698] or a PCN threshold and excess-rate marker [RFC5670] could
could mark ECN at the lower rate and drop at the higher. Or an mark ECN at the lower rate and drop at the higher. Or an existing
existing rate policer could have congestion-rate policing added, rate policer could have congestion-rate policing added, e.g., using
e.g. using the 'local' (non-ConEx) variant of the ConEx aggregate the 'local' (non-ConEx) variant of the ConEx aggregate congestion
congestion policer [I-D.briscoe-conex-policing]. It might also be policer [CONG-POLICING]. It might also be possible to design
possible to design scalable congestion controls to respond less Scalable congestion controls to respond less catastrophically to loss
catastrophically to loss that has not been preceded by a period of that has not been preceded by a period of increasing delay.
increasing delay.
The design of L4S-friendly rate policers will require a separate The design of L4S-friendly rate policers will require a separate,
dedicated document. For further discussion of the interaction dedicated document. For further discussion of the interaction
between L4S and Diffserv, see [I-D.briscoe-tsvwg-l4s-diffserv]. between L4S and Diffserv, see [L4S-DIFFSERV].
8.4. ECN Integrity 8.4. ECN Integrity
Various ways have been developed to protect the integrity of the Various ways have been developed to protect the integrity of the
congestion feedback loop (whether signalled by loss, Classic ECN or congestion feedback loop (whether signalled by loss, Classic ECN, or
L4S ECN) against misbehaviour by the receiver, sender or network (or L4S ECN) against misbehaviour by the receiver, sender, or network (or
all three). Brief details of each including applicability, pros and all three). Brief details of each, including applicability, pros,
cons is given in Appendix C.1 of the L4S ECN and cons, are given in Appendix C.1 of the L4S ECN spec [RFC9331].
spec [I-D.ietf-tsvwg-ecn-l4s-id].
8.5. Privacy Considerations 8.5. Privacy Considerations
As discussed in Section 5.2, the L4S architecture does not preclude As discussed in Section 5.2, the L4S architecture does not preclude
approaches that inspect end-to-end transport layer identifiers. For approaches that inspect end-to-end transport layer identifiers. For
instance, L4S support has been added to FQ-CoDel, which classifies by instance, L4S support has been added to FQ-CoDel, which classifies by
application flow ID in the network. However, the main innovation of application flow identifier in the network. However, the main
L4S is the DualQ AQM framework that does not need to inspect any innovation of L4S is the DualQ AQM framework that does not need to
deeper than the outermost IP header, because the L4S identifier is in inspect any deeper than the outermost IP header, because the L4S
the IP-ECN field. identifier is in the IP-ECN field.
Thus, the L4S architecture enables very low queuing delay without Thus, the L4S architecture enables very low queuing delay without
_requiring_ inspection of information above the IP layer. This means _requiring_ inspection of information above the IP layer. This means
that users who want to encrypt application flow identifiers, e.g. in that users who want to encrypt application flow identifiers, e.g., in
IPSec or other encrypted VPN tunnels, don't have to sacrifice low IPsec or other encrypted VPN tunnels, don't have to sacrifice low
delay [RFC8404]. delay [RFC8404].
Because L4S can provide low delay for a broad set of applications Because L4S can provide low delay for a broad set of applications
that choose to use it, there is no need for individual applications that choose to use it, there is no need for individual applications
or classes within that broad set to be distinguishable in any way or classes within that broad set to be distinguishable in any way
while traversing networks. This removes much of the ability to while traversing networks. This removes much of the ability to
correlate between the delay requirements of traffic and other correlate between the delay requirements of traffic and other
identifying features [RFC6973]. There may be some types of traffic identifying features [RFC6973]. There may be some types of traffic
that prefer not to use L4S, but the coarse binary categorization of that prefer not to use L4S, but the coarse binary categorization of
traffic reveals very little that could be exploited to compromise traffic reveals very little that could be exploited to compromise
privacy. privacy.
9. Informative References 9. Informative References
[ACCECN] Briscoe, B., Kühlewind, M., and R. Scheffenegger, "More
Accurate ECN Feedback in TCP", Work in Progress, Internet-
Draft, draft-ietf-tcpm-accurate-ecn-22, 9 November 2022,
<https://datatracker.ietf.org/doc/html/draft-ietf-tcpm-
accurate-ecn-22>.
[AFCD] Xue, L., Kumar, S., Cui, C., Kondikoppa, P., Chiu, C-H., [AFCD] Xue, L., Kumar, S., Cui, C., Kondikoppa, P., Chiu, C-H.,
and S-J. Park, "Towards fair and low latency next and S-J. Park, "Towards fair and low latency next
generation high speed networks: AFCD queuing", Journal of generation high speed networks: AFCD queuing", Journal of
Network and Computer Applications 70:183--193, July 2016, Network and Computer Applications, Volume 70, pp. 183-193,
DOI 10.1016/j.jnca.2016.03.021, July 2016,
<https://doi.org/10.1016/j.jnca.2016.03.021>. <https://doi.org/10.1016/j.jnca.2016.03.021>.
[BBRv2] Cardwell, N., "TCP BBR v2 Alpha/Preview Release", GitHub [BBR-CC] Cardwell, N., Cheng, Y., Yeganeh, S. H., Swett, I., and V.
repository; Linux congestion control module, Jacobson, "BBR Congestion Control", Work in Progress,
<https://github.com/google/bbr/blob/v2alpha/README.md>. Internet-Draft, draft-cardwell-iccrg-bbr-congestion-
control-02, 7 March 2022,
<https://datatracker.ietf.org/doc/html/draft-cardwell-
iccrg-bbr-congestion-control-02>.
[BDPdata] Briscoe, B., "PI2 Parameters", Technical Report TR-BB- [BBRv2] "TCP BBR v2 Alpha/Preview Release", commit 17700ca, June
2021-001 arXiv:2107.01003 [cs.NI], July 2021, 2022, <https://github.com/google/bbr>.
<https://arxiv.org/abs/2107.01003>.
[BDPdata] Briscoe, B., "PI2 Parameters", TR-BB-2021-001,
arXiv:2107.01003 [cs.NI], DOI 10.48550/arXiv.2107.01003,
October 2021, <https://arxiv.org/abs/2107.01003>.
[BufferSize] [BufferSize]
Appenzeller, G., Keslassy, I., and N. McKeown, "Sizing Appenzeller, G., Keslassy, I., and N. McKeown, "Sizing
Router Buffers", In Proc. SIGCOMM'04 34(4):281--292, Router Buffers", SIGCOMM '04: Proceedings of the 2004
September 2004, <https://doi.org/10.1145/1015467.1015499>. conference on Applications, technologies, architectures,
and protocols for computer communications, pp. 281-292,
DOI 10.1145/1015467.1015499, October 2004,
<https://doi.org/10.1145/1015467.1015499>.
[COBALT] Palmei, J., Gupta, S., Imputato, P., Morton, J., [COBALT] Palmei, J., Gupta, S., Imputato, P., Morton, J.,
Tahiliani, M. P., Avallone, S., and D. Täht, "Design and Tahiliani, M. P., Avallone, S., and D. Täht, "Design and
Evaluation of COBALT Queue Discipline", In Proc. IEEE Evaluation of COBALT Queue Discipline", IEEE International
Int'l Symp. Local and Metropolitan Area Networks Symposium on Local and Metropolitan Area Networks
(LANMAN'19) 2019:1-6, July 2019, (LANMAN), DOI 10.1109/LANMAN.2019.8847054, July 2019,
<https://ieeexplore.ieee.org/abstract/document/8847054>. <https://ieeexplore.ieee.org/abstract/document/8847054>.
[DCttH19] De Schepper, K., Bondarenko, O., Tilmans, O., and B. [CODEL-APPROX-FAIR]
Briscoe, "`Data Centre to the Home': Ultra-Low Latency for Morton, J. and P. G. Heist, "Controlled Delay Approximate
All", Updated RITE project Technical Report , July 2019, Fairness AQM", Work in Progress, Internet-Draft, draft-
<https://bobbriscoe.net/pubs.html#DCttH_TR>. morton-tsvwg-codel-approx-fair-01, 9 March 2020,
<https://datatracker.ietf.org/doc/html/draft-morton-tsvwg-
codel-approx-fair-01>.
[CONG-POLICING]
Briscoe, B., "Network Performance Isolation using
Congestion Policing", Work in Progress, Internet-Draft,
draft-briscoe-conex-policing-01, 14 February 2014,
<https://datatracker.ietf.org/doc/html/draft-briscoe-
conex-policing-01>.
[CTCP] Sridharan, M., Tan, K., Bansal, D., and D. Thaler,
"Compound TCP: A New TCP Congestion Control for High-Speed
and Long Distance Networks", Work in Progress, Internet-
Draft, draft-sridharan-tcpm-ctcp-02, 11 November 2008,
<https://datatracker.ietf.org/doc/html/draft-sridharan-
tcpm-ctcp-02>.
[DOCSIS-Q-PROT]
Briscoe, B., Ed. and G. White, "The DOCSIS(R) Queue
Protection Algorithm to Preserve Low Latency", Work in
Progress, Internet-Draft, draft-briscoe-docsis-q-
protection-06, 13 May 2022,
<https://datatracker.ietf.org/doc/html/draft-briscoe-
docsis-q-protection-06>.
[DOCSIS3.1] [DOCSIS3.1]
CableLabs, "MAC and Upper Layer Protocols Interface CableLabs, "MAC and Upper Layer Protocols Interface
(MULPI) Specification, CM-SP-MULPIv3.1", Data-Over-Cable (MULPI) Specification, CM-SP-MULPIv3.1", Data-Over-Cable
Service Interface Specifications DOCSIS® 3.1 Version i17 Service Interface Specifications DOCSIS 3.1 Version i17 or
or later, 21 January 2019, <https://specification- later, 21 January 2019, <https://specification-
search.cablelabs.com/CM-SP-MULPIv3.1>. search.cablelabs.com/CM-SP-MULPIv3.1>.
[DOCSIS3AQM] [DOCSIS3AQM]
White, G., "Active Queue Management Algorithms for DOCSIS White, G., "Active Queue Management Algorithms for DOCSIS
3.0; A Simulation Study of CoDel, SFQ-CoDel and PIE in 3.0: A Simulation Study of CoDel, SFQ-CoDel and PIE in
DOCSIS 3.0 Networks", CableLabs Technical Report , April DOCSIS 3.0 Networks", CableLabs Technical Report, April
2013, <{https://www.cablelabs.com/wp- 2013, <https://www.cablelabs.com/wp-
content/uploads/2013/11/ content/uploads/2013/11/
Active_Queue_Management_Algorithms_DOCSIS_3_0.pdf>. Active_Queue_Management_Algorithms_DOCSIS_3_0.pdf>.
[DualPI2Linux] [DualPI2Linux]
Albisser, O., De Schepper, K., Briscoe, B., Tilmans, O., Albisser, O., De Schepper, K., Briscoe, B., Tilmans, O.,
and H. Steen, "DUALPI2 - Low Latency, Low Loss and and H. Steen, "DUALPI2 - Low Latency, Low Loss and
Scalable (L4S) AQM", Proc. Linux Netdev 0x13 , March 2019, Scalable (L4S) AQM", Proceedings of Linux Netdev 0x13,
<https://www.netdevconf.org/0x13/session.html?talk- March 2019, <https://www.netdevconf.org/0x13/
DUALPI2-AQM>. session.html?talk-DUALPI2-AQM>.
[Dukkipati06] [Dukkipati06]
Dukkipati, N. and N. McKeown, "Why Flow-Completion Time is Dukkipati, N. and N. McKeown, "Why Flow-Completion Time is
the Right Metric for Congestion Control", ACM CCR the Right Metric for Congestion Control", ACM SIGCOMM
36(1):59--62, January 2006, Computer Communication Review, Volume 36, Issue 1, pp.
59-62, DOI 10.1145/1111322.1111336, January 2006,
<https://dl.acm.org/doi/10.1145/1111322.1111336>. <https://dl.acm.org/doi/10.1145/1111322.1111336>.
[ECN-ENCAP]
Briscoe, B. and J. Kaippallimalil, "Guidelines for Adding
Congestion Notification to Protocols that Encapsulate IP",
Work in Progress, Internet-Draft, draft-ietf-tsvwg-ecn-
encap-guidelines-17, 11 July 2022,
<https://datatracker.ietf.org/doc/html/draft-ietf-tsvwg-
ecn-encap-guidelines-17>.
[ECN-SCTP] Stewart, R. R., Tüxen, M., and X. Dong, "ECN for Stream
Control Transmission Protocol (SCTP)", Work in Progress,
Internet-Draft, draft-stewart-tsvwg-sctpecn-05, 15 January
2014, <https://datatracker.ietf.org/doc/html/draft-
stewart-tsvwg-sctpecn-05>.
[ECN-SHIM] Briscoe, B., "Propagating Explicit Congestion Notification
Across IP Tunnel Headers Separated by a Shim", Work in
Progress, Internet-Draft, draft-ietf-tsvwg-rfc6040update-
shim-15, 11 July 2022,
<https://datatracker.ietf.org/doc/html/draft-ietf-tsvwg-
rfc6040update-shim-15>.
[FQ_CoDel_Thresh] [FQ_CoDel_Thresh]
Høiland-Jørgensen, T., "fq_codel: generalise ce_threshold "fq_codel: generalise ce_threshold marking for subset of
marking for subset of traffic", Linux Patch Commit ID: traffic", commit dfcb63ce1de6b10b, October 2021,
dfcb63ce1de6b10b, 20 October 2021,
<https://git.kernel.org/pub/scm/linux/kernel/git/netdev/ <https://git.kernel.org/pub/scm/linux/kernel/git/netdev/
net-next.git/commit/?id=dfcb63ce1de6b10b>. net-next.git/commit/?id=dfcb63ce1de6b10b>.
[Hohlfeld14] [Hohlfeld14]
Hohlfeld, O., Pujol, E., Ciucu, F., Feldmann, A., and P. Hohlfeld, O., Pujol, E., Ciucu, F., Feldmann, A., and P.
Barford, "A QoE Perspective on Sizing Network Buffers", Barford, "A QoE Perspective on Sizing Network Buffers",
Proc. ACM Internet Measurement Conf (IMC'14) hmm, November IMC '14: Proceedings of the 2014 Conference on Internet
2014, <https://doi.acm.org/10.1145/2663716.2663730>. Measurement, pp. 333-346, DOI 10.1145/2663716.2663730,
November 2014,
[I-D.briscoe-conex-policing] <https://doi.acm.org/10.1145/2663716.2663730>.
Briscoe, B., "Network Performance Isolation using
Congestion Policing", Work in Progress, Internet-Draft,
draft-briscoe-conex-policing-01, 14 February 2014,
<https://www.ietf.org/archive/id/draft-briscoe-conex-
policing-01.txt>.
[I-D.briscoe-docsis-q-protection]
Briscoe, B. and G. White, "The DOCSIS(r) Queue Protection
Algorithm to Preserve Low Latency", Work in Progress,
Internet-Draft, draft-briscoe-docsis-q-protection-06, 13
May 2022,
<https://datatracker.ietf.org/api/v1/doc/document/draft-
briscoe-docsis-q-protection/>.
[I-D.briscoe-iccrg-prague-congestion-control]
Schepper, K. D., Tilmans, O., and B. Briscoe, "Prague
Congestion Control", Work in Progress, Internet-Draft,
draft-briscoe-iccrg-prague-congestion-control-01, 11 July
2022, <https://datatracker.ietf.org/api/v1/doc/document/
draft-briscoe-iccrg-prague-congestion-control/>.
[I-D.briscoe-tsvwg-l4s-diffserv] [L4S-DIFFSERV]
Briscoe, B., "Interactions between Low Latency, Low Loss, Briscoe, B., "Interactions between Low Latency, Low Loss,
Scalable Throughput (L4S) and Differentiated Services", Scalable Throughput (L4S) and Differentiated Services",
Work in Progress, Internet-Draft, draft-briscoe-tsvwg-l4s- Work in Progress, Internet-Draft, draft-briscoe-tsvwg-l4s-
diffserv-02, 2 July 2018, diffserv-02, 4 November 2018,
<https://datatracker.ietf.org/api/v1/doc/document/draft- <https://datatracker.ietf.org/doc/html/draft-briscoe-
briscoe-tsvwg-l4s-diffserv/>. tsvwg-l4s-diffserv-02>.
[I-D.cardwell-iccrg-bbr-congestion-control]
Cardwell, N., Cheng, Y., Yeganeh, S. H., Swett, I., and V.
Jacobson, "BBR Congestion Control", Work in Progress,
Internet-Draft, draft-cardwell-iccrg-bbr-congestion-
control-02, 7 March 2022,
<https://datatracker.ietf.org/api/v1/doc/document/draft-
cardwell-iccrg-bbr-congestion-control/>.
[I-D.ietf-tcpm-accurate-ecn]
Briscoe, B., Kühlewind, M., and R. Scheffenegger, "More
Accurate ECN Feedback in TCP", Work in Progress, Internet-
Draft, draft-ietf-tcpm-accurate-ecn-20, 25 July 2022,
<https://datatracker.ietf.org/api/v1/doc/document/draft-
ietf-tcpm-accurate-ecn/>.
[I-D.ietf-tsvwg-aqm-dualq-coupled]
Schepper, K. D., Briscoe, B., and G. White, "DualQ Coupled
AQMs for Low Latency, Low Loss and Scalable Throughput
(L4S)", Work in Progress, Internet-Draft, draft-ietf-
tsvwg-aqm-dualq-coupled-24, 7 July 2022,
<https://datatracker.ietf.org/api/v1/doc/document/draft-
ietf-tsvwg-aqm-dualq-coupled/>.
[I-D.ietf-tsvwg-ecn-encap-guidelines]
Briscoe, B. and J. Kaippallimalil, "Guidelines for Adding
Congestion Notification to Protocols that Encapsulate IP",
Work in Progress, Internet-Draft, draft-ietf-tsvwg-ecn-
encap-guidelines-17, 11 July 2022,
<https://datatracker.ietf.org/api/v1/doc/document/draft-
ietf-tsvwg-ecn-encap-guidelines/>.
[I-D.ietf-tsvwg-ecn-l4s-id]
Schepper, K. D. and B. Briscoe, "Explicit Congestion
Notification (ECN) Protocol for Very Low Queuing Delay
(L4S)", Work in Progress, Internet-Draft, draft-ietf-
tsvwg-ecn-l4s-id-28, 8 August 2022,
<https://datatracker.ietf.org/api/v1/doc/document/draft-
ietf-tsvwg-ecn-l4s-id/>.
[I-D.ietf-tsvwg-l4sops]
White, G., "Operational Guidance for Deployment of L4S in
the Internet", Work in Progress, Internet-Draft, draft-
ietf-tsvwg-l4sops-03, 28 April 2022,
<https://datatracker.ietf.org/api/v1/doc/document/draft-
ietf-tsvwg-l4sops/>.
[I-D.ietf-tsvwg-nqb]
White, G. and T. Fossati, "A Non-Queue-Building Per-Hop
Behavior (NQB PHB) for Differentiated Services", Work in
Progress, Internet-Draft, draft-ietf-tsvwg-nqb-10, 4 March
2022, <https://datatracker.ietf.org/api/v1/doc/document/
draft-ietf-tsvwg-nqb/>.
[I-D.ietf-tsvwg-rfc6040update-shim]
Briscoe, B., "Propagating Explicit Congestion Notification
Across IP Tunnel Headers Separated by a Shim", Work in
Progress, Internet-Draft, draft-ietf-tsvwg-rfc6040update-
shim-15, 11 July 2022,
<https://datatracker.ietf.org/api/v1/doc/document/draft-
ietf-tsvwg-rfc6040update-shim/>.
[I-D.morton-tsvwg-codel-approx-fair] [L4Sdemo16]
Morton, J. and P. G. Heist, "Controlled Delay Approximate Bondarenko, O., De Schepper, K., Tsang, I., Briscoe, B.,
Fairness AQM", Work in Progress, Internet-Draft, draft- Petlund, A., and C. Griwodz, "Ultra-Low Delay for All:
morton-tsvwg-codel-approx-fair-01, 9 March 2020, Live Experience, Live Analysis", Proceedings of the 7th
<https://www.ietf.org/archive/id/draft-morton-tsvwg-codel- International Conference on Multimedia Systems, Article
approx-fair-01.txt>. No. 33, pp. 1-4, DOI 10.1145/2910017.2910633, May 2016,
<https://dl.acm.org/citation.cfm?doid=2910017.2910633>.
[I-D.sridharan-tcpm-ctcp] [L4Sdemo16-Video]
Sridharan, M., Tan, K., Bansal, D., and D. Thaler, "Videos used in IETF dispatch WG 'Ultra-Low Queuing Delay
"Compound TCP: A New TCP Congestion Control for High-Speed for All Apps' slot",
and Long Distance Networks", Work in Progress, Internet- <https://riteproject.eu/dctth/#1511dispatchwg>.
Draft, draft-sridharan-tcpm-ctcp-02, 29 October 2007,
<https://datatracker.ietf.org/api/v1/doc/document/draft-
sridharan-tcpm-ctcp/>.
[I-D.stewart-tsvwg-sctpecn] [L4Seval22]
Stewart, R. R., Tuexen, M., and X. Dong, "ECN for Stream De Schepper, K., Albisser, O., Tilmans, O., and B.
Control Transmission Protocol (SCTP)", Work in Progress, Briscoe, "Dual Queue Coupled AQM: Deployable Very Low
Internet-Draft, draft-stewart-tsvwg-sctpecn-05, 15 January Queuing Delay for All", TR-BB-2022-001, arXiv:2209.01078
2014, <https://www.ietf.org/archive/id/draft-stewart- [cs.NI], DOI 10.48550/arXiv.2209.01078, September 2022,
tsvwg-sctpecn-05.txt>. <https://arxiv.org/abs/2209.01078>.
[L4Sdemo16] [L4SOPS] White, G., Ed., "Operational Guidance for Deployment of
Bondarenko, O., De Schepper, K., Tsang, I., and B. L4S in the Internet", Work in Progress, Internet-Draft,
Briscoe, "Ultra-Low Delay for All: Live Experience, Live draft-ietf-tsvwg-l4sops-03, 28 April 2022,
Analysis", Proc. MMSYS'16 pp33:1--33:4, May 2016, <https://datatracker.ietf.org/doc/html/draft-ietf-tsvwg-
<https://dl.acm.org/citation.cfm?doid=2910017.2910633 l4sops-03>.
(videos of demos:
https://riteproject.eu/dctth/#1511dispatchwg )>.
[LEDBAT_AQM] [LEDBAT_AQM]
Al-Saadi, R., Armitage, G., and J. But, "Characterising Al-Saadi, R., Armitage, G., and J. But, "Characterising
LEDBAT Performance Through Bottlenecks Using PIE, FQ-CoDel LEDBAT Performance Through Bottlenecks Using PIE, FQ-CoDel
and FQ-PIE Active Queue Management", Proc. IEEE 42nd and FQ-PIE Active Queue Management", IEEE 42nd Conference
Conference on Local Computer Networks (LCN) 278--285, on Local Computer Networks (LCN), DOI 10.1109/LCN.2017.22,
2017, <https://ieeexplore.ieee.org/document/8109367>. October 2017,
<https://ieeexplore.ieee.org/document/8109367>.
[lowat] Meenan, P., "Optimizing HTTP/2 prioritization with BBR and [lowat] Meenan, P., "Optimizing HTTP/2 prioritization with BBR and
tcp_notsent_lowat", Cloudflare Blog , 12 October 2018, tcp_notsent_lowat", Cloudflare Blog, October 2018,
<https://blog.cloudflare.com/http-2-prioritization-with- <https://blog.cloudflare.com/http-2-prioritization-with-
nginx/>. nginx/>.
[Mathis09] Mathis, M., "Relentless Congestion Control", PFLDNeT'09 ,
May 2009, <https://www.gdt.id.au/~gdt/
presentations/2010-07-06-questnet-tcp/reference-
materials/papers/mathis-relentless-congestion-
control.pdf>.
[McIlroy78] [McIlroy78]
McIlroy, M.D., Pinson, E. N., and B. A. Tague, "UNIX Time- McIlroy, M.D., Pinson, E. N., and B. A. Tague, "UNIX Time-
Sharing System: Foreword", The Bell System Technical Sharing System: Foreword", The Bell System Technical
Journal 57:6(1902--1903), July 1978, Journal 57: 6, pp. 1899-1904,
DOI 10.1002/j.1538-7305.1978.tb02135.x, July 1978,
<https://archive.org/details/bstj57-6-1899>. <https://archive.org/details/bstj57-6-1899>.
[Nadas20] Nádas, S., Gombos, G., Fejes, F., and S. Laki, "A [Nadas20] Nádas, S., Gombos, G., Fejes, F., and S. Laki, "A
Congestion Control Independent L4S Scheduler", Proc. Congestion Control Independent L4S Scheduler", ANRW '20:
Applied Networking Research Workshop (ANRW '20) 45--51, Proceedings of the Applied Networking Research Workshop,
July 2020, <https://doi.org/10.1145/3404868.3406669>. pp. 45-51, DOI 10.1145/3404868.3406669, July 2020,
<https://doi.org/10.1145/3404868.3406669>.
[NASA04] Bailey, R.R., Trey Arthur III, J.J., and S.P. Williams, [NASA04] Bailey, R., Trey Arthur III, J., and S. Williams, "Latency
"Latency Requirements for Head-Worn Display S/EVS Requirements for Head-Worn Display S/EVS Applications",
Applications", SPIE Defense and Security Proceedings of SPIE 5424, DOI 10.1117/12.554462, April
Symposium LF99-1955, April 2004, 2004, <https://ntrs.nasa.gov/api/citations/20120009198/
<https://ntrs.nasa.gov/api/citations/20120009198/
downloads/20120009198.pdf?attachment=true>. downloads/20120009198.pdf?attachment=true>.
[NQB-PHB] White, G. and T. Fossati, "A Non-Queue-Building Per-Hop
Behavior (NQB PHB) for Differentiated Services", Work in
Progress, Internet-Draft, draft-ietf-tsvwg-nqb-15, 11
January 2023, <https://datatracker.ietf.org/doc/html/
draft-ietf-tsvwg-nqb-15>.
[PRAGUE-CC]
De Schepper, K., Tilmans, O., and B. Briscoe, Ed., "Prague
Congestion Control", Work in Progress, Internet-Draft,
draft-briscoe-iccrg-prague-congestion-control-01, 11 July
2022, <https://datatracker.ietf.org/doc/html/draft-
briscoe-iccrg-prague-congestion-control-01>.
[PragueLinux] [PragueLinux]
Briscoe, B., De Schepper, K., Albisser, O., Misund, J., Briscoe, B., De Schepper, K., Albisser, O., Misund, J.,
Tilmans, O., Kühlewind, M., and A.S. Ahmed, "Implementing Tilmans, O., Kühlewind, M., and A.S. Ahmed, "Implementing
the `TCP Prague' Requirements for Low Latency Low Loss the 'TCP Prague' Requirements for Low Latency Low Loss
Scalable Throughput (L4S)", Proc. Linux Netdev 0x13 , Scalable Throughput (L4S)", Proceedings Linux Netdev 0x13,
March 2019, <https://www.netdevconf.org/0x13/ March 2019, <https://www.netdevconf.org/0x13/
session.html?talk-tcp-prague-l4s>. session.html?talk-tcp-prague-l4s>.
[QDyn] Briscoe, B., "Rapid Signalling of Queue Dynamics", [QDyn] Briscoe, B., "Rapid Signalling of Queue Dynamics", TR-BB-
bobbriscoe.net Technical Report TR-BB-2017-001; 2017-001, arXiv:1904.07044 [cs.NI],
arXiv:1904.07044 [cs.NI], September 2017, DOI 10.48550/arXiv.1904.07044, April 2019,
<https://arxiv.org/abs/1904.07044>. <https://arxiv.org/abs/1904.07044>.
[Raaen14] Raaen, K. and T-M. Grønli, "Latency thresholds for [Raaen14] Raaen, K. and T-M. Grønli, "Latency Thresholds for
usability in games: A survey", Norsk IKT-konferanse for Usability in Games: A Survey", Norsk IKT-konferanse for
forskning og utdanning , 2014, forskning og utdanning (Norwegian ICT conference for
research and education), 2014,
<http://ojs.bibsys.no/index.php/NIK/article/view/9/6>. <http://ojs.bibsys.no/index.php/NIK/article/view/9/6>.
[Rajiullah15] [Rajiullah15]
Rajiullah, M., "Towards a Low Latency Internet: Rajiullah, M., "Towards a Low Latency Internet:
Understanding and Solutions", Master's Thesis; Karlstad Understanding and Solutions", Dissertation, Karlstad
Uni, Dept of Maths & CS 2015:41, 2015, <https://www.diva- University, 2015, <https://www.diva-
portal.org/smash/get/diva2:846109/FULLTEXT01.pdf>. portal.org/smash/get/diva2:846109/FULLTEXT01.pdf>.
[RELENTLESS]
Mathis, M., "Relentless Congestion Control", Work in
Progress, Internet-Draft, draft-mathis-iccrg-relentless-
tcp-00, 4 March 2009,
<https://datatracker.ietf.org/doc/html/draft-mathis-iccrg-
relentless-tcp-00>.
[RFC0970] Nagle, J., "On Packet Switches With Infinite Storage", [RFC0970] Nagle, J., "On Packet Switches With Infinite Storage",
RFC 970, DOI 10.17487/RFC0970, December 1985, RFC 970, DOI 10.17487/RFC0970, December 1985,
<https://www.rfc-editor.org/info/rfc970>. <https://www.rfc-editor.org/info/rfc970>.
[RFC2475] Blake, S., Black, D., Carlson, M., Davies, E., Wang, Z., [RFC2475] Blake, S., Black, D., Carlson, M., Davies, E., Wang, Z.,
and W. Weiss, "An Architecture for Differentiated and W. Weiss, "An Architecture for Differentiated
Services", RFC 2475, DOI 10.17487/RFC2475, December 1998, Services", RFC 2475, DOI 10.17487/RFC2475, December 1998,
<https://www.rfc-editor.org/info/rfc2475>. <https://www.rfc-editor.org/info/rfc2475>.
[RFC2698] Heinanen, J. and R. Guerin, "A Two Rate Three Color [RFC2698] Heinanen, J. and R. Guerin, "A Two Rate Three Color
skipping to change at page 44, line 6 skipping to change at line 2016
DOI 10.17487/RFC7713, December 2015, DOI 10.17487/RFC7713, December 2015,
<https://www.rfc-editor.org/info/rfc7713>. <https://www.rfc-editor.org/info/rfc7713>.
[RFC8033] Pan, R., Natarajan, P., Baker, F., and G. White, [RFC8033] Pan, R., Natarajan, P., Baker, F., and G. White,
"Proportional Integral Controller Enhanced (PIE): A "Proportional Integral Controller Enhanced (PIE): A
Lightweight Control Scheme to Address the Bufferbloat Lightweight Control Scheme to Address the Bufferbloat
Problem", RFC 8033, DOI 10.17487/RFC8033, February 2017, Problem", RFC 8033, DOI 10.17487/RFC8033, February 2017,
<https://www.rfc-editor.org/info/rfc8033>. <https://www.rfc-editor.org/info/rfc8033>.
[RFC8034] White, G. and R. Pan, "Active Queue Management (AQM) Based [RFC8034] White, G. and R. Pan, "Active Queue Management (AQM) Based
on Proportional Integral Controller Enhanced PIE) for on Proportional Integral Controller Enhanced (PIE) for
Data-Over-Cable Service Interface Specifications (DOCSIS) Data-Over-Cable Service Interface Specifications (DOCSIS)
Cable Modems", RFC 8034, DOI 10.17487/RFC8034, February Cable Modems", RFC 8034, DOI 10.17487/RFC8034, February
2017, <https://www.rfc-editor.org/info/rfc8034>. 2017, <https://www.rfc-editor.org/info/rfc8034>.
[RFC8170] Thaler, D., Ed., "Planning for Protocol Adoption and [RFC8170] Thaler, D., Ed., "Planning for Protocol Adoption and
Subsequent Transitions", RFC 8170, DOI 10.17487/RFC8170, Subsequent Transitions", RFC 8170, DOI 10.17487/RFC8170,
May 2017, <https://www.rfc-editor.org/info/rfc8170>. May 2017, <https://www.rfc-editor.org/info/rfc8170>.
[RFC8257] Bensley, S., Thaler, D., Balasubramanian, P., Eggert, L., [RFC8257] Bensley, S., Thaler, D., Balasubramanian, P., Eggert, L.,
and G. Judd, "Data Center TCP (DCTCP): TCP Congestion and G. Judd, "Data Center TCP (DCTCP): TCP Congestion
skipping to change at page 45, line 10 skipping to change at line 2065
[RFC8511] Khademi, N., Welzl, M., Armitage, G., and G. Fairhurst, [RFC8511] Khademi, N., Welzl, M., Armitage, G., and G. Fairhurst,
"TCP Alternative Backoff with ECN (ABE)", RFC 8511, "TCP Alternative Backoff with ECN (ABE)", RFC 8511,
DOI 10.17487/RFC8511, December 2018, DOI 10.17487/RFC8511, December 2018,
<https://www.rfc-editor.org/info/rfc8511>. <https://www.rfc-editor.org/info/rfc8511>.
[RFC8888] Sarker, Z., Perkins, C., Singh, V., and M. Ramalho, "RTP [RFC8888] Sarker, Z., Perkins, C., Singh, V., and M. Ramalho, "RTP
Control Protocol (RTCP) Feedback for Congestion Control", Control Protocol (RTCP) Feedback for Congestion Control",
RFC 8888, DOI 10.17487/RFC8888, January 2021, RFC 8888, DOI 10.17487/RFC8888, January 2021,
<https://www.rfc-editor.org/info/rfc8888>. <https://www.rfc-editor.org/info/rfc8888>.
[RFC8985] Cheng, Y., Cardwell, N., Dukkipati, N., and P. Jha, "The
RACK-TLP Loss Detection Algorithm for TCP", RFC 8985,
DOI 10.17487/RFC8985, February 2021,
<https://www.rfc-editor.org/info/rfc8985>.
[RFC9000] Iyengar, J., Ed. and M. Thomson, Ed., "QUIC: A UDP-Based [RFC9000] Iyengar, J., Ed. and M. Thomson, Ed., "QUIC: A UDP-Based
Multiplexed and Secure Transport", RFC 9000, Multiplexed and Secure Transport", RFC 9000,
DOI 10.17487/RFC9000, May 2021, DOI 10.17487/RFC9000, May 2021,
<https://www.rfc-editor.org/info/rfc9000>. <https://www.rfc-editor.org/info/rfc9000>.
[RFC9113] Thomson, M., Ed. and C. Benfield, Ed., "HTTP/2", RFC 9113, [RFC9113] Thomson, M., Ed. and C. Benfield, Ed., "HTTP/2", RFC 9113,
DOI 10.17487/RFC9113, June 2022, DOI 10.17487/RFC9113, June 2022,
<https://www.rfc-editor.org/info/rfc9113>. <https://www.rfc-editor.org/info/rfc9113>.
[SCReAM] Johansson, I., "SCReAM", GitHub repository; , [RFC9331] De Schepper, K. and B. Briscoe, Ed., "The Explicit
<https://github.com/EricssonResearch/scream/blob/master/ Congestion Notification (ECN) Protocol for Low Latency,
README.md>. Low Loss, and Scalable Throughput (L4S)", RFC 9331,
DOI 10.17487/RFC9331, January 2023,
<https://www.rfc-editor.org/info/rfc9331>.
[TCP-CA] Jacobson, V. and M.J. Karels, "Congestion Avoidance and [RFC9332] De Schepper, K., Briscoe, B., Ed., and G. White, "Dual-
Queue Coupled Active Queue Management (AQM) for Low
Latency, Low Loss, and Scalable Throughput (L4S)",
RFC 9332, DOI 10.17487/RFC9332, January 2023,
<https://www.rfc-editor.org/info/rfc9332>.
[SCReAM-L4S]
"SCReAM", commit fda6c53, June 2022,
<https://github.com/EricssonResearch/scream>.
[TCP-CA] Jacobson, V. and M. Karels, "Congestion Avoidance and
Control", Laurence Berkeley Labs Technical Report , Control", Laurence Berkeley Labs Technical Report ,
November 1988, <https://ee.lbl.gov/papers/congavoid.pdf>. November 1988, <https://ee.lbl.gov/papers/congavoid.pdf>.
[UnorderedLTE] [UnorderedLTE]
Austrheim, M.V., "Implementing immediate forwarding for 4G Austrheim, M., "Implementing immediate forwarding for 4G
in a network simulator", Master's Thesis, Uni Oslo , June in a network simulator", Master's Thesis, University of
2019. Oslo, 2018.
Acknowledgements Acknowledgements
Thanks to Richard Scheffenegger, Wes Eddy, Karen Nielsen, David Thanks to Richard Scheffenegger, Wes Eddy, Karen Nielsen, David
Black, Jake Holland, Vidhi Goel, Ermin Sakic, Praveen Black, Jake Holland, Vidhi Goel, Ermin Sakic, Praveen
Balasubramanian, Gorry Fairhurst, Mirja Kuehlewind, Philip Eardley, Balasubramanian, Gorry Fairhurst, Mirja Kuehlewind, Philip Eardley,
Neal Cardwell, Pete Heist and Martin Duke for their useful review Neal Cardwell, Pete Heist, and Martin Duke for their useful review
comments. Thanks also to the area reviewers: Marco Tiloca, Lars comments. Thanks also to the area reviewers: Marco Tiloca, Lars
Eggert, Roman Danyliw and Eric Vyncke. Eggert, Roman Danyliw, and Éric Vyncke.
Bob Briscoe and Koen De Schepper were part-funded by the European Bob Briscoe and Koen De Schepper were partly funded by the European
Community under its Seventh Framework Programme through the Reducing Community under its Seventh Framework Programme through the Reducing
Internet Transport Latency (RITE) project (ICT-317700). The Internet Transport Latency (RITE) project (ICT-317700). The
contribution of Koen De Schepper was also part-funded by the 5Growth contribution of Koen De Schepper was also partly funded by the
and DAEMON EU H2020 projects. Bob Briscoe was also part-funded by 5Growth and DAEMON EU H2020 projects. Bob Briscoe was also partly
the Research Council of Norway through the TimeIn project, partly by funded by the Research Council of Norway through the TimeIn project,
CableLabs and partly by the Comcast Innovation Fund. The views partly by CableLabs, and partly by the Comcast Innovation Fund. The
expressed here are solely those of the authors. views expressed here are solely those of the authors.
Authors' Addresses Authors' Addresses
Bob Briscoe (editor) Bob Briscoe (editor)
Independent Independent
United Kingdom United Kingdom
Email: ietf@bobbriscoe.net Email: ietf@bobbriscoe.net
URI: https://bobbriscoe.net/ URI: https://bobbriscoe.net/
Koen De Schepper Koen De Schepper
Nokia Bell Labs Nokia Bell Labs
Antwerp Antwerp
Belgium Belgium
Email: koen.de_schepper@nokia.com Email: koen.de_schepper@nokia.com
URI: https://www.bell-labs.com/about/researcher-profiles/ URI: https://www.bell-labs.com/about/researcher-profiles/
koende_schepper/ koende_schepper/
Marcelo Bagnulo Marcelo Bagnulo
Universidad Carlos III de Madrid Universidad Carlos III de Madrid
Av. Universidad 30 Av. Universidad 30
Leganes, Madrid 28911 28911 Madrid
Spain Spain
Phone: 34 91 6249500 Phone: 34 91 6249500
Email: marcelo@it.uc3m.es Email: marcelo@it.uc3m.es
URI: https://www.it.uc3m.es URI: https://www.it.uc3m.es
Greg White Greg White
CableLabs CableLabs
United States of America United States of America
Email: G.White@CableLabs.com Email: G.White@CableLabs.com
 End of changes. 277 change blocks. 
1009 lines changed or deleted 1019 lines changed or added

This html diff was produced by rfcdiff 1.48.