File: rfc9330.html

package info (click to toggle)
doc-rfc 20230121-1
links: PTS, VCS
area: non-free
in suites: bookworm, forky, sid, trixie
size: 1,609,944 kB
file content (3654 lines) | stat: -rw-r--r-- 227,895 bytes
<!DOCTYPE html>
<html lang="en" class="RFC">
<head>
<meta charset="utf-8">
<meta content="Common,Latin" name="scripts">
<meta content="initial-scale=1.0" name="viewport">
<title>RFC 9330: Low Latency, Low Loss, and Scalable Throughput (L4S) Internet Service: Architecture</title>
<meta content="Bob Briscoe" name="author">
<meta content="Koen De Schepper" name="author">
<meta content="Marcelo Bagnulo" name="author">
<meta content="Greg White" name="author">
<meta content="
       This document describes the L4S architecture, which enables Internet 
    applications to achieve low queuing latency, low congestion loss, and scalable
    throughput control. L4S is based on the insight that the root cause of
    queuing delay is in the capacity-seeking congestion controllers of
    senders, not in the queue itself. With the L4S architecture, all Internet
    applications could (but do not have to) transition away from congestion
    control algorithms that cause substantial queuing delay and instead adopt a new class
    of congestion controls that can seek capacity with very little queuing.
    These are aided by a modified form of Explicit Congestion Notification
    (ECN) from the network. With this new architecture, applications can
    have both low latency and high throughput. 
       The architecture primarily concerns incremental deployment. It
    defines mechanisms that allow the new class of L4S congestion controls
    to coexist with 'Classic' congestion controls in a shared network. The
    aim is for L4S latency and throughput to be usually much better (and
    rarely worse) while typically not impacting Classic performance. 
    " name="description">
<meta content="xml2rfc 3.15.3" name="generator">
<meta content="Performance" name="keyword">
<meta content="Queuing Delay" name="keyword">
<meta content="One Way Delay" name="keyword">
<meta content="Round-Trip Time" name="keyword">
<meta content="RTT" name="keyword">
<meta content="Jitter" name="keyword">
<meta content="Congestion Control" name="keyword">
<meta content="Congestion Avoidance" name="keyword">
<meta content="Quality of Service" name="keyword">
<meta content="QoS" name="keyword">
<meta content="Quality of Experience" name="keyword">
<meta content="QoE" name="keyword">
<meta content="Active Queue Management" name="keyword">
<meta content="AQM" name="keyword">
<meta content="Explicit Congestion Notification" name="keyword">
<meta content="ECN" name="keyword">
<meta content="Pacing" name="keyword">
<meta content="Burstiness" name="keyword">
<meta content="9330" name="rfc.number">
<!-- Generator version information:
  xml2rfc 3.15.3
    Python 3.9.15
    appdirs 1.4.4
    ConfigArgParse 1.5.3
    google-i18n-address 2.5.1
    html5lib 1.1
    intervaltree 3.1.0
    Jinja2 3.1.2
    lxml 4.9.0
    MarkupSafe 2.1.1
    pycountry 22.3.5
    PyYAML 6.0
    requests 2.28.0
    setuptools 44.1.1
    six 1.16.0
    wcwidth 0.2.5
    weasyprint 56.1
-->
<link href="rfc9330.xml" rel="alternate" type="application/rfc+xml">
<link href="#copyright" rel="license">
<style type="text/css">/*

  NOTE: Changes at the bottom of this file overrides some earlier settings.

  Once the style has stabilized and has been adopted as an official RFC style,
  this can be consolidated so that style settings occur only in one place, but
  for now the contents of this file consists first of the initial CSS work as
  provided to the RFC Formatter (xml2rfc) work, followed by itemized and
  commented changes found necessary during the development of the v3
  formatters.

*/

/* fonts */
@import url('https://fonts.googleapis.com/css?family=Noto+Sans'); /* Sans-serif */
@import url('https://fonts.googleapis.com/css?family=Noto+Serif'); /* Serif (print) */
@import url('https://fonts.googleapis.com/css?family=Roboto+Mono'); /* Monospace */

:root {
  --font-sans: 'Noto Sans', Arial, Helvetica, sans-serif;
  --font-serif: 'Noto Serif', 'Times', 'Times New Roman', serif;
  --font-mono: 'Roboto Mono', Courier, 'Courier New', monospace;
}

@viewport {
  zoom: 1.0;
  width: extend-to-zoom;
}
@-ms-viewport {
  width: extend-to-zoom;
  zoom: 1.0;
}
/* general and mobile first */
html {
}
body {
  max-width: 90%;
  margin: 1.5em auto;
  color: #222;
  background-color: #fff;
  font-size: 14px;
  font-family: var(--font-sans);
  line-height: 1.6;
  scroll-behavior: smooth;
}
.ears {
  display: none;
}

/* headings */
#title, h1, h2, h3, h4, h5, h6 {
  margin: 1em 0 0.5em;
  font-weight: bold;
  line-height: 1.3;
}
#title {
  clear: both;
  border-bottom: 1px solid #ddd;
  margin: 0 0 0.5em 0;
  padding: 1em 0 0.5em;
}
.author {
  padding-bottom: 4px;
}
h1 {
  font-size: 26px;
  margin: 1em 0;
}
h2 {
  font-size: 22px;
  margin-top: -20px;  /* provide offset for in-page anchors */
  padding-top: 33px;
}
h3 {
  font-size: 18px;
  margin-top: -36px;  /* provide offset for in-page anchors */
  padding-top: 42px;
}
h4 {
  font-size: 16px;
  margin-top: -36px;  /* provide offset for in-page anchors */
  padding-top: 42px;
}
h5, h6 {
  font-size: 14px;
}
#n-copyright-notice {
  border-bottom: 1px solid #ddd;
  padding-bottom: 1em;
  margin-bottom: 1em;
}
/* general structure */
p {
  padding: 0;
  margin: 0 0 1em 0;
  text-align: left;
}
div, span {
  position: relative;
}
div {
  margin: 0;
}
.alignRight.art-text {
  background-color: #f9f9f9;
  border: 1px solid #eee;
  border-radius: 3px;
  padding: 1em 1em 0;
  margin-bottom: 1.5em;
}
.alignRight.art-text pre {
  padding: 0;
}
.alignRight {
  margin: 1em 0;
}
.alignRight > *:first-child {
  border: none;
  margin: 0;
  float: right;
  clear: both;
}
.alignRight > *:nth-child(2) {
  clear: both;
  display: block;
  border: none;
}
svg {
  display: block;
}
svg[font-family~="serif" i], svg [font-family~="serif" i] {
  font-family: var(--font-serif);
}
svg[font-family~="sans-serif" i], svg [font-family~="sans-serif" i] {
  font-family: var(--font-sans);
}
svg[font-family~="monospace" i], svg [font-family~="monospace" i] {
  font-family: var(--font-mono);
}
.alignCenter.art-text {
  background-color: #f9f9f9;
  border: 1px solid #eee;
  border-radius: 3px;
  padding: 1em 1em 0;
  margin-bottom: 1.5em;
}
.alignCenter.art-text pre {
  padding: 0;
}
.alignCenter {
  margin: 1em 0;
}
.alignCenter > *:first-child {
  display: table;
  border: none;
  margin: 0 auto;
}

/* lists */
ol, ul {
  padding: 0;
  margin: 0 0 1em 2em;
}
ol ol, ul ul, ol ul, ul ol {
  margin-left: 1em;
}
li {
  margin: 0 0 0.25em 0;
}
.ulCompact li {
  margin: 0;
}
ul.empty, .ulEmpty {
  list-style-type: none;
}
ul.empty li, .ulEmpty li {
  margin-top: 0.5em;
}
ul.ulBare, li.ulBare {
  margin-left: 0em !important;
}
ul.compact, .ulCompact,
ol.compact, .olCompact {
  line-height: 100%;
  margin: 0 0 0 2em;
}

/* definition lists */
dl {
}
dl > dt {
  float: left;
  margin-right: 1em;
}
/* 
dl.nohang > dt {
  float: none;
}
*/
dl > dd {
  margin-bottom: .8em;
  min-height: 1.3em;
}
dl.compact > dd, .dlCompact > dd {
  margin-bottom: 0em;
}
dl > dd > dl {
  margin-top: 0.5em;
  margin-bottom: 0em;
}

/* links */
a {
  text-decoration: none;
}
a[href] {
  color: #22e; /* Arlen: WCAG 2019 */
}
a[href]:hover {
  background-color: #f2f2f2;
}
figcaption a[href],
a[href].selfRef {
  color: #222;
}
/* XXX probably not this:
a.selfRef:hover {
  background-color: transparent;
  cursor: default;
} */

/* Figures */
tt, code, pre {
  background-color: #f9f9f9;
  font-family: var(--font-mono);
}
pre {
  border: 1px solid #eee;
  margin: 0;
  padding: 1em;
}
img {
  max-width: 100%;
}
figure {
  margin: 0;
}
figure blockquote {
  margin: 0.8em 0.4em 0.4em;
}
figcaption {
  font-style: italic;
  margin: 0 0 1em 0;
}
@media screen {
  pre {
    overflow-x: auto;
    max-width: 100%;
    max-width: calc(100% - 22px);
  }
}

/* aside, blockquote */
aside, blockquote {
  margin-left: 0;
  padding: 1.2em 2em;
}
blockquote {
  background-color: #f9f9f9;
  color: #111; /* Arlen: WCAG 2019 */
  border: 1px solid #ddd;
  border-radius: 3px;
  margin: 1em 0;
}
cite {
  display: block;
  text-align: right;
  font-style: italic;
}

/* tables */
table {
  width: 100%;
  margin: 0 0 1em;
  border-collapse: collapse;
  border: 1px solid #eee;
}
th, td {
  text-align: left;
  vertical-align: top;
  padding: 0.5em 0.75em;
}
th {
  text-align: left;
  background-color: #e9e9e9;
}
tr:nth-child(2n+1) > td {
  background-color: #f5f5f5;
}
table caption {
  font-style: italic;
  margin: 0;
  padding: 0;
  text-align: left;
}
table p {
  /* XXX to avoid bottom margin on table row signifiers. If paragraphs should
     be allowed within tables more generally, it would be far better to select on a class. */
  margin: 0;
}

/* pilcrow */
a.pilcrow {
  color: #666; /* Arlen: AHDJ 2019 */
  text-decoration: none;
  visibility: hidden;
  user-select: none;
  -ms-user-select: none;
  -o-user-select:none;
  -moz-user-select: none;
  -khtml-user-select: none;
  -webkit-user-select: none;
  -webkit-touch-callout: none;
}
@media screen {
  aside:hover > a.pilcrow,
  p:hover > a.pilcrow,
  blockquote:hover > a.pilcrow,
  div:hover > a.pilcrow,
  li:hover > a.pilcrow,
  pre:hover > a.pilcrow {
    visibility: visible;
  }
  a.pilcrow:hover {
    background-color: transparent;
  }
}

/* misc */
hr {
  border: 0;
  border-top: 1px solid #eee;
}
.bcp14 {
  font-variant: small-caps;
}

.role {
  font-variant: all-small-caps;
}

/* info block */
#identifiers {
  margin: 0;
  font-size: 0.9em;
}
#identifiers dt {
  width: 3em;
  clear: left;
}
#identifiers dd {
  float: left;
  margin-bottom: 0;
}
/* Fix PDF info block run off issue */
@media print {
  #identifiers dd {
    float: none;
  }
}
#identifiers .authors .author {
  display: inline-block;
  margin-right: 1.5em;
}
#identifiers .authors .org {
  font-style: italic;
}

/* The prepared/rendered info at the very bottom of the page */
.docInfo {
  color: #666; /* Arlen: WCAG 2019 */
  font-size: 0.9em;
  font-style: italic;
  margin-top: 2em;
}
.docInfo .prepared {
  float: left;
}
.docInfo .prepared {
  float: right;
}

/* table of contents */
#toc  {
  padding: 0.75em 0 2em 0;
  margin-bottom: 1em;
}
nav.toc ul {
  margin: 0 0.5em 0 0;
  padding: 0;
  list-style: none;
}
nav.toc li {
  line-height: 1.3em;
  margin: 0.75em 0;
  padding-left: 1.2em;
  text-indent: -1.2em;
}
/* references */
.references dt {
  text-align: right;
  font-weight: bold;
  min-width: 7em;
}
.references dd {
  margin-left: 8em;
  overflow: auto;
}

.refInstance {
  margin-bottom: 1.25em;
}

.references .ascii {
  margin-bottom: 0.25em;
}

/* index */
.index ul {
  margin: 0 0 0 1em;
  padding: 0;
  list-style: none;
}
.index ul ul {
  margin: 0;
}
.index li {
  margin: 0;
  text-indent: -2em;
  padding-left: 2em;
  padding-bottom: 5px;
}
.indexIndex {
  margin: 0.5em 0 1em;
}
.index a {
  font-weight: 700;
}
/* make the index two-column on all but the smallest screens */
@media (min-width: 600px) {
  .index ul {
    -moz-column-count: 2;
    -moz-column-gap: 20px;
  }
  .index ul ul {
    -moz-column-count: 1;
    -moz-column-gap: 0;
  }
}

/* authors */
address.vcard {
  font-style: normal;
  margin: 1em 0;
}

address.vcard .nameRole {
  font-weight: 700;
  margin-left: 0;
}
address.vcard .label {
  font-family: var(--font-sans);
  margin: 0.5em 0;
}
address.vcard .type {
  display: none;
}
.alternative-contact {
  margin: 1.5em 0 1em;
}
hr.addr {
  border-top: 1px dashed;
  margin: 0;
  color: #ddd;
  max-width: calc(100% - 16px);
}

/* temporary notes */
.rfcEditorRemove::before {
  position: absolute;
  top: 0.2em;
  right: 0.2em;
  padding: 0.2em;
  content: "The RFC Editor will remove this note";
  color: #9e2a00; /* Arlen: WCAG 2019 */
  background-color: #ffd; /* Arlen: WCAG 2019 */
}
.rfcEditorRemove {
  position: relative;
  padding-top: 1.8em;
  background-color: #ffd; /* Arlen: WCAG 2019 */
  border-radius: 3px;
}
.cref {
  background-color: #ffd; /* Arlen: WCAG 2019 */
  padding: 2px 4px;
}
.crefSource {
  font-style: italic;
}
/* alternative layout for smaller screens */
@media screen and (max-width: 1023px) {
  body {
    padding-top: 2em;
  }
  #title {
    padding: 1em 0;
  }
  h1 {
    font-size: 24px;
  }
  h2 {
    font-size: 20px;
    margin-top: -18px;  /* provide offset for in-page anchors */
    padding-top: 38px;
  }
  #identifiers dd {
    max-width: 60%;
  }
  #toc {
    position: fixed;
    z-index: 2;
    top: 0;
    right: 0;
    padding: 0;
    margin: 0;
    background-color: inherit;
    border-bottom: 1px solid #ccc;
  }
  #toc h2 {
    margin: -1px 0 0 0;
    padding: 4px 0 4px 6px;
    padding-right: 1em;
    min-width: 190px;
    font-size: 1.1em;
    text-align: right;
    background-color: #444;
    color: white;
    cursor: pointer;
  }
  #toc h2::before { /* css hamburger */
    float: right;
    position: relative;
    width: 1em;
    height: 1px;
    left: -164px;
    margin: 6px 0 0 0;
    background: white none repeat scroll 0 0;
    box-shadow: 0 4px 0 0 white, 0 8px 0 0 white;
    content: "";
  }
  #toc nav {
    display: none;
    padding: 0.5em 1em 1em;
    overflow: auto;
    height: calc(100vh - 48px);
    border-left: 1px solid #ddd;
  }
}

/* alternative layout for wide screens */
@media screen and (min-width: 1024px) {
  body {
    max-width: 724px;
    margin: 42px auto;
    padding-left: 1.5em;
    padding-right: 29em;
  }
  #toc {
    position: fixed;
    top: 42px;
    right: 42px;
    width: 25%;
    margin: 0;
    padding: 0 1em;
    z-index: 1;
  }
  #toc h2 {
    border-top: none;
    border-bottom: 1px solid #ddd;
    font-size: 1em;
    font-weight: normal;
    margin: 0;
    padding: 0.25em 1em 1em 0;
  }
  #toc nav {
    display: block;
    height: calc(90vh - 84px);
    bottom: 0;
    padding: 0.5em 0 0;
    overflow: auto;
  }
  img { /* future proofing */
    max-width: 100%;
    height: auto;
  }
}

/* pagination */
@media print {
  body {

    width: 100%;
  }
  p {
    orphans: 3;
    widows: 3;
  }
  #n-copyright-notice {
    border-bottom: none;
  }
  #toc, #n-introduction {
    page-break-before: always;
  }
  #toc {
    border-top: none;
    padding-top: 0;
  }
  figure, pre {
    page-break-inside: avoid;
  }
  figure {
    overflow: scroll;
  }
  .breakable pre {
    break-inside: auto;
  }
  h1, h2, h3, h4, h5, h6 {
    page-break-after: avoid;
  }
  h2+*, h3+*, h4+*, h5+*, h6+* {
    page-break-before: avoid;
  }
  pre {
    white-space: pre-wrap;
    word-wrap: break-word;
    font-size: 10pt;
  }
  table {
    border: 1px solid #ddd;
  }
  td {
    border-top: 1px solid #ddd;
  }
}

/* This is commented out here, as the string-set: doesn't
   pass W3C validation currently */
/*
.ears thead .left {
  string-set: ears-top-left content();
}

.ears thead .center {
  string-set: ears-top-center content();
}

.ears thead .right {
  string-set: ears-top-right content();
}

.ears tfoot .left {
  string-set: ears-bottom-left content();
}

.ears tfoot .center {
  string-set: ears-bottom-center content();
}

.ears tfoot .right {
  string-set: ears-bottom-right content();
}
*/

@page :first {
  padding-top: 0;
  @top-left {
    content: normal;
    border: none;
  }
  @top-center {
    content: normal;
    border: none;
  }
  @top-right {
    content: normal;
    border: none;
  }
}

@page {
  size: A4;
  margin-bottom: 45mm;
  padding-top: 20px;
  /* The following is commented out here, but set appropriately by in code, as
     the content depends on the document */
  /*
  @top-left {
    content: 'Internet-Draft';
    vertical-align: bottom;
    border-bottom: solid 1px #ccc;
  }
  @top-left {
    content: string(ears-top-left);
    vertical-align: bottom;
    border-bottom: solid 1px #ccc;
  }
  @top-center {
    content: string(ears-top-center);
    vertical-align: bottom;
    border-bottom: solid 1px #ccc;
  }
  @top-right {
    content: string(ears-top-right);
    vertical-align: bottom;
    border-bottom: solid 1px #ccc;
  }
  @bottom-left {
    content: string(ears-bottom-left);
    vertical-align: top;
    border-top: solid 1px #ccc;
  }
  @bottom-center {
    content: string(ears-bottom-center);
    vertical-align: top;
    border-top: solid 1px #ccc;
  }
  @bottom-right {
      content: '[Page ' counter(page) ']';
      vertical-align: top;
      border-top: solid 1px #ccc;
  }
  */

}

/* Changes introduced to fix issues found during implementation */
/* Make sure links are clickable even if overlapped by following H* */
a {
  z-index: 2;
}
/* Separate body from document info even without intervening H1 */
section {
  clear: both;
}


/* Top align author divs, to avoid names without organization dropping level with org names */
.author {
  vertical-align: top;
}

/* Leave room in document info to show Internet-Draft on one line */
#identifiers dt {
  width: 8em;
}

/* Don't waste quite as much whitespace between label and value in doc info */
#identifiers dd {
  margin-left: 1em;
}

/* Give floating toc a background color (needed when it's a div inside section */
#toc {
  background-color: white;
}

/* Make the collapsed ToC header render white on gray also when it's a link */
@media screen and (max-width: 1023px) {
  #toc h2 a,
  #toc h2 a:link,
  #toc h2 a:focus,
  #toc h2 a:hover,
  #toc a.toplink,
  #toc a.toplink:hover {
    color: white;
    background-color: #444;
    text-decoration: none;
  }
}

/* Give the bottom of the ToC some whitespace */
@media screen and (min-width: 1024px) {
  #toc {
    padding: 0 0 1em 1em;
  }
}

/* Style section numbers with more space between number and title */
.section-number {
  padding-right: 0.5em;
}

/* prevent monospace from becoming overly large */
tt, code, pre {
  font-size: 95%;
}

/* Fix the height/width aspect for ascii art*/
.sourcecode pre,
.art-text pre {
  line-height: 1.12;
}


/* Add styling for a link in the ToC that points to the top of the document */
a.toplink {
  float: right;
  margin-right: 0.5em;
}

/* Fix the dl styling to match the RFC 7992 attributes */
dl > dt,
dl.dlParallel > dt {
  float: left;
  margin-right: 1em;
}
dl.dlNewline > dt {
  float: none;
}

/* Provide styling for table cell text alignment */
table td.text-left,
table th.text-left {
  text-align: left;
}
table td.text-center,
table th.text-center {
  text-align: center;
}
table td.text-right,
table th.text-right {
  text-align: right;
}

/* Make the alternative author contact information look less like just another
   author, and group it closer with the primary author contact information */
.alternative-contact {
  margin: 0.5em 0 0.25em 0;
}
address .non-ascii {
  margin: 0 0 0 2em;
}

/* With it being possible to set tables with alignment
  left, center, and right, { width: 100%; } does not make sense */
table {
  width: auto;
}

/* Avoid reference text that sits in a block with very wide left margin,
   because of a long floating dt label.*/
.references dd {
  overflow: visible;
}

/* Control caption placement */
caption {
  caption-side: bottom;
}

/* Limit the width of the author address vcard, so names in right-to-left
   script don't end up on the other side of the page. */

address.vcard {
  max-width: 30em;
  margin-right: auto;
}

/* For address alignment dependent on LTR or RTL scripts */
address div.left {
  text-align: left;
}
address div.right {
  text-align: right;
}

/* Provide table alignment support.  We can't use the alignX classes above
   since they do unwanted things with caption and other styling. */
table.right {
 margin-left: auto;
 margin-right: 0;
}
table.center {
 margin-left: auto;
 margin-right: auto;
}
table.left {
 margin-left: 0;
 margin-right: auto;
}

/* Give the table caption label the same styling as the figcaption */
caption a[href] {
  color: #222;
}

@media print {
  .toplink {
    display: none;
  }

  /* avoid overwriting the top border line with the ToC header */
  #toc {
    padding-top: 1px;
  }

  /* Avoid page breaks inside dl and author address entries */
  .vcard {
    page-break-inside: avoid;
  }

}
/* Tweak the bcp14 keyword presentation */
.bcp14 {
  font-variant: small-caps;
  font-weight: bold;
  font-size: 0.9em;
}
/* Tweak the invisible space above H* in order not to overlay links in text above */
 h2 {
  margin-top: -18px;  /* provide offset for in-page anchors */
  padding-top: 31px;
 }
 h3 {
  margin-top: -18px;  /* provide offset for in-page anchors */
  padding-top: 24px;
 }
 h4 {
  margin-top: -18px;  /* provide offset for in-page anchors */
  padding-top: 24px;
 }
/* Float artwork pilcrow to the right */
@media screen {
  .artwork a.pilcrow {
    display: block;
    line-height: 0.7;
    margin-top: 0.15em;
  }
}
/* Make pilcrows on dd visible */
@media screen {
  dd:hover > a.pilcrow {
    visibility: visible;
  }
}
/* Make the placement of figcaption match that of a table's caption
   by removing the figure's added bottom margin */
.alignLeft.art-text,
.alignCenter.art-text,
.alignRight.art-text {
   margin-bottom: 0;
}
.alignLeft,
.alignCenter,
.alignRight {
  margin: 1em 0 0 0;
}
/* In print, the pilcrow won't show on hover, so prevent it from taking up space,
   possibly even requiring a new line */
@media print {
  a.pilcrow {
    display: none;
  }
}
/* Styling for the external metadata */
div#external-metadata {
  background-color: #eee;
  padding: 0.5em;
  margin-bottom: 0.5em;
  display: none;
}
div#internal-metadata {
  padding: 0.5em;                       /* to match the external-metadata padding */
}
/* Styling for title RFC Number */
h1#rfcnum {
  clear: both;
  margin: 0 0 -1em;
  padding: 1em 0 0 0;
}
/* Make .olPercent look the same as <ol><li> */
dl.olPercent > dd {
  margin-bottom: 0.25em;
  min-height: initial;
}
/* Give aside some styling to set it apart */
aside {
  border-left: 1px solid #ddd;
  margin: 1em 0 1em 2em;
  padding: 0.2em 2em;
}
aside > dl,
aside > ol,
aside > ul,
aside > table,
aside > p {
  margin-bottom: 0.5em;
}
/* Additional page break settings */
@media print {
  figcaption, table caption {
    page-break-before: avoid;
  }
}
/* Font size adjustments for print */
@media print {
  body  { font-size: 10pt;      line-height: normal; max-width: 96%; }
  h1    { font-size: 1.72em;    padding-top: 1.5em; } /* 1*1.2*1.2*1.2 */
  h2    { font-size: 1.44em;    padding-top: 1.5em; } /* 1*1.2*1.2 */
  h3    { font-size: 1.2em;     padding-top: 1.5em; } /* 1*1.2 */
  h4    { font-size: 1em;       padding-top: 1.5em; }
  h5, h6 { font-size: 1em;      margin: initial; padding: 0.5em 0 0.3em; }
}
/* Sourcecode margin in print, when there's no pilcrow */
@media print {
  .artwork,
  .artwork > pre,
  .sourcecode {
    margin-bottom: 1em;
  }
}
/* Avoid narrow tables forcing too narrow table captions, which may render badly */
table {
  min-width: 20em;
}
/* ol type a */
ol.type-a { list-style-type: lower-alpha; }
ol.type-A { list-style-type: upper-alpha; }
ol.type-i { list-style-type: lower-roman; }
ol.type-I { list-style-type: lower-roman; }
/* Apply the print table and row borders in general, on request from the RPC,
and increase the contrast between border and odd row background slightly */
table {
  border: 1px solid #ddd;
}
td {
  border-top: 1px solid #ddd;
}
tr {
  break-inside: avoid;
}
tr:nth-child(2n+1) > td {
  background-color: #f8f8f8;
}
/* Use style rules to govern display of the TOC. */
@media screen and (max-width: 1023px) {
  #toc nav { display: none; }
  #toc.active nav { display: block; }
}
/* Add support for keepWithNext */
.keepWithNext {
  break-after: avoid-page;
  break-after: avoid-page;
}
/* Add support for keepWithPrevious */
.keepWithPrevious {
  break-before: avoid-page;
}
/* Change the approach to avoiding breaks inside artwork etc. */
figure, pre, table, .artwork, .sourcecode  {
  break-before: auto;
  break-after: auto;
}
/* Avoid breaks between <dt> and <dd> */
dl {
  break-before: auto;
  break-inside: auto;
}
dt {
  break-before: auto;
  break-after: avoid-page;
}
dd {
  break-before: avoid-page;
  break-after: auto;
  orphans: 3;
  widows: 3
}
span.break, dd.break {
  margin-bottom: 0;
  min-height: 0;
  break-before: auto;
  break-inside: auto;
  break-after: auto;
}
/* Undo break-before ToC */
@media print {
  #toc {
    break-before: auto;
  }
}
/* Text in compact lists should not get extra bottom margin space,
   since that would makes the list not compact */
ul.compact p, .ulCompact p,
ol.compact p, .olCompact p {
 margin: 0;
}
/* But the list as a whole needs the extra space at the end */
section ul.compact,
section .ulCompact,
section ol.compact,
section .olCompact {
  margin-bottom: 1em;                    /* same as p not within ul.compact etc. */
}
/* The tt and code background above interferes with for instance table cell
   backgrounds.  Changed to something a bit more selective. */
tt, code {
  background-color: transparent;
}
p tt, p code, li tt, li code {
  background-color: #f8f8f8;
}
/* Tweak the pre margin -- 0px doesn't come out well */
pre {
   margin-top: 0.5px;
}
/* Tweak the compact list text */
ul.compact, .ulCompact,
ol.compact, .olCompact,
dl.compact, .dlCompact {
  line-height: normal;
}
/* Don't add top margin for nested lists */
li > ul, li > ol, li > dl,
dd > ul, dd > ol, dd > dl,
dl > dd > dl {
  margin-top: initial;
}
/* Elements that should not be rendered on the same line as a <dt> */
/* This should match the element list in writer.text.TextWriter.render_dl() */
dd > div.artwork:first-child,
dd > aside:first-child,
dd > figure:first-child,
dd > ol:first-child,
dd > div.sourcecode:first-child,
dd > table:first-child,
dd > ul:first-child {
  clear: left;
}
/* fix for weird browser behaviour when <dd/> is empty */
dt+dd:empty::before{
  content: "\00a0";
}
/* Make paragraph spacing inside <li> smaller than in body text, to fit better within the list */
li > p {
  margin-bottom: 0.5em
}
/* Don't let p margin spill out from inside list items */
li > p:last-of-type:only-child {
  margin-bottom: 0;
}
</style>
<link href="rfc-local.css" rel="stylesheet" type="text/css">
<link href="https://dx.doi.org/10.17487/rfc9330" rel="alternate">
  <link href="urn:issn:2070-1721" rel="alternate">
  <link href="https://datatracker.ietf.org/doc/draft-ietf-tsvwg-l4s-arch-20" rel="prev">
  </head>
<body class="xml2rfc">
<script src="https://www.rfc-editor.org/js/metadata.min.js"></script>
<table class="ears">
<thead><tr>
<td class="left">RFC 9330</td>
<td class="center">L4S Architecture</td>
<td class="right">January 2023</td>
</tr></thead>
<tfoot><tr>
<td class="left">Briscoe, et al.</td>
<td class="center">Informational</td>
<td class="right">[Page]</td>
</tr></tfoot>
</table>
<div id="external-metadata" class="document-information"></div>
<div id="internal-metadata" class="document-information">
<dl id="identifiers">
<dt class="label-stream">Stream:</dt>
<dd class="stream">Internet Engineering Task Force (IETF)</dd>
<dt class="label-rfc">RFC:</dt>
<dd class="rfc"><a href="https://www.rfc-editor.org/rfc/rfc9330" class="eref">9330</a></dd>
<dt class="label-category">Category:</dt>
<dd class="category">Informational</dd>
<dt class="label-published">Published:</dt>
<dd class="published">
<time datetime="2023-01" class="published">January 2023</time>
    </dd>
<dt class="label-issn">ISSN:</dt>
<dd class="issn">2070-1721</dd>
<dt class="label-authors">Authors:</dt>
<dd class="authors">
<div class="author">
      <div class="author-name">B. Briscoe, <span class="editor">Ed.</span>
</div>
<div class="org">Independent</div>
</div>
<div class="author">
      <div class="author-name">K. De Schepper</div>
<div class="org">Nokia Bell Labs</div>
</div>
<div class="author">
      <div class="author-name">M. Bagnulo</div>
<div class="org">Universidad Carlos III de Madrid</div>
</div>
<div class="author">
      <div class="author-name">G. White</div>
<div class="org">CableLabs</div>
</div>
</dd>
</dl>
</div>
<h1 id="rfcnum">RFC 9330</h1>
<h1 id="title">Low Latency, Low Loss, and Scalable Throughput (L4S) Internet Service: Architecture</h1>
<section id="section-abstract">
      <h2 id="abstract"><a href="#abstract" class="selfRef">Abstract</a></h2>
<p id="section-abstract-1">This document describes the L4S architecture, which enables Internet 
    applications to achieve low queuing latency, low congestion loss, and scalable
    throughput control. L4S is based on the insight that the root cause of
    queuing delay is in the capacity-seeking congestion controllers of
    senders, not in the queue itself. With the L4S architecture, all Internet
    applications could (but do not have to) transition away from congestion
    control algorithms that cause substantial queuing delay and instead adopt a new class
    of congestion controls that can seek capacity with very little queuing.
    These are aided by a modified form of Explicit Congestion Notification
    (ECN) from the network. With this new architecture, applications can
    have both low latency and high throughput.<a href="#section-abstract-1" class="pilcrow">¶</a></p>
<p id="section-abstract-2">The architecture primarily concerns incremental deployment. It
    defines mechanisms that allow the new class of L4S congestion controls
    to coexist with 'Classic' congestion controls in a shared network. The
    aim is for L4S latency and throughput to be usually much better (and
    rarely worse) while typically not impacting Classic performance.<a href="#section-abstract-2" class="pilcrow">¶</a></p>
</section>
<div id="status-of-memo">
<section id="section-boilerplate.1">
        <h2 id="name-status-of-this-memo">
<a href="#name-status-of-this-memo" class="section-name selfRef">Status of This Memo</a>
        </h2>
<p id="section-boilerplate.1-1">
            This document is not an Internet Standards Track specification; it is
            published for informational purposes.<a href="#section-boilerplate.1-1" class="pilcrow">¶</a></p>
<p id="section-boilerplate.1-2">
            This document is a product of the Internet Engineering Task Force
            (IETF).  It represents the consensus of the IETF community.  It has
            received public review and has been approved for publication by the
            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.<a href="#section-boilerplate.1-2" class="pilcrow">¶</a></p>
<p id="section-boilerplate.1-3">
            Information about the current status of this document, any
            errata, and how to provide feedback on it may be obtained at
            <span><a href="https://www.rfc-editor.org/info/rfc9330">https://www.rfc-editor.org/info/rfc9330</a></span>.<a href="#section-boilerplate.1-3" class="pilcrow">¶</a></p>
</section>
</div>
<div id="copyright">
<section id="section-boilerplate.2">
        <h2 id="name-copyright-notice">
<a href="#name-copyright-notice" class="section-name selfRef">Copyright Notice</a>
        </h2>
<p id="section-boilerplate.2-1">
            Copyright (c) 2023 IETF Trust and the persons identified as the
            document authors. All rights reserved.<a href="#section-boilerplate.2-1" class="pilcrow">¶</a></p>
<p id="section-boilerplate.2-2">
            This document is subject to BCP 78 and the IETF Trust's Legal
            Provisions Relating to IETF Documents
            (<span><a href="https://trustee.ietf.org/license-info">https://trustee.ietf.org/license-info</a></span>) in effect on the date of
            publication of this document. Please review these documents
            carefully, as they describe your rights and restrictions with
            respect to this document. Code Components extracted from this
            document must include Revised BSD License text as described in
            Section 4.e of the Trust Legal Provisions and are provided without
            warranty as described in the Revised BSD License.<a href="#section-boilerplate.2-2" class="pilcrow">¶</a></p>
</section>
</div>
<div id="toc">
<section id="section-toc.1">
        <a href="#" onclick="scroll(0,0)" class="toplink">▲</a><h2 id="name-table-of-contents">
<a href="#name-table-of-contents" class="section-name selfRef">Table of Contents</a>
        </h2>
<nav class="toc"><ul class="compact toc ulBare ulEmpty">
<li class="compact toc ulBare ulEmpty" id="section-toc.1-1.1">
            <p id="section-toc.1-1.1.1" class="keepWithNext"><a href="#section-1" class="auto internal xref">1</a>.  <a href="#name-introduction" class="internal xref">Introduction</a></p>
<ul class="compact toc ulBare ulEmpty">
<li class="compact toc ulBare ulEmpty" id="section-toc.1-1.1.2.1">
                <p id="section-toc.1-1.1.2.1.1" class="keepWithNext"><a href="#section-1.1" class="auto internal xref">1.1</a>.  <a href="#name-document-roadmap" class="internal xref">Document Roadmap</a></p>
</li>
            </ul>
</li>
          <li class="compact toc ulBare ulEmpty" id="section-toc.1-1.2">
            <p id="section-toc.1-1.2.1" class="keepWithNext"><a href="#section-2" class="auto internal xref">2</a>.  <a href="#name-l4s-architecture-overview" class="internal xref">L4S Architecture Overview</a></p>
</li>
          <li class="compact toc ulBare ulEmpty" id="section-toc.1-1.3">
            <p id="section-toc.1-1.3.1"><a href="#section-3" class="auto internal xref">3</a>.  <a href="#name-terminology" class="internal xref">Terminology</a></p>
</li>
          <li class="compact toc ulBare ulEmpty" id="section-toc.1-1.4">
            <p id="section-toc.1-1.4.1"><a href="#section-4" class="auto internal xref">4</a>.  <a href="#name-l4s-architecture-components" class="internal xref">L4S Architecture Components</a></p>
<ul class="compact toc ulBare ulEmpty">
<li class="compact toc ulBare ulEmpty" id="section-toc.1-1.4.2.1">
                <p id="section-toc.1-1.4.2.1.1"><a href="#section-4.1" class="auto internal xref">4.1</a>.  <a href="#name-protocol-mechanisms" class="internal xref">Protocol Mechanisms</a></p>
</li>
              <li class="compact toc ulBare ulEmpty" id="section-toc.1-1.4.2.2">
                <p id="section-toc.1-1.4.2.2.1"><a href="#section-4.2" class="auto internal xref">4.2</a>.  <a href="#name-network-components" class="internal xref">Network Components</a></p>
</li>
              <li class="compact toc ulBare ulEmpty" id="section-toc.1-1.4.2.3">
                <p id="section-toc.1-1.4.2.3.1"><a href="#section-4.3" class="auto internal xref">4.3</a>.  <a href="#name-host-mechanisms" class="internal xref">Host Mechanisms</a></p>
</li>
            </ul>
</li>
          <li class="compact toc ulBare ulEmpty" id="section-toc.1-1.5">
            <p id="section-toc.1-1.5.1"><a href="#section-5" class="auto internal xref">5</a>.  <a href="#name-rationale" class="internal xref">Rationale</a></p>
<ul class="compact toc ulBare ulEmpty">
<li class="compact toc ulBare ulEmpty" id="section-toc.1-1.5.2.1">
                <p id="section-toc.1-1.5.2.1.1"><a href="#section-5.1" class="auto internal xref">5.1</a>.  <a href="#name-why-these-primary-component" class="internal xref">Why These Primary Components?</a></p>
</li>
              <li class="compact toc ulBare ulEmpty" id="section-toc.1-1.5.2.2">
                <p id="section-toc.1-1.5.2.2.1"><a href="#section-5.2" class="auto internal xref">5.2</a>.  <a href="#name-what-l4s-adds-to-existing-a" class="internal xref">What L4S Adds to Existing Approaches</a></p>
</li>
            </ul>
</li>
          <li class="compact toc ulBare ulEmpty" id="section-toc.1-1.6">
            <p id="section-toc.1-1.6.1"><a href="#section-6" class="auto internal xref">6</a>.  <a href="#name-applicability" class="internal xref">Applicability</a></p>
<ul class="compact toc ulBare ulEmpty">
<li class="compact toc ulBare ulEmpty" id="section-toc.1-1.6.2.1">
                <p id="section-toc.1-1.6.2.1.1"><a href="#section-6.1" class="auto internal xref">6.1</a>.  <a href="#name-applications" class="internal xref">Applications</a></p>
</li>
              <li class="compact toc ulBare ulEmpty" id="section-toc.1-1.6.2.2">
                <p id="section-toc.1-1.6.2.2.1"><a href="#section-6.2" class="auto internal xref">6.2</a>.  <a href="#name-use-cases" class="internal xref">Use Cases</a></p>
</li>
              <li class="compact toc ulBare ulEmpty" id="section-toc.1-1.6.2.3">
                <p id="section-toc.1-1.6.2.3.1"><a href="#section-6.3" class="auto internal xref">6.3</a>.  <a href="#name-applicability-with-specific" class="internal xref">Applicability with Specific Link Technologies</a></p>
</li>
              <li class="compact toc ulBare ulEmpty" id="section-toc.1-1.6.2.4">
                <p id="section-toc.1-1.6.2.4.1"><a href="#section-6.4" class="auto internal xref">6.4</a>.  <a href="#name-deployment-considerations" class="internal xref">Deployment Considerations</a></p>
<ul class="compact toc ulBare ulEmpty">
<li class="compact toc ulBare ulEmpty" id="section-toc.1-1.6.2.4.2.1">
                    <p id="section-toc.1-1.6.2.4.2.1.1"><a href="#section-6.4.1" class="auto internal xref">6.4.1</a>.  <a href="#name-deployment-topology" class="internal xref">Deployment Topology</a></p>
</li>
                  <li class="compact toc ulBare ulEmpty" id="section-toc.1-1.6.2.4.2.2">
                    <p id="section-toc.1-1.6.2.4.2.2.1"><a href="#section-6.4.2" class="auto internal xref">6.4.2</a>.  <a href="#name-deployment-sequences" class="internal xref">Deployment Sequences</a></p>
</li>
                  <li class="compact toc ulBare ulEmpty" id="section-toc.1-1.6.2.4.2.3">
                    <p id="section-toc.1-1.6.2.4.2.3.1"><a href="#section-6.4.3" class="auto internal xref">6.4.3</a>.  <a href="#name-l4s-flow-but-non-ecn-bottle" class="internal xref">L4S Flow but Non-ECN Bottleneck</a></p>
</li>
                  <li class="compact toc ulBare ulEmpty" id="section-toc.1-1.6.2.4.2.4">
                    <p id="section-toc.1-1.6.2.4.2.4.1"><a href="#section-6.4.4" class="auto internal xref">6.4.4</a>.  <a href="#name-l4s-flow-but-classic-ecn-bo" class="internal xref">L4S Flow but Classic ECN Bottleneck</a></p>
</li>
                  <li class="compact toc ulBare ulEmpty" id="section-toc.1-1.6.2.4.2.5">
                    <p id="section-toc.1-1.6.2.4.2.5.1"><a href="#section-6.4.5" class="auto internal xref">6.4.5</a>.  <a href="#name-l4s-aqm-deployment-within-t" class="internal xref">L4S AQM Deployment within Tunnels</a></p>
</li>
                </ul>
</li>
            </ul>
</li>
          <li class="compact toc ulBare ulEmpty" id="section-toc.1-1.7">
            <p id="section-toc.1-1.7.1"><a href="#section-7" class="auto internal xref">7</a>.  <a href="#name-iana-considerations" class="internal xref">IANA Considerations</a></p>
</li>
          <li class="compact toc ulBare ulEmpty" id="section-toc.1-1.8">
            <p id="section-toc.1-1.8.1"><a href="#section-8" class="auto internal xref">8</a>.  <a href="#name-security-considerations" class="internal xref">Security Considerations</a></p>
<ul class="compact toc ulBare ulEmpty">
<li class="compact toc ulBare ulEmpty" id="section-toc.1-1.8.2.1">
                <p id="section-toc.1-1.8.2.1.1"><a href="#section-8.1" class="auto internal xref">8.1</a>.  <a href="#name-traffic-rate-non-policing" class="internal xref">Traffic Rate (Non-)Policing</a></p>
<ul class="compact toc ulBare ulEmpty">
<li class="compact toc ulBare ulEmpty" id="section-toc.1-1.8.2.1.2.1">
                    <p id="section-toc.1-1.8.2.1.2.1.1"><a href="#section-8.1.1" class="auto internal xref">8.1.1</a>.  <a href="#name-non-policing-rate-per-flow" class="internal xref">(Non-)Policing Rate per Flow</a></p>
</li>
                  <li class="compact toc ulBare ulEmpty" id="section-toc.1-1.8.2.1.2.2">
                    <p id="section-toc.1-1.8.2.1.2.2.1"><a href="#section-8.1.2" class="auto internal xref">8.1.2</a>.  <a href="#name-non-policing-l4s-service-ra" class="internal xref">(Non-)Policing L4S Service Rate</a></p>
</li>
                </ul>
</li>
              <li class="compact toc ulBare ulEmpty" id="section-toc.1-1.8.2.2">
                <p id="section-toc.1-1.8.2.2.1"><a href="#section-8.2" class="auto internal xref">8.2</a>.  <a href="#name-latency-friendliness" class="internal xref">'Latency Friendliness'</a></p>
</li>
              <li class="compact toc ulBare ulEmpty" id="section-toc.1-1.8.2.3">
                <p id="section-toc.1-1.8.2.3.1"><a href="#section-8.3" class="auto internal xref">8.3</a>.  <a href="#name-interaction-between-rate-po" class="internal xref">Interaction between Rate Policing and L4S</a></p>
</li>
              <li class="compact toc ulBare ulEmpty" id="section-toc.1-1.8.2.4">
                <p id="section-toc.1-1.8.2.4.1"><a href="#section-8.4" class="auto internal xref">8.4</a>.  <a href="#name-ecn-integrity" class="internal xref">ECN Integrity</a></p>
</li>
              <li class="compact toc ulBare ulEmpty" id="section-toc.1-1.8.2.5">
                <p id="section-toc.1-1.8.2.5.1"><a href="#section-8.5" class="auto internal xref">8.5</a>.  <a href="#name-privacy-considerations" class="internal xref">Privacy Considerations</a></p>
</li>
            </ul>
</li>
          <li class="compact toc ulBare ulEmpty" id="section-toc.1-1.9">
            <p id="section-toc.1-1.9.1"><a href="#section-9" class="auto internal xref">9</a>.  <a href="#name-informative-references" class="internal xref">Informative References</a></p>
</li>
          <li class="compact toc ulBare ulEmpty" id="section-toc.1-1.10">
            <p id="section-toc.1-1.10.1"><a href="#appendix-A" class="auto internal xref"></a><a href="#name-acknowledgements" class="internal xref">Acknowledgements</a></p>
</li>
          <li class="compact toc ulBare ulEmpty" id="section-toc.1-1.11">
            <p id="section-toc.1-1.11.1"><a href="#appendix-B" class="auto internal xref"></a><a href="#name-authors-addresses" class="internal xref">Authors' Addresses</a></p>
</li>
        </ul>
</nav>
</section>
</div>
<div id="l4sps_intro">
<section id="section-1">
      <h2 id="name-introduction">
<a href="#section-1" class="section-number selfRef">1. </a><a href="#name-introduction" class="section-name selfRef">Introduction</a>
      </h2>
<p id="section-1-1">At any one time, it is increasingly common for all of the traffic in
    a bottleneck link (e.g., a household's Internet access or Wi-Fi) to come from
    applications that prefer low delay: interactive web, web services,
    voice, conversational video, interactive video, interactive remote
    presence, instant messaging, online and cloud-rendered gaming, remote desktop, cloud-based
    applications, cloud-rendered virtual reality or augmented reality, and video-assisted remote control of machinery and
    industrial processes. In the last decade or so, much has been done to
    reduce propagation delay by placing caches or servers closer to users.
    However, queuing remains a major, albeit intermittent, component of
    latency. For instance, spikes of hundreds of milliseconds are not
    uncommon, even with state-of-the-art Active Queue Management
    (AQM) <span>[<a href="#COBALT" class="cite xref">COBALT</a>]</span> <span>[<a href="#DOCSIS3AQM" class="cite xref">DOCSIS3AQM</a>]</span>. A Classic AQM in an
    access network bottleneck is typically configured to buffer the sawteeth of 
    lone flows, which can cause peak overall
    network delay to roughly double during a long-running flow, relative to
    expected base (unloaded) path delay <span>[<a href="#BufferSize" class="cite xref">BufferSize</a>]</span>.
    Low loss is also important because, for interactive applications, losses
    translate into even longer retransmission delays.<a href="#section-1-1" class="pilcrow">¶</a></p>
<p id="section-1-2">It has been demonstrated that, once access network bit rates reach
    levels now common in the developed world, increasing link capacity
    offers diminishing returns if latency (delay) is not addressed <span>[<a href="#Dukkipati06" class="cite xref">Dukkipati06</a>]</span> <span>[<a href="#Rajiullah15" class="cite xref">Rajiullah15</a>]</span>. Therefore, the
    goal is an Internet service with very low queuing latency, very low
    loss, and scalable throughput. Very low queuing latency means less
    than 1 millisecond (ms) on average and less than about 2 ms at
    the 99th percentile. End-to-end delay above 50 ms <span>[<a href="#Raaen14" class="cite xref">Raaen14</a>]</span>, or even above 20 ms <span>[<a href="#NASA04" class="cite xref">NASA04</a>]</span>,
    starts to feel unnatural for more demanding interactive applications. Therefore,
    removing unnecessary delay variability increases the reach of these
    applications (the distance over which they are comfortable 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.<a href="#section-1-2" class="pilcrow">¶</a></p>
<p id="section-1-3">Differentiated services (Diffserv) offers Expedited Forwarding
    (EF) <span>[<a href="#RFC3246" class="cite xref">RFC3246</a>]</span> for some packets at the expense of
    others, but this makes no difference when all (or most) of the traffic
    at a bottleneck at any one time requires low latency. In contrast, L4S
    still works well when all traffic is L4S -- a service that gives without
    taking needs none of the configuration or management baggage (traffic
    policing or traffic contracts) associated with favouring some traffic
    flows over others.<a href="#section-1-3" class="pilcrow">¶</a></p>
<p id="section-1-4">Queuing delay degrades performance intermittently <span>[<a href="#Hohlfeld14" class="cite xref">Hohlfeld14</a>]</span>. 
    It 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 typically in the access network, or ii) when the
    low latency application is itself a large capacity-seeking or adaptive
    rate flow (e.g., interactive video). 
    At these times, the performance
    improvement from L4S must be sufficient for network operators to be motivated 
    to deploy it.<a href="#section-1-4" class="pilcrow">¶</a></p>
<p id="section-1-5">Active Queue Management (AQM) is part of the solution to queuing
    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
    network without addressing the root of the problem.<a href="#section-1-5" class="pilcrow">¶</a></p>
<p id="section-1-6">The root of the problem is the presence of standard congestion
    control (Reno <span>[<a href="#RFC5681" class="cite xref">RFC5681</a>]</span>) or compatible variants
    (e.g., CUBIC <span>[<a href="#RFC8312" class="cite xref">RFC8312</a>]</span>) that are used in TCP and
    in other transports, such as QUIC <span>[<a href="#RFC9000" class="cite xref">RFC9000</a>]</span>.
    We shall use
    the term 'Classic' for these Reno-friendly congestion controls. 
    Classic
    congestion controls induce relatively large sawtooth-shaped excursions
    of queue occupancy. So if a network operator naively
    attempts to reduce queuing delay by configuring an AQM to operate at a
    shallower queue, a Classic congestion control will significantly
    underutilize the link at the bottom of every sawtooth. These sawteeth have 
    also been growing in duration as flow rate scales (see <a href="#l4sps_why_primary_components" class="auto internal xref">Section 5.1</a> 
    and <span>[<a href="#RFC3649" class="cite xref">RFC3649</a>]</span>).<a href="#section-1-6" class="pilcrow">¶</a></p>
<p id="section-1-7">It has been demonstrated that, if the sending host replaces a Classic
    congestion control with a 'Scalable' alternative, the performance under load of all the above
    interactive applications can be significantly improved once a suitable AQM is
    deployed in the network. 
    Taking the example solution cited below that uses Data Center TCP (DCTCP) 
    <span>[<a href="#RFC8257" class="cite xref">RFC8257</a>]</span> and a Dual-Queue Coupled AQM <span>[<a href="#RFC9332" class="cite xref">RFC9332</a>]</span> on a DSL or Ethernet link, 
    queuing delay under heavy load is roughly 1-2 ms at
    the 99th percentile without losing link utilization <span>[<a href="#L4Seval22" class="cite xref">L4Seval22</a>]</span> <span>[<a href="#DualPI2Linux" class="cite xref">DualPI2Linux</a>]</span> (for other link types,
    see <a href="#l4sarch_link-specifics" class="auto internal xref">Section 6.3</a>). 
    This compares with
    5-20 ms on <em>average</em> with a Classic
    congestion control and current state-of-the-art AQMs, such as
    Flow Queue CoDel <span>[<a href="#RFC8290" class="cite xref">RFC8290</a>]</span>, Proportional Integral controller Enhanced (PIE) <span>[<a href="#RFC8033" class="cite xref">RFC8033</a>]</span>, or DOCSIS PIE <span>[<a href="#RFC8034" class="cite xref">RFC8034</a>]</span> and about
    20-30 ms at the 99th percentile <span>[<a href="#DualPI2Linux" class="cite xref">DualPI2Linux</a>]</span>.<a href="#section-1-7" class="pilcrow">¶</a></p>
<p id="section-1-8">L4S is designed for incremental deployment. It is possible to deploy
    the L4S service at a bottleneck link alongside the existing best efforts
    service <span>[<a href="#DualPI2Linux" class="cite xref">DualPI2Linux</a>]</span> so that unmodified
    applications can start using it as soon as the sender's stack is
    updated. Access networks are typically designed with one link as the
    bottleneck for each site (which might be a home, small enterprise, or
    mobile device), so deployment at either or both ends of this link should
    give nearly all the benefit in the respective direction. 
    With some
    transport protocols, namely TCP <span>[<a href="#I-D.ietf-tcpm-accurate-ecn" class="cite xref">ACCECN</a>]</span>, the sender has to check that
    the receiver has been suitably updated to give more accurate feedback,
    whereas with more recent transport protocols, such as QUIC <span>[<a href="#RFC9000" class="cite xref">RFC9000</a>]</span> and Datagram Congestion Control Protocol (DCCP) <span>[<a href="#RFC4340" class="cite xref">RFC4340</a>]</span>, all
    receivers have always been suitable.<a href="#section-1-8" class="pilcrow">¶</a></p>
<p id="section-1-9">This document presents the L4S architecture. It consists of three
    components: network support to isolate L4S traffic from Classic traffic;
    protocol features that allow network elements to identify L4S traffic;
    and host support for L4S congestion controls. The protocol is defined
    separately in <span>[<a href="#RFC9331" class="cite xref">RFC9331</a>]</span> as an experimental
    change to Explicit Congestion Notification (ECN). This document
    describes and justifies the component parts and how they interact to
    provide the low latency, low loss, and scalable Internet service. It also
    details the approach to incremental deployment, as briefly summarized
    above.<a href="#section-1-9" class="pilcrow">¶</a></p>
<section id="section-1.1">
        <h3 id="name-document-roadmap">
<a href="#section-1.1" class="section-number selfRef">1.1. </a><a href="#name-document-roadmap" class="section-name selfRef">Document Roadmap</a>
        </h3>
<p id="section-1.1-1">This document describes the L4S architecture in three passes. First,
      the brief overview in <a href="#l4s-arch_arch_overview" class="auto internal xref">Section 2</a> gives the very high-level idea and states the main
      components with minimal rationale. This is only intended to give some
      context for the terminology definitions that follow in <a href="#l4sps_Terminology" class="auto internal xref">Section 3</a> and to explain the structure of the rest
      of the document. Then, <a href="#l4sps_components" class="auto internal xref">Section 4</a> goes into more
      detail on each component with some rationale but still mostly stating
      what the architecture is, rather than why. Finally, <a href="#l4sps_rationale" class="auto internal xref">Section 5</a> justifies why each element of the solution
      was chosen (<a href="#l4sps_why_primary_components" class="auto internal xref">Section 5.1</a>) and why
      these choices were different from other solutions (<a href="#l4sps_why-not" class="auto internal xref">Section 5.2</a>).<a href="#section-1.1-1" class="pilcrow">¶</a></p>
<p id="section-1.1-2">After the architecture has been described, <a href="#l4sarch_applicability" class="auto internal xref">Section 6</a> 
      clarifies its applicability by describing the applications and use cases 
      that motivated the design, the challenges applying the architecture to 
      various link technologies, and various incremental deployment models 
      (including the two main deployment topologies, different sequences for 
      incremental deployment, and various interactions with preexisting 
      approaches). The document
      ends with the usual tailpieces, including extensive discussion of
      traffic policing and other security considerations in <a href="#l4sps_Security_Considerations" class="auto internal xref">Section 8</a>.<a href="#section-1.1-2" class="pilcrow">¶</a></p>
</section>
</section>
</div>
<div id="l4s-arch_arch_overview">
<section id="section-2">
      <h2 id="name-l4s-architecture-overview">
<a href="#section-2" class="section-number selfRef">2. </a><a href="#name-l4s-architecture-overview" class="section-name selfRef">L4S Architecture Overview</a>
      </h2>
<p id="section-2-1">Below, we outline the three main components to the L4S architecture:
    1) the Scalable congestion control on the sending host; 2) the AQM at
    the network bottleneck; and 3) the protocol between them.<a href="#section-2-1" class="pilcrow">¶</a></p>
<p id="section-2-2">But first, the main point to grasp is that low latency is not
    provided by the network; low latency results from the careful behaviour
    of the Scalable congestion controllers used by L4S senders. The network
    does have a role, primarily to isolate the low latency of the carefully
    behaving L4S traffic from the higher queuing delay needed by traffic
    with preexisting Classic behaviour. The network also alters the way it
    signals queue growth to the transport. It uses the Explicit Congestion
    Notification (ECN) protocol, but it signals the very start of queue
    growth immediately, without the smoothing delay typical of Classic
    AQMs. Because ECN support is essential for L4S, senders use the ECN
    field as the protocol that allows the network to identify which packets
    are L4S and which are Classic.<a href="#section-2-2" class="pilcrow">¶</a></p>
<span class="break"></span><dl class="olPercent" id="section-2-3">
      <dt>1)</dt>
<dd id="section-2-3.1">
          <p id="section-2-3.1.1">Host:<a href="#section-2-3.1.1" class="pilcrow">¶</a></p>
<p id="section-2-3.1.2">Scalable congestion controls already exist.  They solve the 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 <a href="#l4sps_why_primary_components" class="auto internal xref">Section 5.1</a> and <span>[<a href="#RFC3649" class="cite xref">RFC3649</a>]</span>. 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.<a href="#section-2-3.1.2" class="pilcrow">¶</a></p>
<p id="section-2-3.1.3">With a Scalable congestion control, the average time from one
      congestion signal to the next (the recovery time) remains invariant as
      flow rate scales, all other factors being equal.  This maintains
      the same degree of control over queuing and utilization, whatever the
      flow rate, as well as ensuring that high throughput is more robust to
      disturbances. The Scalable control used most widely (in controlled
      environments) is DCTCP <span>[<a href="#RFC8257" class="cite xref">RFC8257</a>]</span>, which has been implemented and deployed in
      Windows Server Editions (since 2012), in Linux, and in
      FreeBSD.
      Although DCTCP as-is functions well over wide-area round-trip
      times (RTTs), most implementations lack certain safety features that would be
      necessary for use outside controlled environments, like data centres
      (see <a href="#l4sarch_sec_non-l4s-neck" class="auto internal xref">Section 6.4.3</a>). Therefore,
      Scalable congestion 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 
      <span>[<a href="#I-D.briscoe-iccrg-prague-congestion-control" class="cite xref">PRAGUE-CC</a>]</span> <span>[<a href="#PragueLinux" class="cite xref">PragueLinux</a>]</span>, an L4S
      variant of the RMCAT SCReAM controller <span>[<a href="#SCReAM-L4S" class="cite xref">SCReAM-L4S</a>]</span>, and the L4S ECN part of Bottleneck Bandwidth and Round-trip propagation time (BBRv2) <span>[<a href="#BBRv2" class="cite xref">BBRv2</a>]</span> intended for TCP and QUIC transports.<a href="#section-2-3.1.3" class="pilcrow">¶</a></p>
</dd>
        <dd class="break"></dd>
<dt>2)</dt>
<dd id="section-2-3.2">
          <p id="section-2-3.2.1">Network:<a href="#section-2-3.2.1" class="pilcrow">¶</a></p>
<p id="section-2-3.2.2">L4S traffic needs to be isolated from the queuing latency of
      Classic traffic. One queue per application flow (FQ) is one way to
      achieve this, e.g., FQ-CoDel <span>[<a href="#RFC8290" class="cite xref">RFC8290</a>]</span>. However, using just two queues is sufficient and
      does not require inspection of transport layer headers in the network,
      which is not always possible (see <a href="#l4sps_why-not" class="auto internal xref">Section 5.2</a>). With just two queues, it might seem impossible to
      know how much capacity to schedule for each queue 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.<a href="#section-2-3.2.2" class="pilcrow">¶</a></p>
<p id="section-2-3.2.3"><a href="#l4sps_components" class="auto internal xref">Section 4</a> gives a high-level
      explanation of how the per-flow queue (FQ) and DualQ variants of
      L4S work, and <span>[<a href="#RFC9332" class="cite xref">RFC9332</a>]</span> gives a full explanation of the DualQ Coupled AQM
      framework. A specific marking algorithm is not mandated for L4S
      AQMs. Appendices of <span>[<a href="#RFC9332" class="cite xref">RFC9332</a>]</span> 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.<a href="#section-2-3.2.3" class="pilcrow">¶</a></p>
</dd>
        <dd class="break"></dd>
<dt>3)</dt>
<dd id="section-2-3.3">
          <p id="section-2-3.3.1">Protocol:<a href="#section-2-3.3.1" class="pilcrow">¶</a></p>
<p id="section-2-3.3.2">A sending host needs to distinguish L4S and Classic packets with an
      identifier so that the network can classify them into their separate
      treatments. The L4S identifier spec <span>[<a href="#RFC9331" class="cite xref">RFC9331</a>]</span> concludes 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.<a href="#section-2-3.3.2" class="pilcrow">¶</a></p>
</dd>
      <dd class="break"></dd>
</dl>
</section>
</div>
<div id="l4sps_Terminology">
<section id="section-3">
      <h2 id="name-terminology">
<a href="#section-3" class="section-number selfRef">3. </a><a href="#name-terminology" class="section-name selfRef">Terminology</a>
      </h2>
<span class="break"></span><dl class="dlParallel" id="section-3-1">
        <dt id="section-3-1.1">Classic Congestion Control:</dt>
        <dd style="margin-left: 1.5em" id="section-3-1.2">A congestion control
      behaviour that can coexist with standard Reno <span>[<a href="#RFC5681" class="cite xref">RFC5681</a>]</span> without causing significantly negative impact on
      its flow rate <span>[<a href="#RFC5033" class="cite xref">RFC5033</a>]</span>. The scaling problem
      with Classic congestion control is explained, with examples, in
      <a href="#l4sps_why_primary_components" class="auto internal xref">Section 5.1</a> and in <span>[<a href="#RFC3649" class="cite xref">RFC3649</a>]</span>.<a href="#section-3-1.2" class="pilcrow">¶</a>
</dd>
        <dd class="break"></dd>
<dt id="section-3-1.3">Scalable Congestion Control:</dt>
        <dd style="margin-left: 1.5em" id="section-3-1.4">A congestion control
      where the average time from one congestion signal to the next (the
      recovery time) remains invariant as flow rate scales, all other
      factors being equal. 
      For instance, DCTCP averages 2 congestion
      signals per round trip, whatever the flow rate, as do other recently
      developed Scalable congestion controls, e.g., Relentless
      TCP <span>[<a href="#I-D.mathis-iccrg-relentless-tcp" class="cite xref">RELENTLESS</a>]</span>, Prague for TCP and QUIC <span>[<a href="#I-D.briscoe-iccrg-prague-congestion-control" class="cite xref">PRAGUE-CC</a>]</span> <span>[<a href="#PragueLinux" class="cite xref">PragueLinux</a>]</span>, BBRv2 <span>[<a href="#BBRv2" class="cite xref">BBRv2</a>]</span> <span>[<a href="#I-D.cardwell-iccrg-bbr-congestion-control" class="cite xref">BBR-CC</a>]</span>, and the L4S
      variant of SCReAM for real-time media <span>[<a href="#SCReAM-L4S" class="cite xref">SCReAM-L4S</a>]</span> <span>[<a href="#RFC8298" class="cite xref">RFC8298</a>]</span>. See 
      <span><a href="https://www.rfc-editor.org/rfc/rfc9331#section-4.3" class="relref">Section 4.3</a> of [<a href="#RFC9331" class="cite xref">RFC9331</a>]</span> for more
      explanation.<a href="#section-3-1.4" class="pilcrow">¶</a>
</dd>
        <dd class="break"></dd>
<dt id="section-3-1.5">Classic Service:</dt>
        <dd style="margin-left: 1.5em" id="section-3-1.6">The Classic service is intended for
      all the congestion control behaviours that coexist with
      Reno <span>[<a href="#RFC5681" class="cite xref">RFC5681</a>]</span> (e.g., Reno itself,
      CUBIC <span>[<a href="#RFC8312" class="cite xref">RFC8312</a>]</span>, Compound <span>[<a href="#I-D.sridharan-tcpm-ctcp" class="cite xref">CTCP</a>]</span>, and TFRC <span>[<a href="#RFC5348" class="cite xref">RFC5348</a>]</span>). The term 'Classic queue' means a queue
          providing the Classic service.<a href="#section-3-1.6" class="pilcrow">¶</a>
</dd>
        <dd class="break"></dd>
<dt id="section-3-1.7">Low Latency, Low Loss, and Scalable throughput (L4S) service:</dt>
        <dd style="margin-left: 1.5em" id="section-3-1.8">
          <p id="section-3-1.8.1">The
          'L4S' service is intended for traffic from Scalable congestion
          control algorithms, such as the Prague congestion control <span>[<a href="#I-D.briscoe-iccrg-prague-congestion-control" class="cite xref">PRAGUE-CC</a>]</span>, which was
          derived from DCTCP  <span>[<a href="#RFC8257" class="cite xref">RFC8257</a>]</span>. The L4S service
          is for more general traffic than just Prague -- it allows the
          set of congestion controls with similar scaling properties to Prague
          to evolve, such as the examples listed above (Relentless, SCReAM, etc.).
          The term 'L4S queue' means a queue providing the L4S service.<a href="#section-3-1.8.1" class="pilcrow">¶</a></p>
<p id="section-3-1.8.2">The terms Classic or L4S can also qualify other
          nouns, such as 'queue', 'codepoint', 'identifier', 'classification',
          'packet', and 'flow'. For example, an L4S packet means a packet with an
          L4S identifier sent from an L4S congestion control.<a href="#section-3-1.8.2" class="pilcrow">¶</a></p>
<p id="section-3-1.8.3">Both Classic and L4S services can cope with a
          proportion of 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 queue (e.g., DNS, Voice over IP (VoIP), game sync datagrams,
          etc.).<a href="#section-3-1.8.3" class="pilcrow">¶</a></p>
</dd>
        <dd class="break"></dd>
<dt id="section-3-1.9">Reno-friendly:</dt>
        <dd style="margin-left: 1.5em" id="section-3-1.10">The subset of Classic traffic that is
          friendly to the standard Reno congestion control defined for TCP in
          <span>[<a href="#RFC5681" class="cite xref">RFC5681</a>]</span>. The TFRC spec <span>[<a href="#RFC5348" class="cite xref">RFC5348</a>]</span> indirectly implies that 'friendly' is defined as
          "generally within a factor of two of the sending rate of a TCP flow
          under the same conditions". Reno-friendly is used here in place of
          'TCP-friendly', given the latter has become imprecise, because the
          TCP protocol is now used with so many different congestion control
          behaviours, and Reno is used in non-TCP transports, such as
          QUIC <span>[<a href="#RFC9000" class="cite xref">RFC9000</a>]</span>.<a href="#section-3-1.10" class="pilcrow">¶</a>
</dd>
        <dd class="break"></dd>
<dt id="section-3-1.11">Classic ECN:</dt>
        <dd style="margin-left: 1.5em" id="section-3-1.12">
          <p id="section-3-1.12.1">The original Explicit Congestion
          Notification (ECN) protocol <span>[<a href="#RFC3168" class="cite xref">RFC3168</a>]</span> that
          requires ECN signals to be treated as equivalent to drops, both when
          generated in the network and when responded to by the sender.<a href="#section-3-1.12.1" class="pilcrow">¶</a></p>
<p id="section-3-1.12.2">For L4S, the names used for the four codepoints of the 2-bit
   IP-ECN field are unchanged from those defined in the ECN spec
   <span>[<a href="#RFC3168" class="cite xref">RFC3168</a>]</span>, i.e., Not-ECT, ECT(0),
   ECT(1), and CE, where ECT stands for ECN-Capable Transport and CE
   stands for Congestion Experienced. A packet marked with the CE
   codepoint is termed 'ECN-marked' or sometimes just 'marked' where
   the context makes ECN obvious.<a href="#section-3-1.12.2" class="pilcrow">¶</a></p>
</dd>
        <dd class="break"></dd>
<dt id="section-3-1.13">Site:</dt>
        <dd style="margin-left: 1.5em" id="section-3-1.14">A home, mobile device, small enterprise, or
          campus where the network bottleneck is typically the access link to
          the site. Not all network arrangements fit this model, but it is a
          useful, widely applicable generalization.<a href="#section-3-1.14" class="pilcrow">¶</a>
</dd>
        <dd class="break"></dd>
<dt id="section-3-1.15">Traffic Policing:</dt>
        <dd style="margin-left: 1.5em" id="section-3-1.16">Limiting traffic by dropping packets
          or shifting them to a lower service class (as opposed to introducing
          delay, which is termed 'traffic shaping'). Policing can involve
          limiting the average rate and/or burst size. Policing focused on
          limiting queuing but not the average flow rate is termed 'congestion
          policing', 'latency policing', 'burst policing', or 'queue protection' in
          this document. Otherwise, the term rate policing is used.<a href="#section-3-1.16" class="pilcrow">¶</a>
</dd>
      <dd class="break"></dd>
</dl>
</section>
</div>
<div id="l4sps_components">
<section id="section-4">
      <h2 id="name-l4s-architecture-components">
<a href="#section-4" class="section-number selfRef">4. </a><a href="#name-l4s-architecture-components" class="section-name selfRef">L4S Architecture Components</a>
      </h2>
<p id="section-4-1">The L4S architecture is composed of the elements in the following
      three subsections.<a href="#section-4-1" class="pilcrow">¶</a></p>
<div id="l4sps_protocol_components">
<section id="section-4.1">
        <h3 id="name-protocol-mechanisms">
<a href="#section-4.1" class="section-number selfRef">4.1. </a><a href="#name-protocol-mechanisms" class="section-name selfRef">Protocol Mechanisms</a>
        </h3>
<p id="section-4.1-1">The L4S architecture involves: a) unassignment of the previous use
        of the identifier; b) reassignment of the same identifier; and c)
        optional further identifiers:<a href="#section-4.1-1" class="pilcrow">¶</a></p>
<ol start="1" type="a" class="normal type-a" id="section-4.1-2">
<li id="section-4.1-2.1">
            <p id="section-4.1-2.1.1">An essential aspect of a Scalable congestion control is the use
            of explicit congestion signals. Classic ECN <span>[<a href="#RFC3168" class="cite xref">RFC3168</a>]</span> requires 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. L4S needs networks and hosts to
            support a more fine-grained meaning for each ECN signal that is
            less severe than a drop, so that the L4S signals:<a href="#section-4.1-2.1.1" class="pilcrow">¶</a></p>
<ul class="normal">
<li class="normal" id="section-4.1-2.1.2.1">can be much more frequent and<a href="#section-4.1-2.1.2.1" class="pilcrow">¶</a>
</li>
              <li class="normal" id="section-4.1-2.1.2.2">can be signalled immediately, without the significant delay
                required to smooth out fluctuations in the queue.<a href="#section-4.1-2.1.2.2" class="pilcrow">¶</a>
</li>
            </ul>
<p id="section-4.1-2.1.3">To enable L4S, the Standards Track Classic ECN
            spec <span>[<a href="#RFC3168" class="cite xref">RFC3168</a>]</span> has had to be updated to allow
            L4S packets to depart from the 'equivalent-to-drop' constraint.
            <span>[<a href="#RFC8311" class="cite xref">RFC8311</a>]</span> is a Standards Track update to
     relax specific requirements in <span>[<a href="#RFC3168" class="cite xref">RFC3168</a>]</span>
     (and certain other Standards
            Track RFCs), which clears the way for the experimental changes
            proposed for L4S. Also, the ECT(1) codepoint was previously
            assigned as the experimental ECN nonce <span>[<a href="#RFC3540" class="cite xref">RFC3540</a>]</span>, which <span>[<a href="#RFC8311" class="cite xref">RFC8311</a>]</span> recategorizes as historic to
            make the codepoint available again.<a href="#section-4.1-2.1.3" class="pilcrow">¶</a></p>
</li>
          <li id="section-4.1-2.2">
            <p id="section-4.1-2.2.1"><span>[<a href="#RFC9331" class="cite xref">RFC9331</a>]</span> specifies that
            ECT(1) is used as the identifier to classify L4S packets into a
            separate treatment from Classic packets. This satisfies the
            requirement for identifying an alternative ECN treatment in <span>[<a href="#RFC4774" class="cite xref">RFC4774</a>]</span>.<a href="#section-4.1-2.2.1" class="pilcrow">¶</a></p>
<p id="section-4.1-2.2.2">The CE codepoint is
            used to indicate Congestion Experienced by both L4S and Classic
            treatments. This raises the concern that a Classic AQM earlier on
            the path might have marked some ECT(0) packets as CE. Then, these
            packets will be erroneously classified into the L4S queue.
            <span><a href="https://www.rfc-editor.org/rfc/rfc9331#appendix-B" class="relref">Appendix B</a> of [<a href="#RFC9331" class="cite xref">RFC9331</a>]</span> explains why five unlikely
            eventualities all have to coincide for this to have any
            detrimental effect, which even then would only involve a
            vanishingly small likelihood of a spurious retransmission.<a href="#section-4.1-2.2.2" class="pilcrow">¶</a></p>
</li>
          <li id="section-4.1-2.3">A network operator might wish to include certain unresponsive,
            non-L4S traffic in the L4S queue if it is deemed to be paced smoothly
            enough and at a low enough rate not to build a queue, for
            instance, VoIP, low rate datagrams to sync online games,
            relatively low rate application-limited traffic, DNS, Lightweight Directory Access Protocol (LDAP), etc.
            This traffic would need to be tagged with specific identifiers,
            e.g., a low-latency Diffserv codepoint such as Expedited
            Forwarding (EF) <span>[<a href="#RFC3246" class="cite xref">RFC3246</a>]</span>, Non-Queue-Building
            (NQB) <span>[<a href="#I-D.ietf-tsvwg-nqb" class="cite xref">NQB-PHB</a>]</span>, or
            operator-specific identifiers.<a href="#section-4.1-2.3" class="pilcrow">¶</a>
</li>
        </ol>
</section>
</div>
<div id="l4sps_network_components">
<section id="section-4.2">
        <h3 id="name-network-components">
<a href="#section-4.2" class="section-number selfRef">4.2. </a><a href="#name-network-components" class="section-name selfRef">Network Components</a>
        </h3>
<p id="section-4.2-1">The L4S architecture aims to provide low latency without the <em>need</em> for per-flow operations in network
        components. Nonetheless, the architecture does not preclude per-flow
        solutions. The following 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; b) per-flow queues with an instance of a
        Classic and an L4S AQM in each queue; and c) Dual queues with per-flow
        AQMs but no per-flow queues:<a href="#section-4.2-1" class="pilcrow">¶</a></p>
<ol start="1" type="a" class="normal type-a" id="section-4.2-2">
<li id="section-4.2-2.1">
            <p id="section-4.2-2.1.1">The Dual-Queue Coupled AQM (illustrated in <a href="#l4sps_fig_components" class="auto internal xref">Figure 1</a>) achieves the 'semi-permeable'
            membrane property mentioned earlier as follows:<a href="#section-4.2-2.1.1" class="pilcrow">¶</a></p>
<ul class="normal">
<li class="normal" id="section-4.2-2.1.2.1">Latency isolation: Two separate queues are used to isolate
                L4S queuing delay from the larger queue that Classic traffic
                needs to maintain full utilization.<a href="#section-4.2-2.1.2.1" class="pilcrow">¶</a>
</li>
              <li class="normal" id="section-4.2-2.1.2.2">Bandwidth pooling: The two queues act as if they are a
                single pool of bandwidth in which flows of either type get
                roughly equal throughput without the scheduler needing to
                identify any flows. This is achieved by having an AQM in each
                queue, but the Classic AQM provides a congestion signal to
                both queues in a manner that ensures a consistent response
                from the two classes of congestion control. Specifically, the
                Classic AQM generates a drop/mark probability based on
                congestion in its own queue, which it uses both to drop/mark
                packets in its own queue and to affect the marking probability
                in the L4S queue. The strength of the coupling of the
                congestion signalling between the two queues is enough to make
                the L4S flows slow down to leave the right amount of capacity
                for the Classic flows (as they would if they were the same
                type of traffic sharing the same queue).<a href="#section-4.2-2.1.2.2" class="pilcrow">¶</a>
</li>
            </ul>
<p id="section-4.2-2.1.3">Then, the scheduler can serve the L4S queue with priority
            (denoted by the '1' on the higher priority input), because the L4S
            traffic isn't offering up enough traffic to use all the priority
            that it is given. Therefore:<a href="#section-4.2-2.1.3" class="pilcrow">¶</a></p>
<ul class="normal">
<li class="normal" id="section-4.2-2.1.4.1">for latency isolation on short timescales (sub-round-trip),
                the prioritization of the L4S queue protects its low latency
                by allowing bursts to dissipate quickly;<a href="#section-4.2-2.1.4.1" class="pilcrow">¶</a>
</li>
              <li class="normal" id="section-4.2-2.1.4.2">but for bandwidth pooling on longer timescales (round-trip
                and longer), the Classic queue creates an equal and opposite
                pressure against the L4S traffic to ensure that neither has
                priority when it comes to bandwidth -- the tension between
                prioritizing L4S and coupling the marking from the Classic AQM
                results in approximate per-flow fairness.<a href="#section-4.2-2.1.4.2" class="pilcrow">¶</a>
</li>
            </ul>
<p id="section-4.2-2.1.5">To protect against the prioritization of persistent L4S traffic
            deadlocking the Classic queue for a while in some implementations, 
            it is advisable for the priority to be conditional, not
            strict (see <span><a href="https://www.rfc-editor.org/rfc/rfc9332#appendix-A" class="relref">Appendix A</a> of the DualQ spec [<a href="#RFC9332" class="cite xref">RFC9332</a>]</span>).<a href="#section-4.2-2.1.5" class="pilcrow">¶</a></p>
<p id="section-4.2-2.1.6">When there is no Classic traffic, the L4S
            queue's own AQM comes into play. It starts congestion
            marking with a very shallow queue, so L4S traffic maintains very
            low queuing delay.<a href="#section-4.2-2.1.6" class="pilcrow">¶</a></p>
<p id="section-4.2-2.1.7">If either queue becomes persistently overloaded, drop of some
            ECN-capable packets is introduced, as recommended in <span><a href="https://www.rfc-editor.org/rfc/rfc3168#section-7" class="relref">Section 7</a> of the ECN
            spec [<a href="#RFC3168" class="cite xref">RFC3168</a>]</span> and <span><a href="https://www.rfc-editor.org/rfc/rfc7567#section-4.2.1" class="relref">Section 4.2.1</a> of the AQM recommendations [<a href="#RFC7567" class="cite xref">RFC7567</a>]</span>. The trade-offs with different approaches 
            are discussed in <span><a href="https://www.rfc-editor.org/rfc/rfc9332#section-4.2.3" class="relref">Section 4.2.3</a> of the DualQ
            spec [<a href="#RFC9332" class="cite xref">RFC9332</a>]</span> (not shown in the figure here).<a href="#section-4.2-2.1.7" class="pilcrow">¶</a></p>
<p id="section-4.2-2.1.8">The Dual-Queue Coupled AQM has been specified as
            generically as possible <span>[<a href="#RFC9332" class="cite xref">RFC9332</a>]</span> without specifying the
            particular AQMs to use in the two queues so that designers are
            free to implement diverse ideas. Informational appendices in that
            document give pseudocode examples of two different specific AQM
            approaches: one called DualPI2 (pronounced Dual PI
            Squared) <span>[<a href="#DualPI2Linux" class="cite xref">DualPI2Linux</a>]</span> that uses the PI2
            variant of PIE and a zero-config variant of Random Early Detection (RED) called Curvy RED.
            A DualQ Coupled AQM based on PIE has also been specified and
            implemented for Low Latency DOCSIS <span>[<a href="#DOCSIS3.1" class="cite xref">DOCSIS3.1</a>]</span>.<a href="#section-4.2-2.1.8" class="pilcrow">¶</a></p>
<span id="name-components-of-an-l4s-dualq-"></span><div id="l4sps_fig_components">
<figure id="figure-1">
              <div class="alignCenter art-text artwork" id="section-4.2-2.1.9.1">
<pre>
              (3)                  (2)
              .-------^------..------------^------------------.
 ,-(1)-----.                               _____
; ________  :            L4S  -------.    |     |
:|Scalable| :               _\      ||__\_|mark |
:| sender | :  __________  / /      ||  / |_____|\   _________
:|________|\; |          |/   -------'       ^    \1|condit'nl|
 `---------'\_|  IP-ECN  |          Coupling :     \|priority |_\
  ________  / |Classifier|                   :     /|scheduler| /
 |Classic |/  |__________|\   -------.     __:__  / |_________|
 | sender |                \_\ || | ||__\_|mark/|/
 |________|                  / || | ||  / |drop |
                      Classic -------'    |_____|


(1) Scalable sending host
(2) Isolation in separate network queues
(3) Packet identification protocol
</pre>
</div>
<figcaption><a href="#figure-1" class="selfRef">Figure 1</a>:
<a href="#name-components-of-an-l4s-dualq-" class="selfRef">Components of an L4S DualQ Coupled AQM Solution</a>
              </figcaption></figure>
</div>
</li>
          <li id="section-4.2-2.2">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 is
            typically also the option of ECN marking at an immediate
            (unsmoothed) shallow threshold to support use in data centres (see
            <span><a href="https://www.rfc-editor.org/rfc/rfc8290#section-5.2.7" class="relref">Section 5.2.7</a> of the FQ-CoDel spec [<a href="#RFC8290" class="cite xref">RFC8290</a>]</span>). In
            Linux, this has been modified so that the shallow threshold can be
            solely applied to ECT(1) packets <span>[<a href="#FQ_CoDel_Thresh" class="cite xref">FQ_CoDel_Thresh</a>]</span>. Then, if there is a flow of Not-ECT or
            ECT(0) packets in the per-flow queue, the Classic AQM
            (e.g., CoDel) is applied; whereas, if there is a flow of ECT(1)
            packets in the queue, the shallower (typically sub-millisecond)
            threshold is applied.
     In addition, ECT(0) and Not-ECT packets
            could potentially be classified into a separate flow queue from
            ECT(1) and CE packets to avoid them mixing if they share a common
            flow identifier (e.g., in a VPN).<a href="#section-4.2-2.2" class="pilcrow">¶</a>
</li>
          <li id="section-4.2-2.3">
            <p id="section-4.2-2.3.1">Dual queues but per-flow AQMs: It should also be possible to
            use dual queues for isolation but with per-flow marking to
            control flow rates (instead of the coupled per-queue marking of
            the Dual-Queue Coupled AQM). One of the two queues would be for
            isolating L4S packets, which would be classified by the ECN
            codepoint. Flow rates could be controlled by flow-specific
            marking. The policy goal of the marking could be to differentiate
            flow rates (e.g., <span>[<a href="#Nadas20" class="cite xref">Nadas20</a>]</span>, which requires
            additional signalling of a per-flow 'value') or to equalize
            flow rates (perhaps in a similar way to Approx Fair
            CoDel <span>[<a href="#AFCD" class="cite xref">AFCD</a>]</span> <span>[<a href="#I-D.morton-tsvwg-codel-approx-fair" class="cite xref">CODEL-APPROX-FAIR</a>]</span> but with two queues
            not one).<a href="#section-4.2-2.3.1" class="pilcrow">¶</a></p>
<p id="section-4.2-2.3.2">Note that, whenever the term
            'DualQ' is used loosely without saying whether marking is
            per queue or per flow, it means a dual-queue AQM with per-queue
            marking.<a href="#section-4.2-2.3.2" class="pilcrow">¶</a></p>
</li>
        </ol>
</section>
</div>
<div id="l4sps_host_components">
<section id="section-4.3">
        <h3 id="name-host-mechanisms">
<a href="#section-4.3" class="section-number selfRef">4.3. </a><a href="#name-host-mechanisms" class="section-name selfRef">Host Mechanisms</a>
        </h3>
<p id="section-4.3-1">The L4S architecture includes two main mechanisms in the end host
        that we enumerate next:<a href="#section-4.3-1" class="pilcrow">¶</a></p>
<ol start="1" type="a" class="normal type-a" id="section-4.3-2">
<li id="section-4.3-2.1">
            <p id="section-4.3-2.1.1">Scalable congestion control at the sender: <a href="#l4s-arch_arch_overview" class="auto internal xref">Section 2</a> defines a Scalable congestion
            control as one where the average time from one congestion signal
            to the next (the recovery time) remains invariant as flow rate
            scales, all other factors being equal. DCTCP is the most
            widely used example. It has been documented as an informational
            record of the protocol currently in use in controlled
            environments <span>[<a href="#RFC8257" class="cite xref">RFC8257</a>]</span>. A list of safety
            and performance improvements for a Scalable congestion control to
            be usable on the public Internet has been drawn up (see the so-called
            'Prague L4S requirements' in <span><a href="https://www.rfc-editor.org/rfc/rfc9331#appendix-A" class="relref">Appendix A</a> of [<a href="#RFC9331" class="cite xref">RFC9331</a>]</span>). 
            The subset that involve
            risk of harm to others have been captured as normative
            requirements in <span><a href="https://www.rfc-editor.org/rfc/rfc9331#section-4" class="relref">Section 4</a> of [<a href="#RFC9331" class="cite xref">RFC9331</a>]</span>. TCP Prague <span>[<a href="#I-D.briscoe-iccrg-prague-congestion-control" class="cite xref">PRAGUE-CC</a>]</span> has been
            implemented in Linux as a reference implementation to address
            these requirements <span>[<a href="#PragueLinux" class="cite xref">PragueLinux</a>]</span>.<a href="#section-4.3-2.1.1" class="pilcrow">¶</a></p>
<p id="section-4.3-2.1.2">Transport protocols other than TCP use various
            congestion controls that are designed to be friendly with Reno.
            Before they can use the L4S service, they will need to be updated
            to implement a Scalable congestion response, which they will have
            to indicate by using the ECT(1) codepoint. Scalable variants are
            under consideration for more recent transport protocols
            (e.g., QUIC), and the L4S ECN part of BBRv2 <span>[<a href="#BBRv2" class="cite xref">BBRv2</a>]</span> <span>[<a href="#I-D.cardwell-iccrg-bbr-congestion-control" class="cite xref">BBR-CC</a>]</span> is a Scalable
            congestion control intended for the TCP and QUIC transports,
            amongst others. Also, an L4S variant of the RMCAT SCReAM
            controller <span>[<a href="#RFC8298" class="cite xref">RFC8298</a>]</span> has been
            implemented <span>[<a href="#SCReAM-L4S" class="cite xref">SCReAM-L4S</a>]</span> for media transported
            over RTP.<a href="#section-4.3-2.1.2" class="pilcrow">¶</a></p>
<p id="section-4.3-2.1.3"> <span><a href="https://www.rfc-editor.org/rfc/rfc9331#section-4.3" class="relref">Section 4.3</a> of the L4S ECN spec [<a href="#RFC9331" class="cite xref">RFC9331</a>]</span> defines
            Scalable congestion control in more detail and specifies the
            requirements that an L4S Scalable congestion control has to comply
            with.<a href="#section-4.3-2.1.3" class="pilcrow">¶</a></p>
</li>
          <li id="section-4.3-2.2">
            <p id="section-4.3-2.2.1">The ECN feedback in some transport protocols is already
            sufficiently fine-grained for L4S (specifically DCCP <span>[<a href="#RFC4340" class="cite xref">RFC4340</a>]</span> and QUIC <span>[<a href="#RFC9000" class="cite xref">RFC9000</a>]</span>). But
            others either require updates or are in the process of being
            updated:<a href="#section-4.3-2.2.1" class="pilcrow">¶</a></p>
<ul class="normal">
<li class="normal" id="section-4.3-2.2.2.1">For the case of TCP, the feedback protocol for ECN embeds
                the assumption from Classic ECN <span>[<a href="#RFC3168" class="cite xref">RFC3168</a>]</span>
                that an ECN mark is equivalent to a drop, making it unusable
                for a Scalable TCP. Therefore, the implementation of TCP
                receivers will have to be upgraded <span>[<a href="#RFC7560" class="cite xref">RFC7560</a>]</span>. 
                Work to standardize and implement more
                accurate ECN feedback for TCP (AccECN) is in
                progress <span>[<a href="#I-D.ietf-tcpm-accurate-ecn" class="cite xref">ACCECN</a>]</span>
                <span>[<a href="#PragueLinux" class="cite xref">PragueLinux</a>]</span>.<a href="#section-4.3-2.2.2.1" class="pilcrow">¶</a>
</li>
              <li class="normal" id="section-4.3-2.2.2.2">ECN feedback was only roughly sketched in the appendix of
                the now obsoleted second specification of SCTP <span>[<a href="#RFC4960" class="cite xref">RFC4960</a>]</span>, while a fuller specification was proposed
                in a long-expired document <span>[<a href="#I-D.stewart-tsvwg-sctpecn" class="cite xref">ECN-SCTP</a>]</span>. A new design would need
                to be implemented and deployed before SCTP could support
                L4S.<a href="#section-4.3-2.2.2.2" class="pilcrow">¶</a>
</li>
              <li class="normal" id="section-4.3-2.2.2.3">For RTP, sufficient ECN feedback was defined in <span>[<a href="#RFC6679" class="cite xref">RFC6679</a>]</span>, but <span>[<a href="#RFC8888" class="cite xref">RFC8888</a>]</span> defines the
                latest Standards Track improvements.<a href="#section-4.3-2.2.2.3" class="pilcrow">¶</a>
</li>
            </ul>
</li>
        </ol>
</section>
</div>
</section>
</div>
<div id="l4sps_rationale">
<section id="section-5">
      <h2 id="name-rationale">
<a href="#section-5" class="section-number selfRef">5. </a><a href="#name-rationale" class="section-name selfRef">Rationale</a>
      </h2>
<div id="l4sps_why_primary_components">
<section id="section-5.1">
        <h3 id="name-why-these-primary-component">
<a href="#section-5.1" class="section-number selfRef">5.1. </a><a href="#name-why-these-primary-component" class="section-name selfRef">Why These Primary Components?</a>
        </h3>
<span class="break"></span><dl class="dlParallel" id="section-5.1-1">
          <dt id="section-5.1-1.1">Explicit congestion signalling (protocol):</dt>
          <dd style="margin-left: 1.5em" id="section-5.1-1.2">
            <p id="section-5.1-1.2.1">Explicit
            congestion signalling is a key part of the L4S approach. In
            contrast, use of drop as a congestion signal creates tension
            because drop is both an impairment (less would be better) and a
            useful signal (more would be better):<a href="#section-5.1-1.2.1" class="pilcrow">¶</a></p>
<ul class="normal">
<li class="normal" id="section-5.1-1.2.2.1">Explicit congestion signals can be used many times per
                round trip to keep tight control without any impairment.
                Under heavy load, even more explicit signals can be applied
                so that the queue can be kept short whatever the load. In
                contrast, Classic AQMs have to introduce very high packet drop
                at high load to keep the queue short. By using ECN, an L4S
                congestion control's sawtooth reduction can be smaller and
                therefore return to the operating point more often, without
                worrying that more sawteeth will cause more signals. The
                consequent smaller amplitude sawteeth fit between an empty
                queue and a very shallow marking threshold (~1 ms in the
                public Internet), so queue delay variation can be very low,
                without risk of underutilization.<a href="#section-5.1-1.2.2.1" class="pilcrow">¶</a>
</li>
              <li class="normal" id="section-5.1-1.2.2.2">Explicit congestion signals can be emitted immediately to
                track fluctuations of the queue. L4S shifts smoothing from the
                network to the host. The network doesn't know the round-trip
                times (RTTs) of any of the flows. So if the network is responsible
                for smoothing (as in the Classic approach), it has to assume a
                worst case RTT, otherwise long RTT flows would become
                unstable. This delays Classic congestion signals by 100-200
                ms. In contrast, each host knows its own RTT. So,
                in the L4S approach, the host can smooth each flow over its
                own RTT, introducing no more smoothing delay than strictly
                necessary (usually only a few milliseconds). A host can also
                choose not to introduce any smoothing delay if appropriate,
                e.g., during flow start-up.<a href="#section-5.1-1.2.2.2" class="pilcrow">¶</a>
</li>
            </ul>
<p id="section-5.1-1.2.3">Neither of the above are feasible if explicit congestion
            signalling has to be considered 'equivalent to drop' (as was
            required with Classic ECN <span>[<a href="#RFC3168" class="cite xref">RFC3168</a>]</span>), because
            drop is an impairment as well as a signal. So drop cannot be
            excessively frequent, and drop cannot be immediate; otherwise, too
            many drops would turn out to have been due to only a transient
            fluctuation 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 variant of ECN that is not equivalent to drop (see
            <span><a href="https://www.rfc-editor.org/rfc/rfc9331#section-5.2" class="relref">Section 5.2</a> of the L4S ECN spec [<a href="#RFC9331" class="cite xref">RFC9331</a>]</span>), while the Classic queue
            uses either Classic ECN <span>[<a href="#RFC3168" class="cite xref">RFC3168</a>]</span> or drop,
            which are still equivalent to each other.<a href="#section-5.1-1.2.3" class="pilcrow">¶</a></p>
<p id="section-5.1-1.2.4">Before
            Classic ECN was standardized, there were various proposals to give
            an ECN mark a different meaning from drop. However, there was no
            particular reason to agree on any one of the alternative meanings,
            so 'equivalent to drop' was the only compromise that could be
            reached. <span>[<a href="#RFC3168" class="cite xref">RFC3168</a>]</span> contains a statement that:<a href="#section-5.1-1.2.4" class="pilcrow">¶</a></p>
<ul class="normal ulEmpty">
<li class="normal ulEmpty" id="section-5.1-1.2.5.1">
                <p style="margin-left: 0.5em" id="section-5.1-1.2.5.1.1">An environment where all end nodes were
 ECN-Capable could allow new criteria to be developed for
 setting the CE codepoint, and new congestion control
 mechanisms for end-node reaction to CE packets. However, this
 is a research issue, and as such is not addressed in this
 document.<a href="#section-5.1-1.2.5.1.1" class="pilcrow">¶</a></p>
</li>
            </ul>
</dd>
          <dd class="break"></dd>
<dt id="section-5.1-1.3">Latency isolation (network):</dt>
          <dd style="margin-left: 1.5em" id="section-5.1-1.4">L4S congestion controls
            keep queue delay low, whereas Classic congestion controls need a
            queue of the order of the RTT to avoid underutilization. One
            queue cannot have two lengths; therefore, L4S traffic needs to be
            isolated in a separate queue (e.g., DualQ) or queues
            (e.g., FQ).<a href="#section-5.1-1.4" class="pilcrow">¶</a>
</dd>
          <dd class="break"></dd>
<dt id="section-5.1-1.5">Coupled congestion notification:</dt>
          <dd style="margin-left: 1.5em" id="section-5.1-1.6">Coupling the
            congestion notification between two queues as in the DualQ Coupled
            AQM is not necessarily essential, but it is a simple way to allow
            senders to determine their rate packet by packet, rather than be
            overridden by a network scheduler. An alternative is for a network
            scheduler to control the rate of each application flow (see the
            discussion in <a href="#l4sps_why-not" class="auto internal xref">Section 5.2</a>).<a href="#section-5.1-1.6" class="pilcrow">¶</a>
</dd>
          <dd class="break"></dd>
<dt id="section-5.1-1.7">L4S packet identifier (protocol):</dt>
          <dd style="margin-left: 1.5em" id="section-5.1-1.8">Once there are at
            least two treatments in the network, hosts need an identifier at
            the IP layer to distinguish which treatment they intend to
            use.<a href="#section-5.1-1.8" class="pilcrow">¶</a>
</dd>
          <dd class="break"></dd>
<dt id="section-5.1-1.9">Scalable congestion notification:</dt>
          <dd style="margin-left: 1.5em" id="section-5.1-1.10">A Scalable
            congestion control in the host keeps the signalling frequency from
            the network high, whatever the flow rate, so that queue delay
            variations can be small when conditions are stable, and rate can
            track variations in available capacity as rapidly as possible
            otherwise.<a href="#section-5.1-1.10" class="pilcrow">¶</a>
</dd>
          <dd class="break"></dd>
<dt id="section-5.1-1.11">Low loss:</dt>
          <dd style="margin-left: 1.5em" id="section-5.1-1.12">Latency is not the only concern of L4S.
            The 'Low Loss' part of the name denotes that L4S generally
            achieves zero congestion loss due to its use of ECN. Otherwise,
            loss would itself cause delay, particularly for short flows, due
            to retransmission delay <span>[<a href="#RFC2884" class="cite xref">RFC2884</a>]</span>.<a href="#section-5.1-1.12" class="pilcrow">¶</a>
</dd>
          <dd class="break"></dd>
<dt id="section-5.1-1.13">Scalable throughput:</dt>
          <dd style="margin-left: 1.5em" id="section-5.1-1.14">
            <p id="section-5.1-1.14.1">The 'Scalable throughput' part
            of the name denotes that the per-flow throughput of Scalable
            congestion controls should scale indefinitely, avoiding the
            imminent scaling problems with Reno-friendly congestion control
            algorithms <span>[<a href="#RFC3649" class="cite xref">RFC3649</a>]</span>. It was known when TCP
            congestion avoidance was first developed in 1988 that it would not
            scale to high bandwidth-delay products (see footnote 6 in <span>[<a href="#TCP-CA" class="cite xref">TCP-CA</a>]</span>). Today, regular broadband flow rates over WAN
            distances are already beyond the scaling range of Classic Reno
            congestion control. So 'less unscalable' CUBIC <span>[<a href="#RFC8312" class="cite xref">RFC8312</a>]</span> and Compound <span>[<a href="#I-D.sridharan-tcpm-ctcp" class="cite xref">CTCP</a>]</span> variants of TCP have been
            successfully deployed. However, these are now approaching their
            scaling limits.<a href="#section-5.1-1.14.1" class="pilcrow">¶</a></p>
<p id="section-5.1-1.14.2">For instance, we will
            consider a scenario with a maximum RTT of 30 ms at the peak
            of each sawtooth. As Reno packet rate scales 8 times from 1,250 to
            10,000 packet/s (from 15 to 120 Mb/s with 1500 B
            packets), the time to recover from a congestion event rises
            proportionately by 8 times as well, from 422 ms to 3.38 s. It
            is clearly problematic for a congestion control to take multiple
            seconds to recover from each congestion event. CUBIC <span>[<a href="#RFC8312" class="cite xref">RFC8312</a>]</span> 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 is still fully in its
            Reno-friendly mode, so it takes about 4.3 s 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 12.2 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, the
            recovery time is 24.3 s. In contrast, a Scalable congestion
            control like DCTCP or Prague induces 2 congestion signals per
            round trip on average, which remains invariant for any flow rate,
            keeping dynamic control very tight.<a href="#section-5.1-1.14.2" class="pilcrow">¶</a></p>
<p id="section-5.1-1.14.3">For a
            feel of where the global average lone-flow download sits on this
            scale at the time of writing (2021), according to <span>[<a href="#BDPdata" class="cite xref">BDPdata</a>]</span>, the global average fixed access capacity was 103
            Mb/s in 2020 and the average base RTT to a CDN was 25 to 34 ms in 2019.
            Averaging of per-country data was weighted by Internet user
            population (data collected globally is necessarily of variable
            quality, but the paper does double-check that the outcome compares
            well against a 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, if the flow even lasted that long. This is
            described as 'at best' because it assumes everyone uses an AQM,
            whereas in reality, most users still have a (probably bloated)
            tail-drop buffer. 
            In the tail-drop case, the likely average recovery
            time would be at least 4 times 5 s, if not more, because RTT under load
            would be at least double that of an AQM, and the recovery time of Reno-friendly flows depends
            on the square of RTT.<a href="#section-5.1-1.14.3" class="pilcrow">¶</a></p>
<p id="section-5.1-1.14.4">Although work on
            scaling congestion controls tends to start with TCP as the
            transport, the above is not intended to exclude other transports
            (e.g., SCTP and QUIC) or less elastic algorithms
            (e.g., RMCAT), which all tend to adopt the same or similar
            developments.<a href="#section-5.1-1.14.4" class="pilcrow">¶</a></p>
</dd>
        <dd class="break"></dd>
</dl>
</section>
</div>
<div id="l4sps_why-not">
<section id="section-5.2">
        <h3 id="name-what-l4s-adds-to-existing-a">
<a href="#section-5.2" class="section-number selfRef">5.2. </a><a href="#name-what-l4s-adds-to-existing-a" class="section-name selfRef">What L4S Adds to Existing Approaches</a>
        </h3>
<p id="section-5.2-1">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
        improves on them, rather than being a mutually exclusive
        alternative:<a href="#section-5.2-1" class="pilcrow">¶</a></p>
<span class="break"></span><dl class="dlParallel" id="section-5.2-2">
          <dt id="section-5.2-2.1">Diffserv:</dt>
          <dd style="margin-left: 1.5em" id="section-5.2-2.2">
            <p id="section-5.2-2.2.1">Diffserv addresses the problem of
            bandwidth apportionment for important traffic as well as queuing
            latency for delay-sensitive traffic. Of these, L4S solely
            addresses the problem of queuing latency. Diffserv will still be
            necessary where important traffic requires priority (e.g., for
            commercial reasons or for protection of critical infrastructure
            traffic) -- see <span>[<a href="#I-D.briscoe-tsvwg-l4s-diffserv" class="cite xref">L4S-DIFFSERV</a>]</span>.
            Nonetheless, the L4S approach can provide low latency for all
            traffic within each Diffserv class (including the case where there
            is only the one default Diffserv class).<a href="#section-5.2-2.2.1" class="pilcrow">¶</a></p>
<p id="section-5.2-2.2.2">Also, Diffserv can only provide a latency benefit
            if a small subset of the traffic on a bottleneck link requests low
            latency. As already explained, it has no effect when all the
            applications in use at one time at a single site (e.g., a home, small
            business, or mobile device) require low latency. In contrast,
            because L4S works for all traffic, it needs none of the management
            baggage (traffic policing or traffic contracts) associated with
            favouring some packets over others. This lack of management
            baggage ought to give L4S a better chance of end-to-end
            deployment.<a href="#section-5.2-2.2.2" class="pilcrow">¶</a></p>
<p id="section-5.2-2.2.3">In particular, if networks do not trust end systems to identify which 
            packets should be favoured, they assign packets to Diffserv classes 
            themselves. However, the techniques available to such networks, like 
            inspection of flow identifiers or deeper inspection of application 
            signatures, do not always sit well with encryption of the layers above 
            IP <span>[<a href="#RFC8404" class="cite xref">RFC8404</a>]</span>. In these cases, users 
            can have either privacy or Quality of Service (QoS), but not both.<a href="#section-5.2-2.2.3" class="pilcrow">¶</a></p>
<p id="section-5.2-2.2.4">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 a need for
            a certain level of quality. Rather, it promises a certain
            behaviour (Scalable congestion response), which networks can
            objectively verify if they need to. This is because low delay
            depends on collective host behaviour, whereas bandwidth priority
            depends on network behaviour.<a href="#section-5.2-2.2.4" class="pilcrow">¶</a></p>
</dd>
          <dd class="break"></dd>
<dt id="section-5.2-2.3">State-of-the-art AQMs:</dt>
          <dd style="margin-left: 1.5em" id="section-5.2-2.4">AQMs for Classic traffic, such as PIE and FQ-CoDel,
            give a significant reduction in queuing delay relative to no AQM
            at all. L4S is intended to complement these AQMs and should not
            distract from the need to deploy them as widely as possible.
            Nonetheless, AQMs alone cannot reduce queuing delay too far
            without significantly reducing link utilization, because the root
            cause of the problem is on the host -- where Classic congestion
            controls use large sawtoothing rate variations. The L4S approach
            resolves this tension between delay and utilization by enabling
            hosts to minimize the amplitude of their sawteeth. A single-queue
            Classic AQM is not sufficient to allow hosts to use small sawteeth
            for two reasons: i) smaller sawteeth would not get lower delay in
            an AQM designed for larger amplitude Classic sawteeth, because a
            queue can only have one length at a time and ii) much smaller
            sawteeth implies much more frequent sawteeth, so L4S flows would
            drive a Classic AQM into a high level of ECN-marking, which would
            appear as heavy congestion to Classic flows, which in turn would
            greatly reduce their rate as a result (see <a href="#l4sarch_sec_classic-ecn-neck" class="auto internal xref">Section 6.4.4</a>).<a href="#section-5.2-2.4" class="pilcrow">¶</a>
</dd>
          <dd class="break"></dd>
<dt id="section-5.2-2.5">Per-flow queuing or marking:</dt>
          <dd style="margin-left: 1.5em" id="section-5.2-2.6">
            <p id="section-5.2-2.6.1">Similarly, per-flow
            approaches, such as FQ-CoDel or Approx Fair CoDel <span>[<a href="#AFCD" class="cite xref">AFCD</a>]</span>, are not incompatible with the L4S approach.
            However, per-flow queuing alone is not enough -- it only isolates
            the queuing of one flow from others, not from itself. Per-flow
            implementations need to have support for Scalable congestion
            control added, which has already been done for FQ-CoDel in Linux
            (see <span><a href="https://www.rfc-editor.org/rfc/rfc8290#section-5.2.7" class="relref">Section 5.2.7</a> of [<a href="#RFC8290" class="cite xref">RFC8290</a>]</span> and <span>[<a href="#FQ_CoDel_Thresh" class="cite xref">FQ_CoDel_Thresh</a>]</span>). Without this simple modification,
            per-flow AQMs, like FQ-CoDel, would still not be able to support
            applications that need both very low delay and high bandwidth,
            e.g., video-based control of remote procedures or interactive
            cloud-based video (see Note <a href="#l4sarch_note_app_shuffle" class="auto internal xref">1</a> below).<a href="#section-5.2-2.6.1" class="pilcrow">¶</a></p>
<p id="section-5.2-2.6.2">Although per-flow techniques are not incompatible
            with L4S, it is important to have the DualQ alternative. This is
            because handling end-to-end (layer 4) flows in the network (layer
            3 or 2) precludes some important end-to-end functions. For
            instance:<a href="#section-5.2-2.6.2" class="pilcrow">¶</a></p>
<ol start="1" type="a" class="normal type-a" id="section-5.2-2.6.3">
<li id="section-5.2-2.6.3.1">
                <p id="section-5.2-2.6.3.1.1">Per-flow forms of L4S, like FQ-CoDel, are incompatible with
                full end-to-end encryption of transport layer identifiers for
                privacy and confidentiality (e.g., IPsec or encrypted VPN
                tunnels, as opposed to DTLS over UDP), because they require
                packet inspection to access the end-to-end transport flow
                identifiers.<a href="#section-5.2-2.6.3.1.1" class="pilcrow">¶</a></p>
<p id="section-5.2-2.6.3.1.2">In contrast, the DualQ
                form of L4S requires no deeper inspection than the IP layer.
                So as long as operators take the DualQ approach, their users
                can have both very low queuing delay and full end-to-end
                encryption <span>[<a href="#RFC8404" class="cite xref">RFC8404</a>]</span>.<a href="#section-5.2-2.6.3.1.2" class="pilcrow">¶</a></p>
</li>
              <li id="section-5.2-2.6.3.2">
                <p id="section-5.2-2.6.3.2.1">With per-flow forms of L4S, the network takes over control of
                the relative rates of each application flow. Some see it as
                an advantage that the network will prevent some flows running
                faster than others. Others consider it an inherent part of the
                Internet's appeal that applications can control their rate
                while taking account of the needs of others via congestion
                signals.
                They maintain that this has allowed applications with
                interesting rate behaviours to evolve, for instance: i) a variable
                bit-rate video that varies around an equal share, rather than
                being forced to remain equal at every instant or ii) end-to-end
                scavenger behaviours <span>[<a href="#RFC6817" class="cite xref">RFC6817</a>]</span> that use
                less than an equal share of capacity <span>[<a href="#LEDBAT_AQM" class="cite xref">LEDBAT_AQM</a>]</span>.<a href="#section-5.2-2.6.3.2.1" class="pilcrow">¶</a></p>
<p id="section-5.2-2.6.3.2.2">The L4S
                architecture does not require the IETF to commit to one
                approach over the other, because it supports both so that the
                'market' can decide. Nonetheless, in the spirit of 'Do one
                thing and do it well' <span>[<a href="#McIlroy78" class="cite xref">McIlroy78</a>]</span>, the
                DualQ option provides low delay without prejudging the issue
                of flow-rate control. Then, flow rate policing can be added
                separately if desired. In contrast to scheduling, a policer would allow application control up to a
                point, but the network would still be able to set the point at
                which it intervened to prevent one flow completely starving
                another.<a href="#section-5.2-2.6.3.2.2" class="pilcrow">¶</a></p>
</li>
            </ol>
<p id="section-5.2-2.6.4">Note:<a href="#section-5.2-2.6.4" class="pilcrow">¶</a></p>
<ol start="1" type="1" class="normal type-1" id="section-5.2-2.6.5">
       <li id="section-5.2-2.6.5.1">
<div id="l4sarch_note_app_shuffle">It might seem that
                self-inflicted queuing delay within a per-flow queue should
                not be counted, because if the delay wasn't in the network, it
                would just shift to the sender. However, modern adaptive
                applications, e.g., HTTP/2 <span>[<a href="#RFC9113" class="cite xref">RFC9113</a>]</span>
                or some interactive media applications (see <a href="#l4sarch_apps" class="auto internal xref">Section 6.1</a>), can keep low latency objects at the
                front of their local send queue by shuffling priorities of
                other objects dependent on the progress of other transfers
                (for example, see <span>[<a href="#lowat" class="cite xref">lowat</a>]</span>). They cannot shuffle
                objects once they have released them into the network.<a href="#l4sarch_note_app_shuffle" class="pilcrow">¶</a>
</div>
            </li>
</ol>
</dd>
          <dd class="break"></dd>
<dt id="section-5.2-2.7">Alternative Back-off ECN (ABE):</dt>
          <dd style="margin-left: 1.5em" id="section-5.2-2.8">Here again, L4S is
            not an alternative to ABE but a complement that introduces much
            lower queuing delay. ABE <span>[<a href="#RFC8511" class="cite xref">RFC8511</a>]</span> alters the
            host behaviour in response to ECN marking to utilize a link better
            and give ECN flows faster throughput. It uses ECT(0) and assumes
            the network still treats ECN and drop the same. Therefore, ABE
            exploits any lower queuing delay that AQMs can provide. But, as
            explained above, AQMs still cannot reduce queuing delay too much
            without losing link utilization (to allow for other, non-ABE,
            flows).<a href="#section-5.2-2.8" class="pilcrow">¶</a>
</dd>
          <dd class="break"></dd>
<dt id="section-5.2-2.9">BBR:</dt>
          <dd style="margin-left: 1.5em" id="section-5.2-2.10">
            <p id="section-5.2-2.10.1">Bottleneck Bandwidth and Round-trip propagation
            time (BBR) <span>[<a href="#I-D.cardwell-iccrg-bbr-congestion-control" class="cite xref">BBR-CC</a>]</span> controls
            queuing delay end-to-end without needing any special logic in the
            network, such as an AQM. So it works pretty much on any path. BBR
            keeps queuing delay reasonably low, but perhaps not quite as low
            as with state-of-the-art AQMs, such as PIE or FQ-CoDel, and
            certainly nowhere near as low as with L4S. Queuing delay is also
            not consistently low, due to BBR's regular bandwidth probing
            spikes and its aggressive flow start-up phase.<a href="#section-5.2-2.10.1" class="pilcrow">¶</a></p>
<p id="section-5.2-2.10.2">L4S complements BBR. Indeed, BBRv2 can use L4S ECN
            where available and a Scalable L4S congestion control behaviour in
            response to any ECN signalling from the path <span>[<a href="#BBRv2" class="cite xref">BBRv2</a>]</span>. The L4S ECN signal complements the delay-based
            congestion control aspects of BBR with an explicit indication that
            hosts can use, both to 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 be assumed or estimated.<a href="#section-5.2-2.10.2" class="pilcrow">¶</a></p>
</dd>
        <dd class="break"></dd>
</dl>
</section>
</div>
</section>
</div>
<div id="l4sarch_applicability">
<section id="section-6">
      <h2 id="name-applicability">
<a href="#section-6" class="section-number selfRef">6. </a><a href="#name-applicability" class="section-name selfRef">Applicability</a>
      </h2>
<div id="l4sarch_apps">
<section id="section-6.1">
        <h3 id="name-applications">
<a href="#section-6.1" class="section-number selfRef">6.1. </a><a href="#name-applications" class="section-name selfRef">Applications</a>
        </h3>
<p id="section-6.1-1">A transport layer that solves the current latency issues will
        provide new service, product, and application opportunities.<a href="#section-6.1-1" class="pilcrow">¶</a></p>
<p id="section-6.1-2">With the L4S approach, the following existing applications also
        experience significantly better quality of experience under load:<a href="#section-6.1-2" class="pilcrow">¶</a></p>
<ul class="normal">
<li class="normal" id="section-6.1-3.1">gaming, including cloud-based gaming;<a href="#section-6.1-3.1" class="pilcrow">¶</a>
</li>
          <li class="normal" id="section-6.1-3.2">VoIP;<a href="#section-6.1-3.2" class="pilcrow">¶</a>
</li>
          <li class="normal" id="section-6.1-3.3">video conferencing;<a href="#section-6.1-3.3" class="pilcrow">¶</a>
</li>
          <li class="normal" id="section-6.1-3.4">web browsing;<a href="#section-6.1-3.4" class="pilcrow">¶</a>
</li>
          <li class="normal" id="section-6.1-3.5">(adaptive) video streaming; and<a href="#section-6.1-3.5" class="pilcrow">¶</a>
</li>
          <li class="normal" id="section-6.1-3.6">instant messaging.<a href="#section-6.1-3.6" class="pilcrow">¶</a>
</li>
        </ul>
<p id="section-6.1-4">The significantly lower queuing latency also enables some
        interactive application functions to be offloaded to the cloud that
        would hardly even be usable today, including:<a href="#section-6.1-4" class="pilcrow">¶</a></p>
<ul class="normal">
<li class="normal" id="section-6.1-5.1">cloud-based interactive video and<a href="#section-6.1-5.1" class="pilcrow">¶</a>
</li>
          <li class="normal" id="section-6.1-5.2">cloud-based virtual and augmented reality.<a href="#section-6.1-5.2" class="pilcrow">¶</a>
</li>
        </ul>
<p id="section-6.1-6">The above two applications have been successfully demonstrated with
        L4S, both running together over a 40 Mb/s broadband access link
        loaded up with the numerous other latency-sensitive applications in
        the previous list, as well as numerous downloads, with all sharing the same
        bottleneck queue simultaneously <span>[<a href="#L4Sdemo16" class="cite xref">L4Sdemo16</a>]</span> <span>[<a href="#L4Sdemo16-Video" class="cite xref">L4Sdemo16-Video</a>]</span>. For
        the former, a panoramic video of a football stadium could be swiped
        and pinched so that, on the fly, a proxy in the cloud could generate a
        sub-window of the match video under the finger-gesture control of each
        user. For the latter, a virtual reality headset displayed a viewport
        taken from a 360-degree camera in a racing car. The user's head
        movements controlled the viewport extracted by a cloud-based proxy. In
        both cases, with a 7 ms end-to-end base delay, the additional
        queuing delay of roughly 1 ms was so low that it seemed the video
        was generated locally.<a href="#section-6.1-6" class="pilcrow">¶</a></p>
<p id="section-6.1-7">Using a swiping finger gesture or head movement to pan a video are
        extremely latency-demanding actions -- far more demanding than
        VoIP -- because human vision can detect extremely low delays of the
        order of single milliseconds when delay is translated into a visual
        lag between a video and a reference point (the finger or the
        orientation of the head sensed by the balance system in the inner ear,
        i.e., the vestibular system). With an alternative AQM, the video
        noticeably lagged behind the finger gestures and head movements.<a href="#section-6.1-7" class="pilcrow">¶</a></p>
<p id="section-6.1-8">Without the low queuing delay of L4S, cloud-based applications like
        these would not be credible without significantly more access-network bandwidth 
        (to deliver all possible areas of the video that might be viewed) and
        more local processing, which would increase the weight and power
        consumption of head-mounted displays. When all interactive processing
        can be done in the cloud, only the data to be rendered for the end
        user needs to be sent.<a href="#section-6.1-8" class="pilcrow">¶</a></p>
<p id="section-6.1-9">Other low latency high bandwidth applications, such as:<a href="#section-6.1-9" class="pilcrow">¶</a></p>
<ul class="normal">
<li class="normal" id="section-6.1-10.1">interactive remote presence and<a href="#section-6.1-10.1" class="pilcrow">¶</a>
</li>
          <li class="normal" id="section-6.1-10.2">video-assisted remote control of machinery or industrial
            processes<a href="#section-6.1-10.2" class="pilcrow">¶</a>
</li>
        </ul>
<p id="section-6.1-11">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.<a href="#section-6.1-11" class="pilcrow">¶</a></p>
</section>
</div>
<section id="section-6.2">
        <h3 id="name-use-cases">
<a href="#section-6.2" class="section-number selfRef">6.2. </a><a href="#name-use-cases" class="section-name selfRef">Use Cases</a>
        </h3>
<p id="section-6.2-1">The following use cases for L4S are being considered by various
        interested parties:<a href="#section-6.2-1" class="pilcrow">¶</a></p>
<ul class="normal">
<li class="normal" id="section-6.2-2.1">where the bottleneck is one of various types of access network,
            e.g., DSL, Passive Optical Networks (PONs), DOCSIS cable,
            mobile, satellite; or where it's a Wi-Fi link (see <a href="#l4sarch_link-specifics" class="auto internal xref">Section 6.3</a> for
            some technology-specific details)<a href="#section-6.2-2.1" class="pilcrow">¶</a>
</li>
          <li class="normal" id="section-6.2-2.2">
            <p id="section-6.2-2.2.1">private networks of heterogeneous data centres, where there is
            no single administrator that can arrange for all the simultaneous
            changes to senders, receivers, and networks needed to deploy
            DCTCP:<a href="#section-6.2-2.2.1" class="pilcrow">¶</a></p>
<ul class="normal">
<li class="normal" id="section-6.2-2.2.2.1">a set of private data centres interconnected over a wide
                area with separate administrations but within the same
                company<a href="#section-6.2-2.2.2.1" class="pilcrow">¶</a>
</li>
              <li class="normal" id="section-6.2-2.2.2.2">a set of data centres operated by separate companies
                interconnected by a community of interest network
                (e.g., for the finance sector)<a href="#section-6.2-2.2.2.2" class="pilcrow">¶</a>
</li>
              <li class="normal" id="section-6.2-2.2.2.3">multi-tenant (cloud) data centres where tenants choose
                their operating system stack (Infrastructure as a Service
                (IaaS))<a href="#section-6.2-2.2.2.3" class="pilcrow">¶</a>
</li>
            </ul>
</li>
          <li class="normal" id="section-6.2-2.3">
            <p id="section-6.2-2.3.1">different types of transport (or application) congestion
            control:<a href="#section-6.2-2.3.1" class="pilcrow">¶</a></p>
<ul class="normal">
<li class="normal" id="section-6.2-2.3.2.1">elastic (TCP/SCTP);<a href="#section-6.2-2.3.2.1" class="pilcrow">¶</a>
</li>
              <li class="normal" id="section-6.2-2.3.2.2">real-time (RTP, RMCAT); and<a href="#section-6.2-2.3.2.2" class="pilcrow">¶</a>
</li>
              <li class="normal" id="section-6.2-2.3.2.3">query-response (DNS/LDAP).<a href="#section-6.2-2.3.2.3" class="pilcrow">¶</a>
</li>
            </ul>
</li>
          <li class="normal" id="section-6.2-2.4">
            <p id="section-6.2-2.4.1">where low delay QoS is required but without
            inspecting or intervening above the IP layer <span>[<a href="#RFC8404" class="cite xref">RFC8404</a>]</span>:<a href="#section-6.2-2.4.1" class="pilcrow">¶</a></p>
<ul class="normal">
<li class="normal" id="section-6.2-2.4.2.1">Mobile and other networks have tended to inspect higher
                layers in order to guess application QoS requirements.
                However, with growing demand for support of privacy and
                encryption, L4S offers an alternative. There is no need to
                select which traffic to favour for queuing when L4S can give
                favourable queuing to all traffic.<a href="#section-6.2-2.4.2.1" class="pilcrow">¶</a>
</li>
            </ul>
</li>
          <li class="normal" id="section-6.2-2.5">If queuing delay is minimized, applications with a fixed delay
            budget can communicate over longer distances or via more circuitous paths, e.g., longer
            chains of service functions <span>[<a href="#RFC7665" class="cite xref">RFC7665</a>]</span> or of onion
            routers.<a href="#section-6.2-2.5" class="pilcrow">¶</a>
</li>
          <li class="normal" id="section-6.2-2.6">If delay jitter is minimized, it is possible to reduce the
            dejitter buffers on the receiving end of video streaming, which
            should improve the interactive experience.<a href="#section-6.2-2.6" class="pilcrow">¶</a>
</li>
        </ul>
</section>
<div id="l4sarch_link-specifics">
<section id="section-6.3">
        <h3 id="name-applicability-with-specific">
<a href="#section-6.3" class="section-number selfRef">6.3. </a><a href="#name-applicability-with-specific" class="section-name selfRef">Applicability with Specific Link Technologies</a>
        </h3>
<p id="section-6.3-1">Certain link technologies aggregate data from multiple packets into
        bursts and buffer incoming packets while building each burst. Wi-Fi,
        PON, and cable all involve such packet aggregation, whereas fixed
        Ethernet and DSL do not. No sender, whether L4S or not, can do
        anything to reduce the buffering needed for packet aggregation. So an
        AQM should not count this buffering as part of the queue that it
        controls, given no amount of congestion signals will reduce it.<a href="#section-6.3-1" class="pilcrow">¶</a></p>
<p id="section-6.3-2">Certain link technologies also add buffering for other reasons,
        specifically:<a href="#section-6.3-2" class="pilcrow">¶</a></p>
<ul class="normal">
<li class="normal" id="section-6.3-3.1">Radio links (cellular, Wi-Fi, or satellite) that are distant from
            the source are particularly challenging. The radio link capacity
            can vary rapidly by orders of magnitude, so it is considered
            desirable to hold a standing queue that can utilize sudden
            increases of capacity.<a href="#section-6.3-3.1" class="pilcrow">¶</a>
</li>
          <li class="normal" id="section-6.3-3.2">Cellular networks are further complicated by a perceived need
            to buffer in order to make hand-overs imperceptible.<a href="#section-6.3-3.2" class="pilcrow">¶</a>
</li>
        </ul>
<p id="section-6.3-4">L4S cannot remove the need for all these different forms of
        buffering. However, by removing 'the longest pole in the tent'
        (buffering for the large sawteeth of Classic congestion controls), L4S
        exposes all these 'shorter poles' to greater scrutiny.<a href="#section-6.3-4" class="pilcrow">¶</a></p>
<p id="section-6.3-5">Until now, the buffering needed for these additional reasons tended
        to be over-specified -- with the excuse that none were 'the longest
        pole in the tent'. But having removed the 'longest pole', it becomes
        worthwhile to minimize them, for instance, reducing packet aggregation
        burst sizes and MAC scheduling intervals.<a href="#section-6.3-5" class="pilcrow">¶</a></p>
<p id="section-6.3-6">Also, certain link types, particularly radio-based links, are far
        more prone to transmission losses. <a href="#l4sarch_sec_non-l4s-neck" class="auto internal xref">Section 6.4.3</a> explains how an L4S response to
        loss has to be as drastic as a Classic response. Nonetheless, research
        referred to in the same section has demonstrated potential for
        considerably more effective loss repair at the link layer, due to the
        relaxed ordering constraints of L4S packets.<a href="#section-6.3-6" class="pilcrow">¶</a></p>
</section>
</div>
<section id="section-6.4">
        <h3 id="name-deployment-considerations">
<a href="#section-6.4" class="section-number selfRef">6.4. </a><a href="#name-deployment-considerations" class="section-name selfRef">Deployment Considerations</a>
        </h3>
<p id="section-6.4-1">L4S AQMs, whether DualQ <span>[<a href="#RFC9332" class="cite xref">RFC9332</a>]</span> or FQ <span>[<a href="#RFC8290" class="cite xref">RFC8290</a>]</span>, are in themselves an incremental deployment
        mechanism for L4S -- so that L4S traffic can coexist with existing
        Classic (Reno-friendly) traffic. <a href="#l4sarch_deploy_top" class="auto internal xref">Section 6.4.1</a>
        explains why only deploying an L4S AQM in one node at each end of the
        access link will realize nearly all the benefit of L4S.<a href="#section-6.4-1" class="pilcrow">¶</a></p>
<p id="section-6.4-2">L4S involves both the network and end systems, so <a href="#l4s_arch_deploy_seq" class="auto internal xref">Section 6.4.2</a> suggests some typical sequences to
        deploy each part and why there will be an immediate and significant
        benefit after deploying just one part.<a href="#section-6.4-2" class="pilcrow">¶</a></p>
<p id="section-6.4-3">Sections <a href="#l4sarch_sec_non-l4s-neck" class="auto internal xref">6.4.3</a> and <a href="#l4sarch_sec_classic-ecn-neck" class="auto internal xref">6.4.4</a> describe the converse
        incremental deployment case where there is no L4S AQM at the network
        bottleneck, so any L4S flow traversing this bottleneck has to take
        care in case it is competing with Classic traffic.<a href="#section-6.4-3" class="pilcrow">¶</a></p>
<div id="l4sarch_deploy_top">
<section id="section-6.4.1">
          <h4 id="name-deployment-topology">
<a href="#section-6.4.1" class="section-number selfRef">6.4.1. </a><a href="#name-deployment-topology" class="section-name selfRef">Deployment Topology</a>
          </h4>
<p id="section-6.4.1-1">L4S AQMs will not have to be deployed throughout the Internet
          before L4S can benefit anyone. Operators of public Internet access
          networks typically design their networks so that the bottleneck will
          nearly always occur at one known (logical) link. This confines the
          cost of queue management technology to one place.<a href="#section-6.4.1-1" class="pilcrow">¶</a></p>
<p id="section-6.4.1-2">The case of mesh networks is different and will be discussed
          later in this section. 
          However, the known-bottleneck case is generally
          true for Internet access to all sorts of different 'sites', where
          the word 'site' includes home networks, small- to medium-sized
          campus or enterprise networks and even cellular devices (<a href="#l4sarch_fig_access_topology" class="auto internal xref">Figure 2</a>).
   Also, this known-bottleneck
          case tends to be applicable whatever the access link technology,
          whether xDSL, cable, PON, cellular, line of sight wireless, or
          satellite.<a href="#section-6.4.1-2" class="pilcrow">¶</a></p>
<p id="section-6.4.1-3">Therefore, the full benefit of the L4S service should be
          available in the downstream direction when an L4S AQM is deployed at
          the ingress to this bottleneck link. And similarly, the full
          upstream service will typically be available once an L4S AQM is deployed at
          the ingress into the upstream link. (Of course, multihomed sites
          would only see the full benefit once all their access links were
          covered.)<a href="#section-6.4.1-3" class="pilcrow">¶</a></p>
<span id="name-likely-location-of-dualq-dq"></span><div id="l4sarch_fig_access_topology">
<figure id="figure-2">
            <div class="alignLeft art-text artwork" id="section-6.4.1-4.1">
<pre>
                                         ______
                                        (      )
                      __          __  (          )
                     |DQ\________/DQ|( enterprise )
                 ___ |__/        \__| ( /campus  )
                (   )                   (______)
              (      )                           ___||_
+----+      (          )  __                 __ /      \
| DC |-----(    Core    )|DQ\_______________/DQ|| home |
+----+      (          ) |__/               \__||______|
               (_____) __
                      |DQ\__/\        __ ,===.
                      |__/    \  ____/DQ||| ||mobile
                               \/    \__|||_||device
                                         | o |
                                         `---'

</pre>
</div>
<figcaption><a href="#figure-2" class="selfRef">Figure 2</a>:
<a href="#name-likely-location-of-dualq-dq" class="selfRef">Likely Location of DualQ (DQ) Deployments in Common Access Topologies</a>
            </figcaption></figure>
</div>
<p id="section-6.4.1-5">Deployment in mesh topologies depends on how overbooked the core
          is. If the core is non-blocking, or at least generously provisioned
          so that the edges are nearly always the bottlenecks, it would only
          be necessary to deploy an L4S AQM at the edge bottlenecks.
   For
          example, some data-centre networks are designed with the bottleneck
          in the hypervisor or host Network Interface Controllers (NICs), while others
   bottleneck at the
          top-of-rack switch (both the output ports facing hosts and those
          facing the core).<a href="#section-6.4.1-5" class="pilcrow">¶</a></p>
<p id="section-6.4.1-6">An L4S AQM would often next be needed where the Wi-Fi links in a
          home sometimes become the bottleneck. Also an L4S AQM would
          eventually need to be deployed at any other persistent
          bottlenecks, such as network interconnections, e.g., some public
          Internet exchange points and the ingress and egress to WAN links
          interconnecting data centres.<a href="#section-6.4.1-6" class="pilcrow">¶</a></p>
</section>
</div>
<div id="l4s_arch_deploy_seq">
<section id="section-6.4.2">
          <h4 id="name-deployment-sequences">
<a href="#section-6.4.2" class="section-number selfRef">6.4.2. </a><a href="#name-deployment-sequences" class="section-name selfRef">Deployment Sequences</a>
          </h4>
<p id="section-6.4.2-1">For any one L4S flow to provide benefit, it requires three (or
          sometimes two) parts to have been deployed: i) the congestion
          control at the sender; ii) the AQM at the bottleneck; and iii) older
          transports (namely TCP) need upgraded receiver feedback too. This
          was the same deployment problem that ECN faced <span>[<a href="#RFC8170" class="cite xref">RFC8170</a>]</span>, so we have learned from that experience.<a href="#section-6.4.2-1" class="pilcrow">¶</a></p>
<p id="section-6.4.2-2">Firstly, L4S deployment exploits the fact that DCTCP already
          exists on many Internet hosts (e.g., Windows, FreeBSD, and Linux), both
          servers and clients. Therefore, an L4S AQM can be deployed at a
          network bottleneck to immediately give a working deployment of all
          the L4S parts for testing, as long as the ECT(0) codepoint is
          switched to ECT(1). DCTCP needs some safety concerns to be fixed for
          general use over the public Internet (see <span><a href="https://www.rfc-editor.org/rfc/rfc9331#section-4.3" class="relref">Section 4.3</a> of the L4S ECN spec [<a href="#RFC9331" class="cite xref">RFC9331</a>]</span>), but DCTCP is
          not on by default, so these issues can be managed within controlled
          deployments or controlled trials.<a href="#section-6.4.2-2" class="pilcrow">¶</a></p>
<p id="section-6.4.2-3">Secondly, the performance improvement with L4S is so significant
          that it enables new interactive services and products that were not
          previously possible. It is much easier for companies to initiate new
          work on deployment if there is budget for a new product trial.
          In contrast, if there were only an incremental performance improvement
          (as with Classic ECN), spending on deployment tends to be much
          harder to justify.<a href="#section-6.4.2-3" class="pilcrow">¶</a></p>
<p id="section-6.4.2-4">Thirdly, the L4S identifier is defined so that network
          operators can initially enable L4S exclusively for certain customers or
          certain applications. However, this is carefully defined so that it does
          not compromise future evolution towards L4S as an Internet-wide
          service. 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 packet header or some status of a customer or their
          access link (see <span><a href="https://www.rfc-editor.org/rfc/rfc9331#section-5.4" class="relref">Section 5.4</a> of [<a href="#RFC9331" class="cite xref">RFC9331</a>]</span>). Operators could do this
          anyway, even if it were not blessed by the IETF. However, it is best
          for the IETF to specify that, if they use their own local
          identifier, it must be in combination with the IETF's identifier, ECT(1).
          Then, if an operator has opted for an exclusive local-use approach,
          they only have to remove this extra rule later to make the service
          work across the Internet -- it will already traverse middleboxes, peerings,
          etc.<a href="#section-6.4.2-4" class="pilcrow">¶</a></p>
<span id="name-example-l4s-deployment-sequ"></span><div id="l4s_arch_fig_deploy_seq">
<figure id="figure-3">
            <div class="alignLeft art-text artwork" id="section-6.4.2-5.1">
<pre>
+-+--------------------+----------------------+---------------------+
| | Servers or proxies |      Access link     |             Clients |
+-+--------------------+----------------------+---------------------+
|0| DCTCP (existing)   |                      |    DCTCP (existing) |
+-+--------------------+----------------------+---------------------+
|1|                    |Add L4S AQM downstream|                     |
| |       WORKS DOWNSTREAM FOR CONTROLLED DEPLOYMENTS/TRIALS        |
+-+--------------------+----------------------+---------------------+
|2| Upgrade DCTCP to   |                      |Replace DCTCP feedb'k|
| | TCP Prague         |                      |         with AccECN |
| |                 FULLY     WORKS     DOWNSTREAM                  |
+-+--------------------+----------------------+---------------------+
| |                    |                      |    Upgrade DCTCP to |
|3|                    | Add L4S AQM upstream |          TCP Prague |
| |                    |                      |                     |
| |              FULLY WORKS UPSTREAM AND DOWNSTREAM                |
+-+--------------------+----------------------+---------------------+

</pre>
</div>
<figcaption><a href="#figure-3" class="selfRef">Figure 3</a>:
<a href="#name-example-l4s-deployment-sequ" class="selfRef">Example L4S Deployment Sequence</a>
            </figcaption></figure>
</div>
<p id="section-6.4.2-6"><a href="#l4s_arch_fig_deploy_seq" class="auto internal xref">Figure 3</a> illustrates some example
          sequences in which the parts of L4S might be deployed. It consists
          of the following stages, preceded by a presumption that DCTCP is
          already installed at both ends:<a href="#section-6.4.2-6" class="pilcrow">¶</a></p>
<ol start="1" type="1" class="normal type-1" id="section-6.4.2-7">
<li id="section-6.4.2-7.1">
              <p id="section-6.4.2-7.1.1">DCTCP is not applicable for use over the public Internet, so
              it is emphasized here that any DCTCP flow has to be completely
              contained within a controlled trial environment.<a href="#section-6.4.2-7.1.1" class="pilcrow">¶</a></p>
<p id="section-6.4.2-7.1.2">Within this trial environment, once an L4S AQM
              has been deployed, the trial DCTCP flow will experience
              immediate benefit, without any other deployment being needed. In
              this example, downstream deployment is first, but in other
              scenarios, the upstream might be deployed first. If no AQM at all
              was previously deployed for the downstream access, an L4S AQM
              greatly improves the Classic service (as well as adding the L4S
              service). If an AQM was already deployed, the Classic service
              will be unchanged (and L4S will add an improvement on top).<a href="#section-6.4.2-7.1.2" class="pilcrow">¶</a></p>
</li>
            <li id="section-6.4.2-7.2">
              <p id="section-6.4.2-7.2.1">In this stage, the name 'TCP Prague' <span>[<a href="#I-D.briscoe-iccrg-prague-congestion-control" class="cite xref">PRAGUE-CC</a>]</span> is used
              to represent a variant of DCTCP that is designed to be used in a
              production Internet environment (that is, it has to comply with
              all the requirements in <span><a href="https://www.rfc-editor.org/rfc/rfc9331#section-4" class="relref">Section 4</a> of the L4S ECN spec [<a href="#RFC9331" class="cite xref">RFC9331</a>]</span>, which then means it can be
              used over the public Internet). If the application is primarily
              unidirectional, 'TCP Prague' at the sending end will provide all 
              the benefit needed, as long as the receiving end supports Accurate ECN (AccECN) 
              feedback <span>[<a href="#I-D.ietf-tcpm-accurate-ecn" class="cite xref">ACCECN</a>]</span>.<a href="#section-6.4.2-7.2.1" class="pilcrow">¶</a></p>
<p id="section-6.4.2-7.2.2">For TCP transports,
               AccECN feedback is needed at the other
              end, but it is a generic ECN feedback facility that is already
              planned to be deployed for other purposes, e.g., DCTCP and BBR.
              The two ends can be deployed in either order because, in TCP,
              an L4S congestion control only enables itself if it has
              negotiated the use of AccECN feedback with the other end during
              the connection handshake. Thus, deployment of TCP Prague on a
              server enables L4S trials to move to a production service in one
              direction, wherever AccECN is deployed at the other end. This
              stage might be further motivated by the performance improvements
              of TCP Prague relative to DCTCP (see  <span><a href="https://www.rfc-editor.org/rfc/rfc9331#appendix-A.2" class="relref">Appendix A.2</a> of the L4S ECN spec [<a href="#RFC9331" class="cite xref">RFC9331</a>]</span>).<a href="#section-6.4.2-7.2.2" class="pilcrow">¶</a></p>
<p id="section-6.4.2-7.2.3">Unlike TCP, from the outset, QUIC ECN
              feedback <span>[<a href="#RFC9000" class="cite xref">RFC9000</a>]</span> has supported L4S.
              Therefore, if the transport is QUIC, one-ended deployment of a
              Prague congestion control at this stage is simple and
              sufficient.<a href="#section-6.4.2-7.2.3" class="pilcrow">¶</a></p>
<p id="section-6.4.2-7.2.4">For QUIC, if a proxy sits in
              the path between multiple origin servers and the access
              bottlenecks to multiple clients, then upgrading the proxy with a
              Scalable congestion control would provide the benefits of L4S
              over all the clients' downstream bottlenecks in one go --
              whether or not all the origin servers were upgraded. Conversely,
              where a proxy has not been upgraded, 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 upgraded to support
              L4S.<a href="#section-6.4.2-7.2.4" class="pilcrow">¶</a></p>
<p id="section-6.4.2-7.2.5">For TCP, a proxy upgraded to support
              'TCP Prague' would provide the benefits of L4S downstream to all
              clients that support AccECN (whether or not they support L4S as
              well). And in the upstream, the proxy would also support AccECN
              as a receiver, so that any client deploying its own L4S support
              would benefit in the upstream direction, irrespective of whether
              any origin server beyond the proxy supported AccECN.<a href="#section-6.4.2-7.2.5" class="pilcrow">¶</a></p>
</li>
            <li id="section-6.4.2-7.3">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. To motivate the first of two independent moves, the
              deferred benefit of enabling new services after the second move
              has to be worth it to cover the first mover's investment risk.
              As explained already, the potential for new interactive services
              provides this motivation. An L4S AQM also improves the upstream
              Classic service significantly if no other AQM has already been
              deployed.<a href="#section-6.4.2-7.3" class="pilcrow">¶</a>
</li>
          </ol>
<p id="section-6.4.2-8">Note that other deployment sequences might occur. For
          instance, the upstream might be deployed first; a non-TCP protocol
          might be used end to end, e.g., QUIC and RTP; a body, such as the
          3GPP, might require L4S to be implemented in 5G user equipment; or
          other random acts of kindness might arise.<a href="#section-6.4.2-8" class="pilcrow">¶</a></p>
</section>
</div>
<div id="l4sarch_sec_non-l4s-neck">
<section id="section-6.4.3">
          <h4 id="name-l4s-flow-but-non-ecn-bottle">
<a href="#section-6.4.3" class="section-number selfRef">6.4.3. </a><a href="#name-l4s-flow-but-non-ecn-bottle" class="section-name selfRef">L4S Flow but Non-ECN Bottleneck</a>
          </h4>
<p id="section-6.4.3-1">If L4S is enabled between two hosts, the L4S sender is required
          to coexist safely with Reno in response to any drop (see <span><a href="https://www.rfc-editor.org/rfc/rfc9331#section-4.3" class="relref">Section 4.3</a> of the L4S ECN spec [<a href="#RFC9331" class="cite xref">RFC9331</a>]</span>).<a href="#section-6.4.3-1" class="pilcrow">¶</a></p>
<p id="section-6.4.3-2">Unfortunately, as well as protecting Classic traffic, this rule
          degrades the L4S service whenever there is any loss, even if the
          cause is not persistent congestion at a bottleneck, for example:<a href="#section-6.4.3-2" class="pilcrow">¶</a></p>
<ul class="normal">
<li class="normal" id="section-6.4.3-3.1">congestion loss at other transient bottlenecks, e.g., due
              to bursts in shallower queues;<a href="#section-6.4.3-3.1" class="pilcrow">¶</a>
</li>
            <li class="normal" id="section-6.4.3-3.2">transmission errors, e.g., due to electrical
              interference; and<a href="#section-6.4.3-3.2" class="pilcrow">¶</a>
</li>
            <li class="normal" id="section-6.4.3-3.3">rate policing.<a href="#section-6.4.3-3.3" class="pilcrow">¶</a>
</li>
          </ul>
<p id="section-6.4.3-4">Three complementary approaches are in progress to address this
          issue, but they are all currently research:<a href="#section-6.4.3-4" class="pilcrow">¶</a></p>
<ul class="normal">
<li class="normal" id="section-6.4.3-5.1">In Prague congestion control, ignore certain losses deemed
              unlikely to be due to congestion (using some ideas from
              BBR <span>[<a href="#I-D.cardwell-iccrg-bbr-congestion-control" class="cite xref">BBR-CC</a>]</span> regarding
              isolated losses). This could mask any of the above types of loss
              while still coexisting with drop-based congestion controls.<a href="#section-6.4.3-5.1" class="pilcrow">¶</a>
</li>
            <li class="normal" id="section-6.4.3-5.2">A combination of Recent Acknowledgement (RACK) <span>[<a href="#RFC8985" class="cite xref">RFC8985</a>]</span>, L4S, and link retransmission without
              resequencing could repair transmission errors without the head
              of line blocking delay usually associated with link-layer
              retransmission <span>[<a href="#UnorderedLTE" class="cite xref">UnorderedLTE</a>]</span> <span>[<a href="#RFC9331" class="cite xref">RFC9331</a>]</span>.<a href="#section-6.4.3-5.2" class="pilcrow">¶</a>
</li>
            <li class="normal" id="section-6.4.3-5.3">Hybrid ECN/drop rate policers (see <a href="#l4s_arch_sec_policing" class="auto internal xref">Section 8.3</a>).<a href="#section-6.4.3-5.3" class="pilcrow">¶</a>
</li>
          </ul>
<p id="section-6.4.3-6">L4S deployment scenarios that minimize these issues
          (e.g., over wireline networks) can proceed in parallel to this
          research, in the expectation that research success could continually
          widen L4S applicability.<a href="#section-6.4.3-6" class="pilcrow">¶</a></p>
</section>
</div>
<div id="l4sarch_sec_classic-ecn-neck">
<section id="section-6.4.4">
          <h4 id="name-l4s-flow-but-classic-ecn-bo">
<a href="#section-6.4.4" class="section-number selfRef">6.4.4. </a><a href="#name-l4s-flow-but-classic-ecn-bo" class="section-name selfRef">L4S Flow but Classic ECN Bottleneck</a>
          </h4>
<p id="section-6.4.4-1">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 all due to the addition of support for ECN in implementations of
          FQ-CoDel and/or FQ-COBALT, which is not generally problematic,
          because flow queue (FQ) scheduling inherently prevents a flow from
          exceeding the 'fair' rate irrespective of its aggressiveness.
          However, some of this Classic ECN marking might be due to
          single-queue ECN deployment. This case is discussed in
          <span><a href="https://www.rfc-editor.org/rfc/rfc9331#section-4.3" class="relref">Section 4.3</a> of the L4S ECN spec [<a href="#RFC9331" class="cite xref">RFC9331</a>]</span>.<a href="#section-6.4.4-1" class="pilcrow">¶</a></p>
</section>
</div>
<section id="section-6.4.5">
          <h4 id="name-l4s-aqm-deployment-within-t">
<a href="#section-6.4.5" class="section-number selfRef">6.4.5. </a><a href="#name-l4s-aqm-deployment-within-t" class="section-name selfRef">L4S AQM Deployment within Tunnels</a>
          </h4>
<p id="section-6.4.5-1">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,
          correct functioning of ECN signalling requires standards-compliant propagation
          of the ECN field up the layers <span>[<a href="#RFC6040" class="cite xref">RFC6040</a>]</span> <span>[<a href="#I-D.ietf-tsvwg-rfc6040update-shim" class="cite xref">ECN-SHIM</a>]</span> <span>[<a href="#I-D.ietf-tsvwg-ecn-encap-guidelines" class="cite xref">ECN-ENCAP</a>]</span>.<a href="#section-6.4.5-1" class="pilcrow">¶</a></p>
</section>
</section>
</section>
</div>
<div id="l4sps_IANA">
<section id="section-7">
      <h2 id="name-iana-considerations">
<a href="#section-7" class="section-number selfRef">7. </a><a href="#name-iana-considerations" class="section-name selfRef">IANA Considerations</a>
      </h2>
<p id="section-7-1">This document has no IANA actions.<a href="#section-7-1" class="pilcrow">¶</a></p>
</section>
</div>
<div id="l4sps_Security_Considerations">
<section id="section-8">
      <h2 id="name-security-considerations">
<a href="#section-8" class="section-number selfRef">8. </a><a href="#name-security-considerations" class="section-name selfRef">Security Considerations</a>
      </h2>
<section id="section-8.1">
        <h3 id="name-traffic-rate-non-policing">
<a href="#section-8.1" class="section-number selfRef">8.1. </a><a href="#name-traffic-rate-non-policing" class="section-name selfRef">Traffic Rate (Non-)Policing</a>
        </h3>
<section id="section-8.1.1">
          <h4 id="name-non-policing-rate-per-flow">
<a href="#section-8.1.1" class="section-number selfRef">8.1.1. </a><a href="#name-non-policing-rate-per-flow" class="section-name selfRef">(Non-)Policing Rate per Flow</a>
          </h4>
<p id="section-8.1.1-1">In the current Internet, ISPs usually enforce separation between
          the capacity of shared links assigned to different 'sites'
          (e.g., households, businesses, or mobile users -- see terminology
          in <a href="#l4sps_Terminology" class="auto internal xref">Section 3</a>) using some form of
          scheduler <span>[<a href="#RFC0970" class="cite xref">RFC0970</a>]</span>. And they use various
          techniques, like redirection to traffic scrubbing facilities, to deal
          with flooding attacks. However, there has never been a universal
          need to police the rate of individual application flows -- the
          Internet has generally always relied on self-restraint of congestion
          controls at senders for sharing intra-'site' capacity.<a href="#section-8.1.1-1" class="pilcrow">¶</a></p>
<p id="section-8.1.1-2">L4S has been designed not to upset this status quo. If a DualQ is
          used to provide L4S service, <span><a href="https://www.rfc-editor.org/rfc/rfc9332#section-4.2" class="relref">Section 4.2</a> of [<a href="#RFC9332" class="cite xref">RFC9332</a>]</span> explains how it is
          designed to give no more rate advantage to unresponsive flows than a
          single-queue AQM would, whether or not there is traffic
          overload.<a href="#section-8.1.1-2" class="pilcrow">¶</a></p>
<p id="section-8.1.1-3">Also, in case per-flow rate policing is ever required, it can be
          added because it is orthogonal to the distinction between L4S and
          Classic. As explained in <a href="#l4sps_why-not" class="auto internal xref">Section 5.2</a>, the DualQ
          variant of L4S provides low delay without prejudging the issue of
          flow-rate control. 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 modular addition to a DualQ.
          However, per-flow rate control is not usually deployed as a security
          mechanism, because an active attacker can just shard its traffic
          over more flow identifiers if the rate of each is restricted.<a href="#section-8.1.1-3" class="pilcrow">¶</a></p>
</section>
<section id="section-8.1.2">
          <h4 id="name-non-policing-l4s-service-ra">
<a href="#section-8.1.2" class="section-number selfRef">8.1.2. </a><a href="#name-non-policing-l4s-service-ra" class="section-name selfRef">(Non-)Policing L4S Service Rate</a>
          </h4>
<p id="section-8.1.2-1"><a href="#l4sps_why-not" class="auto internal xref">Section 5.2</a> explains how Diffserv only makes a
          difference if some packets get less favourable treatment than
          others, which typically requires traffic rate policing for a low
          latency class. In contrast, it should not be necessary to
          rate-police access to the L4S service to protect the Classic
          service, because L4S is designed to reduce delay without harming the
          delay or rate of any Classic traffic.<a href="#section-8.1.2-1" class="pilcrow">¶</a></p>
<p id="section-8.1.2-2">During early deployment (and perhaps always), some networks will
          not offer the L4S service. In general, these networks should not
          need to police L4S traffic. They are required (by both the ECN
          spec <span>[<a href="#RFC3168" class="cite xref">RFC3168</a>]</span> and the L4S ECN spec <span>[<a href="#RFC9331" class="cite xref">RFC9331</a>]</span>) not to change the L4S
          identifier, which would interfere with end-to-end congestion
          control. If they already treat ECN traffic as Not-ECT, they can
          merely treat L4S traffic as Not-ECT too. At a bottleneck, such
          networks will introduce some queuing and dropping. When a Scalable
          congestion control detects a drop, it will have to respond safely
          with respect to Classic congestion controls (as required in 
          <span><a href="https://www.rfc-editor.org/rfc/rfc9331#section-4.3" class="relref">Section 4.3</a> of [<a href="#RFC9331" class="cite xref">RFC9331</a>]</span>). This will
          degrade the L4S service to be no better (but never worse) than
          Classic best efforts whenever a non-ECN bottleneck is encountered
          on a path (see <a href="#l4sarch_sec_non-l4s-neck" class="auto internal xref">Section 6.4.3</a>).<a href="#section-8.1.2-2" class="pilcrow">¶</a></p>
<p id="section-8.1.2-3">In cases that are expected to be rare, networks that solely
          support Classic ECN <span>[<a href="#RFC3168" class="cite xref">RFC3168</a>]</span> in a single queue
          bottleneck might opt to police L4S traffic so as to protect
          competing Classic ECN traffic (for instance, see
          <span><a href="https://datatracker.ietf.org/doc/html/draft-ietf-tsvwg-l4sops-03#section-6.1.3" class="relref">Section 6.1.3</a> of the L4S operational guidance [<a href="#I-D.ietf-tsvwg-l4sops" class="cite xref">L4SOPS</a>]</span>). However, <span><a href="https://www.rfc-editor.org/rfc/rfc9331#section-4.3" class="relref">Section 4.3</a> of the L4S ECN spec [<a href="#RFC9331" class="cite xref">RFC9331</a>]</span> recommends
          that the sender adapts its congestion response to properly coexist
          with Classic ECN flows, i.e., reverting to the self-restraint
          approach.<a href="#section-8.1.2-3" class="pilcrow">¶</a></p>
<p id="section-8.1.2-4">Certain network operators might choose to restrict access to the
          L4S service, perhaps only to selected premium customers as a
          value-added service. Their packet classifier (item 2 in <a href="#l4sps_fig_components" class="auto internal xref">Figure 1</a>) could identify such customers
          against some other field (e.g., source address range), as well as
          classifying on the ECN field. If only the ECN L4S identifier
          matched, but not (say) the source address, the classifier could
          direct these packets (from non-premium customers) into the Classic
          queue. Explaining clearly how operators can use additional local
          classifiers (see <span><a href="https://www.rfc-editor.org/rfc/rfc9331#section-5.4" class="relref">Section 5.4</a> of [<a href="#RFC9331" class="cite xref">RFC9331</a>]</span>) is intended to remove any
          motivation to clear the L4S identifier. Then at least the L4S ECN
          identifier will be more likely to survive end to end, even though the
          service may not be supported at every hop.
   Such local arrangements
          would only require simple registered/not-registered packet
          classification, rather than the managed, application-specific
          traffic policing against customer-specific traffic contracts that
          Diffserv uses.<a href="#section-8.1.2-4" class="pilcrow">¶</a></p>
</section>
</section>
<section id="section-8.2">
        <h3 id="name-latency-friendliness">
<a href="#section-8.2" class="section-number selfRef">8.2. </a><a href="#name-latency-friendliness" class="section-name selfRef">'Latency Friendliness'</a>
        </h3>
<p id="section-8.2-1">Like the Classic service, the L4S service relies on self-restraint to
        limit the rate in response to congestion. In addition, the L4S
        service requires self-restraint in terms of limiting latency
        (burstiness). It is hoped that self-interest and guidance on dynamic
        behaviour (especially flow start-up, which might need to be
        standardized) will be sufficient to prevent transports from sending
        excessive bursts of L4S traffic, given the application's own latency
        will suffer most from such behaviour.<a href="#section-8.2-1" class="pilcrow">¶</a></p>
<p id="section-8.2-2">Because the L4S service can reduce delay without discernibly
        increasing the delay of any Classic traffic, it should not be
        necessary to police L4S traffic to protect the delay of Classic traffic.
        However, whether burst policing becomes necessary to protect other L4S
        traffic remains to be seen. Without it, there will be potential for
        attacks on the low latency of the L4S service.<a href="#section-8.2-2" class="pilcrow">¶</a></p>
<p id="section-8.2-3">If needed, various arrangements could be used to address this
        concern:<a href="#section-8.2-3" class="pilcrow">¶</a></p>
<span class="break"></span><dl class="dlParallel" id="section-8.2-4">
          <dt id="section-8.2-4.1">Local bottleneck queue protection:</dt>
          <dd style="margin-left: 1.5em" id="section-8.2-4.2">A per-flow
            (5-tuple) queue protection function <span>[<a href="#I-D.briscoe-docsis-q-protection" class="cite xref">DOCSIS-Q-PROT</a>]</span> has been developed for
            the low latency queue in DOCSIS, which has adopted the DualQ L4S
            architecture. It protects the low latency service from any
            queue-building flows that accidentally or maliciously classify
            themselves into the low latency queue. It is designed to score
            flows based solely on their contribution to queuing (not flow rate
            in itself). Then, if the shared low latency queue is at risk of
            exceeding a threshold, the function redirects enough packets of
            the highest scoring flow(s) into the Classic queue to preserve low
            latency.<a href="#section-8.2-4.2" class="pilcrow">¶</a>
</dd>
          <dd class="break"></dd>
<dt id="section-8.2-4.3">Distributed traffic scrubbing:</dt>
          <dd style="margin-left: 1.5em" id="section-8.2-4.4">Rather than policing
            locally at each bottleneck, it may only be necessary to address
            problems reactively, e.g., punitively target any deployments
            of new bursty malware, in a similar way to how traffic from 
            flooding attack sources is rerouted via scrubbing facilities.<a href="#section-8.2-4.4" class="pilcrow">¶</a>
</dd>
          <dd class="break"></dd>
<dt id="section-8.2-4.5">Local bottleneck per-flow scheduling:</dt>
          <dd style="margin-left: 1.5em" id="section-8.2-4.6">Per-flow
            scheduling should inherently isolate non-bursty flows from bursty flows
            (see <a href="#l4sps_why-not" class="auto internal xref">Section 5.2</a> for discussion of the merits
            of per-flow scheduling relative to per-flow policing).<a href="#section-8.2-4.6" class="pilcrow">¶</a>
</dd>
          <dd class="break"></dd>
<dt id="section-8.2-4.7">Distributed access subnet queue protection:</dt>
          <dd style="margin-left: 1.5em" id="section-8.2-4.8">Per-flow
            queue protection could be arranged for a queue structure
            distributed across a subnet intercommunicating using lower layer
            control messages (see Section 2.1.4 of <span>[<a href="#QDyn" class="cite xref">QDyn</a>]</span>). For
            instance, in a radio access network, user equipment already sends
            regular buffer status reports to a radio network controller, which
            could use this information to remotely police individual
            flows.<a href="#section-8.2-4.8" class="pilcrow">¶</a>
</dd>
          <dd class="break"></dd>
<dt id="section-8.2-4.9">Distributed Congestion Exposure to ingress policers:</dt>
          <dd style="margin-left: 1.5em" id="section-8.2-4.10">The
            Congestion Exposure (ConEx) architecture <span>[<a href="#RFC7713" class="cite xref">RFC7713</a>]</span> uses an egress audit to motivate senders to
            truthfully signal path congestion in-band, where it can be used by
            ingress policers. An edge-to-edge variant of this architecture is
            also possible.<a href="#section-8.2-4.10" class="pilcrow">¶</a>
</dd>
          <dd class="break"></dd>
<dt id="section-8.2-4.11">Distributed domain-edge traffic conditioning:</dt>
          <dd style="margin-left: 1.5em" id="section-8.2-4.12">An
            architecture similar to Diffserv <span>[<a href="#RFC2475" class="cite xref">RFC2475</a>]</span> may
            be preferred, where traffic is proactively conditioned on entry to
            a domain, rather than reactively policed only if it leads to
            queuing once combined with other traffic at a bottleneck.<a href="#section-8.2-4.12" class="pilcrow">¶</a>
</dd>
          <dd class="break"></dd>
<dt id="section-8.2-4.13">Distributed core network queue protection:</dt>
          <dd style="margin-left: 1.5em" id="section-8.2-4.14">The
            policing function could be divided between per-flow mechanisms at
            the network ingress that characterize the burstiness of each flow
            into a signal carried with the traffic and per-class mechanisms
            at bottlenecks that act on these signals if queuing actually
            occurs once the traffic converges. This would be somewhat similar
            to <span>[<a href="#Nadas20" class="cite xref">Nadas20</a>]</span>, which is in turn similar to the idea
            behind core stateless fair queuing.<a href="#section-8.2-4.14" class="pilcrow">¶</a>
</dd>
        <dd class="break"></dd>
</dl>
<p id="section-8.2-5">No single one of these possible queue protection capabilities is
        considered an essential part of the L4S architecture, which works
        without any of them under non-attack conditions (much as the Internet
        normally works without per-flow rate policing). 
        Indeed, even where
        latency policers are deployed, under normal circumstances, they would
        not intervene, and if operators found they were not necessary, they
        could disable them. Part of the L4S experiment will be to see whether
        such a function is necessary and which arrangements are most
        appropriate to the size of the problem.<a href="#section-8.2-5" class="pilcrow">¶</a></p>
</section>
<div id="l4s_arch_sec_policing">
<section id="section-8.3">
        <h3 id="name-interaction-between-rate-po">
<a href="#section-8.3" class="section-number selfRef">8.3. </a><a href="#name-interaction-between-rate-po" class="section-name selfRef">Interaction between Rate Policing and L4S</a>
        </h3>
<p id="section-8.3-1">As mentioned in <a href="#l4sps_why-not" class="auto internal xref">Section 5.2</a>, L4S should remove
        the need for low latency Diffserv classes. However, those Diffserv
        classes that give certain applications or users priority over
        capacity would still be applicable in certain scenarios
        (e.g., corporate networks). Then, within such Diffserv classes,
        L4S would often be applicable to give traffic low latency and low loss
        as well. Within such a Diffserv class, the bandwidth available to a
        user or application is often limited by a rate policer. Similarly, in
        the default Diffserv class, rate policers are sometimes used to
        partition shared capacity.<a href="#section-8.3-1" class="pilcrow">¶</a></p>
<p id="section-8.3-2">A Classic rate policer drops any packets exceeding a set rate,
        usually also giving a burst allowance (variants exist where the
        policer re-marks noncompliant traffic to a discard-eligible Diffserv
        codepoint, so they can be dropped elsewhere during contention).
        Whenever L4S traffic encounters one of these rate policers, it will
        experience drops and the source will have to fall back to a Classic
        congestion control, thus losing the benefits of L4S (<a href="#l4sarch_sec_non-l4s-neck" class="auto internal xref">Section 6.4.3</a>). So in networks that already use
        rate policers and plan to deploy L4S, it will be preferable to
        redesign these rate policers to be more friendly to the L4S
        service.<a href="#section-8.3-2" class="pilcrow">¶</a></p>
<p id="section-8.3-3">L4S-friendly rate policing is currently a research area (note that
        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
        just under the policed rate or just under the burst allowance where
        drop is introduced. For instance, the two-rate, three-colour
        marker <span>[<a href="#RFC2698" class="cite xref">RFC2698</a>]</span> or a PCN threshold and
        excess-rate marker <span>[<a href="#RFC5670" class="cite xref">RFC5670</a>]</span> could mark ECN at the
        lower rate and drop at the higher. Or an existing rate policer could
        have congestion-rate policing added, e.g., using the 'local'
        (non-ConEx) variant of the ConEx aggregate congestion
        policer <span>[<a href="#I-D.briscoe-conex-policing" class="cite xref">CONG-POLICING</a>]</span>. It might
        also be possible to design Scalable congestion controls to respond
        less catastrophically to loss that has not been preceded by a period
        of increasing delay.<a href="#section-8.3-3" class="pilcrow">¶</a></p>
<p id="section-8.3-4">The design of L4S-friendly rate policers will require a separate,
        dedicated document. For further discussion of the interaction between
        L4S and Diffserv, see <span>[<a href="#I-D.briscoe-tsvwg-l4s-diffserv" class="cite xref">L4S-DIFFSERV</a>]</span>.<a href="#section-8.3-4" class="pilcrow">¶</a></p>
</section>
</div>
<section id="section-8.4">
        <h3 id="name-ecn-integrity">
<a href="#section-8.4" class="section-number selfRef">8.4. </a><a href="#name-ecn-integrity" class="section-name selfRef">ECN Integrity</a>
        </h3>
<p id="section-8.4-1">Various ways have been developed to protect the integrity of the
        congestion feedback loop (whether signalled by loss, Classic ECN, or
        L4S ECN) against misbehaviour by the receiver, sender, or network (or
        all three). Brief details of each, including applicability, pros, and
        cons, are given in <span><a href="https://www.rfc-editor.org/rfc/rfc9331#appendix-C.1" class="relref">Appendix C.1</a> of the L4S ECN spec [<a href="#RFC9331" class="cite xref">RFC9331</a>]</span>.<a href="#section-8.4-1" class="pilcrow">¶</a></p>
</section>
<section id="section-8.5">
        <h3 id="name-privacy-considerations">
<a href="#section-8.5" class="section-number selfRef">8.5. </a><a href="#name-privacy-considerations" class="section-name selfRef">Privacy Considerations</a>
        </h3>
<p id="section-8.5-1">As discussed in <a href="#l4sps_why-not" class="auto internal xref">Section 5.2</a>, the L4S
        architecture does not preclude approaches that inspect end-to-end
        transport layer identifiers. For instance, L4S support has been added
        to FQ-CoDel, which classifies by application flow identifier in the network.
        However, the main innovation of L4S is the DualQ AQM framework that
        does not need to inspect any deeper than the outermost IP header,
        because the L4S identifier is in the IP-ECN field.<a href="#section-8.5-1" class="pilcrow">¶</a></p>
<p id="section-8.5-2">Thus, the L4S architecture enables very low queuing delay without
        <em>requiring</em> inspection of information above
        the IP layer. This means that users who want to encrypt application
        flow identifiers, e.g., in IPsec or other encrypted VPN tunnels,
        don't have to sacrifice low delay <span>[<a href="#RFC8404" class="cite xref">RFC8404</a>]</span>.<a href="#section-8.5-2" class="pilcrow">¶</a></p>
<p id="section-8.5-3">Because L4S can provide low delay for a broad set of 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 while
        traversing networks. This removes much of the ability to correlate
        between the delay requirements of traffic and other identifying
        features <span>[<a href="#RFC6973" class="cite xref">RFC6973</a>]</span>. There may be some types of
        traffic that prefer not to use L4S, but the coarse binary
        categorization of traffic reveals very little that could be exploited
        to compromise privacy.<a href="#section-8.5-3" class="pilcrow">¶</a></p>
</section>
</section>
</div>
<section id="section-9">
      <h2 id="name-informative-references">
<a href="#section-9" class="section-number selfRef">9. </a><a href="#name-informative-references" class="section-name selfRef">Informative References</a>
      </h2>
<dl class="references">
<dt id="I-D.ietf-tcpm-accurate-ecn">[ACCECN]</dt>
      <dd>
<span class="refAuthor">Briscoe, B.</span>, <span class="refAuthor">Kühlewind, M.</span>, and <span class="refAuthor">R. Scheffenegger</span>, <span class="refTitle">"More Accurate ECN Feedback in TCP"</span>, <span class="refContent">Work in Progress</span>, <span class="seriesInfo">Internet-Draft, draft-ietf-tcpm-accurate-ecn-22</span>, <time datetime="2022-11-09" class="refDate">9 November 2022</time>, <span>&lt;<a href="https://datatracker.ietf.org/doc/html/draft-ietf-tcpm-accurate-ecn-22">https://datatracker.ietf.org/doc/html/draft-ietf-tcpm-accurate-ecn-22</a>&gt;</span>. </dd>
<dd class="break"></dd>
<dt id="AFCD">[AFCD]</dt>
      <dd>
<span class="refAuthor">Xue, L.</span>, <span class="refAuthor">Kumar, S.</span>, <span class="refAuthor">Cui, C.</span>, <span class="refAuthor">Kondikoppa, P.</span>, <span class="refAuthor">Chiu, C-H.</span>, and <span class="refAuthor">S-J. Park</span>, <span class="refTitle">"Towards fair and low latency next generation high speed networks: AFCD queuing"</span>, <span class="refContent">Journal of Network and Computer Applications, Volume 70, pp. 183-193</span>, <span class="seriesInfo">DOI 10.1016/j.jnca.2016.03.021</span>, <time datetime="2016-07" class="refDate">July 2016</time>, <span>&lt;<a href="https://doi.org/10.1016/j.jnca.2016.03.021">https://doi.org/10.1016/j.jnca.2016.03.021</a>&gt;</span>. </dd>
<dd class="break"></dd>
<dt id="I-D.cardwell-iccrg-bbr-congestion-control">[BBR-CC]</dt>
      <dd>
<span class="refAuthor">Cardwell, N.</span>, <span class="refAuthor">Cheng, Y.</span>, <span class="refAuthor">Hassas Yeganeh, S.</span>, <span class="refAuthor">Swett, I.</span>, and <span class="refAuthor">V. Jacobson</span>, <span class="refTitle">"BBR Congestion Control"</span>, <span class="refContent">Work in Progress</span>, <span class="seriesInfo">Internet-Draft, draft-cardwell-iccrg-bbr-congestion-control-02</span>, <time datetime="2022-03-07" class="refDate">7 March 2022</time>, <span>&lt;<a href="https://datatracker.ietf.org/doc/html/draft-cardwell-iccrg-bbr-congestion-control-02">https://datatracker.ietf.org/doc/html/draft-cardwell-iccrg-bbr-congestion-control-02</a>&gt;</span>. </dd>
<dd class="break"></dd>
<dt id="BBRv2">[BBRv2]</dt>
      <dd>
<span class="refTitle">"TCP BBR v2 Alpha/Preview Release"</span>, <span class="refContent">commit 17700ca</span>, <time datetime="2022-06" class="refDate">June 2022</time>, <span>&lt;<a href="https://github.com/google/bbr">https://github.com/google/bbr</a>&gt;</span>. </dd>
<dd class="break"></dd>
<dt id="BDPdata">[BDPdata]</dt>
      <dd>
<span class="refAuthor">Briscoe, B.</span>, <span class="refTitle">"PI2 Parameters"</span>, <span class="refContent">TR-BB-2021-001, arXiv:2107.01003 [cs.NI]</span>, <span class="seriesInfo">DOI 10.48550/arXiv.2107.01003</span>, <time datetime="2021-10" class="refDate">October 2021</time>, <span>&lt;<a href="https://arxiv.org/abs/2107.01003">https://arxiv.org/abs/2107.01003</a>&gt;</span>. </dd>
<dd class="break"></dd>
<dt id="BufferSize">[BufferSize]</dt>
      <dd>
<span class="refAuthor">Appenzeller, G.</span>, <span class="refAuthor">Keslassy, I.</span>, and <span class="refAuthor">N. McKeown</span>, <span class="refTitle">"Sizing Router Buffers"</span>, <span class="refContent">SIGCOMM '04: Proceedings of the 2004 conference on Applications, technologies, architectures, and protocols for computer communications, pp. 281-292</span>, <span class="seriesInfo">DOI 10.1145/1015467.1015499</span>, <time datetime="2004-10" class="refDate">October 2004</time>, <span>&lt;<a href="https://doi.org/10.1145/1015467.1015499">https://doi.org/10.1145/1015467.1015499</a>&gt;</span>. </dd>
<dd class="break"></dd>
<dt id="COBALT">[COBALT]</dt>
      <dd>
<span class="refAuthor">Palmei, J.</span>, <span class="refAuthor">Gupta, S.</span>, <span class="refAuthor">Imputato, P.</span>, <span class="refAuthor">Morton, J.</span>, <span class="refAuthor">Tahiliani, M. P.</span>, <span class="refAuthor">Avallone, S.</span>, and <span class="refAuthor">D. Täht</span>, <span class="refTitle">"Design and Evaluation of COBALT Queue Discipline"</span>, <span class="refContent">IEEE International Symposium on Local and Metropolitan Area Networks (LANMAN)</span>, <span class="seriesInfo">DOI 10.1109/LANMAN.2019.8847054</span>, <time datetime="2019-07" class="refDate">July 2019</time>, <span>&lt;<a href="https://ieeexplore.ieee.org/abstract/document/8847054">https://ieeexplore.ieee.org/abstract/document/8847054</a>&gt;</span>. </dd>
<dd class="break"></dd>
<dt id="I-D.morton-tsvwg-codel-approx-fair">[CODEL-APPROX-FAIR]</dt>
      <dd>
<span class="refAuthor">Morton, J.</span> and <span class="refAuthor">P. Heist</span>, <span class="refTitle">"Controlled Delay Approximate Fairness AQM"</span>, <span class="refContent">Work in Progress</span>, <span class="seriesInfo">Internet-Draft, draft-morton-tsvwg-codel-approx-fair-01</span>, <time datetime="2020-03-09" class="refDate">9 March 2020</time>, <span>&lt;<a href="https://datatracker.ietf.org/doc/html/draft-morton-tsvwg-codel-approx-fair-01">https://datatracker.ietf.org/doc/html/draft-morton-tsvwg-codel-approx-fair-01</a>&gt;</span>. </dd>
<dd class="break"></dd>
<dt id="I-D.briscoe-conex-policing">[CONG-POLICING]</dt>
      <dd>
<span class="refAuthor">Briscoe, B.</span>, <span class="refTitle">"Network Performance Isolation using Congestion Policing"</span>, <span class="refContent">Work in Progress</span>, <span class="seriesInfo">Internet-Draft, draft-briscoe-conex-policing-01</span>, <time datetime="2014-02-14" class="refDate">14 February 2014</time>, <span>&lt;<a href="https://datatracker.ietf.org/doc/html/draft-briscoe-conex-policing-01">https://datatracker.ietf.org/doc/html/draft-briscoe-conex-policing-01</a>&gt;</span>. </dd>
<dd class="break"></dd>
<dt id="I-D.sridharan-tcpm-ctcp">[CTCP]</dt>
      <dd>
<span class="refAuthor">Sridharan, M.</span>, <span class="refAuthor">Tan, K.</span>, <span class="refAuthor">Bansal, D.</span>, and <span class="refAuthor">D. Thaler</span>, <span class="refTitle">"Compound TCP: A New TCP Congestion Control for High-Speed and Long Distance Networks"</span>, <span class="refContent">Work in Progress</span>, <span class="seriesInfo">Internet-Draft, draft-sridharan-tcpm-ctcp-02</span>, <time datetime="2008-11-11" class="refDate">11 November 2008</time>, <span>&lt;<a href="https://datatracker.ietf.org/doc/html/draft-sridharan-tcpm-ctcp-02">https://datatracker.ietf.org/doc/html/draft-sridharan-tcpm-ctcp-02</a>&gt;</span>. </dd>
<dd class="break"></dd>
<dt id="I-D.briscoe-docsis-q-protection">[DOCSIS-Q-PROT]</dt>
      <dd>
<span class="refAuthor">Briscoe, B., Ed.</span> and <span class="refAuthor">G. White</span>, <span class="refTitle">"The DOCSIS® Queue Protection Algorithm to Preserve Low Latency"</span>, <span class="refContent">Work in Progress</span>, <span class="seriesInfo">Internet-Draft, draft-briscoe-docsis-q-protection-06</span>, <time datetime="2022-05-13" class="refDate">13 May 2022</time>, <span>&lt;<a href="https://datatracker.ietf.org/doc/html/draft-briscoe-docsis-q-protection-06">https://datatracker.ietf.org/doc/html/draft-briscoe-docsis-q-protection-06</a>&gt;</span>. </dd>
<dd class="break"></dd>
<dt id="DOCSIS3.1">[DOCSIS3.1]</dt>
      <dd>
<span class="refAuthor">CableLabs</span>, <span class="refTitle">"MAC and Upper Layer Protocols Interface (MULPI) Specification, CM-SP-MULPIv3.1"</span>, <span class="seriesInfo">Data-Over-Cable Service Interface Specifications DOCSIS 3.1 Version i17 or later</span>, <time datetime="2019-01-21" class="refDate">21 January 2019</time>, <span>&lt;<a href="https://specification-search.cablelabs.com/CM-SP-MULPIv3.1">https://specification-search.cablelabs.com/CM-SP-MULPIv3.1</a>&gt;</span>. </dd>
<dd class="break"></dd>
<dt id="DOCSIS3AQM">[DOCSIS3AQM]</dt>
      <dd>
<span class="refAuthor">White, G.</span>, <span class="refTitle">"Active Queue Management Algorithms for DOCSIS 3.0: A Simulation Study of CoDel, SFQ-CoDel and PIE in DOCSIS 3.0 Networks"</span>, <span class="refContent">CableLabs Technical Report</span>, <time datetime="2013-04" class="refDate">April 2013</time>, <span>&lt;<a href="https://www.cablelabs.com/wp-content/uploads/2013/11/Active_Queue_Management_Algorithms_DOCSIS_3_0.pdf">https://www.cablelabs.com/wp-content/uploads/2013/11/Active_Queue_Management_Algorithms_DOCSIS_3_0.pdf</a>&gt;</span>. </dd>
<dd class="break"></dd>
<dt id="DualPI2Linux">[DualPI2Linux]</dt>
      <dd>
<span class="refAuthor">Albisser, O.</span>, <span class="refAuthor">De Schepper, K.</span>, <span class="refAuthor">Briscoe, B.</span>, <span class="refAuthor">Tilmans, O.</span>, and <span class="refAuthor">H. Steen</span>, <span class="refTitle">"DUALPI2 - Low Latency, Low Loss and Scalable (L4S) AQM"</span>, <span class="refContent">Proceedings of Linux Netdev 0x13</span>, <time datetime="2019-03" class="refDate">March 2019</time>, <span>&lt;<a href="https://www.netdevconf.org/0x13/session.html?talk-DUALPI2-AQM">https://www.netdevconf.org/0x13/session.html?talk-DUALPI2-AQM</a>&gt;</span>. </dd>
<dd class="break"></dd>
<dt id="Dukkipati06">[Dukkipati06]</dt>
      <dd>
<span class="refAuthor">Dukkipati, N.</span> and <span class="refAuthor">N. McKeown</span>, <span class="refTitle">"Why Flow-Completion Time is the Right Metric for Congestion Control"</span>, <span class="refContent">ACM SIGCOMM Computer Communication Review, Volume 36, Issue 1, pp. 59-62</span>, <span class="seriesInfo">DOI 10.1145/1111322.1111336</span>, <time datetime="2006-01" class="refDate">January 2006</time>, <span>&lt;<a href="https://dl.acm.org/doi/10.1145/1111322.1111336">https://dl.acm.org/doi/10.1145/1111322.1111336</a>&gt;</span>. </dd>
<dd class="break"></dd>
<dt id="I-D.ietf-tsvwg-ecn-encap-guidelines">[ECN-ENCAP]</dt>
      <dd>
<span class="refAuthor">Briscoe, B.</span> and <span class="refAuthor">J. Kaippallimalil</span>, <span class="refTitle">"Guidelines for Adding Congestion Notification to Protocols that Encapsulate IP"</span>, <span class="refContent">Work in Progress</span>, <span class="seriesInfo">Internet-Draft, draft-ietf-tsvwg-ecn-encap-guidelines-17</span>, <time datetime="2022-07-11" class="refDate">11 July 2022</time>, <span>&lt;<a href="https://datatracker.ietf.org/doc/html/draft-ietf-tsvwg-ecn-encap-guidelines-17">https://datatracker.ietf.org/doc/html/draft-ietf-tsvwg-ecn-encap-guidelines-17</a>&gt;</span>. </dd>
<dd class="break"></dd>
<dt id="I-D.stewart-tsvwg-sctpecn">[ECN-SCTP]</dt>
      <dd>
<span class="refAuthor">Stewart, R.</span>, <span class="refAuthor">Tuexen, M.</span>, and <span class="refAuthor">X. Dong</span>, <span class="refTitle">"ECN for Stream Control Transmission Protocol (SCTP)"</span>, <span class="refContent">Work in Progress</span>, <span class="seriesInfo">Internet-Draft, draft-stewart-tsvwg-sctpecn-05</span>, <time datetime="2014-01-15" class="refDate">15 January 2014</time>, <span>&lt;<a href="https://datatracker.ietf.org/doc/html/draft-stewart-tsvwg-sctpecn-05">https://datatracker.ietf.org/doc/html/draft-stewart-tsvwg-sctpecn-05</a>&gt;</span>. </dd>
<dd class="break"></dd>
<dt id="I-D.ietf-tsvwg-rfc6040update-shim">[ECN-SHIM]</dt>
      <dd>
<span class="refAuthor">Briscoe, B.</span>, <span class="refTitle">"Propagating Explicit Congestion Notification Across IP Tunnel Headers Separated by a Shim"</span>, <span class="refContent">Work in Progress</span>, <span class="seriesInfo">Internet-Draft, draft-ietf-tsvwg-rfc6040update-shim-15</span>, <time datetime="2022-07-11" class="refDate">11 July 2022</time>, <span>&lt;<a href="https://datatracker.ietf.org/doc/html/draft-ietf-tsvwg-rfc6040update-shim-15">https://datatracker.ietf.org/doc/html/draft-ietf-tsvwg-rfc6040update-shim-15</a>&gt;</span>. </dd>
<dd class="break"></dd>
<dt id="FQ_CoDel_Thresh">[FQ_CoDel_Thresh]</dt>
      <dd>
<span class="refTitle">"fq_codel: generalise ce_threshold marking for subset of traffic"</span>, <span class="refContent">commit dfcb63ce1de6b10b</span>, <time datetime="2021-10" class="refDate">October 2021</time>, <span>&lt;<a href="https://git.kernel.org/pub/scm/linux/kernel/git/netdev/net-next.git/commit/?id=dfcb63ce1de6b10b">https://git.kernel.org/pub/scm/linux/kernel/git/netdev/net-next.git/commit/?id=dfcb63ce1de6b10b</a>&gt;</span>. </dd>
<dd class="break"></dd>
<dt id="Hohlfeld14">[Hohlfeld14]</dt>
      <dd>
<span class="refAuthor">Hohlfeld, O.</span>, <span class="refAuthor">Pujol, E.</span>, <span class="refAuthor">Ciucu, F.</span>, <span class="refAuthor">Feldmann, A.</span>, and <span class="refAuthor">P. Barford</span>, <span class="refTitle">"A QoE Perspective on Sizing Network Buffers"</span>, <span class="refContent">IMC '14: Proceedings of the 2014 Conference on Internet Measurement, pp. 333-346</span>, <span class="seriesInfo">DOI 10.1145/2663716.2663730</span>, <time datetime="2014-11" class="refDate">November 2014</time>, <span>&lt;<a href="https://doi.acm.org/10.1145/2663716.2663730">https://doi.acm.org/10.1145/2663716.2663730</a>&gt;</span>. </dd>
<dd class="break"></dd>
<dt id="I-D.briscoe-tsvwg-l4s-diffserv">[L4S-DIFFSERV]</dt>
      <dd>
<span class="refAuthor">Briscoe, B.</span>, <span class="refTitle">"Interactions between Low Latency, Low Loss, Scalable Throughput (L4S) and Differentiated Services"</span>, <span class="refContent">Work in Progress</span>, <span class="seriesInfo">Internet-Draft, draft-briscoe-tsvwg-l4s-diffserv-02</span>, <time datetime="2018-11-04" class="refDate">4 November 2018</time>, <span>&lt;<a href="https://datatracker.ietf.org/doc/html/draft-briscoe-tsvwg-l4s-diffserv-02">https://datatracker.ietf.org/doc/html/draft-briscoe-tsvwg-l4s-diffserv-02</a>&gt;</span>. </dd>
<dd class="break"></dd>
<dt id="L4Sdemo16">[L4Sdemo16]</dt>
      <dd>
<span class="refAuthor">Bondarenko, O.</span>, <span class="refAuthor">De Schepper, K.</span>, <span class="refAuthor">Tsang, I.</span>, <span class="refAuthor">Briscoe, B.</span>, <span class="refAuthor">Petlund, A.</span>, and <span class="refAuthor">C. Griwodz</span>, <span class="refTitle">"Ultra-Low Delay for All: Live Experience, Live Analysis"</span>, <span class="refContent">Proceedings of the 7th International Conference on Multimedia Systems, Article No. 33, pp. 1-4</span>, <span class="seriesInfo">DOI 10.1145/2910017.2910633</span>, <time datetime="2016-05" class="refDate">May 2016</time>, <span>&lt;<a href="https://dl.acm.org/citation.cfm?doid=2910017.2910633">https://dl.acm.org/citation.cfm?doid=2910017.2910633</a>&gt;</span>. </dd>
<dd class="break"></dd>
<dt id="L4Sdemo16-Video">[L4Sdemo16-Video]</dt>
      <dd>
<span class="refTitle">"Videos used in IETF dispatch WG 'Ultra-Low Queuing Delay for All Apps' slot"</span>, <span>&lt;<a href="https://riteproject.eu/dctth/#1511dispatchwg">https://riteproject.eu/dctth/#1511dispatchwg</a>&gt;</span>. </dd>
<dd class="break"></dd>
<dt id="L4Seval22">[L4Seval22]</dt>
      <dd>
<span class="refAuthor">De Schepper, K.</span>, <span class="refAuthor">Albisser, O.</span>, <span class="refAuthor">Tilmans, O.</span>, and <span class="refAuthor">B. Briscoe</span>, <span class="refTitle">"Dual Queue Coupled AQM: Deployable Very Low Queuing Delay for All"</span>, <span class="refContent">TR-BB-2022-001, arXiv:2209.01078 [cs.NI]</span>, <span class="seriesInfo">DOI 10.48550/arXiv.2209.01078</span>, <time datetime="2022-09" class="refDate">September 2022</time>, <span>&lt;<a href="https://arxiv.org/abs/2209.01078">https://arxiv.org/abs/2209.01078</a>&gt;</span>. </dd>
<dd class="break"></dd>
<dt id="I-D.ietf-tsvwg-l4sops">[L4SOPS]</dt>
      <dd>
<span class="refAuthor">White, G., Ed.</span>, <span class="refTitle">"Operational Guidance for Deployment of L4S in the Internet"</span>, <span class="refContent">Work in Progress</span>, <span class="seriesInfo">Internet-Draft, draft-ietf-tsvwg-l4sops-03</span>, <time datetime="2022-04-28" class="refDate">28 April 2022</time>, <span>&lt;<a href="https://datatracker.ietf.org/doc/html/draft-ietf-tsvwg-l4sops-03">https://datatracker.ietf.org/doc/html/draft-ietf-tsvwg-l4sops-03</a>&gt;</span>. </dd>
<dd class="break"></dd>
<dt id="LEDBAT_AQM">[LEDBAT_AQM]</dt>
      <dd>
<span class="refAuthor">Al-Saadi, R.</span>, <span class="refAuthor">Armitage, G.</span>, and <span class="refAuthor">J. But</span>, <span class="refTitle">"Characterising LEDBAT Performance Through Bottlenecks Using PIE, FQ-CoDel and FQ-PIE Active Queue Management"</span>, <span class="refContent">IEEE 42nd Conference on Local Computer Networks (LCN)</span>, <span class="seriesInfo">DOI 10.1109/LCN.2017.22</span>, <time datetime="2017-10" class="refDate">October 2017</time>, <span>&lt;<a href="https://ieeexplore.ieee.org/document/8109367">https://ieeexplore.ieee.org/document/8109367</a>&gt;</span>. </dd>
<dd class="break"></dd>
<dt id="lowat">[lowat]</dt>
      <dd>
<span class="refAuthor">Meenan, P.</span>, <span class="refTitle">"Optimizing HTTP/2 prioritization with BBR and tcp_notsent_lowat"</span>, <span class="refContent">Cloudflare Blog</span>, <time datetime="2018-10" class="refDate">October 2018</time>, <span>&lt;<a href="https://blog.cloudflare.com/http-2-prioritization-with-nginx/">https://blog.cloudflare.com/http-2-prioritization-with-nginx/</a>&gt;</span>. </dd>
<dd class="break"></dd>
<dt id="McIlroy78">[McIlroy78]</dt>
      <dd>
<span class="refAuthor">McIlroy, M.D.</span>, <span class="refAuthor">Pinson, E. N.</span>, and <span class="refAuthor">B. A. Tague</span>, <span class="refTitle">"UNIX Time-Sharing System: Foreword"</span>, <span class="refContent">The Bell System Technical Journal 57: 6, pp. 1899-1904</span>, <span class="seriesInfo">DOI 10.1002/j.1538-7305.1978.tb02135.x</span>, <time datetime="1978-07" class="refDate">July 1978</time>, <span>&lt;<a href="https://archive.org/details/bstj57-6-1899">https://archive.org/details/bstj57-6-1899</a>&gt;</span>. </dd>
<dd class="break"></dd>
<dt id="Nadas20">[Nadas20]</dt>
      <dd>
<span class="refAuthor">Nádas, S.</span>, <span class="refAuthor">Gombos, G.</span>, <span class="refAuthor">Fejes, F.</span>, and <span class="refAuthor">S. Laki</span>, <span class="refTitle">"A Congestion Control Independent L4S Scheduler"</span>, <span class="refContent">ANRW '20: Proceedings of the Applied Networking Research Workshop, pp. 45-51</span>, <span class="seriesInfo">DOI 10.1145/3404868.3406669</span>, <time datetime="2020-07" class="refDate">July 2020</time>, <span>&lt;<a href="https://doi.org/10.1145/3404868.3406669">https://doi.org/10.1145/3404868.3406669</a>&gt;</span>. </dd>
<dd class="break"></dd>
<dt id="NASA04">[NASA04]</dt>
      <dd>
<span class="refAuthor">Bailey, R.</span>, <span class="refAuthor">Trey Arthur III, J.</span>, and <span class="refAuthor">S. Williams</span>, <span class="refTitle">"Latency Requirements for Head-Worn Display S/EVS Applications"</span>, <span class="refContent">Proceedings of SPIE 5424</span>, <span class="seriesInfo">DOI 10.1117/12.554462</span>, <time datetime="2004-04" class="refDate">April 2004</time>, <span>&lt;<a href="https://ntrs.nasa.gov/api/citations/20120009198/downloads/20120009198.pdf?attachment=true">https://ntrs.nasa.gov/api/citations/20120009198/downloads/20120009198.pdf?attachment=true</a>&gt;</span>. </dd>
<dd class="break"></dd>
<dt id="I-D.ietf-tsvwg-nqb">[NQB-PHB]</dt>
      <dd>
<span class="refAuthor">White, G.</span> and <span class="refAuthor">T. Fossati</span>, <span class="refTitle">"A Non-Queue-Building Per-Hop Behavior (NQB PHB) for Differentiated Services"</span>, <span class="refContent">Work in Progress</span>, <span class="seriesInfo">Internet-Draft, draft-ietf-tsvwg-nqb-15</span>, <time datetime="2023-01-11" class="refDate">11 January 2023</time>, <span>&lt;<a href="https://datatracker.ietf.org/doc/html/draft-ietf-tsvwg-nqb-15">https://datatracker.ietf.org/doc/html/draft-ietf-tsvwg-nqb-15</a>&gt;</span>. </dd>
<dd class="break"></dd>
<dt id="I-D.briscoe-iccrg-prague-congestion-control">[PRAGUE-CC]</dt>
      <dd>
<span class="refAuthor">De Schepper, K.</span>, <span class="refAuthor">Tilmans, O.</span>, and <span class="refAuthor">B. Briscoe, Ed.</span>, <span class="refTitle">"Prague Congestion Control"</span>, <span class="refContent">Work in Progress</span>, <span class="seriesInfo">Internet-Draft, draft-briscoe-iccrg-prague-congestion-control-01</span>, <time datetime="2022-07-11" class="refDate">11 July 2022</time>, <span>&lt;<a href="https://datatracker.ietf.org/doc/html/draft-briscoe-iccrg-prague-congestion-control-01">https://datatracker.ietf.org/doc/html/draft-briscoe-iccrg-prague-congestion-control-01</a>&gt;</span>. </dd>
<dd class="break"></dd>
<dt id="PragueLinux">[PragueLinux]</dt>
      <dd>
<span class="refAuthor">Briscoe, B.</span>, <span class="refAuthor">De Schepper, K.</span>, <span class="refAuthor">Albisser, O.</span>, <span class="refAuthor">Misund, J.</span>, <span class="refAuthor">Tilmans, O.</span>, <span class="refAuthor">Kühlewind, M.</span>, and <span class="refAuthor">A.S. Ahmed</span>, <span class="refTitle">"Implementing the 'TCP Prague' Requirements for Low Latency Low Loss Scalable Throughput (L4S)"</span>, <span class="refContent">Proceedings Linux Netdev 0x13</span>, <time datetime="2019-03" class="refDate">March 2019</time>, <span>&lt;<a href="https://www.netdevconf.org/0x13/session.html?talk-tcp-prague-l4s">https://www.netdevconf.org/0x13/session.html?talk-tcp-prague-l4s</a>&gt;</span>. </dd>
<dd class="break"></dd>
<dt id="QDyn">[QDyn]</dt>
      <dd>
<span class="refAuthor">Briscoe, B.</span>, <span class="refTitle">"Rapid Signalling of Queue Dynamics"</span>, <span class="refContent">TR-BB-2017-001, arXiv:1904.07044 [cs.NI]</span>, <span class="seriesInfo">DOI 10.48550/arXiv.1904.07044</span>, <time datetime="2019-04" class="refDate">April 2019</time>, <span>&lt;<a href="https://arxiv.org/abs/1904.07044">https://arxiv.org/abs/1904.07044</a>&gt;</span>. </dd>
<dd class="break"></dd>
<dt id="Raaen14">[Raaen14]</dt>
      <dd>
<span class="refAuthor">Raaen, K.</span> and <span class="refAuthor">T-M. Grønli</span>, <span class="refTitle">"Latency Thresholds for Usability in Games: A Survey"</span>, <span class="refContent">Norsk IKT-konferanse for forskning og utdanning (Norwegian 
        ICT conference for research and education)</span>, <time datetime="2014" class="refDate">2014</time>, <span>&lt;<a href="http://ojs.bibsys.no/index.php/NIK/article/view/9/6">http://ojs.bibsys.no/index.php/NIK/article/view/9/6</a>&gt;</span>. </dd>
<dd class="break"></dd>
<dt id="Rajiullah15">[Rajiullah15]</dt>
      <dd>
<span class="refAuthor">Rajiullah, M.</span>, <span class="refTitle">"Towards a Low Latency Internet: Understanding and Solutions"</span>, <span class="refContent">Dissertation, Karlstad University</span>, <time datetime="2015" class="refDate">2015</time>, <span>&lt;<a href="https://www.diva-portal.org/smash/get/diva2:846109/FULLTEXT01.pdf">https://www.diva-portal.org/smash/get/diva2:846109/FULLTEXT01.pdf</a>&gt;</span>. </dd>
<dd class="break"></dd>
<dt id="I-D.mathis-iccrg-relentless-tcp">[RELENTLESS]</dt>
      <dd>
<span class="refAuthor">Mathis, M.</span>, <span class="refTitle">"Relentless Congestion Control"</span>, <span class="refContent">Work in Progress</span>, <span class="seriesInfo">Internet-Draft, draft-mathis-iccrg-relentless-tcp-00</span>, <time datetime="2009-03-04" class="refDate">4 March 2009</time>, <span>&lt;<a href="https://datatracker.ietf.org/doc/html/draft-mathis-iccrg-relentless-tcp-00">https://datatracker.ietf.org/doc/html/draft-mathis-iccrg-relentless-tcp-00</a>&gt;</span>. </dd>
<dd class="break"></dd>
<dt id="RFC0970">[RFC0970]</dt>
      <dd>
<span class="refAuthor">Nagle, J.</span>, <span class="refTitle">"On Packet Switches With Infinite Storage"</span>, <span class="seriesInfo">RFC 970</span>, <span class="seriesInfo">DOI 10.17487/RFC0970</span>, <time datetime="1985-12" class="refDate">December 1985</time>, <span>&lt;<a href="https://www.rfc-editor.org/info/rfc970">https://www.rfc-editor.org/info/rfc970</a>&gt;</span>. </dd>
<dd class="break"></dd>
<dt id="RFC2475">[RFC2475]</dt>
      <dd>
<span class="refAuthor">Blake, S.</span>, <span class="refAuthor">Black, D.</span>, <span class="refAuthor">Carlson, M.</span>, <span class="refAuthor">Davies, E.</span>, <span class="refAuthor">Wang, Z.</span>, and <span class="refAuthor">W. Weiss</span>, <span class="refTitle">"An Architecture for Differentiated Services"</span>, <span class="seriesInfo">RFC 2475</span>, <span class="seriesInfo">DOI 10.17487/RFC2475</span>, <time datetime="1998-12" class="refDate">December 1998</time>, <span>&lt;<a href="https://www.rfc-editor.org/info/rfc2475">https://www.rfc-editor.org/info/rfc2475</a>&gt;</span>. </dd>
<dd class="break"></dd>
<dt id="RFC2698">[RFC2698]</dt>
      <dd>
<span class="refAuthor">Heinanen, J.</span> and <span class="refAuthor">R. Guerin</span>, <span class="refTitle">"A Two Rate Three Color Marker"</span>, <span class="seriesInfo">RFC 2698</span>, <span class="seriesInfo">DOI 10.17487/RFC2698</span>, <time datetime="1999-09" class="refDate">September 1999</time>, <span>&lt;<a href="https://www.rfc-editor.org/info/rfc2698">https://www.rfc-editor.org/info/rfc2698</a>&gt;</span>. </dd>
<dd class="break"></dd>
<dt id="RFC2884">[RFC2884]</dt>
      <dd>
<span class="refAuthor">Hadi Salim, J.</span> and <span class="refAuthor">U. Ahmed</span>, <span class="refTitle">"Performance Evaluation of Explicit Congestion Notification (ECN) in IP Networks"</span>, <span class="seriesInfo">RFC 2884</span>, <span class="seriesInfo">DOI 10.17487/RFC2884</span>, <time datetime="2000-07" class="refDate">July 2000</time>, <span>&lt;<a href="https://www.rfc-editor.org/info/rfc2884">https://www.rfc-editor.org/info/rfc2884</a>&gt;</span>. </dd>
<dd class="break"></dd>
<dt id="RFC3168">[RFC3168]</dt>
      <dd>
<span class="refAuthor">Ramakrishnan, K.</span>, <span class="refAuthor">Floyd, S.</span>, and <span class="refAuthor">D. Black</span>, <span class="refTitle">"The Addition of Explicit Congestion Notification (ECN) to IP"</span>, <span class="seriesInfo">RFC 3168</span>, <span class="seriesInfo">DOI 10.17487/RFC3168</span>, <time datetime="2001-09" class="refDate">September 2001</time>, <span>&lt;<a href="https://www.rfc-editor.org/info/rfc3168">https://www.rfc-editor.org/info/rfc3168</a>&gt;</span>. </dd>
<dd class="break"></dd>
<dt id="RFC3246">[RFC3246]</dt>
      <dd>
<span class="refAuthor">Davie, B.</span>, <span class="refAuthor">Charny, A.</span>, <span class="refAuthor">Bennet, J.C.R.</span>, <span class="refAuthor">Benson, K.</span>, <span class="refAuthor">Le Boudec, J.Y.</span>, <span class="refAuthor">Courtney, W.</span>, <span class="refAuthor">Davari, S.</span>, <span class="refAuthor">Firoiu, V.</span>, and <span class="refAuthor">D. Stiliadis</span>, <span class="refTitle">"An Expedited Forwarding PHB (Per-Hop Behavior)"</span>, <span class="seriesInfo">RFC 3246</span>, <span class="seriesInfo">DOI 10.17487/RFC3246</span>, <time datetime="2002-03" class="refDate">March 2002</time>, <span>&lt;<a href="https://www.rfc-editor.org/info/rfc3246">https://www.rfc-editor.org/info/rfc3246</a>&gt;</span>. </dd>
<dd class="break"></dd>
<dt id="RFC3540">[RFC3540]</dt>
      <dd>
<span class="refAuthor">Spring, N.</span>, <span class="refAuthor">Wetherall, D.</span>, and <span class="refAuthor">D. Ely</span>, <span class="refTitle">"Robust Explicit Congestion Notification (ECN) Signaling with Nonces"</span>, <span class="seriesInfo">RFC 3540</span>, <span class="seriesInfo">DOI 10.17487/RFC3540</span>, <time datetime="2003-06" class="refDate">June 2003</time>, <span>&lt;<a href="https://www.rfc-editor.org/info/rfc3540">https://www.rfc-editor.org/info/rfc3540</a>&gt;</span>. </dd>
<dd class="break"></dd>
<dt id="RFC3649">[RFC3649]</dt>
      <dd>
<span class="refAuthor">Floyd, S.</span>, <span class="refTitle">"HighSpeed TCP for Large Congestion Windows"</span>, <span class="seriesInfo">RFC 3649</span>, <span class="seriesInfo">DOI 10.17487/RFC3649</span>, <time datetime="2003-12" class="refDate">December 2003</time>, <span>&lt;<a href="https://www.rfc-editor.org/info/rfc3649">https://www.rfc-editor.org/info/rfc3649</a>&gt;</span>. </dd>
<dd class="break"></dd>
<dt id="RFC4340">[RFC4340]</dt>
      <dd>
<span class="refAuthor">Kohler, E.</span>, <span class="refAuthor">Handley, M.</span>, and <span class="refAuthor">S. Floyd</span>, <span class="refTitle">"Datagram Congestion Control Protocol (DCCP)"</span>, <span class="seriesInfo">RFC 4340</span>, <span class="seriesInfo">DOI 10.17487/RFC4340</span>, <time datetime="2006-03" class="refDate">March 2006</time>, <span>&lt;<a href="https://www.rfc-editor.org/info/rfc4340">https://www.rfc-editor.org/info/rfc4340</a>&gt;</span>. </dd>
<dd class="break"></dd>
<dt id="RFC4774">[RFC4774]</dt>
      <dd>
<span class="refAuthor">Floyd, S.</span>, <span class="refTitle">"Specifying Alternate Semantics for the Explicit Congestion Notification (ECN) Field"</span>, <span class="seriesInfo">BCP 124</span>, <span class="seriesInfo">RFC 4774</span>, <span class="seriesInfo">DOI 10.17487/RFC4774</span>, <time datetime="2006-11" class="refDate">November 2006</time>, <span>&lt;<a href="https://www.rfc-editor.org/info/rfc4774">https://www.rfc-editor.org/info/rfc4774</a>&gt;</span>. </dd>
<dd class="break"></dd>
<dt id="RFC4960">[RFC4960]</dt>
      <dd>
<span class="refAuthor">Stewart, R., Ed.</span>, <span class="refTitle">"Stream Control Transmission Protocol"</span>, <span class="seriesInfo">RFC 4960</span>, <span class="seriesInfo">DOI 10.17487/RFC4960</span>, <time datetime="2007-09" class="refDate">September 2007</time>, <span>&lt;<a href="https://www.rfc-editor.org/info/rfc4960">https://www.rfc-editor.org/info/rfc4960</a>&gt;</span>. </dd>
<dd class="break"></dd>
<dt id="RFC5033">[RFC5033]</dt>
      <dd>
<span class="refAuthor">Floyd, S.</span> and <span class="refAuthor">M. Allman</span>, <span class="refTitle">"Specifying New Congestion Control Algorithms"</span>, <span class="seriesInfo">BCP 133</span>, <span class="seriesInfo">RFC 5033</span>, <span class="seriesInfo">DOI 10.17487/RFC5033</span>, <time datetime="2007-08" class="refDate">August 2007</time>, <span>&lt;<a href="https://www.rfc-editor.org/info/rfc5033">https://www.rfc-editor.org/info/rfc5033</a>&gt;</span>. </dd>
<dd class="break"></dd>
<dt id="RFC5348">[RFC5348]</dt>
      <dd>
<span class="refAuthor">Floyd, S.</span>, <span class="refAuthor">Handley, M.</span>, <span class="refAuthor">Padhye, J.</span>, and <span class="refAuthor">J. Widmer</span>, <span class="refTitle">"TCP Friendly Rate Control (TFRC): Protocol Specification"</span>, <span class="seriesInfo">RFC 5348</span>, <span class="seriesInfo">DOI 10.17487/RFC5348</span>, <time datetime="2008-09" class="refDate">September 2008</time>, <span>&lt;<a href="https://www.rfc-editor.org/info/rfc5348">https://www.rfc-editor.org/info/rfc5348</a>&gt;</span>. </dd>
<dd class="break"></dd>
<dt id="RFC5670">[RFC5670]</dt>
      <dd>
<span class="refAuthor">Eardley, P., Ed.</span>, <span class="refTitle">"Metering and Marking Behaviour of PCN-Nodes"</span>, <span class="seriesInfo">RFC 5670</span>, <span class="seriesInfo">DOI 10.17487/RFC5670</span>, <time datetime="2009-11" class="refDate">November 2009</time>, <span>&lt;<a href="https://www.rfc-editor.org/info/rfc5670">https://www.rfc-editor.org/info/rfc5670</a>&gt;</span>. </dd>
<dd class="break"></dd>
<dt id="RFC5681">[RFC5681]</dt>
      <dd>
<span class="refAuthor">Allman, M.</span>, <span class="refAuthor">Paxson, V.</span>, and <span class="refAuthor">E. Blanton</span>, <span class="refTitle">"TCP Congestion Control"</span>, <span class="seriesInfo">RFC 5681</span>, <span class="seriesInfo">DOI 10.17487/RFC5681</span>, <time datetime="2009-09" class="refDate">September 2009</time>, <span>&lt;<a href="https://www.rfc-editor.org/info/rfc5681">https://www.rfc-editor.org/info/rfc5681</a>&gt;</span>. </dd>
<dd class="break"></dd>
<dt id="RFC6040">[RFC6040]</dt>
      <dd>
<span class="refAuthor">Briscoe, B.</span>, <span class="refTitle">"Tunnelling of Explicit Congestion Notification"</span>, <span class="seriesInfo">RFC 6040</span>, <span class="seriesInfo">DOI 10.17487/RFC6040</span>, <time datetime="2010-11" class="refDate">November 2010</time>, <span>&lt;<a href="https://www.rfc-editor.org/info/rfc6040">https://www.rfc-editor.org/info/rfc6040</a>&gt;</span>. </dd>
<dd class="break"></dd>
<dt id="RFC6679">[RFC6679]</dt>
      <dd>
<span class="refAuthor">Westerlund, M.</span>, <span class="refAuthor">Johansson, I.</span>, <span class="refAuthor">Perkins, C.</span>, <span class="refAuthor">O'Hanlon, P.</span>, and <span class="refAuthor">K. Carlberg</span>, <span class="refTitle">"Explicit Congestion Notification (ECN) for RTP over UDP"</span>, <span class="seriesInfo">RFC 6679</span>, <span class="seriesInfo">DOI 10.17487/RFC6679</span>, <time datetime="2012-08" class="refDate">August 2012</time>, <span>&lt;<a href="https://www.rfc-editor.org/info/rfc6679">https://www.rfc-editor.org/info/rfc6679</a>&gt;</span>. </dd>
<dd class="break"></dd>
<dt id="RFC6817">[RFC6817]</dt>
      <dd>
<span class="refAuthor">Shalunov, S.</span>, <span class="refAuthor">Hazel, G.</span>, <span class="refAuthor">Iyengar, J.</span>, and <span class="refAuthor">M. Kuehlewind</span>, <span class="refTitle">"Low Extra Delay Background Transport (LEDBAT)"</span>, <span class="seriesInfo">RFC 6817</span>, <span class="seriesInfo">DOI 10.17487/RFC6817</span>, <time datetime="2012-12" class="refDate">December 2012</time>, <span>&lt;<a href="https://www.rfc-editor.org/info/rfc6817">https://www.rfc-editor.org/info/rfc6817</a>&gt;</span>. </dd>
<dd class="break"></dd>
<dt id="RFC6973">[RFC6973]</dt>
      <dd>
<span class="refAuthor">Cooper, A.</span>, <span class="refAuthor">Tschofenig, H.</span>, <span class="refAuthor">Aboba, B.</span>, <span class="refAuthor">Peterson, J.</span>, <span class="refAuthor">Morris, J.</span>, <span class="refAuthor">Hansen, M.</span>, and <span class="refAuthor">R. Smith</span>, <span class="refTitle">"Privacy Considerations for Internet Protocols"</span>, <span class="seriesInfo">RFC 6973</span>, <span class="seriesInfo">DOI 10.17487/RFC6973</span>, <time datetime="2013-07" class="refDate">July 2013</time>, <span>&lt;<a href="https://www.rfc-editor.org/info/rfc6973">https://www.rfc-editor.org/info/rfc6973</a>&gt;</span>. </dd>
<dd class="break"></dd>
<dt id="RFC7560">[RFC7560]</dt>
      <dd>
<span class="refAuthor">Kuehlewind, M., Ed.</span>, <span class="refAuthor">Scheffenegger, R.</span>, and <span class="refAuthor">B. Briscoe</span>, <span class="refTitle">"Problem Statement and Requirements for Increased Accuracy in Explicit Congestion Notification (ECN) Feedback"</span>, <span class="seriesInfo">RFC 7560</span>, <span class="seriesInfo">DOI 10.17487/RFC7560</span>, <time datetime="2015-08" class="refDate">August 2015</time>, <span>&lt;<a href="https://www.rfc-editor.org/info/rfc7560">https://www.rfc-editor.org/info/rfc7560</a>&gt;</span>. </dd>
<dd class="break"></dd>
<dt id="RFC7567">[RFC7567]</dt>
      <dd>
<span class="refAuthor">Baker, F., Ed.</span> and <span class="refAuthor">G. Fairhurst, Ed.</span>, <span class="refTitle">"IETF Recommendations Regarding Active Queue Management"</span>, <span class="seriesInfo">BCP 197</span>, <span class="seriesInfo">RFC 7567</span>, <span class="seriesInfo">DOI 10.17487/RFC7567</span>, <time datetime="2015-07" class="refDate">July 2015</time>, <span>&lt;<a href="https://www.rfc-editor.org/info/rfc7567">https://www.rfc-editor.org/info/rfc7567</a>&gt;</span>. </dd>
<dd class="break"></dd>
<dt id="RFC7665">[RFC7665]</dt>
      <dd>
<span class="refAuthor">Halpern, J., Ed.</span> and <span class="refAuthor">C. Pignataro, Ed.</span>, <span class="refTitle">"Service Function Chaining (SFC) Architecture"</span>, <span class="seriesInfo">RFC 7665</span>, <span class="seriesInfo">DOI 10.17487/RFC7665</span>, <time datetime="2015-10" class="refDate">October 2015</time>, <span>&lt;<a href="https://www.rfc-editor.org/info/rfc7665">https://www.rfc-editor.org/info/rfc7665</a>&gt;</span>. </dd>
<dd class="break"></dd>
<dt id="RFC7713">[RFC7713]</dt>
      <dd>
<span class="refAuthor">Mathis, M.</span> and <span class="refAuthor">B. Briscoe</span>, <span class="refTitle">"Congestion Exposure (ConEx) Concepts, Abstract Mechanism, and Requirements"</span>, <span class="seriesInfo">RFC 7713</span>, <span class="seriesInfo">DOI 10.17487/RFC7713</span>, <time datetime="2015-12" class="refDate">December 2015</time>, <span>&lt;<a href="https://www.rfc-editor.org/info/rfc7713">https://www.rfc-editor.org/info/rfc7713</a>&gt;</span>. </dd>
<dd class="break"></dd>
<dt id="RFC8033">[RFC8033]</dt>
      <dd>
<span class="refAuthor">Pan, R.</span>, <span class="refAuthor">Natarajan, P.</span>, <span class="refAuthor">Baker, F.</span>, and <span class="refAuthor">G. White</span>, <span class="refTitle">"Proportional Integral Controller Enhanced (PIE): A Lightweight Control Scheme to Address the Bufferbloat Problem"</span>, <span class="seriesInfo">RFC 8033</span>, <span class="seriesInfo">DOI 10.17487/RFC8033</span>, <time datetime="2017-02" class="refDate">February 2017</time>, <span>&lt;<a href="https://www.rfc-editor.org/info/rfc8033">https://www.rfc-editor.org/info/rfc8033</a>&gt;</span>. </dd>
<dd class="break"></dd>
<dt id="RFC8034">[RFC8034]</dt>
      <dd>
<span class="refAuthor">White, G.</span> and <span class="refAuthor">R. Pan</span>, <span class="refTitle">"Active Queue Management (AQM) Based on Proportional Integral Controller Enhanced (PIE) for Data-Over-Cable Service Interface Specifications (DOCSIS) Cable Modems"</span>, <span class="seriesInfo">RFC 8034</span>, <span class="seriesInfo">DOI 10.17487/RFC8034</span>, <time datetime="2017-02" class="refDate">February 2017</time>, <span>&lt;<a href="https://www.rfc-editor.org/info/rfc8034">https://www.rfc-editor.org/info/rfc8034</a>&gt;</span>. </dd>
<dd class="break"></dd>
<dt id="RFC8170">[RFC8170]</dt>
      <dd>
<span class="refAuthor">Thaler, D., Ed.</span>, <span class="refTitle">"Planning for Protocol Adoption and Subsequent Transitions"</span>, <span class="seriesInfo">RFC 8170</span>, <span class="seriesInfo">DOI 10.17487/RFC8170</span>, <time datetime="2017-05" class="refDate">May 2017</time>, <span>&lt;<a href="https://www.rfc-editor.org/info/rfc8170">https://www.rfc-editor.org/info/rfc8170</a>&gt;</span>. </dd>
<dd class="break"></dd>
<dt id="RFC8257">[RFC8257]</dt>
      <dd>
<span class="refAuthor">Bensley, S.</span>, <span class="refAuthor">Thaler, D.</span>, <span class="refAuthor">Balasubramanian, P.</span>, <span class="refAuthor">Eggert, L.</span>, and <span class="refAuthor">G. Judd</span>, <span class="refTitle">"Data Center TCP (DCTCP): TCP Congestion Control for Data Centers"</span>, <span class="seriesInfo">RFC 8257</span>, <span class="seriesInfo">DOI 10.17487/RFC8257</span>, <time datetime="2017-10" class="refDate">October 2017</time>, <span>&lt;<a href="https://www.rfc-editor.org/info/rfc8257">https://www.rfc-editor.org/info/rfc8257</a>&gt;</span>. </dd>
<dd class="break"></dd>
<dt id="RFC8290">[RFC8290]</dt>
      <dd>
<span class="refAuthor">Hoeiland-Joergensen, T.</span>, <span class="refAuthor">McKenney, P.</span>, <span class="refAuthor">Taht, D.</span>, <span class="refAuthor">Gettys, J.</span>, and <span class="refAuthor">E. Dumazet</span>, <span class="refTitle">"The Flow Queue CoDel Packet Scheduler and Active Queue Management Algorithm"</span>, <span class="seriesInfo">RFC 8290</span>, <span class="seriesInfo">DOI 10.17487/RFC8290</span>, <time datetime="2018-01" class="refDate">January 2018</time>, <span>&lt;<a href="https://www.rfc-editor.org/info/rfc8290">https://www.rfc-editor.org/info/rfc8290</a>&gt;</span>. </dd>
<dd class="break"></dd>
<dt id="RFC8298">[RFC8298]</dt>
      <dd>
<span class="refAuthor">Johansson, I.</span> and <span class="refAuthor">Z. Sarker</span>, <span class="refTitle">"Self-Clocked Rate Adaptation for Multimedia"</span>, <span class="seriesInfo">RFC 8298</span>, <span class="seriesInfo">DOI 10.17487/RFC8298</span>, <time datetime="2017-12" class="refDate">December 2017</time>, <span>&lt;<a href="https://www.rfc-editor.org/info/rfc8298">https://www.rfc-editor.org/info/rfc8298</a>&gt;</span>. </dd>
<dd class="break"></dd>
<dt id="RFC8311">[RFC8311]</dt>
      <dd>
<span class="refAuthor">Black, D.</span>, <span class="refTitle">"Relaxing Restrictions on Explicit Congestion Notification (ECN) Experimentation"</span>, <span class="seriesInfo">RFC 8311</span>, <span class="seriesInfo">DOI 10.17487/RFC8311</span>, <time datetime="2018-01" class="refDate">January 2018</time>, <span>&lt;<a href="https://www.rfc-editor.org/info/rfc8311">https://www.rfc-editor.org/info/rfc8311</a>&gt;</span>. </dd>
<dd class="break"></dd>
<dt id="RFC8312">[RFC8312]</dt>
      <dd>
<span class="refAuthor">Rhee, I.</span>, <span class="refAuthor">Xu, L.</span>, <span class="refAuthor">Ha, S.</span>, <span class="refAuthor">Zimmermann, A.</span>, <span class="refAuthor">Eggert, L.</span>, and <span class="refAuthor">R. Scheffenegger</span>, <span class="refTitle">"CUBIC for Fast Long-Distance Networks"</span>, <span class="seriesInfo">RFC 8312</span>, <span class="seriesInfo">DOI 10.17487/RFC8312</span>, <time datetime="2018-02" class="refDate">February 2018</time>, <span>&lt;<a href="https://www.rfc-editor.org/info/rfc8312">https://www.rfc-editor.org/info/rfc8312</a>&gt;</span>. </dd>
<dd class="break"></dd>
<dt id="RFC8404">[RFC8404]</dt>
      <dd>
<span class="refAuthor">Moriarty, K., Ed.</span> and <span class="refAuthor">A. Morton, Ed.</span>, <span class="refTitle">"Effects of Pervasive Encryption on Operators"</span>, <span class="seriesInfo">RFC 8404</span>, <span class="seriesInfo">DOI 10.17487/RFC8404</span>, <time datetime="2018-07" class="refDate">July 2018</time>, <span>&lt;<a href="https://www.rfc-editor.org/info/rfc8404">https://www.rfc-editor.org/info/rfc8404</a>&gt;</span>. </dd>
<dd class="break"></dd>
<dt id="RFC8511">[RFC8511]</dt>
      <dd>
<span class="refAuthor">Khademi, N.</span>, <span class="refAuthor">Welzl, M.</span>, <span class="refAuthor">Armitage, G.</span>, and <span class="refAuthor">G. Fairhurst</span>, <span class="refTitle">"TCP Alternative Backoff with ECN (ABE)"</span>, <span class="seriesInfo">RFC 8511</span>, <span class="seriesInfo">DOI 10.17487/RFC8511</span>, <time datetime="2018-12" class="refDate">December 2018</time>, <span>&lt;<a href="https://www.rfc-editor.org/info/rfc8511">https://www.rfc-editor.org/info/rfc8511</a>&gt;</span>. </dd>
<dd class="break"></dd>
<dt id="RFC8888">[RFC8888]</dt>
      <dd>
<span class="refAuthor">Sarker, Z.</span>, <span class="refAuthor">Perkins, C.</span>, <span class="refAuthor">Singh, V.</span>, and <span class="refAuthor">M. Ramalho</span>, <span class="refTitle">"RTP Control Protocol (RTCP) Feedback for Congestion Control"</span>, <span class="seriesInfo">RFC 8888</span>, <span class="seriesInfo">DOI 10.17487/RFC8888</span>, <time datetime="2021-01" class="refDate">January 2021</time>, <span>&lt;<a href="https://www.rfc-editor.org/info/rfc8888">https://www.rfc-editor.org/info/rfc8888</a>&gt;</span>. </dd>
<dd class="break"></dd>
<dt id="RFC8985">[RFC8985]</dt>
      <dd>
<span class="refAuthor">Cheng, Y.</span>, <span class="refAuthor">Cardwell, N.</span>, <span class="refAuthor">Dukkipati, N.</span>, and <span class="refAuthor">P. Jha</span>, <span class="refTitle">"The RACK-TLP Loss Detection Algorithm for TCP"</span>, <span class="seriesInfo">RFC 8985</span>, <span class="seriesInfo">DOI 10.17487/RFC8985</span>, <time datetime="2021-02" class="refDate">February 2021</time>, <span>&lt;<a href="https://www.rfc-editor.org/info/rfc8985">https://www.rfc-editor.org/info/rfc8985</a>&gt;</span>. </dd>
<dd class="break"></dd>
<dt id="RFC9000">[RFC9000]</dt>
      <dd>
<span class="refAuthor">Iyengar, J., Ed.</span> and <span class="refAuthor">M. Thomson, Ed.</span>, <span class="refTitle">"QUIC: A UDP-Based Multiplexed and Secure Transport"</span>, <span class="seriesInfo">RFC 9000</span>, <span class="seriesInfo">DOI 10.17487/RFC9000</span>, <time datetime="2021-05" class="refDate">May 2021</time>, <span>&lt;<a href="https://www.rfc-editor.org/info/rfc9000">https://www.rfc-editor.org/info/rfc9000</a>&gt;</span>. </dd>
<dd class="break"></dd>
<dt id="RFC9113">[RFC9113]</dt>
      <dd>
<span class="refAuthor">Thomson, M., Ed.</span> and <span class="refAuthor">C. Benfield, Ed.</span>, <span class="refTitle">"HTTP/2"</span>, <span class="seriesInfo">RFC 9113</span>, <span class="seriesInfo">DOI 10.17487/RFC9113</span>, <time datetime="2022-06" class="refDate">June 2022</time>, <span>&lt;<a href="https://www.rfc-editor.org/info/rfc9113">https://www.rfc-editor.org/info/rfc9113</a>&gt;</span>. </dd>
<dd class="break"></dd>
<dt id="RFC9331">[RFC9331]</dt>
      <dd>
<span class="refAuthor">De Schepper, K.</span> and <span class="refAuthor">B. Briscoe, Ed.</span>, <span class="refTitle">"The Explicit Congestion Notification (ECN) Protocol for Low Latency, Low Loss, and Scalable Throughput (L4S)"</span>, <span class="seriesInfo">RFC 9331</span>, <span class="seriesInfo">DOI 10.17487/RFC9331</span>, <time datetime="2023-01" class="refDate">January 2023</time>, <span>&lt;<a href="https://www.rfc-editor.org/info/rfc9331">https://www.rfc-editor.org/info/rfc9331</a>&gt;</span>. </dd>
<dd class="break"></dd>
<dt id="RFC9332">[RFC9332]</dt>
      <dd>
<span class="refAuthor">De Schepper, K.</span>, <span class="refAuthor">Briscoe, B., Ed.</span>, and <span class="refAuthor">G. White</span>, <span class="refTitle">"Dual-Queue Coupled Active Queue Management (AQM) for Low Latency, Low Loss, and Scalable Throughput (L4S)"</span>, <span class="seriesInfo">RFC 9332</span>, <span class="seriesInfo">DOI 10.17487/RFC9332</span>, <time datetime="2023-01" class="refDate">January 2023</time>, <span>&lt;<a href="https://www.rfc-editor.org/info/rfc9332">https://www.rfc-editor.org/info/rfc9332</a>&gt;</span>. </dd>
<dd class="break"></dd>
<dt id="SCReAM-L4S">[SCReAM-L4S]</dt>
      <dd>
<span class="refTitle">"SCReAM"</span>, <span class="refContent">commit fda6c53</span>, <time datetime="2022-06" class="refDate">June 2022</time>, <span>&lt;<a href="https://github.com/EricssonResearch/scream">https://github.com/EricssonResearch/scream</a>&gt;</span>. </dd>
<dd class="break"></dd>
<dt id="TCP-CA">[TCP-CA]</dt>
      <dd>
<span class="refAuthor">Jacobson, V.</span> and <span class="refAuthor">M. Karels</span>, <span class="refTitle">"Congestion Avoidance and Control"</span>, <span class="seriesInfo">Laurence Berkeley Labs Technical Report </span>, <time datetime="1988-11" class="refDate">November 1988</time>, <span>&lt;<a href="https://ee.lbl.gov/papers/congavoid.pdf">https://ee.lbl.gov/papers/congavoid.pdf</a>&gt;</span>. </dd>
<dd class="break"></dd>
<dt id="UnorderedLTE">[UnorderedLTE]</dt>
    <dd>
<span class="refAuthor">Austrheim, M.</span>, <span class="refTitle">"Implementing immediate forwarding for 4G in a network simulator"</span>, <span class="refContent">Master's Thesis, University of Oslo</span>, <time datetime="2018" class="refDate">2018</time>. </dd>
<dd class="break"></dd>
</dl>
</section>
<section id="appendix-A">
      <h2 id="name-acknowledgements">
<a href="#name-acknowledgements" class="section-name selfRef">Acknowledgements</a>
      </h2>
<p id="appendix-A-1">Thanks to <span class="contact-name">Richard Scheffenegger</span>, <span class="contact-name">Wes Eddy</span>, <span class="contact-name">Karen Nielsen</span>, <span class="contact-name">David Black</span>, <span class="contact-name">Jake Holland</span>, <span class="contact-name">Vidhi Goel</span>, <span class="contact-name">Ermin Sakic</span>, <span class="contact-name">Praveen Balasubramanian</span>, <span class="contact-name">Gorry       Fairhurst</span>, <span class="contact-name">Mirja Kuehlewind</span>, <span class="contact-name">Philip Eardley</span>, <span class="contact-name">Neal Cardwell</span>,
      <span class="contact-name">Pete Heist</span>, and <span class="contact-name">Martin Duke</span>
      for their useful review comments. Thanks also to the area reviewers:
      <span class="contact-name">Marco Tiloca</span>, <span class="contact-name">Lars Eggert</span>,
      <span class="contact-name">Roman Danyliw</span>, and <span class="contact-name">Éric       Vyncke</span>.<a href="#appendix-A-1" class="pilcrow">¶</a></p>
<p id="appendix-A-2"><span class="contact-name">Bob Briscoe</span> and <span class="contact-name">Koen De       Schepper</span> were partly funded by the European Community under its Seventh
      Framework Programme through the Reducing Internet Transport Latency
      (RITE) project (ICT-317700). The contribution of <span class="contact-name">Koen       De Schepper</span> was also partly funded by the 5Growth and DAEMON EU H2020
      projects. <span class="contact-name">Bob Briscoe</span> was also partly funded by the
      Research Council of Norway through the TimeIn project, partly by
      CableLabs, and partly by the Comcast Innovation Fund. The views expressed
      here are solely those of the authors.<a href="#appendix-A-2" class="pilcrow">¶</a></p>
</section>
<div id="authors-addresses">
<section id="appendix-B">
      <h2 id="name-authors-addresses">
<a href="#name-authors-addresses" class="section-name selfRef">Authors' Addresses</a>
      </h2>
<address class="vcard">
        <div dir="auto" class="left"><span class="fn nameRole">Bob Briscoe (<span class="role">editor</span>)</span></div>
<div dir="auto" class="left"><span class="org">Independent</span></div>
<div dir="auto" class="left"><span class="country-name">United Kingdom</span></div>
<div class="email">
<span>Email:</span>
<a href="mailto:ietf@bobbriscoe.net" class="email">ietf@bobbriscoe.net</a>
</div>
<div class="url">
<span>URI:</span>
<a href="https://bobbriscoe.net/" class="url">https://bobbriscoe.net/</a>
</div>
</address>
<address class="vcard">
        <div dir="auto" class="left"><span class="fn nameRole">Koen De Schepper</span></div>
<div dir="auto" class="left"><span class="org">Nokia Bell Labs</span></div>
<div dir="auto" class="left"><span class="locality">Antwerp</span></div>
<div dir="auto" class="left"><span class="country-name">Belgium</span></div>
<div class="email">
<span>Email:</span>
<a href="mailto:koen.de_schepper@nokia.com" class="email">koen.de_schepper@nokia.com</a>
</div>
<div class="url">
<span>URI:</span>
<a href="https://www.bell-labs.com/about/researcher-profiles/koende_schepper/" class="url">https://www.bell-labs.com/about/researcher-profiles/koende_schepper/</a>
</div>
</address>
<address class="vcard">
        <div dir="auto" class="left"><span class="fn nameRole">Marcelo Bagnulo</span></div>
<div dir="auto" class="left"><span class="org">Universidad Carlos III de Madrid</span></div>
<div dir="auto" class="left"><span class="street-address">Av. Universidad 30</span></div>
<div dir="auto" class="left">
<span class="postal-code">28911</span> <span class="locality">Madrid</span> </div>
<div dir="auto" class="left"><span class="country-name">Spain</span></div>
<div class="tel">
<span>Phone:</span>
<a href="tel:34%2091%206249500" class="tel">34 91 6249500</a>
</div>
<div class="email">
<span>Email:</span>
<a href="mailto:marcelo@it.uc3m.es" class="email">marcelo@it.uc3m.es</a>
</div>
<div class="url">
<span>URI:</span>
<a href="https://www.it.uc3m.es" class="url">https://www.it.uc3m.es</a>
</div>
</address>
<address class="vcard">
        <div dir="auto" class="left"><span class="fn nameRole">Greg White</span></div>
<div dir="auto" class="left"><span class="org">CableLabs</span></div>
<div dir="auto" class="left"><span class="country-name">United States of America</span></div>
<div class="email">
<span>Email:</span>
<a href="mailto:G.White@CableLabs.com" class="email">G.White@CableLabs.com</a>
</div>
</address>
</section>
</div>
<script>const toc = document.getElementById("toc");
toc.querySelector("h2").addEventListener("click", e => {
  toc.classList.toggle("active");
});
toc.querySelector("nav").addEventListener("click", e => {
  toc.classList.remove("active");
});
</script>
</body>
</html>