<?xmlversion='1.0' encoding='utf-8'?>version="1.0" encoding="UTF-8"?> <!DOCTYPE rfc [ <!ENTITY nbsp " "> <!ENTITY zwsp "​"> <!ENTITY nbhy "‑"> <!ENTITY wj "⁠"> ]><!-- This template is for creating an Internet Draft using xml2rfc, which is available here: http://xml.resource.org. --> <?xml-stylesheet type='text/xsl' href='rfc2629.xslt' ?> <!-- used by XSLT processors --> <!-- For a complete list and description of processing instructions (PIs), please see http://xml.resource.org/authoring/README.html. --> <!-- Below are generally applicable Processing Instructions (PIs) that most I-Ds might want to use. (Here they are set differently than their defaults in xml2rfc v1.32) --> <?rfc strict="yes" ?> <!-- give errors regarding ID-nits and DTD validation --> <!-- control the table of contents (ToC) --> <?rfc toc="yes"?> <!-- generate a ToC --> <?rfc tocdepth="4"?> <!-- the number of levels of subsections in ToC. default: 3 --> <!-- control references --> <?rfc symrefs="yes"?> <!-- use symbolic references tags, i.e, [RFC2119] instead of [1] --> <?rfc sortrefs="yes" ?> <!-- sort the reference entries alphabetically --> <!-- control vertical white space (using these PIs as follows is recommended by the RFC Editor) --> <?rfc compact="yes" ?> <!-- do not start each main section on a new page --> <?rfc subcompact="no" ?> <!-- keep one blank line between list items --> <!-- end of list of popular I-D processing instructions --><rfc xmlns:xi="http://www.w3.org/2001/XInclude" submissionType="IETF" category="exp" consensus="true" docName="draft-ietf-tsvwg-aqm-dualq-coupled-25" number="9332" ipr="trust200902" updates="" obsoletes=""submissionType="IETF"xml:lang="en" tocInclude="true" tocDepth="4" symRefs="true" sortRefs="true" version="3"> <!-- xml2rfc v2v3 conversion 3.14.1 --><!-- category values: std, bcp, info, exp, and historic ipr values: trust200902, noModificationTrust200902, noDerivativesTrust200902, or pre5378Trust200902 you can add the attributes updates="NNNN" and obsoletes="NNNN" they will automatically be output with "(if approved)" --> <!-- ***** FRONT MATTER ***** --><front><!-- The abbreviated title is used in the page header - it is only necessary if the full title is longer than 39 characters --><title abbrev="DualQ CoupledAQMs">DualQAQMs">Dual-Queue CoupledAQMsActive Queue Management (AQM) for Low Latency, LowLossLoss, and Scalable Throughput (L4S)</title> <seriesInfoname="Internet-Draft" value="draft-ietf-tsvwg-aqm-dualq-coupled-25"/>name="RFC" value="9332"/> <author fullname="Koen De Schepper" initials="K." surname="De Schepper"> <organization>Nokia Bell Labs</organization> <address> <postal><street/><city>Antwerp</city> <country>Belgium</country> </postal> <email>koen.de_schepper@nokia.com</email> <uri>https://www.bell-labs.com/about/researcher-profiles/koende_schepper/</uri> </address> </author> <author fullname="Bob Briscoe" initials="B." role="editor" surname="Briscoe"> <organization>Independent</organization> <address> <postal><street/> <country>UK</country><country>United Kingdom</country> </postal> <email>ietf@bobbriscoe.net</email> <uri>https://bobbriscoe.net/</uri> </address> </author> <author fullname="Greg White" initials="G." surname="White"> <organization>CableLabs</organization> <address> <postal><street/> <city>Louisville, CO</city> <country>US</country><city>Louisville</city> <region>CO</region> <country>United States of America</country> </postal> <email>G.White@CableLabs.com</email> </address> </author><!-- <author fullname="Olga Albisser" initials="O." surname="Albisser"> <organization>Simula Research Lab</organization> <address> <postal> <street/> <city>Lysaker</city> <country>Norway</country> </postal> <email>olga@albisser.org</email> <uri>https://www.simula.no/people/olgabo</uri> </address> </author> <author fullname="Ing Jyh Tsang" initials="I." surname="Tsang"> <organization>Nokia</organization> <address> <postal> <street/> <city>Antwerp</city> <country>Belgium</country> </postal> <email>ing-jyh.tsang@nokia.com</email> </address> </author> --><datemonth="" year=""/> <area>Transport</area> <workgroup>Transport Area working group (tsvwg)</workgroup> <keyword>Internet-Draft</keyword> <keyword>I-D</keyword>year="2023" month="January" /> <area>tsv</area> <workgroup>tsvwg</workgroup> <keyword>Performance</keyword> <keyword>Queuing Delay</keyword> <keyword>One Way Delay</keyword> <keyword>Round-Trip Time</keyword> <keyword>RTT</keyword> <keyword>Jitter</keyword> <keyword>Congestion Control</keyword> <keyword>Congestion Avoidance</keyword> <keyword>Quality of Service</keyword> <keyword>QoS</keyword> <keyword>Quality of Experience</keyword> <keyword>QoE</keyword> <keyword>Active Queue Management</keyword> <keyword>AQM</keyword> <keyword>Explicit Congestion Notification</keyword> <keyword>ECN</keyword> <keyword>Pacing</keyword> <keyword>Burstiness</keyword> <abstract> <t>This specification defines a framework for coupling the Active Queue Management (AQM) algorithms in two queues intended for flows with different responses to congestion. This provides a way for the Internet to transition from the scaling problems of standardTCP Reno-friendlyTCP-Reno-friendly ('Classic') congestion controls to the family of 'Scalable' congestion controls. These are designed for consistently veryLowlow queuingLatency,latency, veryLowlow congestionLossloss, andScalingscaling of per-flow throughput(L4S)by using Explicit Congestion Notification (ECN) in a modified way. Until the CoupledDualQ,Dual Queue (DualQ), thesescalableScalable L4S congestion controls could only be deployed where a clean-slate environment could be arranged, such as in private data centres.</t><t>The<t>This specification first explains how a Coupled DualQ works. It then gives the normative requirements that are necessary for it to work well. All this is independent of which two AQMs are used, but pseudocode examples of specific AQMs are given in appendices.</t> </abstract> </front> <middle> <section anchor="dualq_intro" numbered="true" toc="default"> <name>Introduction</name> <t>This document specifies a framework for DualQ Coupled AQMs, which can serve as the network part of the L4Sarchitecture <xref target="I-D.ietf-tsvwg-l4s-arch"architecture <xref target="RFC9330" format="default"/>. ACoupledDualQ Coupled AQM consists of twoqueues;queues: L4S and Classic. The L4S queue is intended for Scalable congestion controls that can maintain very low queuing latency (sub-millisecond on average) and high throughput at the same time. The Coupled DualQ acts like a semi-permeable membrane: the L4S queue isolates the sub-millisecond average queuing delay of L4S from Classiclatency;latency, while the coupling between the queues pools the capacity between both queues so that ad hoc numbers of capacity-seeking applications all sharing the same capacity can have roughly equivalent throughput per flow, whichever queue they use. The DualQ achieves this indirectly, without having to inspecttransport layertransport-layer flow identifiers and without compromising the performance of the Classic traffic, relative to a single queue. The DualQ design has low complexity and requires no configuration for the public Internet.</t> <section anchor="dualq_problem" numbered="true" toc="default"> <name>Outline of the Problem</name> <t>Latency is becoming the critical performance factor for many(most?)(perhaps most) applications on the public Internet,e.g. interactive Web, Webe.g., interactive web, web services, voice, conversational video, interactive video, interactive remote presence, instant messaging, online gaming, remote desktop, cloud-based applications, and video-assisted remote control of machinery and industrial processes. Once access networkbit ratesbitrates reach levels now common in the developed world, further increases offer diminishing returns unless latency is also addressed <xref target="Dukkipati06" format="default"/>. In the last decade or so, much has been done to reduce propagation time by placing caches or servers closer to users. However, queuing remains a major intermittent component of latency.</t><t>Traditionally<t>Previously, very low latency has only been available for a few selectedlow ratelow-rate applications, that confine their sending rate within a specially carved-off portion of capacity, which is prioritized over other traffic,e.g. Diffserv EF <xrefe.g., Diffserv Expedited Forwarding (EF) <xref target="RFC3246" format="default"/>. Up tonownow, it has not been possible to allow any number oflow latency,low-latency, high throughput applications to seek to fully utilize available capacity, because the capacity-seeking process itself causes too much queuing delay.</t> <t>To reduce this queuing delay caused by thecapacity seekingcapacity-seeking process, changes either to the network alone or toend-systemsend systems alone are in progress. L4S involves a recognition that both approaches are yielding diminishing returns:</t> <ul spacing="normal"> <li>Recent state-of-the-artactive queue management (AQM)AQM in the network,e.g. FQ-CoDel <xrefe.g., Flow Queue CoDel <xref target="RFC8290" format="default"/>,PIE <xrefProportional Integral controller Enhanced (PIE) <xref target="RFC8033" format="default"/>, and AdaptiveRED <xrefRandom Early Detection (ARED) <xref target="ARED01"format="default"/> )format="default"/>), has reduced queuing delay for all traffic, not just a select few applications. However, no matter how good the AQM, the capacity-seeking (sawtoothing) rate of TCP-like congestion controls represents a lower limit that willeithercause either the queuing delay to vary orcausethe link to beunder-utilized.underutilized. These AQMs are tuned to allow a typical capacity-seekingReno-friendlyTCP-Reno-friendly flow to induce an average queue that roughly doubles the baseRTT,round-trip time (RTT), adding 5-15 ms of queuing on average for a mix of long-running flows and web traffic (cf. 500 microseconds with L4S for the same traffic mixof long-running and web traffic).<xref target="L4Seval22" format="default"/>). However, for manyapplicationsapplications, low delay is not useful unless it is consistently low. With these AQMs, 99th percentile queuing delay is 20-30 ms (cf. 2 ms with the same traffic over L4S).</li> <li>Similarly, recent research into usinge2eend-to-end congestion control without needing an AQM in the network(e.g. BBR(e.g., Bottleneck Bandwidth and Round-trip propagation time (BBR) <xref target="I-D.cardwell-iccrg-bbr-congestion-control" format="default"/>) seems to have hit a similarlower limit toqueuing delay floor of about20ms20 ms on average, but there are also regular25ms25 ms delay spikes due to bandwidth probes and60ms60 ms spikes due to flow-starts.</li> </ul> <t>L4S learns from the experience of Data CenterTCP <xrefTCP (DCTCP) <xref target="RFC8257" format="default"/>, which shows the power of complementary changes both in the network and onend-systems.end systems. DCTCP teaches us that two small but radical changes to congestion control are needed to cut the two major outstanding causes of queuing delay variability:</t> <ol spacing="normal" type="1"><li>Far smaller rate variations (sawteeth) than Reno-friendly congestioncontrols;</li>controls.</li> <li>A shift of smoothing and hence smoothing delay from network to sender.</li> </ol> <t>Without the former, a 'Classic'(e.g. Reno-friendly)(e.g., Reno-friendly) flow'sround trip time (RTT)RTT varies between roughly 1 and 2 times the base RTT between the machines in question. Without thelatterlatter, a 'Classic' flow's response to changing events is delayed by a worst-case (transcontinental) RTT, which could be hundreds of times the actual smoothing delay needed for the RTT of typical traffic from localizedCDNs.</t>Content Delivery Networks (CDNs).</t> <t>These changes are the two main features of the family of so-called 'Scalable' congestion controls (whichincludesinclude DCTCP,TCP PraguePrague, andSCReAM).Self-Clocked Rate Adaptation for Multimedia (SCReAM)). Both of these changes only reduce delay in combination with a complementary change in thenetworknetwork, and they are both only feasible with ECN, not drop, for the signalling:</t> <olspacing="normal" type="1"><li>Thespacing="normal"> <li>The smaller sawteeth allow an extremely shallow ECN packet-marking threshold in the queue.</li><li>And no<li>No smoothing in the network means that every fluctuation of the queue is signalled immediately.</li> </ol> <t>Without ECN, either of these would lead to very high loss levels.But,In contrast, with ECN, the resulting high marking levels are just signals, not impairments. (Note thatBBRv2 <xrefBBRv2 <xref target="BBRv2" format="default"/> combines the best of both worlds--- it works as ascalableScalable congestion control when ECN is available, but it also aims to minimize delay whenit isn't.)</t>ECN is absent.)</t> <t>However, until now, Scalable congestion controls (like DCTCP) did notco-existcoexist well in a shared ECN-capable queue with existing Classic(e.g. Reno <xref(e.g., Reno <xref target="RFC5681" format="default"/> orCubic <xrefCUBIC <xref target="RFC8312" format="default"/>) congestion controls -- Scalable controls are so aggressive that these 'Classic' algorithms would drive themselves to a small capacity share. Therefore, until now, L4S controls could only be deployed where a clean-slate environment could be arranged, such as in private data centres (hence the name DCTCP).</t> <t>One way to solve the problem of coexistence between Scalable and Classic flows is to use a per-flow-queuing (FQ) approach such asFQ-CoDel <xrefFQ-CoDel <xref target="RFC8290" format="default"/>. It classifies packets by flow identifier into separate queues in order to isolate sparse flows from the higher latency in the queues assigned to heavier flows. However, if a Classic flow needs both low delay and high throughput, having a queue to itself does not isolate it from the harm it causes to itself. Also FQ approaches need to inspect flow identifiers, which is not always practical.</t> <t>In summary, Scalable congestion controls address the root cause of the latency, loss and scaling problems with Classic congestion controls. Both FQ and DualQ AQMs can be enablers for this smoothlow latencylow-latency scalable behaviour. The DualQ approach is particularly useful because identifying flows is sometimes not practical or desirable.</t> </section> <section anchor="dualq_scope" numbered="true" toc="default"> <name>Context,Scope &Scope, and Applicability</name> <t>L4S involves complementary changes in the network and onend-systems:</t>end systems:</t> <dlnewline="false"newline="true" spacing="normal"> <dt>Network:</dt> <dd>A DualQ Coupled AQM (defined in the present document) or a modification toflow-queueflow queue AQMs (described insection 4.2.bparagraph "b" in Section <xref target="RFC9330" sectionFormat="bare" section="4.2"/> of the L4Sarchitecture <xref target="I-D.ietf-tsvwg-l4s-arch" format="default"/>);</dd> <dt>End-system:</dt>architecture <xref target="RFC9330" format="default"/>).</dd> <dt>End system:</dt> <dd>A Scalable congestion control (defined insection 4Section <xref target="RFC9331" sectionFormat="bare" section="4"/> of the L4S ECNprotocol <xref target="I-D.ietf-tsvwg-ecn-l4s-id"protocol spec <xref target="RFC9331" format="default"/>).</dd> <dt>Packet identifier:</dt> <dd>The network and end-system parts of L4S can be deployed incrementally, because they both identify L4S packets using the experimentally assignedexplicit congestion notification (ECN)ECN codepoints in the IP header: ECT(1) andCE <xrefCE <xref target="RFC8311" format="default"/> <xreftarget="I-D.ietf-tsvwg-ecn-l4s-id"target="RFC9331" format="default"/>.</dd> </dl><t>Data Center TCP (DCTCP <xref<t>DCTCP <xref target="RFC8257"format="default"/>)format="default"/> is an example of a Scalable congestion control for controlled environments that has been deployed for some time in Linux,WindowsWindows, and FreeBSD operating systems. During the progress of this document through theIETFIETF, a number of other Scalable congestion controls were implemented,e.g. TCPe.g., Prague over TCP and QUIC <xref target="I-D.briscoe-iccrg-prague-congestion-control"format="default"/> <xrefformat="default"/> <xref target="PragueLinux" format="default"/>, BBRv2 <xref target="BBRv2"format="default"/>,format="default"/> <xref target="I-D.cardwell-iccrg-bbr-congestion-control" format="default"/>,QUIC Pragueand the L4S variant ofSCREAMSCReAM for real-timemedia <xrefmedia <xref target="SCReAM-L4S" format="default"/> <xref target="RFC8298" format="default"/>.</t> <t>The focus of this specification is to enable deployment of the network part of the L4S service. Then, without any management intervention, applications can exploit this new network capability as the applications or their operating systems migrate to Scalable congestion controls, which can then evolve <em>while</em> their benefits are being enjoyed by everyone on the Internet.</t> <t>The DualQ Coupled AQM framework can incorporate any AQM designed for a single queue that generates a statistical or deterministic mark/drop probability driven by the queue dynamics. Pseudocode examples of two different DualQ Coupled AQMs are given in the appendices. In many cases the framework simplifies the basic controlalgorithm,algorithm and requires little extra processing. Therefore, it is believed the Coupled AQM would be applicable and easy to deploy in all types ofbuffers;buffers such as buffers in cost-reduced mass-market residential equipment; buffers in end-system stacks; buffers in carrier-scale equipment including remote access servers, routers,firewallsfirewalls, and Ethernet switches; buffers in network interfacecards,cards; buffers in virtualized network appliances,hypervisors,hypervisors; and so on.</t> <t>For the public Internet, nearly all the benefit will typically be achieved by deploying the Coupled AQM into either end of the access link between a 'site' and the Internet, which is invariably the bottleneck (seesection 6.4 of<xref target="I-D.ietf-tsvwg-l4s-arch" format="default"/><xref target="RFC9330" sectionFormat="of" section="6.4"/> about deployment, which also defines the term 'site' to mean a home, an office, acampuscampus, or mobile user equipment).</t> <t>Latency is not the only concern of L4S:</t> <ul spacing="normal"> <li>The"Low Loss"'Low Loss' part of the name denotes that L4S generally achieves zero congestion loss (which would otherwise cause retransmission delays), due to its use of ECN.</li> <li>The"Scalable throughput"'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 'TCP-Friendly' congestion controlalgorithms <xrefalgorithms <xref target="RFC3649" format="default"/>.</li> </ul> <t>The former is clearly in scope of this AQM document. However, the latter is an outcome of the end-systembehaviour,behaviour and is therefore outside the scope of this AQM document, even though the AQM is an enabler.</t> <t>The overall L4Sarchitecture <xref target="I-D.ietf-tsvwg-l4s-arch"architecture <xref target="RFC9330" format="default"/> gives more detail, including on wider deployment aspects such as backwards compatibility of Scalable congestion controls in bottlenecks where a DualQ Coupled AQM has not been deployed. The supporting papers <xref target="L4Seval22"/>, <xref target="DualPI2Linux" format="default"/>, <xref target="PI2" format="default"/>,<xref target="DCttH19" format="default"/>and <xref target="PI2param" format="default"/> give the full rationale for theAQM'sAQM design, both discursively and in more precise mathematical form, as well as the results of performance evaluations. The main results have been validated independently when using the Prague congestion control <xref target="Boru20" format="default"/> (experiments are run using Prague and DCTCP, but only the formerareis relevant for validation, because Prague fixes a number of problems with the Linux DCTCP code that make it unsuitable for the public Internet).</t> </section> <section anchor="dualq_Terminology" numbered="true" toc="default"> <name>Terminology</name><t>The<t> The key words"MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY","<bcp14>MUST</bcp14>", "<bcp14>MUST NOT</bcp14>", "<bcp14>REQUIRED</bcp14>", "<bcp14>SHALL</bcp14>", "<bcp14>SHALL NOT</bcp14>", "<bcp14>SHOULD</bcp14>", "<bcp14>SHOULD NOT</bcp14>", "<bcp14>RECOMMENDED</bcp14>", "<bcp14>NOT RECOMMENDED</bcp14>", "<bcp14>MAY</bcp14>", and"OPTIONAL""<bcp14>OPTIONAL</bcp14>" in this document are to be interpreted as described in BCP 14 <xreftarget="RFC2119" format="default"/>target="RFC2119"/> <xreftarget="RFC8174" format="default"/>target="RFC8174"/> when, and only when, they appear in all capitals, as shownhere.</t>here. </t> <t>The DualQ Coupled AQM uses two queues for twoservices. Each of the following terms identifies both the service and the queue that provides the service:</t>services:</t> <dl newline="false" spacing="normal"> <dt>Classicservice/queue:</dt>Service/Queue:</dt> <dd>The Classic service is intended for all the congestion control behaviours thatco-existcoexist withReno <xrefReno <xref target="RFC5681" format="default"/>(e.g. Reno(e.g., Reno itself,Cubic <xrefCUBIC <xref target="RFC8312" format="default"/>,TFRC <xrefand TFRC <xref target="RFC5348"format="default"/>).</dd> <dt>Low-Latency, Low-Lossformat="default"/>). The term 'Classic queue' means a queue providing the Classic service.</dd> <dt>Low Latency, Low Loss, and Scalable throughput (L4S)service/queue:</dt>Service/Queue:</dt> <dd>The 'L4S' service is intended for traffic fromscalableScalable congestion control algorithms, such asTCPthe Prague congestion control <xref target="I-D.briscoe-iccrg-prague-congestion-control" format="default"/>, which was derived from Data CenterTCP <xrefTCP <xref target="RFC8257" format="default"/>. The L4S service is for more general traffic than justTCPPrague -- it allows the set of congestion controls with similar scaling properties to Prague to evolve, such as the examplesof Scalable congestion controlslisted below (Relentless, SCReAM,etc.).</dd>etc.). The term 'L4S queue' means a queue providing the L4S service.</dd> <dt>Classic Congestion Control:</dt> <dd>A congestion control behaviour that canco-existcoexist with standardTCP Reno <xrefReno <xref target="RFC5681" format="default"/> without causing significantly negative impact on its flowrate <xrefrate <xref target="RFC5033" format="default"/>. With Classic congestion controls, such as Reno orCubic,CUBIC, because flow rate has scaled since TCP congestion control was first designed in 1988, 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 insection 5.1Section <xref target="RFC9330" sectionFormat="bare" section="5.1"/> of the L4Sarchitecture <xref target="I-D.ietf-tsvwg-l4s-arch" format="default"/>architecture <xref target="RFC9330"/> and in <xref target="RFC3649" format="default"/>. Therefore, control of queuing and utilization becomes very slack, and the slightest disturbances(e.g. from(e.g., from new flows starting) prevent a high rate from being attained.</dd> <dt>Scalable Congestion Control:</dt> <dd>A congestion control where the average time from one congestion signal to the next (the recovery time) remains invariant astheflow rate scales, all other factors being equal. This maintains the same degree of control overqueueingqueuing and utilization whatever the flow rate, as well as ensuring that high throughput is robust to disturbances. For instance, DCTCP averages 2 congestion signals perround-tripround trip, whatever the flow rate, as do other recently developedscalableScalable congestion controls,e.g. Relentless TCP <xrefe.g., Relentless TCP <xref target="I-D.mathis-iccrg-relentless-tcp" format="default"/>,TCPPrague <xref target="I-D.briscoe-iccrg-prague-congestion-control"format="default"/>, <xrefformat="default"/> <xref target="PragueLinux" format="default"/>, BBRv2 <xref target="BBRv2"format="default"/>,format="default"/> <xref target="I-D.cardwell-iccrg-bbr-congestion-control"format="default"/>format="default"/>, and the L4S variant ofSCREAMSCReAM for real-timemedia <xref target="SCReAM" format="default"/>, <xrefmedia <xref target="SCReAM-L4S" format="default"/> <xref target="RFC8298"format="default"/>).format="default"/>. For the publicInternetInternet, a Scalable transport has to comply with the requirements inSection 4 of<xreftarget="I-D.ietf-tsvwg-ecn-l4s-id" format="default"/> (aka. thetarget="RFC9331" sectionFormat="of" section="4"/> (a.k.a. the 'Prague L4S requirements').</dd> <dt>C:</dt> <dd>Abbreviation for Classic,e.g. whene.g., when used as a subscript.</dd> <dt>L:</dt> <dd> <t>Abbreviation for L4S,e.g. whene.g., when used as a subscript.</t> <t>The terms Classic or L4S can also qualify other nouns, such as 'codepoint', 'identifier', 'classification', 'packet', and 'flow'. Forexample:example, an L4S packet means a packet with an L4S identifier sent from an L4S congestion control.</t> <t>Both Classic and L4S services can cope with a proportion of unresponsive or less-responsive traffic aswell, butwell but, in the L4Scasecase, its rate has to be smooth enough or low enoughnotto not build a queue(e.g. DNS, VoIP,(e.g., DNS, Voice over IP (VoIP), game sync datagrams, etc.). The DualQ Coupled AQM behaviour is defined to be similar to a singleFIFOFirst-In, First-Out (FIFO) queue with respect to unresponsive and overload traffic.</t> </dd> <dt>Reno-friendly:</dt> <dd>The subset of Classic traffic that is friendly to the standard Reno congestion control defined for TCP in <xref target="RFC5681" format="default"/>.Reno-friendlyThe TFRC spec <xref target="RFC5348"/> 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-TCPtransportstransports, such asQUIC.</dd>QUIC <xref target="RFC9000"/>.</dd> <dt>DualQ or DualQ AQM:</dt> <dd>Used loosely as shorthand for a Dual-Queue Coupled AQM, where the context makes 'Coupled AQM' obvious.</dd> <dt>Classic ECN:</dt> <dd> <t>The original Explicit Congestion Notification (ECN)protocol <xrefprotocol <xref target="RFC3168"format="default"/>, whichformat="default"/> that requires ECN signals to be treatedthe sameas equivalent to drops, both when generated in the network and when responded to by the sender.</t> <t>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 <xref target="RFC3168"format="default"/>: Not ECT,format="default"/>, i.e., Not-ECT, ECT(0),ECT(1)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.</t> </dd> </dl> </section> <section numbered="true" toc="default"> <name>Features</name> <t>The AQM couples marking and/or dropping from the Classic queue to the L4S queue in such a way that a flow will get roughly the same throughput whichever it uses. Therefore, both queues can feed into the full capacity of alinklink, and no rates need to be configured for the queues. The L4S queue enables Scalable congestion controls like DCTCP orTCPPrague to give very low andpredictablyconsistently low latency, without compromising the performance of competing 'Classic' Internet traffic.</t> <t>Thousands of tests have been conducted in a typical fixed residential broadband setting. Experiments used a range of baseround tripround-trip delays up to100ms100 ms and link rates up to 200 Mb/s between the data centre and home network, with varying amounts of background traffic in both queues. For every L4S packet, the AQM kept the average queuing delay below1ms1 ms (or 2 packets where serialization delay exceeded1ms1 ms on slower links), with the 99th percentile being no worse than2ms.2 ms. No losses at all were introduced by the L4S AQM. Details of the extensive experiments areavailable <xref target="DualPI2Linux" format="default"/>,available in <xreftarget="PI2" format="default"/>,target="L4Seval22" format="default"/> and <xreftarget="DCttH19"target="DualPI2Linux" format="default"/>. Subjective testing using very demandinghigh bandwidth low latencyhigh-bandwidth low-latency applications over a single shared access link is also describedin <xrefin <xref target="L4Sdemo16" format="default"/> and summarized inthe section about applications inSection <xref target="RFC9330" sectionFormat="bare" section="6.1"/> of the L4Sarchitecture <xref target="I-D.ietf-tsvwg-l4s-arch" format="default"/> .</t>architecture <xref target="RFC9330" format="default"/>. </t> <t>In all these experiments, the host was connected to the home network by fixed Ethernet, in order to quantify the queuing delay that can be achieved by a user who cares about delay. It should be emphasized that L4S support at the bottleneck link cannot 'undelay' bursts introduced by another link on the path, for instance by legacy Wi-Fi equipment. However, if L4S support is added to the queue feeding the <em>outgoing</em> WAN link of a home gateway, it would be counterproductive not to also reduce the burstiness of the <em>incoming</em> Wi-Fi. Also, trials of Wi-Fi equipment with an L4S DualQ Coupled AQM on the <em>outgoing</em> Wi-Fi interface are in progress, and early results of an L4S DualQ Coupled AQM in a 5G radio access network testbed with emulated outdoor cell edge radio fading are given in <xref target="L4S_5G" format="default"/>.</t> <t>Unlike DiffservExpedited Forwarding,EF, the L4S queue does not have to be limited to a small proportion of the link capacity in order to achieve low delay. The L4S queue can be filled with a heavy load of capacity-seeking flows(TCP Prague(Prague, BBRv2, etc.) and still achieve low delay. The L4S queue does not rely on the presence of other traffic in the Classic queue that can be 'overtaken'. It gives low latency to L4S traffic whether or not there is Classic traffic. The tail latency of traffic served by the Classic AQM is sometimes a littlebetterbetter, sometimes a little worse, when a proportion of the traffic is L4S.</t> <t>The two queues are only necessary because:</t> <ul spacing="normal"><li>the<li>The large variations (sawteeth) of Classic flows need roughly a base RTT of queuing delay to ensure fullutilization</li>utilization.</li> <li>Scalable flows do not need a queue to keep utilization high, but they cannot keep latencypredictablyconsistently low if they are mixed with Classictraffic,</li>traffic.</li> </ul> <t>The L4S queue has latency priority withinsub-round tripsub-round-trip timescales, but over longer periods the coupling from the Classic to the L4S AQM (explained below) ensures that it does not have bandwidth priority over the Classic queue.</t> </section> </section> <section anchor="dualq_algo" numbered="true" toc="default"> <name>DualQ Coupled AQM</name> <t>There are two main aspects to the DualQ Coupled AQM approach:</t><ul<ol spacing="normal"> <li>The Coupled AQM that addresses throughput equivalence between Classic(e.g. Reno, Cubic)(e.g., Reno or CUBIC) flows and L4S flows (that satisfy the Prague L4S requirements).</li> <li>TheDual QueueDual-Queue structure that provides latency separation for L4S flows to isolate them from the typically large Classic queue.</li></ul> <!--<t>The following subsections descrbe these two aspects, and how packets are classified between the two queues, then a likely overall structure of a DualQ Coupled AQM is given. The present document applies irrespective of which particular AQMs are used for each queue. So, although the structure is intended to be generic, it might not fit well around types of AQM yet to be considered. Finally normative requirements are given that apply to any specific DualQ Coupled AQM implementation, irrespective of which AQMs it uses. Pseudocode of specific examples are given in non-normative appendices.</t>--></ol> <section anchor="dualq_coupled" numbered="true" toc="default"> <name>Coupled AQM</name> <t>In the 1990s, the`TCP'TCP formula' was derived for the relationship between the steady-state congestion window, cwnd, and the drop probability, p of standard Reno congestioncontrol <xrefcontrol <xref target="RFC5681" format="default"/>. To afirst orderfirst-order approximation, the steady-state cwnd of Reno is inversely proportional to the square root of p.</t> <t>The design focuses on Reno as the worst case, because if it does no harm to Reno, it will not harmCubicCUBIC or any traffic designed to be friendly to Reno. TCPCubicCUBIC implements aReno-compatibilityReno-friendly mode, which is relevant for typical RTTs under20ms20 ms as long as the throughput of a single flow is less than about350Mb/s.350 Mb/s. In suchcasescases, it can be assumed thatCubicCUBIC traffic behaves similarly to Reno. The term 'Classic' will be used for the collection of Reno-friendly traffic includingCubicCUBIC and potentially other experimental congestion controls intended not to significantly impact the flow rate of Reno.</t> <t>A supportingpaper <xrefpaper <xref target="PI2" format="default"/> includes the derivation of the equivalent rate equation for DCTCP, for which cwnd is inversely proportional to p (not the square root), where in this case p is theECN markingECN-marking probability. DCTCP is not the only congestion control that behaves like this, so the term 'Scalable' will be used for all similar congestion control behaviours (see examples in <xref target="dualq_scope" format="default"/>). The term 'L4S' is used for traffic driven by a Scalable congestion control that also complies with the additional 'PragueL4S' requirements <xref target="I-D.ietf-tsvwg-ecn-l4s-id"L4S requirements' <xref target="RFC9331" format="default"/>.</t> <t>For safeco-existence,coexistence, under stationary conditions, a Scalable flow has to run at roughly the same rate as a Reno TCP flow (all other factors being equal). So the drop or marking probability for Classic traffic,p_Cp_C, has to be distinct from the marking probability for L4S traffic, p_L. The original ECNspecification <xrefspec <xref target="RFC3168" format="default"/> required these probabilities to be the same, but <xref target="RFC8311" format="default"/> updatesRFC 3168<xref target="RFC3168" format="default"/> to enable experiments in which these probabilities are different.</t> <t>Also, to remain stable, Classic sources need the network to smooth p_C so it changes relatively slowly. It is hard for a network node to know the RTTs of all the flows, so a Classic AQM adds a <em>worst-case</em> RTT of smoothing delay (about 100-200 ms). In contrast, L4S shifts responsibility for smoothing ECN feedback to the sender, which only delays its response by its <em>own</em> RTT, as well as allowing a more immediate response if necessary.</t> <t>The Coupled AQM achieves safe coexistence by making the Classic drop probability p_C proportional to the square of the coupled L4S probability p_CL. p_CL is an input to the instantaneous L4S marking probabilityp_Lp_L, but it changes as slowly as p_C. This makes the Reno flow rate roughly equal the DCTCP flow rate, because the squaring of p_CL counterbalances the square root of p_C in the 'TCP formula' of Classic Reno congestion control.</t> <t>Stating this as a formula, the relation between Classic drop probability, p_C, and the coupled L4S probability p_CL needs to take the following form:</t><artwork name="" type="" align="left" alt=""><![CDATA[<sourcecode><![CDATA[ p_C = ( p_CL / k)^2 (1)]]></artwork>)^2, (1)]]></sourcecode> <t>where k is the constant of proportionality, which is termed thecoupling factor.</t>'coupling factor'.</t> </section> <section anchor="dualq" numbered="true" toc="default"> <name>Dual Queue</name> <t>Classic traffic needs to build a large queue to preventunder-utilization.underutilization. Therefore, a separate queue is provided for L4S traffic, and it is scheduled with priority over the Classic queue. Priority is conditional to prevent starvation of Classic traffic in certain conditions (see <xref target="dualq_coupled_structure" format="default"/>).</t> <t>Nonetheless, coupled marking ensures that giving priority to L4S traffic still leaves the right amount of spare scheduling time for Classic flows to each get equivalent throughput to DCTCP flows (all otherfactorsfactors, such asRTTRTT, being equal).</t> </section> <section anchor="dualq_classification" numbered="true" toc="default"> <name>Traffic Classification</name> <t>Both the Coupled AQM and DualQ mechanisms need an identifier to distinguish L4S (L) and Classic (C) packets. Then the coupling algorithm can achieve coexistence without having to inspect flow identifiers, because it can apply the appropriate marking or dropping probability to all flows of each type. A separatespecification <xref target="I-D.ietf-tsvwg-ecn-l4s-id"specification <xref target="RFC9331" format="default"/> requires the network to treat the ECT(1) and CE codepoints of the ECN field as this identifier. An additional process document has proved necessary to make the ECT(1) codepoint available forexperimentation <xrefexperimentation <xref target="RFC8311" format="default"/>.</t> <t>For policy reasons, an operator might choose to steer certain packets(e.g. from(e.g., from certain flows or with certain addresses) out of the L queue, even though they identify themselves as L4S by their ECN codepoints. In such cases, the L4S ECNprotocol <xref target="I-D.ietf-tsvwg-ecn-l4s-id"protocol <xref target="RFC9331" format="default"/>saysstates that the device"MUST NOT"<bcp14>MUST NOT</bcp14> alter the end-to-end L4S ECNidentifier",identifier" so that it is preservedend-to-end.end to end. The aim is that each operator can choose how it treats L4S traffic locally, but an individual operator does not alter the identification of L4S packets, which would prevent other operators downstream from making their own choices on how to treat L4S traffic.</t> <t>In addition, an operator could use other identifiers to classify certain additional packet types into the L queue that it deems will not risk harm to the L4Sservice. For instanceservice, for instance, addresses of specific applications or hosts; specific Diffserv codepoints such asEF (Expedited Forwarding), Voice-AdmitEF, Voice-Admit, or the Non-Queue-Building (NQB) per-hop behaviour; or certain protocols(e.g. ARP,(e.g., ARP and DNS) (seeSection 5.4.1 of<xreftarget="I-D.ietf-tsvwg-ecn-l4s-id" format="default"/>).target="RFC9331" sectionFormat="of" section="5.4.1"/>. Note thatthe mechanism only reads these identifiers.<xreftarget="I-D.ietf-tsvwg-ecn-l4s-id"target="RFC9331" format="default"/>says it "MUST NOT alter these non-ECN identifiers".states that "a network node <bcp14>MUST NOT</bcp14> change Not-ECT or ECT(0) in the IP-ECN field into an L4S identifier." Thus, the L queue is not solely an L4Squeue,queue; it can be considered more generally as alow latencylow-latency queue.</t> </section> <section anchor="dualq_coupled_structure" numbered="true" toc="default"> <name>Overall DualQ Coupled AQM Structure</name> <t><xref target="dualq_fig_structure" format="default"/> shows the overall structure that any DualQ Coupled AQM is likely to have. This schematic is intended to aid understanding of the current designs of DualQ Coupled AQMs. However, it is not intended to preclude other innovative ways of satisfying the normative requirements in <xref target="dualq_norm_reqs" format="default"/> that minimally define a DualQ Coupled AQM. Also, the schematic only illustrates operation under normally expected circumstances; behaviour under overload or with operator-specific classifiers is deferred to <xref target="dualq_unexpected" format="default"/>.</t> <t>The classifier on the left separates incoming traffic between the two queues (L and C). Each queue has its own AQM that determines the likelihood of marking or dropping (p_L and p_C).ItIn <xref target="PI2" format="default"/>, it has beenproved <xref target="PI2" format="default"/>proved that it is preferable to control load with a linear controller, then square the output before applying it as a drop probability to Reno-friendly traffic (because Reno congestion control decreases its load proportional to thesquare-rootsquare root of the increase in drop). So, the AQM for Classic traffic needs to be implemented in two stages: i) a base stage that outputs an internal probability p' (pronouncedp-prime);'p-prime') and ii) a squaring stage that outputs p_C, where</t><artwork name="" type="" align="left" alt=""><![CDATA[<sourcecode><![CDATA[ p_C = (p')^2.(2)]]></artwork>(2)]]></sourcecode> <t>Substituting for p_C inEqnequation (1)gives:</t> <artwork name="" type="" align="left" alt=""><![CDATA[gives</t> <sourcecode><![CDATA[ p' = p_CL /k]]></artwork>k.]]></sourcecode> <t>So the slow-moving input to ECN marking in the L queue (the coupled L4S probability)is:</t> <artwork name="" type="" align="left" alt=""><![CDATA[is</t> <sourcecode><![CDATA[ p_CL = k*p'.(3)]]></artwork>(3)]]></sourcecode> <t>The actualECN markingECN-marking probability p_L that is applied to the L queue needs to track the immediate L queue delay under L-only congestion conditions, as well as track p_CL under coupled congestion conditions. So the L queue uses anative AQM'Native AQM' that calculates a probability p'_L as a function of the instantaneous L queue delay.And,And given the L queue has conditional priority over the C queue, whenever the L queue grows, the AQM ought to apply marking probability p'_L, but p_L oughtnotto not fall below p_CL. Thissuggests:</t> <artwork name="" type="" align="left" alt=""><![CDATA[suggests</t> <sourcecode><![CDATA[ p_L = max(p'_L, p_CL),(4)]]></artwork>(4)]]></sourcecode> <t>which has also been found to work very well in practice.</t> <t>The two transformations of p' in equations (2) and (3) implement the required coupling given in equation (1) earlier.</t> <t>The constant of proportionality or coupling factor, k, in equation (1) determines the ratio between the congestion probabilities (loss or marking) experienced by L4S and Classic traffic. Thus, k indirectly determines the ratio between L4S and Classic flow rates, because flows (assuming they are responsive) adjust their rate in response to congestion probability. <xref target="dualq_Choosing_k" format="default"/> gives guidance on the choice of k and its effect on relative flow rates.</t> <figure anchor="dualq_fig_structure"> <name>DualQ Coupled AQM Schematic</name> <artwork name="" type="" align="left" alt=""><![CDATA[ _________ | | ,------. L4S (L) queue | |===>| ECN | ,'| _______|_| |marker|\ <' | | `------'\\ //`' v ^ p_L \\ // ,-------. | \\ // |Native |p'_L | \\,. // | L4S |--->(MAX) < | ___ ,----------.// | AQM | ^ p_CL `\|.'Cond-`. | IP-ECN |/ `-------' | / itional \ ==>|Classifier| ,-------. (k*p') [ priority]==> | |\ | Base | | \scheduler/ `----------'\\ | AQM |---->: ,'|`-.___.-' \\ | |p' | <' | \\ `-------' (p'^2) //`' \\ ^ | // \\,. | v p_C // < | _________ .------.// `\| | | | Drop |/ Classic (C) |queue |===>|/mark | __|______|`------' ]]></artwork> </figure> <t keepWithPrevious="true">Legend: ===>`------']]> Legend: ===> trafficflow; --->flow ---> controldependency.</t>dependency </artwork> </figure> <t>After the AQMs have applied their dropping or marking, the scheduler forwards their packets to the link. Even though the scheduler gives priority to the L queue, it is not as strong as the coupling from the C queue. This is because, as the C queue grows, thebase AQM'Base AQM' applies more congestion signals to L traffic (as well as to C). As L flows reduce their rate in response, they use less than the scheduling share for L traffic. So, because the scheduler is work preserving, it schedules any C traffic in the gaps.</t> <t>Giving priority to the L queue has the benefit of very low L queue delay, because the L queue is kept empty whenever L traffic is controlled by the coupling. Also, there only has to be a coupling in one direction--- from Classic to L4S. Priority has to be conditional in some way to prevent the C queue from being starved in theshort-termshort term (see <xref target="dualq_Overload_Starvation" format="default"/>) to give C traffic a means to push in, as explained next. With normal responsive L traffic, the coupled ECN marking gives C traffic the ability to push back against even strict priority, by congestion marking the L traffic to make it yield some space. However, if there is just a small finite set of C packets(e.g. a(e.g., a DNS request or an initial window ofdata)data), some Classic AQMs will not induce enough ECN marking in the L queue, no matter how long the small set of C packets waits. Then, if the L queue happens to remain busy, the C traffic would never get a scheduling opportunity from a strict priority scheduler.IdeallyIdeally, the Classic AQM would be designed to increase the coupled marking the longer that C packets have been waiting, but this is not always practical--- hence the need for L priority to be conditional. Giving a small weight or limited waiting time for C traffic improves response times for short Classic messages, such as DNS requests, and improves Classic flow startup because immediate capacity is available.</t> <t>Example DualQ Coupled AQM algorithms calledDualPI2'DualPI2' andCurvy RED'Curvy RED' are given in Appendices <xref target="dualq_Ex_algo_pi2"format="default"/>format="counter"/> and <xref target="dualq_Ex_algo"format="default"/>.format="counter"/>. Either example AQM can be used to couple packet marking and dropping across adual Q.</t> <t>DualPI2DualQ:</t> <ul spacing="normal"> <li><t>DualPI2 uses aProportional-IntegralProportional Integral (PI) controller as the Base AQM. Indeed, this Base AQM with just the squared output and no L4S queue can be used as a drop-in replacement forPIE <xrefPIE <xref target="RFC8033" format="default"/>, in which case it is just calledPI2 <xrefPI2 <xref target="PI2" format="default"/>. PI2 is a principled simplification of PIE that is both more responsive and more stable in the face of dynamically varyingload.</t> <t>Curvyload.</t></li> <li><t>Curvy RED is derived fromRED <xref target="RFC2309"RED <xref target="RED" format="default"/>, except its configuration parameters are delay-based to make them insensitive to linkraterate, and it requires fewer operations per packet than RED. However, DualPI2 is more responsive and stable over a wider range of RTTs than Curvy RED. As a consequence, at the time of writing, DualPI2 has attracted more development and evaluation attention than Curvy RED, leaving the Curvy RED design not so fullyevaluated.</t>evaluated.</t></li> </ul> <t>Both AQMs regulate their queue against targets configured in units of time rather than bytes. As already explained, this ensures configuration can be invariant for different drain rates. With AQMs in adualQDualQ structure this is particularly important because the drain rate of each queue can vary rapidly as flows for the two queues arrive and depart, even if the combined link rate is constant.</t> <t>It would be possible to control the queues with other alternative AQMs, as long as the normative requirements (those expressed in capitals) in <xref target="dualq_norm_reqs" format="default"/> are observed.</t> <t>The two queues could optionally be part of a larger queuing hierarchy, such as the initial example ideas in <xref target="I-D.briscoe-tsvwg-l4s-diffserv" format="default"/>.</t> </section> <section anchor="dualq_norm_reqs" numbered="true" toc="default"> <name>Normative Requirements for a DualQ Coupled AQM</name> <t>The following requirements are intended to capture only the essential aspects of a DualQ Coupled AQM. They are intended to be independent of the particular AQMs implemented for eachqueue,queue but to still define the DualQ framework built around those AQMs.</t> <section anchor="dualq_functional_reqs" numbered="true" toc="default"> <name>Functional Requirements</name> <t>ADual QueueDualQ Coupled AQM implementationMUST<bcp14>MUST</bcp14> comply with the prerequisite L4S behaviours for any L4S network node (not just a DualQ) as specified insection 5 of<xreftarget="I-D.ietf-tsvwg-ecn-l4s-id" format="default"/>.target="RFC9331" sectionFormat="of" section="5"/>. These primarily concern classification andremarkingre-marking as briefly summarized earlier in <xref target="dualq_classification"format="default"/> earlier.format="default"/>. Butthere is<xref target="RFC9331" sectionFormat="of" section="5.5"/> alsoa subsection (5.5) givinggives guidance on reducing the burstiness of the link technology underlying any L4S AQM.</t> <t>ADual QueueDualQ Coupled AQM implementationMUST<bcp14>MUST</bcp14> utilize two queues, each with an AQM algorithm.</t> <t>The AQM algorithm for thelow latencylow-latency (L) queueMUST<bcp14>MUST</bcp14> be able to apply ECN marking to ECN-capable packets.</t> <t>The scheduler draining the two queuesMUST<bcp14>MUST</bcp14> give L4S packets priority over Classic, although priorityMUST<bcp14>MUST</bcp14> be bounded in order not to starve Classic traffic (see <xref target="dualq_Overload_Starvation" format="default"/>). The schedulerSHOULD<bcp14>SHOULD</bcp14> be work-conserving, or otherwise close to work-conserving. This is because Classic traffic needs to be able to efficiently fill any space left by L4S traffic even though the scheduler would otherwise allocate it to L4S.</t> <t><xreftarget="I-D.ietf-tsvwg-ecn-l4s-id"target="RFC9331" format="default"/> defines the meaning of an ECN marking on L4S traffic, relative to drop of Classic traffic. In order to ensure coexistence of Classic and Scalable L4S traffic, it says,"The"the likelihood thatanthe AQM drops a Not-ECT Classic packet (p_C)MUST<bcp14>MUST</bcp14> be roughly proportional to the square of the likelihood that it would have marked it if it had been an L4S packet (p_L)." The term 'likelihood' is used to allow for marking and dropping to be either probabilistic or deterministic.</t> <t>For the current specification, this translates into the following requirement. A DualQ Coupled AQMMUST<bcp14>MUST</bcp14> apply ECN marking to traffic in the L queue that is no lower than that derived from the likelihood of drop (or ECN marking) in the Classic queue usingEqn.equation (1).</t> <t>The constant of proportionality, k, inEqnequation (1) determines the relative flow rates of Classic and L4S flows when the AQM concerned is the bottleneck (all other factors being equal). The L4S ECNprotocol <xref target="I-D.ietf-tsvwg-ecn-l4s-id"protocol <xref target="RFC9331" format="default"/> says, "The constant of proportionality (k) does not have to be standardised for interoperability, but a value of 2 isRECOMMENDED."</t><bcp14>RECOMMENDED</bcp14>." </t> <t>Assuming Scalable congestion controls for the Internet will be as aggressive as DCTCP, this will ensure their congestion window will be roughly the same as that of astandards trackStandards Track TCP Reno congestion control(Reno) <xref(Reno) <xref target="RFC5681" format="default"/> and other Reno-friendly controls, such as TCPCubicCUBIC in itsReno-compatibilityReno-friendly mode.</t><!--{ToDo: The TCP Prague requirements are not necessarily final. If the aggressiveness of DCTCP is not defined as the benchmark for Scalable controls on the Internet, the recommended value of k will also be subject to change.}--><t>The choice of k is a matter of operator policy, and operatorsMAY<bcp14>MAY</bcp14> choose a different value using the guidelines in <xref target="dualq_Choosing_k" format="default"/>.</t> <t>If multiple customers or users share capacity at a bottleneck(e.g. in(e.g., in the Internet access link of a campus network), the operator's choice of k will determine capacity sharing between the flows of different customers. However, on the public Internet, access network operators typically isolate customers from each other with some form oflayer-2Layer 2 multiplexing (OFDM(A) inDOCSIS3.1,DOCSIS 3.1, CDMA in 3G, and SC-FDMA in LTE) orL3Layer 3 scheduling(WRR in DSL),(Weighted Round Robin (WRR) for DSL) rather than relying on host congestion controls to share capacity betweencustomers <xrefcustomers <xref target="RFC0970" format="default"/>. In such cases, the choice of k will solely affect relative flow rates within each customer's access capacity, not between customers. Also, k will not affect relative flow rates at any times when all flows are Classic or all flows are L4S, and it will not affect the relative throughput of small flows.</t> <t/> <section anchor="dualq_unexpected" numbered="true" toc="default"> <name>Requirements in Unexpected Cases</name> <t>The flexibility to allow operator-specific classifiers (<xref target="dualq_classification" format="default"/>) leads to the need to specify what the AQM in each queue ought to do with packets that do not carry the ECN field expected for that queue. It is expected that the AQM in each queue will inspect the ECN field to determine what sort of congestion notification to signal, then it will decide whether to apply congestion notification to this particular packet, as follows:</t> <ul spacing="normal"> <li> <t>If a packet that does not carry an ECT(1) or a CE codepoint is classified into the Lqueue:</t>queue, then:</t> <ul spacing="normal"> <li>if the packet is ECT(0), the L AQMSHOULD<bcp14>SHOULD</bcp14> applyCE-markingCE marking using a probability appropriate to Classic congestion control and appropriate to the target delay in the L queue</li> <li> <t>if the packet is Not-ECT, the appropriate action depends on whether some other function is protecting the L queue from misbehaving flows(e.g. per-flow(e.g., per-flow queue protection <xref target="I-D.briscoe-docsis-q-protection" format="default"/> or latency policing):</t> <ul spacing="normal"><li>If<li>if separate queue protection is provided, the L AQMSHOULD<bcp14>SHOULD</bcp14> ignore the packet and forward it unchanged, meaning it should not calculate whether to apply congestionnotificationnotification, and it should neither drop norCE-markCE mark the packet (for instance, the operator might classify EF traffic that is unresponsive to drop into the L queue, alongside responsive L4S-ECN traffic)</li> <li>if separate queue protection is not provided, the L AQMSHOULD<bcp14>SHOULD</bcp14> apply drop using a drop probability appropriate to Classic congestion control andappropriateto the target delay in the L queue</li> </ul> </li> </ul> </li> <li> <t>If a packet that carries an ECT(1) codepoint is classified into the C queue:</t> <ul spacing="normal"> <li>the C AQMSHOULD<bcp14>SHOULD</bcp14> applyCE-markingCE marking using thecoupledCoupled AQM probability p_CL (= k*p').</li> </ul> </li> </ul> <t>The above requirements are worded as"SHOULDs","<bcp14>SHOULD</bcp14>"s, because operator-specific classifiers are for flexibility, by definition. Therefore, alternative actions might be appropriate in the operator's specific circumstances. An example would be where the operator knows that certain legacy trafficmarked withset to one codepoint actually has a congestion response associated with another codepoint.</t> <t>If the DualQ Coupled AQM has detected overload, itMUST<bcp14>MUST</bcp14> introduce Classic drop to both types of ECN-capable traffic until the overload episode has subsided. Introducing drop if ECN marking is persistently high is recommendedbyin Section7<xref target="RFC3168" sectionFormat="bare" section="7"/> of the ECNspecification <xref target="RFC3168" format="default"/>spec <xref target="RFC3168"/> and in Section4.2.1<xref target="RFC7567" sectionFormat="bare" section="4.2.1"/> of the AQMRecommendations <xref target="RFC7567" format="default"/>.</t>Recommendations <xref target="RFC7567"/>.</t> </section> </section> <section numbered="true" toc="default"> <name>Management Requirements</name> <t/> <section anchor="dualq_config" numbered="true" toc="default"> <name>Configuration</name> <t>By default, a DualQ Coupled AQMSHOULD NOT<bcp14>SHOULD NOT</bcp14> need any configuration for use at a bottleneck on the publicInternet <xrefInternet <xref target="RFC7567" format="default"/>. The following parametersMAY<bcp14>MAY</bcp14> be operator-configurable,e.g. toe.g., to tune for non-Internet settings:</t> <ul spacing="normal"> <li>Optional packet classifier(s) to use in addition to the ECN field (see <xref target="dualq_classification"format="default"/>);</li>format="default"/>).</li> <li> <t>Expected typical RTT, which can be used to determine the queuing delay of the Classic AQM at its operating point, in order to prevent typical lone flows fromunder-utilizingunderutilizing capacity. For example:</t> <ul spacing="normal"> <li>for the PI2 algorithm (<xref target="dualq_Ex_algo_pi2"format="default"/>)format="default"/>), the queuing delay target is dependent on the typicalRTT;</li>RTT.</li> <li>for the Curvy RED algorithm (<xref target="dualq_Ex_algo"format="default"/>)format="default"/>), the queuing delay at the desired operating point of the curvy ramp is configured to encompass a typicalRTT;</li>RTT.</li> <li>if another Classic AQM was used, it would be likely to need an operating point for the queue based on the typical RTT, and ifsoso, itSHOULD<bcp14>SHOULD</bcp14> be expressed in units of time.</li> </ul> <t>An operating point that is manually calculated might be directly configurable instead,e.g. fore.g., for links with large numbers of flows whereunder-utilizationunderutilization by a single flow would be unlikely.</t> </li> <li> <t>Expected maximum RTT, which can be used to set the stability parameter(s) of the Classic AQM. For example:</t> <ul spacing="normal"> <li>for the PI2 algorithm (<xref target="dualq_Ex_algo_pi2" format="default"/>), the gain parameters of the PI algorithm depend on the maximum RTT.</li> <li>for the Curvy RED algorithm (<xref target="dualq_Ex_algo"format="default"/>)format="default"/>), the smoothing parameter is chosen to filter out transients in the queue within a maximum RTT.</li> </ul><t>Stability parameter(s)<t>Any stability parameter thatareis manually calculated assuming a maximum RTT might be directly configurable instead.</t> </li> <li>Coupling factor, k (see <xref target="dualq_Choosing_k"format="default"/>);</li>format="default"/>).</li> <li> <t>A limit to the conditional priority of L4S. This is scheduler-dependent, but itSHOULD<bcp14>SHOULD</bcp14> be expressed as a relation between the max delay of a C packet and an L packet. For example:</t> <ul spacing="normal"> <li>for a WRRschedulerscheduler, a weight ratio between L and C of w:1 means that the maximum delaytoof a C packet is w times that of an L packet.</li> <li>for a time-shifted FIFO (TS-FIFO) scheduler (see <xref target="dualq_Overload_Starvation"format="default"/>)format="default"/>), a time-shift of tshift means that the maximum delay to a C packet is tshift greater than that of an L packet. tshift could be expressed as a multiple of the typical RTT rather than as an absolute delay.</li> </ul> </li> <li>The maximum ClassicECN markingECN-marking probability, p_Cmax, before introducing drop.</li> </ul> </section> <section numbered="true" toc="default"> <name>Monitoring</name> <t>An experimental DualQ Coupled AQMSHOULD<bcp14>SHOULD</bcp14> allow the operator to monitor each of the following operational statistics on demand, per queue and per configurable sample interval, for performance monitoring and perhaps also for accounting in some cases:</t> <ul spacing="normal"><li>Bits<li>bits forwarded, from which utilization can be calculated;</li><li>Total<li>total packets in the three categories: arrived, presented to the AQM, and forwarded. The difference between the first two will measure any non-AQM tail discard. The difference between the last two will measure proactive AQM discard;</li> <li>ECN packets marked, non-ECN packets dropped, and ECN packets dropped, which can be combined with the three total packet counts above to calculate marking and droppingprobabilities;</li>probabilities; and</li> <li><t>Queue<t>queue delay (not including serialization delay of the head packet or medium acquisition delay)--- see further notes below.</t> <t>Unlike the other statistics, queue delay cannot be captured in a simple accumulating counter. Therefore, the type of queue delay statistics produced (mean, percentiles, etc.) will depend on implementation constraints. To facilitate comparative evaluation of different implementations and approaches, an implementationSHOULD<bcp14>SHOULD</bcp14> allow mean and 99th percentile queue delay to be derived (per queue per sample interval). A relatively simple way to do this would be to store a coarse-grained histogram of queue delay. This could be done with a small number of bins with configurable edges that represent contiguous ranges of queue delay. Then, over a sample interval, each bin would accumulate a count of the number of packets that had fallen within each range. The maximum queue delay per queue per intervalMAY<bcp14>MAY</bcp14> also be recorded, to aid diagnosis of faults and anomalous events.</t> </li> </ul> </section> <section numbered="true" toc="default"> <name>Anomaly Detection</name> <t>An experimental DualQ Coupled AQMSHOULD<bcp14>SHOULD</bcp14> asynchronously report the following data about anomalous conditions:</t> <ul spacing="normal"> <li><t>Start-time<t>Start time and duration of overload state.</t> <t>A hysteresis mechanismSHOULD<bcp14>SHOULD</bcp14> be used to prevent flapping in and out of overload causing an event storm. For instance,exitexiting from overload state could trigger onereport,report but also latch a timer. Then, during that time, if the AQM enters and exits overload state any number of times, the duration in overload state is accumulated, but no new report is generated until the first time the AQM is out of overload once the timer has expired.</t> </li> </ul> </section> <section numbered="true" toc="default"> <name>Deployment,CoexistenceCoexistence, and Scaling</name> <t><xref target="RFC5706" format="default"/> suggests that deployment,coexistencecoexistence, and scaling should also be covered as management requirements. The raison d'etre of the DualQ Coupled AQM is to enable deployment and coexistence of Scalable congestion controls- as(as incremental replacements for today's Reno-friendly controls that do not scale with bandwidth-delayproduct.product). Therefore, there is no need to repeat these motivating issues here given they are already explained in the Introduction and detailed in the L4Sarchitecture <xref target="I-D.ietf-tsvwg-l4s-arch"architecture <xref target="RFC9330" format="default"/>.</t> <t>The descriptions of specific DualQ Coupled AQM algorithms in the appendices cover scaling of their configuration parameters,e.g. withe.g., with respect to RTT and sampling frequency.</t> </section> </section> </section> </section> <section anchor="dualq_IANA" numbered="true" toc="default"> <name>IANAConsiderations (to be removed by RFC Editor)</name>Considerations</name> <t>Thisspecification containsdocument has no IANAconsiderations.</t>actions.</t> </section> <section anchor="dualq_Security_Considerations" numbered="true" toc="default"> <name>Security Considerations</name> <t/> <section numbered="true" toc="default"> <name>Low Delay without RequiringPer-FlowPer-flow Processing</name> <t>The L4Sarchitecture <xref target="I-D.ietf-tsvwg-l4s-arch"architecture <xref target="RFC9330" format="default"/> compares the DualQ andper-flow-queuing (FQ)FQ approaches to L4S. The privacy considerations section in that document motivates the DualQ on the grounds that users who want to encrypt application flow identifiers,e.g. in IPSece.g., in IPsec or other encrypted VPN tunnels, don't have to sacrifice low delay (<xref target="RFC8404" format="default"/> encourages avoidance of such privacy compromises).</t> <t>The security considerations section of the L4S architecture <xref target="RFC9330" format="default"/> also includes subsections on policing of relativeflow-rates (section 8.1)flow rates (Section <xref target="RFC9330" sectionFormat="bare" section="8.1"/>) and on policing of flows that cause excessive queuing delay(section 8.2).(Section <xref target="RFC9330" sectionFormat="bare" section="8.2"/>). It explains that the interests of users do not collide in the same way for delay as they do for bandwidth. For someone to get more of the bandwidth of a shared link, someone else necessarily gets less (a 'zero-sum game'), whereas queuing delay can be reduced for everyone, without any need for someone else to lose out. It also explains that, on the current Internet, scheduling usually enforces separation of bandwidth between 'sites'(e.g. households, businesses(e.g., households, businesses, or mobile users), but it is not common to need to schedule or police the bandwidth used by individual application flows.</t> <t>By the above arguments, per-flow rate policing might not benecessarynecessary, and in trusted environments(e.g. private(e.g., private datacentres)centres), it is certainly unlikely to be needed. Therefore, because it is hard to avoid complexity and unintended side effects with per-flow rate policing, it needs to be separable from a basic AQM, as an option, under policy control. On this basis, the DualQ Coupled AQM provides low delay without prejudging the question of per-flow rate policing.</t> <t>Nonetheless, the interests of users or flows might conflict,e.g. ine.g., in case of accident or malice. Then per-flow rate control could be necessary. Ifflow-rateper-flow rate control is needed, it can be provided as a modular addition to a DualQ. And similarly, if protection against excessive queue delay is needed, a per-flow queue protection option can be added to a DualQ(e.g. <xref(e.g., <xref target="I-D.briscoe-docsis-q-protection" format="default"/>).</t> </section> <section anchor="dualq_Overload" numbered="true" toc="default"> <name>Handling Unresponsive Flows and Overload</name> <t>In the absence of any per-flow control, it is important that the basic DualQ Coupled AQM gives unresponsive flows no more throughput advantage than a single-queue AQM would, and that it at least handles overload situations. Overload means that incoming load significantly or persistently exceeds output capacity, but it is not intended to be a precise term -- significant and persistent are matters of degree.</t> <t>A trade-off needs to be made between complexity and the risk of either traffic class harming the other. In overloadedconditionsconditions, the higher priority L4S service will have to sacrifice some aspect of its performance. Depending on the degree of overload, alternative solutions may relax a different factor:e.g. throughput,for example, throughput, delay, or drop. These choices need to be made either by the developer or by operator policy, rather than by the IETF. Subsequent subsections discussaspects relating tohandlingofdifferent degrees of overload: </t> <ul spacing="normal"> <li> <t>Unresponsive flows (L and/or C) but not overloaded,i.e. thei.e., the sum of unresponsive load before adding any responsive traffic is belowcapacity;</t>capacity.</t> <ul empty="true" spacing="normal"> <li>This case is handled by the regular Coupled DualQ (<xref target="dualq_coupled" format="default"/>) but not discussed there. So below, <xref target="dualq_unresponsive_wo_overload" format="default"/> explains the designgoal,goal and how it is achieved inpractice;</li>practice.</li> </ul> </li> <li> <t>Unresponsive flows (L and/or C) causing persistent overload,i.e. thei.e., the sum of unresponsive load even before adding any responsive traffic persistently exceedscapacity;</t>capacity.</t> <ul empty="true" spacing="normal"> <li>This case is not covered by the regular Coupled DualQ mechanism (<xref target="dualq_coupled"format="default"/>)format="default"/>), but the lastparaparagraph in <xref target="dualq_unexpected" format="default"/> sets out a requirement to handle the case where ECN-capable traffic could starve non-ECN-capable traffic. <xref target="dualq_Overload_Saturation" format="default"/> below discusses the general options and gives specific examples.</li> </ul> </li> <li> <t>Short-term overload that lies between the 'not overloaded' and 'persistently overloaded'cases. </t>cases.</t> <ul empty="true" spacing="normal"> <li>For the period before overload is deemed persistent, <xref target="dualq_Overload_Starvation" format="default"/> discusses options for more immediate mechanisms at the scheduler timescale. These prevent short-term starvation of the C queue by making the priority of the L queue conditional, as required in <xref target="dualq_functional_reqs" format="default"/>.</li> </ul> </li> </ul> <section anchor="dualq_unresponsive_wo_overload" numbered="true" toc="default"> <name>Unresponsive Traffic without Overload</name> <t>When one or more L flows and/or C flows are unresponsive, but their total load is within the link capacity so that they do not saturate the coupled marking (below 100%), the goal of a DualQ AQM is to behave no worse than a single-queue AQM.</t> <t>Tests have shown that this is indeed the case with no additional mechanism beyond the regular Coupled DualQ of <xref target="dualq_coupled" format="default"/> (see the results of 'overload experiments' in <xreftarget="DCttH19"target="L4Seval22" format="default"/>). Perhapscounter-intuitively,counterintuitively, whether the unresponsive flow classifies itself into the L or the C queue, the DualQ system behaves as if it has subtracted from the overall link capacity. Then, the coupling shares out the remaining capacity between any competing responsive flows (in either queue). See also <xref target="dualq_Overload_Starvation" format="default"/>, which discusses scheduler-specific details.</t> </section> <section anchor="dualq_Overload_Starvation" numbered="true" toc="default"> <name>Avoiding Short-Term Classic Starvation: Sacrifice L4S Throughput or Delay?</name> <t>Priority of L4S is required to be conditional (see Sections <xref target="dualq_coupled_structure"format="default"/> &format="counter"/> and <xref target="dualq_functional_reqs"format="default"/>)format="counter"/>) to avoid short-term starvation of Classic. Otherwise, as explained in <xref target="dualq_coupled_structure" format="default"/>, even a lone responsive L4S flow could temporarily block a small finite set of C packets(e.g. an(e.g., an initial window or DNS request). The blockage would only be brief, but it could be longer for certain AQM implementations that can only increase the congestion signal coupled from the C queue when C packets are actually being dequeued. There is then the question of whether to sacrifice L4S throughput or L4S delay (or some other policy) to make the priority conditional:</t> <dlnewline="false"newline="true" spacing="normal"> <dt>Sacrifice L4S throughput: </dt> <dd anchor="dualq_Minimum_Service"> <t>By usingweighted round-robinWRR as the conditional priority scheduler, the L4S service can sacrifice some throughput during overload. This caneitherbe thought of as guaranteeing either a minimum throughput service for Classictraffic,traffic oras guaranteeinga maximum delay for a packet at the head of the Classic queue.</t><t>Cautionary<aside><t>Cautionary note: a WRR scheduler can only guarantee Classic throughput if Classic sources are sending enough to use it -- congestion signals can undermine scheduling because they determine how much responsive traffic of each class arrives for scheduling in the first place. This is why scheduling is only relied on to handle short-termstarvation;starvation, until congestion signals build up and the sources react. Even during long-term overload (discussed more fully in <xref target="dualq_Overload_Saturation" format="default"/>), it's pragmatic to discard packets from both queues, which again thins the traffic before it reaches the scheduler. This is because a scheduler cannot be relied on to handle long-term overload since the right scheduler weight cannot be known for everyscenario.</t>scenario.</t></aside> <t>The scheduling weight of the Classic queue should be small(e.g. 1/16).(e.g., 1/16). In most trafficscenariosscenarios, the scheduler will not interfere and it will not need to, because the coupling mechanism and theend-systemsend systems will determine the share of capacity across both queues as if it were a single pool. However, if L4S traffic is over-aggressive or unresponsive, the scheduler weight for Classic traffic will at least be large enough to ensure it does not starve in theshort-term.short term. </t> <t>Although WRR scheduling is only expected to address short-term overload, there are (somewhat rare) cases when WRR has an effect on capacity shares over longertime-scales.timescales. But its effect is minor, and it certainly does no harm. Specifically, in cases where the ratio of L4S to Classic flows(e.g. 19:1)(e.g., 19:1) is greater than the ratio of their scheduler weights(e.g. 15:1),(e.g., 15:1), the L4S flows will get less than an equal share of the capacity, but only slightly. For instance, with the example numbers given, each L4S flow will get (15/16)/19 = 4.9% when ideally each would get1/20=5%.1/20 = 5%. In the rather specific case of an unresponsive flow taking up just less than the capacity set aside for L4S(e.g. 14/16(e.g., 14/16 in the above example), using WRR could significantly reduce the capacity left for any responsive L4S flows.</t> <t>The scheduling weight of the Classic queue should not be too small, otherwise a C packet at the head of the queue could be excessively delayed by a continually busy L queue. Forinstanceinstance, if the Classic weight is 1/16, the maximum that a Classic packet at the head of the queue can be delayed by L traffic is the serialization delay of 15 MTU-sized packets.</t> </dd> <dt>Sacrifice L4SDelay:</dt>delay:</dt> <dd anchor="dualq_Delay_Overload"> <t>The operator could choose to control overload of the Classic queue by allowing some delay to 'leak' across to the L4S queue. The scheduler can be made to behave like a singleFirst-In First-Out (FIFO)FIFO queue with different service times by implementing a very simple conditional priority scheduler that could be called a "time-shifted FIFO" (TS-FIFO) (see the Modifier Earliest Deadline First (MEDF)scheduler <xrefscheduler <xref target="MEDF" format="default"/>). This scheduler adds tshift to the queue delay of the next L4S packet, before comparing it with the queue delay of the next Classic packet, then it selects the packet with the greater adjusted queue delay.</t> <t>Under regular conditions,this time-shifted FIFOthe TS-FIFO scheduler behaves just like a strict priority scheduler. But under moderate or highoverloadoverload, it prevents starvation of the Classic queue, because the time-shift (tshift) defines the maximum extra queuing delay of Classic packets relative to L4S. This would control milder overload of responsive traffic by introducing delay to defer invoking the overload mechanisms in <xref target="dualq_Overload_Saturation" format="default"/>, particularly when close to the maximum congestion signal.</t> </dd> </dl> <t>The example implementations in Appendices <xref target="dualq_Ex_algo_pi2"format="default"/>format="counter"/> and <xref target="dualq_Ex_algo"format="default"/>format="counter"/> could both be implemented with either policy.</t> </section> <section anchor="dualq_Overload_Saturation" numbered="true" toc="default"> <name>L4S ECN Saturation: Introduce Drop or Delay?</name> <t>This section concerns persistent overload caused by unresponsive L and/or C flows. To keep the throughput of both L4S and Classic flows roughly equal over the full load range, a different control strategy needs to be defined above the point where the L4S AQM persistently saturates to an ECN marking probability of100%100%, leaving no room to push back the load any harder. L4S ECN marking will saturate first (assuming the coupling factor k>1), even though saturation could be caused by the sum of unresponsive traffic in either or both queues exceeding the link capacity.</t> <t>The term 'unresponsive' includes cases where a flow becomes temporarily unresponsive, for instance, a real-time flow that takes a while to adapt its rate in response to congestion, or a standard Reno flow that is normally responsive, but above a certain congestion level it will not be able to reduce its congestion window below the allowed minimum of 2segments <xrefsegments <xref target="RFC5681" format="default"/>, effectively becoming unresponsive. (Note that L4S traffic ought to remain responsive below a window of 2segments (seesegments. See the L4Srequirements <xref target="I-D.ietf-tsvwg-ecn-l4s-id" format="default"/>).</t>requirements <xref target="RFC9331" format="default"/>.)</t> <t>Saturation raises the question of whether to relieve congestion by introducing some drop into the L4S queue or by allowing delay to grow in both queues (which could eventually lead to drop due to buffer exhaustion anyway):</t> <dlnewline="false"newline="true" spacing="normal"> <dt>Drop on Saturation:</dt> <dd>Persistent saturation can be defined by a maximum threshold for coupled L4S ECN marking (assuming k>1) before saturation starts to make the flow rates of the different traffic types diverge. Above that, the drop probability of Classic traffic is applied to all packets of all traffic types. Then experiments have shown thatqueueingqueuing delay can be kept at the target in any overload situation, including with unresponsive traffic, and no further measures are required (<xref target="dualq_overload_unresp_ect" format="default"/>).</dd> <dt>Delay on Saturation:</dt> <dd>When L4S marking saturates, instead of introducing L4S drop, the drop and marking probabilities of both queues could be capped. Beyond that, delay will grow either solely in the queue with unresponsive traffic (if WRR isused),used) or in both queues (iftime-shifted FIFOTS-FIFO is used). In either case, the higher delay ought to control temporary high congestion. If the overload is more persistent, eventually the combined DualQ will overflow and tail drop will control congestion.</dd> </dl> <t>The example implementation in <xref target="dualq_Ex_algo_pi2" format="default"/> solely applies the "drop on saturation" policy. The DOCSIS specification of a DualQ CoupledAQM <xrefAQM <xref target="DOCSIS3.1" format="default"/> also implements the 'drop on saturation' policy with a very shallow L buffer. However, the addition of DOCSIS per-flow Queue Protection <xref target="I-D.briscoe-docsis-q-protection" format="default"/> turns this into 'delay on saturation' by redirecting some packets of theflow(s)flow or flows that are most responsible for L queue overload into the C queue, which has a higher delay target. If overload continues, this again becomes 'drop on saturation' as the level of drop in the C queue rises to maintain the target delay of the C queue.</t> <section anchor="dualq_overload_unresp_ect" numbered="true" toc="default"> <name>Protecting against Overload by Unresponsive ECN-Capable Traffic</name> <t>Without a specific overload mechanism, unresponsive traffic would have a greater advantage if it were also ECN-capable. The advantage is undetectable at normal low levels of marking. However, it would become significant with the higher levels of marking typical during overload, when it could evade a significant degree of drop. This is an issue whether the ECN-capable traffic is L4S or Classic.</t> <t>This raises the question of whether and when to introduce drop of ECN-capable traffic, as required by both Section7<xref target="RFC3168" sectionFormat="bare" section="7"/> of the ECNspec <xrefspec <xref target="RFC3168" format="default"/> and Section4.2.1<xref target="RFC7567" sectionFormat="bare" section="4.2.1"/> of the AQMrecommendations <xrefrecommendations <xref target="RFC7567" format="default"/>.</t> <t>As an example, experiments with the DualPI2 AQM (<xref target="dualq_Ex_algo_pi2" format="default"/>) have shown that introducing 'drop on saturation' at 100% coupled L4S marking addresses this problem with unresponsiveECN as well as addressingECN, and it also addresses the saturation problem. At saturation, DualPI2 switches into overload mode, where thebaseBase AQM is driven by the max delay of bothqueuesqueues, and it introduces probabilistic drop to both queues equally. It leaves only a small range of congestion levels just below saturation where unresponsive traffic gains any advantage from using the ECN capability (relative to being unresponsive without ECN), and the advantage is hardly detectable (see <xref target="DualQ-Test" format="default"/> and sectionIV-EIV-G of <xreftarget="DCttH19" format="default"/>. Alsotarget="L4Seval22" format="default"/>). Also, overload with an unresponsive ECT(1) flow gets no more bandwidth advantage than with ECT(0).</t> </section> </section> </section> </section> </middle><!-- *****BACK MATTER ***** --><back> <displayreference target="I-D.briscoe-tsvwg-l4s-diffserv" to="L4S-DIFFSERV"/> <displayreference target="I-D.briscoe-docsis-q-protection" to="DOCSIS-Q-PROT"/> <displayreference target="I-D.cardwell-iccrg-bbr-congestion-control" to="BBR-CC"/> <displayreference target="I-D.briscoe-iccrg-prague-congestion-control" to="PRAGUE-CC"/> <displayreference target="I-D.mathis-iccrg-relentless-tcp" to="RELENTLESS"/> <references> <name>References</name> <references> <name>Normative References</name><reference anchor="RFC2119" target="https://www.rfc-editor.org/info/rfc2119" xml:base="https://bib.ietf.org/public/rfc/bibxml/reference.RFC.2119.xml"> <front> <title>Key words for use in RFCs to Indicate Requirement Levels</title> <author initials="S." surname="Bradner" fullname="S. Bradner"> <organization/> </author> <date year="1997" month="March"/> <abstract> <t>In many standards track documents several words are used to signify the requirements in the specification. These words are often capitalized. This document defines these words<xi:include href="https://xml2rfc.ietf.org/public/rfc/bibxml/reference.RFC.2119.xml"/> <xi:include href="https://xml2rfc.ietf.org/public/rfc/bibxml/reference.RFC.3168.xml"/> <xi:include href="https://xml2rfc.ietf.org/public/rfc/bibxml/reference.RFC.8311.xml"/> <!-- [I-D.ietf-tsvwg-ecn-l4s-id] companion doc 9331 - title matches asthey should be interpreted in IETF documents. This document specifies an Internet Best Current Practices for the Internet Community, and requests discussion and suggestions for improvements.</t> </abstract> </front> <seriesInfo name="BCP" value="14"/> <seriesInfo name="RFC" value="2119"/> <seriesInfo name="DOI" value="10.17487/RFC2119"/> </reference> <reference anchor="RFC3168" target="https://www.rfc-editor.org/info/rfc3168" xml:base="https://bib.ietf.org/public/rfc/bibxml/reference.RFC.3168.xml"> <front> <title>The AdditionofExplicit Congestion Notification (ECN) to IP</title> <author initials="K." surname="Ramakrishnan" fullname="K. Ramakrishnan"> <organization/> </author> <author initials="S." surname="Floyd" fullname="S. Floyd"> <organization/> </author> <author initials="D." surname="Black" fullname="D. Black"> <organization/> </author> <date year="2001" month="September"/> <abstract> <t>This memo specifies the incorporation of ECN (Explicit Congestion Notification) to TCP and IP, including ECN's use of two bits in the IP header. [STANDARDS-TRACK]</t> </abstract> </front> <seriesInfo name="RFC" value="3168"/> <seriesInfo name="DOI" value="10.17487/RFC3168"/> </reference>1/17/23--> <referenceanchor="RFC8311" target="https://www.rfc-editor.org/info/rfc8311" xml:base="https://bib.ietf.org/public/rfc/bibxml/reference.RFC.8311.xml">anchor='RFC9331' target='https://www.rfc-editor.org/info/rfc9331'> <front><title>Relaxing Restrictions on Explicit Congestion Notification (ECN) Experimentation</title> <author initials="D." surname="Black" fullname="D. Black"> <organization/> </author> <date year="2018" month="January"/> <abstract> <t>This memo updates RFC 3168, which specifies<title>The Explicit Congestion Notification (ECN)as an alternative to packet drops for indicating network congestion to endpoints. It relaxes restrictions in RFC 3168 that hinder experimentation towards benefits beyond just removal of loss. This memo summarizes the anticipated areas of experimentation and updates RFC 3168 to enable experimentation in these areas. An Experimental RFC in the IETF document stream is required to take advantage of any of these enabling updates. In addition, this memo makes related updates to the ECN specifications for RTP in RFC 6679 and for the Datagram Congestion Control Protocol (DCCP) in RFCs 4341, 4342, and 5622. This memo also records the conclusion of the ECN nonce experiment in RFC 3540 and provides the rationale for reclassification of RFC 3540 from Experimental to Historic; this reclassification enables new experimental use of the ECT(1) codepoint.</t> </abstract> </front> <seriesInfo name="RFC" value="8311"/> <seriesInfo name="DOI" value="10.17487/RFC8311"/> </reference> <reference anchor="I-D.ietf-tsvwg-ecn-l4s-id" target="https://datatracker.ietf.org/api/v1/doc/document/draft-ietf-tsvwg-ecn-l4s-id/" xml:base="https://bib.ietf.org/public/rfc/bibxml-ids/reference.I-D.ietf-tsvwg-ecn-l4s-id.xml"> <front> <title>Explicit Congestion Notification (ECN)Protocol forVeryLowQueuing DelayLatency, Low Loss, and Scalable Throughput (L4S)</title> <authorfullname="Koeninitials='K' surname='De Schepper' fullname='Koen DeSchepper"/> <author fullname="Bob Briscoe"/> <date day="8" month="August" year="2022"/> <abstract> <t>This specification defines the protocol to be used for a new network service called low latency, low loss and scalable throughput (L4S). L4S uses an Explicit Congestion Notification (ECN) scheme at the IP layer that is similar to the original (or 'Classic') ECN approach, except as specified within. L4S uses 'scalable' congestion control, which induces much more frequent control signals from the network and it responds to them with much more fine-grained adjustments, so that very low (typically sub-millisecond on average) and consistently low queuing delay becomes possible for L4S traffic without compromising link utilization. Thus even capacity-seeking (TCP-like) traffic can have high bandwidth and very low delay at the same time, even during periods of high traffic load.</t> <t>The L4S identifier defined in this document distinguishes L4S from 'Classic' (e.g. TCP-Reno-friendly) traffic. Then, network bottlenecks can be incrementally modified to distinguish and isolate existing traffic that still follows the Classic behaviour, to prevent it degrading the low queuing delay and low loss of L4S traffic. This experimental track specification defines the rules that L4S transports and network elements need to follow, with the intention that L4S flows neither harm each other's performance nor that of Classic traffic. It also suggests open questions to be investigated during experimentation. Examples of new active queue management (AQM) marking algorithms and examples of new transports (whether TCP- like or real-time) are specified separately.</t> </abstract> </front> <seriesInfo name="Internet-Draft" value="draft-ietf-tsvwg-ecn-l4s-id-28"/> </reference> </references> <references> <name>Informative References</name> <reference anchor="RFC0970" target="https://www.rfc-editor.org/info/rfc970" xml:base="https://bib.ietf.org/public/rfc/bibxml/reference.RFC.0970.xml"> <front> <title>On Packet Switches With Infinite Storage</title> <author initials="J." surname="Nagle" fullname="J. Nagle"> <organization/> </author> <date year="1985" month="December"/> <abstract> <t>The purpose of this RFC is to focus discussion on a particular problem in the ARPA-Internet and possible methods of solution. Most prior work on congestion in datagram systems focuses on buffer management. In this memo the case of a packet switch with infinite storage is considered. Such a packet switch can never run out of buffers. It can, however, still become congested. The meaning of congestion in an infinite-storage system is explored. An unexpected result is found that shows a datagram network with infinite storage, first-in-first-out queuing, at least two packet switches, and a finite packet lifetime will, under overload, drop all packets. By attacking the problem of congestion for the infinite-storage case, new solutions applicable to switches with finite storage may be found. No proposed solutions this document are intended as standards for the ARPA-Internet at this time.</t> </abstract> </front> <seriesInfo name="RFC" value="970"/> <seriesInfo name="DOI" value="10.17487/RFC0970"/> </reference> <reference anchor="RFC2309" target="https://www.rfc-editor.org/info/rfc2309" xml:base="https://bib.ietf.org/public/rfc/bibxml/reference.RFC.2309.xml"> <front> <title>Recommendations on Queue Management and Congestion Avoidance in the Internet</title> <author initials="B." surname="Braden" fullname="B. Braden"> <organization/> </author> <author initials="D." surname="Clark" fullname="D. Clark"> <organization/> </author> <author initials="J." surname="Crowcroft" fullname="J. Crowcroft"> <organization/> </author> <author initials="B." surname="Davie" fullname="B. Davie"> <organization/> </author> <author initials="S." surname="Deering" fullname="S. Deering"> <organization/> </author> <author initials="D." surname="Estrin" fullname="D. Estrin"> <organization/> </author> <author initials="S." surname="Floyd" fullname="S. Floyd"> <organization/> </author> <author initials="V." surname="Jacobson" fullname="V. Jacobson"> <organization/> </author> <author initials="G." surname="Minshall" fullname="G. Minshall"> <organization/> </author> <author initials="C." surname="Partridge" fullname="C. Partridge"> <organization/>Schepper'> <organization /> </author> <authorinitials="L." surname="Peterson" fullname="L. Peterson"> <organization/> </author> <author initials="K." surname="Ramakrishnan" fullname="K. Ramakrishnan"> <organization/> </author> <author initials="S." surname="Shenker" fullname="S. Shenker"> <organization/> </author> <author initials="J." surname="Wroclawski" fullname="J. Wroclawski"> <organization/> </author> <author initials="L." surname="Zhang" fullname="L. Zhang"> <organization/>initials='B' surname='Briscoe' fullname='Bob Briscoe' role='editor'> <organization /> </author> <dateyear="1998" month="April"/> <abstract> <t>This memo presents two recommendations to the Internet community concerning measures to improve and preserve Internet performance. It presents a strong recommendation for testing, standardization, and widespread deployment of active queue management in routers, to improve the performance of today's Internet. It also urges a concerted effort of research, measurement, and ultimate deployment of router mechanisms to protect the Internet from flows that are not sufficiently responsive to congestion notification. This memo provides information for the Internet community. It does not specify an Internet standard of any kind.</t> </abstract> </front> <seriesInfo name="RFC" value="2309"/> <seriesInfo name="DOI" value="10.17487/RFC2309"/> </reference> <reference anchor="RFC2914" target="https://www.rfc-editor.org/info/rfc2914" xml:base="https://bib.ietf.org/public/rfc/bibxml/reference.RFC.2914.xml"> <front> <title>Congestion Control Principles</title> <author initials="S." surname="Floyd" fullname="S. Floyd"> <organization/> </author> <date year="2000" month="September"/> <abstract> <t>The goal of this document is to explain the need for congestion control in the Internet, and to discuss what constitutes correct congestion control. This document specifies an Internet Best Current Practices for the Internet Community, and requests discussion and suggestions for improvements.</t> </abstract>month='January' year='2023'/> </front> <seriesInfoname="BCP" value="41"/> <seriesInfoname="RFC"value="2914"/>value="9331"/> <seriesInfo name="DOI"value="10.17487/RFC2914"/> </reference> <reference anchor="RFC3246" target="https://www.rfc-editor.org/info/rfc3246" xml:base="https://bib.ietf.org/public/rfc/bibxml/reference.RFC.3246.xml"> <front> <title>An Expedited Forwarding PHB (Per-Hop Behavior)</title> <author initials="B." surname="Davie" fullname="B. Davie"> <organization/> </author> <author initials="A." surname="Charny" fullname="A. Charny"> <organization/> </author> <author initials="J.C.R." surname="Bennet" fullname="J.C.R. Bennet"> <organization/> </author> <author initials="K." surname="Benson" fullname="K. Benson"> <organization/> </author> <author initials="J.Y." surname="Le Boudec" fullname="J.Y. Le Boudec"> <organization/> </author> <author initials="W." surname="Courtney" fullname="W. Courtney"> <organization/> </author> <author initials="S." surname="Davari" fullname="S. Davari"> <organization/> </author> <author initials="V." surname="Firoiu" fullname="V. Firoiu"> <organization/> </author> <author initials="D." surname="Stiliadis" fullname="D. Stiliadis"> <organization/> </author> <date year="2002" month="March"/> <abstract> <t>This document defines a PHB (per-hop behavior) called Expedited Forwarding (EF). The PHB is a basic building block in the Differentiated Services architecture. EF is intended to provide a building block for low delay, low jitter and low loss services by ensuring that the EF aggregate is served at a certain configured rate. This document obsoletes RFC 2598. [STANDARDS-TRACK]</t> </abstract> </front> <seriesInfo name="RFC" value="3246"/> <seriesInfo name="DOI" value="10.17487/RFC3246"/>value="10.17487/RFC9331"/> </reference> </references> <references> <name>Informative References</name> <xi:include href="https://xml2rfc.ietf.org/public/rfc/bibxml/reference.RFC.0970.xml"/> <referenceanchor="RFC3649" target="https://www.rfc-editor.org/info/rfc3649" xml:base="https://bib.ietf.org/public/rfc/bibxml/reference.RFC.3649.xml">anchor="RED" target="https://dl.acm.org/doi/10.1109/90.251892"> <front><title>HighSpeed TCP for Large Congestion Windows</title> <author initials="S." surname="Floyd" fullname="S. Floyd"> <organization/> </author> <date year="2003" month="December"/> <abstract> <t>The proposals in this document are experimental. While they may be deployed in the current Internet, they do not represent a consensus that this is the best method for high-speed congestion control. In particular, we note that alternative experimental proposals are likely to be forthcoming, and it is not well understood how the proposals in this document will interact with such alternative proposals. This document proposes HighSpeed TCP, a modification to TCP's congestion control mechanism for use with TCP connections with large congestion windows. The congestion control mechanisms of the current Standard TCP constrains the congestion windows that can be achieved by TCP in realistic environments. For example,<title>Random Early Detection Gateways fora Standard TCP connection with 1500-byte packets and a 100 ms round-trip time, achieving a steady-state throughput of 10 Gbps would require an average congestion window of 83,333 segments, and a packet drop rate of at most one congestion event every 5,000,000,000 packets (or equivalently, at most one congestion event every 1 2/3 hours). This is widely acknowledged as an unrealistic constraint. To address his limitation of TCP, this document proposes HighSpeed TCP, and solicits experimentation and feedback from the wider community.</t> </abstract> </front> <seriesInfo name="RFC" value="3649"/> <seriesInfo name="DOI" value="10.17487/RFC3649"/> </reference> <reference anchor="RFC5033" target="https://www.rfc-editor.org/info/rfc5033" xml:base="https://bib.ietf.org/public/rfc/bibxml/reference.RFC.5033.xml"> <front> <title>Specifying NewCongestionControl Algorithms</title> <author initials="S." surname="Floyd" fullname="S. Floyd"> <organization/> </author> <author initials="M." surname="Allman" fullname="M. Allman"> <organization/> </author> <date year="2007" month="August"/> <abstract> <t>The IETF's standard congestion control schemes have been widely shown to be inadequate for various environments (e.g., high-speed networks). Recent research has yielded many alternate congestion control schemes that significantly differ from the IETF's congestion control principles. Using these new congestion control schemes in the global Internet has possible ramifications to both the traffic using the new congestion control and to traffic using the currently standardized congestion control. Therefore, the IETF must proceed with caution when dealing with alternate congestion control proposals. The goal of this document is to provide guidance for considering alternate congestion control algorithms within the IETF. This document specifies an Internet Best Current Practices for the Internet Community, and requests discussion and suggestions for improvements.</t> </abstract> </front> <seriesInfo name="BCP" value="133"/> <seriesInfo name="RFC" value="5033"/> <seriesInfo name="DOI" value="10.17487/RFC5033"/> </reference> <reference anchor="RFC5348" target="https://www.rfc-editor.org/info/rfc5348" xml:base="https://bib.ietf.org/public/rfc/bibxml/reference.RFC.5348.xml"> <front> <title>TCP Friendly Rate Control (TFRC): Protocol Specification</title>Avoidance</title> <author fullname="Sally Floyd" initials="S."surname="Floyd" fullname="S. Floyd"> <organization/> </author> <author initials="M." surname="Handley" fullname="M. Handley"> <organization/> </author> <author initials="J." surname="Padhye" fullname="J. Padhye"> <organization/> </author> <author initials="J." surname="Widmer" fullname="J. Widmer"> <organization/> </author> <date year="2008" month="September"/> <abstract> <t>This document specifies TCP Friendly Rate Control (TFRC). TFRC is a congestion control mechanism for unicast flows operating in a best-effort Internet environment. It is reasonably fair when competing for bandwidth with TCP flows, but has a much lower variation of throughput over time compared with TCP, making it more suitable for applications such as streaming media where a relatively smooth sending rate is of importance.</t> <t>This document obsoletes RFC 3448 and updates RFC 4342. [STANDARDS-TRACK]</t> </abstract> </front> <seriesInfo name="RFC" value="5348"/> <seriesInfo name="DOI" value="10.17487/RFC5348"/> </reference> <reference anchor="RFC5681" target="https://www.rfc-editor.org/info/rfc5681" xml:base="https://bib.ietf.org/public/rfc/bibxml/reference.RFC.5681.xml"> <front> <title>TCP Congestion Control</title> <author initials="M." surname="Allman" fullname="M. Allman"> <organization/>surname="Floyd"> <organization>UC Berkeley</organization> </author> <author fullname="Van Jacobson" initials="V."surname="Paxson" fullname="V. Paxson"> <organization/> </author> <author initials="E." surname="Blanton" fullname="E. Blanton"> <organization/> </author> <date year="2009" month="September"/> <abstract> <t>This document defines TCP's four intertwined congestion control algorithms: slow start, congestion avoidance, fast retransmit, and fast recovery. In addition, the document specifies how TCP should begin transmission after a relatively long idle period, as well as discussing various acknowledgment generation methods. This document obsoletes RFC 2581. [STANDARDS-TRACK]</t> </abstract> </front> <seriesInfo name="RFC" value="5681"/> <seriesInfo name="DOI" value="10.17487/RFC5681"/> </reference> <reference anchor="RFC5706" target="https://www.rfc-editor.org/info/rfc5706" xml:base="https://bib.ietf.org/public/rfc/bibxml/reference.RFC.5706.xml"> <front> <title>Guidelines for Considering Operations and Management of New Protocols and Protocol Extensions</title> <author initials="D." surname="Harrington" fullname="D. Harrington"> <organization/> </author> <date year="2009" month="November"/> <abstract> <t>New protocols or protocol extensions are best designed with due consideration of the functionality needed to operate and manage the protocols. Retrofitting operations and management is sub-optimal. The purpose of this document is to provide guidance to authors and reviewers of documents that define new protocols or protocol extensions regarding aspects of operations and management that should be considered. This memo provides information for the Internet community.</t> </abstract> </front> <seriesInfo name="RFC" value="5706"/> <seriesInfo name="DOI" value="10.17487/RFC5706"/> </reference> <reference anchor="RFC7567" target="https://www.rfc-editor.org/info/rfc7567" xml:base="https://bib.ietf.org/public/rfc/bibxml/reference.RFC.7567.xml"> <front> <title>IETF Recommendations Regarding Active Queue Management</title> <author initials="F." surname="Baker" fullname="F. Baker" role="editor"> <organization/> </author> <author initials="G." surname="Fairhurst" fullname="G. Fairhurst" role="editor"> <organization/> </author> <date year="2015" month="July"/> <abstract> <t>This memo presents recommendations to the Internet community concerning measures to improve and preserve Internet performance. It presents a strong recommendation for testing, standardization, and widespread deployment of active queue management (AQM) in network devices to improve the performance of today's Internet. It also urges a concerted effort of research, measurement, and ultimate deployment of AQM mechanisms to protect the Internet from flows that are not sufficiently responsive to congestion notification.</t> <t>Based on 15 years of experience and new research, this document replaces the recommendations of RFC 2309.</t> </abstract> </front> <seriesInfo name="BCP" value="197"/> <seriesInfo name="RFC" value="7567"/> <seriesInfo name="DOI" value="10.17487/RFC7567"/> </reference> <reference anchor="RFC8033" target="https://www.rfc-editor.org/info/rfc8033" xml:base="https://bib.ietf.org/public/rfc/bibxml/reference.RFC.8033.xml"> <front> <title>Proportional Integral Controller Enhanced (PIE): A Lightweight Control Scheme to Address the Bufferbloat Problem</title> <author initials="R." surname="Pan" fullname="R. Pan"> <organization/> </author> <author initials="P." surname="Natarajan" fullname="P. Natarajan"> <organization/> </author> <author initials="F." surname="Baker" fullname="F. Baker"> <organization/> </author> <author initials="G." surname="White" fullname="G. White"> <organization/> </author> <date year="2017" month="February"/> <abstract> <t>Bufferbloat is a phenomenon in which excess buffers in the network cause high latency and latency variation. As more and more interactive applications (e.g., voice over IP, real-time video streaming, and financial transactions) run in the Internet, high latency and latency variation degrade application performance. There is a pressing need to design intelligent queue management schemes that can control latency and latency variation, and hence provide desirable quality of service to users.</t> <t>This document presents a lightweight active queue management design called "PIE" (Proportional Integral controller Enhanced) that can effectively control the average queuing latency to a target value. Simulation results, theoretical analysis, and Linux testbed results have shown that PIE can ensure low latency and achieve high link utilization under various congestion situations. The design does not require per-packet timestamps, so it incurs very little overhead and is simple enough to implement in both hardware and software.</t> </abstract> </front> <seriesInfo name="RFC" value="8033"/> <seriesInfo name="DOI" value="10.17487/RFC8033"/> </reference> <reference anchor="RFC8034" target="https://www.rfc-editor.org/info/rfc8034" xml:base="https://bib.ietf.org/public/rfc/bibxml/reference.RFC.8034.xml"> <front> <title>Active Queue Management (AQM) Based on Proportional Integral Controller Enhanced PIE) for Data-Over-Cable Service Interface Specifications (DOCSIS) Cable Modems</title> <author initials="G." surname="White" fullname="G. White"> <organization/> </author> <author initials="R." surname="Pan" fullname="R. Pan"> <organization/> </author> <date year="2017" month="February"/> <abstract> <t>Cable modems based on Data-Over-Cable Service Interface Specifications (DOCSIS) provide broadband Internet access to over one hundred million users worldwide. In some cases, the cable modem connection is the bottleneck (lowest speed) link between the customer and the Internet. As a result, the impact of buffering and bufferbloat in the cable modem can have a significant effect on user experience. The CableLabs DOCSIS 3.1 specification introduces requirements for cable modems to support an Active Queue Management (AQM) algorithm that is intended to alleviate the impact that buffering has on latency-sensitive traffic, while preserving bulk throughput performance. In addition, the CableLabs DOCSIS 3.0 specifications have also been amended to contain similar requirements. This document describes the requirements on AQM that apply to DOCSIS equipment, including a description of the "DOCSIS-PIE" algorithm that is required on DOCSIS 3.1 cable modems.</t> </abstract> </front> <seriesInfo name="RFC" value="8034"/> <seriesInfo name="DOI" value="10.17487/RFC8034"/> </reference> <reference anchor="RFC8174" target="https://www.rfc-editor.org/info/rfc8174" xml:base="https://bib.ietf.org/public/rfc/bibxml/reference.RFC.8174.xml"> <front> <title>Ambiguity of Uppercase vs Lowercase in RFC 2119 Key Words</title> <author initials="B." surname="Leiba" fullname="B. Leiba"> <organization/> </author> <date year="2017" month="May"/> <abstract> <t>RFC 2119 specifies common key words that may be used in protocol specifications. This document aims to reduce the ambiguity by clarifying that only UPPERCASE usage of the key words have the defined special meanings.</t> </abstract> </front> <seriesInfo name="BCP" value="14"/> <seriesInfo name="RFC" value="8174"/> <seriesInfo name="DOI" value="10.17487/RFC8174"/> </reference> <reference anchor="RFC8257" target="https://www.rfc-editor.org/info/rfc8257" xml:base="https://bib.ietf.org/public/rfc/bibxml/reference.RFC.8257.xml"> <front> <title>Data Center TCP (DCTCP): TCP Congestion Control for Data Centers</title> <author initials="S." surname="Bensley" fullname="S. Bensley"> <organization/> </author> <author initials="D." surname="Thaler" fullname="D. Thaler"> <organization/> </author> <author initials="P." surname="Balasubramanian" fullname="P. Balasubramanian"> <organization/> </author> <author initials="L." surname="Eggert" fullname="L. Eggert"> <organization/> </author> <author initials="G." surname="Judd" fullname="G. Judd"> <organization/> </author> <date year="2017" month="October"/> <abstract> <t>This Informational RFC describes Data Center TCP (DCTCP): a TCP congestion control scheme for data-center traffic. DCTCP extends the Explicit Congestion Notification (ECN) processing to estimate the fraction of bytes that encounter congestion rather than simply detecting that some congestion has occurred. DCTCP then scales the TCP congestion window based on this estimate. This method achieves high-burst tolerance, low latency, and high throughput with shallow- buffered switches. This memo also discusses deployment issues related to the coexistence of DCTCP and conventional TCP, discusses the lack of a negotiating mechanism between sender and receiver, and presents some possible mitigations. This memo documents DCTCP as currently implemented by several major operating systems. DCTCP, as described in this specification, is applicable to deployments in controlled environments like data centers, but it must not be deployed over the public Internet without additional measures.</t> </abstract> </front> <seriesInfo name="RFC" value="8257"/> <seriesInfo name="DOI" value="10.17487/RFC8257"/> </reference> <reference anchor="RFC8298" target="https://www.rfc-editor.org/info/rfc8298" xml:base="https://bib.ietf.org/public/rfc/bibxml/reference.RFC.8298.xml"> <front> <title>Self-Clocked Rate Adaptation for Multimedia</title> <author initials="I." surname="Johansson" fullname="I. Johansson"> <organization/> </author> <author initials="Z." surname="Sarker" fullname="Z. Sarker"> <organization/> </author> <date year="2017" month="December"/> <abstract> <t>This memo describes a rate adaptation algorithm for conversational media services such as interactive video. The solution conforms to the packet conservation principle and uses a hybrid loss-and-delay- based congestion control algorithm. The algorithm is evaluated over both simulated Internet bottleneck scenarios as well as in a Long Term Evolution (LTE) system simulator and is shown to achieve both low latency and high video throughput in these scenarios.</t> </abstract> </front> <seriesInfo name="RFC" value="8298"/> <seriesInfo name="DOI" value="10.17487/RFC8298"/> </reference> <reference anchor="RFC8290" target="https://www.rfc-editor.org/info/rfc8290" xml:base="https://bib.ietf.org/public/rfc/bibxml/reference.RFC.8290.xml"> <front> <title>The Flow Queue CoDel Packet Scheduler and Active Queue Management Algorithm</title> <author initials="T." surname="Hoeiland-Joergensen" fullname="T. Hoeiland-Joergensen"> <organization/> </author> <author initials="P." surname="McKenney" fullname="P. McKenney"> <organization/> </author> <author initials="D." surname="Taht" fullname="D. Taht"> <organization/> </author> <author initials="J." surname="Gettys" fullname="J. Gettys"> <organization/> </author> <author initials="E." surname="Dumazet" fullname="E. Dumazet"> <organization/> </author> <date year="2018" month="January"/> <abstract> <t>This memo presents the FQ-CoDel hybrid packet scheduler and Active Queue Management (AQM) algorithm, a powerful tool for fighting bufferbloat and reducing latency.</t> <t>FQ-CoDel mixes packets from multiple flows and reduces the impact of head-of-line blocking from bursty traffic. It provides isolation for low-rate traffic such as DNS, web, and videoconferencing traffic. It improves utilisation across the networking fabric, especially for bidirectional traffic, by keeping queue lengths short, and it can be implemented in a memory- and CPU-efficient fashion across a wide range of hardware.</t> </abstract> </front> <seriesInfo name="RFC" value="8290"/> <seriesInfo name="DOI" value="10.17487/RFC8290"/> </reference> <reference anchor="RFC8312" target="https://www.rfc-editor.org/info/rfc8312" xml:base="https://bib.ietf.org/public/rfc/bibxml/reference.RFC.8312.xml"> <front> <title>CUBIC for Fast Long-Distance Networks</title> <author initials="I." surname="Rhee" fullname="I. Rhee"> <organization/> </author> <author initials="L." surname="Xu" fullname="L. Xu"> <organization/> </author> <author initials="S." surname="Ha" fullname="S. Ha"> <organization/> </author> <author initials="A." surname="Zimmermann" fullname="A. Zimmermann"> <organization/> </author> <author initials="L." surname="Eggert" fullname="L. Eggert"> <organization/> </author> <author initials="R." surname="Scheffenegger" fullname="R. Scheffenegger"> <organization/>surname="Jacobson"> <organization>UC Berkeley</organization> </author> <dateyear="2018" month="February"/> <abstract> <t>CUBIC is an extension to the current TCP standards. It differs from the current TCP standards only in the congestion control algorithm on the sender side. In particular, it uses a cubic function instead of a linear window increase function of the current TCP standards to improve scalability and stability under fast and long-distance networks. CUBIC and its predecessor algorithm have been adopted as defaults by Linux and have been used for many years. This document provides a specification of CUBIC to enable third-party implementations and to solicit community feedback through experimentation on the performance of CUBIC.</t> </abstract>month="August" year="1993"/> </front> <seriesInfoname="RFC" value="8312"/> <seriesInfoname="DOI"value="10.17487/RFC8312"/> </reference> <reference anchor="RFC8404" target="https://www.rfc-editor.org/info/rfc8404" xml:base="https://bib.ietf.org/public/rfc/bibxml/reference.RFC.8404.xml"> <front> <title>Effects of Pervasive Encryptionvalue="10.1109/90.251892"/> <refcontent>IEEE/ACM Transactions onOperators</title> <author initials="K." surname="Moriarty" fullname="K. Moriarty" role="editor"> <organization/> </author> <author initials="A." surname="Morton" fullname="A. Morton" role="editor"> <organization/> </author> <date year="2018" month="July"/> <abstract> <t>Pervasive monitoring attacks on the privacy of Internet users are of serious concern to both user and operator communities. RFC 7258 discusses the critical need to protect users' privacy when developing IETF specifications and also recognizes that making networks unmanageable to mitigate pervasive monitoring is not an acceptable outcome: an appropriate balance is needed. This document discusses current security and network operations as well as management practices that may be impacted by the shift to increased use of encryption to help guide protocol development in support of manageable and secure networks.</t> </abstract> </front> <seriesInfo name="RFC" value="8404"/> <seriesInfo name="DOI" value="10.17487/RFC8404"/>Networking, Volume 1, Issue 4, pp. 397-413</refcontent> </reference> <xi:include href="https://xml2rfc.ietf.org/public/rfc/bibxml/reference.RFC.2914.xml"/> <xi:include href="https://xml2rfc.ietf.org/public/rfc/bibxml/reference.RFC.3246.xml"/> <xi:include href="https://xml2rfc.ietf.org/public/rfc/bibxml/reference.RFC.3649.xml"/> <xi:include href="https://xml2rfc.ietf.org/public/rfc/bibxml/reference.RFC.5033.xml"/> <xi:include href="https://xml2rfc.ietf.org/public/rfc/bibxml/reference.RFC.5348.xml"/> <xi:include href="https://xml2rfc.ietf.org/public/rfc/bibxml/reference.RFC.5681.xml"/> <xi:include href="https://xml2rfc.ietf.org/public/rfc/bibxml/reference.RFC.5706.xml"/> <xi:include href="https://xml2rfc.ietf.org/public/rfc/bibxml/reference.RFC.7567.xml"/> <xi:include href="https://xml2rfc.ietf.org/public/rfc/bibxml/reference.RFC.8033.xml"/> <xi:include href="https://xml2rfc.ietf.org/public/rfc/bibxml/reference.RFC.8034.xml"/> <xi:include href="https://xml2rfc.ietf.org/public/rfc/bibxml/reference.RFC.8174.xml"/> <xi:include href="https://xml2rfc.ietf.org/public/rfc/bibxml/reference.RFC.8257.xml"/> <xi:include href="https://xml2rfc.ietf.org/public/rfc/bibxml/reference.RFC.8298.xml"/> <xi:include href="https://xml2rfc.ietf.org/public/rfc/bibxml/reference.RFC.8290.xml"/> <xi:include href="https://xml2rfc.ietf.org/public/rfc/bibxml/reference.RFC.8312.xml"/> <xi:include href="https://xml2rfc.ietf.org/public/rfc/bibxml/reference.RFC.8404.xml"/> <xi:include href="https://xml2rfc.ietf.org/public/rfc/bibxml/reference.RFC.9000.xml"/> <reference anchor="ARED01"target="https://www.icir.org/floyd/red.html">target="https://www.icsi.berkeley.edu/icsi/node/2032"> <front> <title>Adaptive RED: An Algorithm for Increasing the Robustness of RED's Active Queue Management</title> <author fullname="Sally Floyd" initials="S." surname="Floyd"> <organization>ACIRI</organization> </author> <author fullname="Ramakrishna Gummadi" initials="R." surname="Gummadi"> <organization>ACIRI</organization> </author> <author fullname="S. Shenker" initials="S." surname="Shenker"> <organization>ACIRI</organization> </author> <date month="August" year="2001"/> </front><seriesInfo name="ACIRI<refcontent>ACIRI TechnicalReport" value=""/>Report 301</refcontent> </reference> <!-- [I-D.ietf-tsvwg-l4s-arch] companion doc 9330 - title matches as of 1/17/23--> <referenceanchor="I-D.ietf-tsvwg-l4s-arch" target="https://datatracker.ietf.org/api/v1/doc/document/draft-ietf-tsvwg-l4s-arch/" xml:base="https://bib.ietf.org/public/rfc/bibxml-ids/reference.I-D.ietf-tsvwg-l4s-arch.xml">anchor='RFC9330' target='https://www.rfc-editor.org/info/rfc9330'> <front> <title>Low Latency, Low Loss, and Scalable Throughput (L4S) Internet Service: Architecture</title> <authorfullname="Bob Briscoe"/>initials='B' surname='Briscoe' fullname='Bob Briscoe' role='editor'> </author> <authorfullname="Koeninitials='K' surname='De Schepper' fullname='Koen DeSchepper"/> <author fullname="Marcelo Bagnulo"/>Schepper'> </author> <authorfullname="Greg White"/> <date day="27" month="July" year="2022"/> <abstract> <t>This document describes the L4S architecture, which enables Internet applications to achieve Low queuing Latency, Low Loss, and Scalable throughput (L4S). The insight on which L4S is based is that the root cause of queuing delay is in the 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, to a new class of congestion controls that induce very little queuing, aided by explicit congestion signalling from the network. This new class of congestion controls can provide low latency for capacity-seeking flows, so applications can achieve both high bandwidth and low latency.</t> <t>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. These mechanisms aim to ensure that the latency and throughput performance using an L4S-compliant congestion controller is usually much better (and rarely worse) than performance would have been using a 'Classic' congestion controller, and that competing flows continuing to use 'Classic' controllers are typically not impacted by the presence of L4S. These characteristics are important to encourage adoption of L4S congestion control algorithms and L4S compliant network elements.</t> <t>The L4S architecture 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 as an experimental change to Explicit Congestion Notification (ECN).</t> </abstract> </front> <seriesInfo name="Internet-Draft" value="draft-ietf-tsvwg-l4s-arch-19"/> </reference> <reference anchor="I-D.briscoe-tsvwg-l4s-diffserv" target="https://datatracker.ietf.org/api/v1/doc/document/draft-briscoe-tsvwg-l4s-diffserv/" xml:base="https://bib.ietf.org/public/rfc/bibxml-ids/reference.I-D.briscoe-tsvwg-l4s-diffserv.xml"> <front> <title>Interactions between Low Latency, Low Loss, Scalable Throughput (L4S) and Differentiated Services</title>initials='M' surname='Bagnulo' fullname='Marcelo Bagnulo'> </author> <authorfullname="Bob Briscoe"/>initials='G' surname='White' fullname='Greg White'> </author> <dateday="2" month="July" year="2018"/> <abstract> <t>L4S and Diffserv offer somewhat overlapping services (low latency and low loss), but bandwidth allocation is out of scope for L4S. Therefore there is scope for the two approaches to complement each other, but also to conflict. This informational document explains how the two approaches interact, how they can be arranged to complement each other and in which cases one can stand alone without needing the other.</t> </abstract>year='2023' month='January'/> </front> <seriesInfoname="Internet-Draft" value="draft-briscoe-tsvwg-l4s-diffserv-02"/> </reference> <reference anchor="I-D.briscoe-docsis-q-protection" target="https://datatracker.ietf.org/api/v1/doc/document/draft-briscoe-docsis-q-protection/" xml:base="https://bib.ietf.org/public/rfc/bibxml-ids/reference.I-D.briscoe-docsis-q-protection.xml"> <front> <title>The DOCSIS(r) Queue Protection Algorithm to Preserve Low Latency</title> <author fullname="Bob Briscoe"/> <author fullname="Greg White"/> <date day="13" month="May" year="2022"/> <abstract> <t>This informational document explains the specification of the queue protection algorithm used in DOCSIS technology since version 3.1. A shared low latency queue relies on the non-queue-building behaviour of every traffic flow using it. However, some flows might not take such care, either accidentally or maliciously. If a queue is about to exceed a threshold level of delay, the queue protection algorithm can rapidly detect the flows most likely to be responsible. It can then prevent harm to other traffic in the low latency queue by ejecting selected packets (or all packets) of these flows. The document is designed for four types of audience: a) congestion control designers who need to understand how to keep on the 'good' side of the algorithm; b) implementers of the algorithm who want to understand it in more depth; c) designers of algorithms with similar goals, perhaps for non-DOCSIS scenarios; and d) researchers interested in evaluating the algorithm.</t> </abstract> </front>name="RFC" value="9330"/> <seriesInfoname="Internet-Draft" value="draft-briscoe-docsis-q-protection-06"/>name="DOI" value="10.17487/RFC9330"/> </reference><reference anchor="I-D.cardwell-iccrg-bbr-congestion-control" target="https://datatracker.ietf.org/api/v1/doc/document/draft-cardwell-iccrg-bbr-congestion-control/" xml:base="https://bib.ietf.org/public/rfc/bibxml-ids/reference.I-D.cardwell-iccrg-bbr-congestion-control.xml"> <front> <title>BBR Congestion Control</title> <author fullname="Neal Cardwell"/> <author fullname="Yuchung Cheng"/> <author fullname="Soheil Hassas Yeganeh"/> <author fullname="Ian Swett"/> <author fullname="Van Jacobson"/> <date day="7" month="March" year="2022"/> <abstract> <t>This document specifies the BBR congestion control algorithm. BBR ("Bottleneck Bandwidth and Round-trip propagation time") uses recent measurements of a transport connection's delivery rate, round-trip time, and packet loss rate to build an explicit model of the network path. BBR then uses this model to control both how fast it sends data and the maximum volume of data it allows in flight in the network at any time. Relative to loss-based congestion control algorithms such<!-- [I-D.briscoe-tsvwg-l4s-diffserv] IESG state Expired asReno [RFC5681] or CUBIC [RFC8312], BBR offers substantially higher throughput for bottlenecks with shallow buffers or random losses, and substantially lower queueing delays for bottlenecks with deep buffers (avoiding "bufferbloat"). BBR can be implemented in any transport protocol that supports packet-delivery acknowledgment. Thus far, open source implementations are available for TCP [RFC793] and QUIC [RFC9000]. This document specifies version 2ofthe BBR algorithm, also sometimes referred to1/17/23 --> <xi:include href="https://datatracker.ietf.org/doc/bibxml3/reference.I-D.briscoe-tsvwg-l4s-diffserv.xml"/> <!-- [I-D.briscoe-docsis-q-protection] in MISSREF state asBBRv2 or bbr2.</t> </abstract> </front> <seriesInfo name="Internet-Draft" value="draft-cardwell-iccrg-bbr-congestion-control-02"/> </reference> <reference anchor="I-D.briscoe-iccrg-prague-congestion-control" target="https://datatracker.ietf.org/api/v1/doc/document/draft-briscoe-iccrg-prague-congestion-control/" xml:base="https://bib.ietf.org/public/rfc/bibxml-ids/reference.I-D.briscoe-iccrg-prague-congestion-control.xml"> <front> <title>Prague Congestion Control</title> <author fullname="Koen De Schepper"/> <author fullname="Olivier Tilmans"/> <author fullname="Bob Briscoe"/> <date day="11" month="July" year="2022"/> <abstract> <t>This specification defines the Prague congestion control scheme, which is derived from DCTCP and adapted for Internet traffic by implementing the Prague L4S requirements. Over paths with L4S support at the bottleneck, it adapts the DCTCP mechanisms to achieve consistently low latency and full throughput. It is defined independently of any particular transport protocol or operating system, but notes are added that highlight issues specific to certain transports and OSs. It is mainly based on the current default optionsofthe reference Linux implementation of TCP Prague, but it includes experience from other implementations where available. It separately describes non-default and optional parts, as well as future plans.</t> <t>The implementation does not satisfy all the Prague requirements (yet) and the IETF might decide that certain requirements need to be relaxed1/17/23 --> <xi:include href="https://datatracker.ietf.org/doc/bibxml3/reference.I-D.briscoe-docsis-q-protection.xml"/> <!-- [I-D.cardwell-iccrg-bbr-congestion-control] IESG state Expired asan outcome of the process of trying to satisfy them all. In two cases, research code is replaced by placeholders until full evaluation is complete.</t> </abstract> </front> <seriesInfo name="Internet-Draft" value="draft-briscoe-iccrg-prague-congestion-control-01"/> </reference> <reference anchor="I-D.mathis-iccrg-relentless-tcp" target="https://www.ietf.org/archive/id/draft-mathis-iccrg-relentless-tcp-00.txt" xml:base="https://bib.ietf.org/public/rfc/bibxml-ids/reference.I-D.mathis-iccrg-relentless-tcp.xml"> <front> <title>Relentless Congestion Control</title> <author fullname="Matt Mathis"/> <date day="4" month="March" year="2009"/> <abstract> <t>Relentless congestion control is a simple modification that can be applied to almost any AIMD style congestion control: insteadofapplying a multiplicative reduction to cwnd after a loss, cwnd is reduced by the number of lost segments. It can be modeled1/17/23 --> <xi:include href="https://datatracker.ietf.org/doc/bibxml3/reference.I-D.cardwell-iccrg-bbr-congestion-control.xml"/> <!-- [I-D.briscoe-iccrg-prague-congestion-control] IESG state Expired asa strict implementation of van Jacobson's Packet Conservation Principle. During recovery, new segments are injected into the network in exact accordance with the segments that are reported to have been delivered to the receiver by the returning ACKs. This algorithm offers a valuable new congestion control property: the TCP portion of the control loop has exactly unity gain, which should make it easier to implement simple controllers in network devices to accurately control queue sizes across a huge range of scales. Relentless Congestion Control conforms to neither the details nor the philosophyofcurrent congestion control standards. These standards are based on the idea that the Internet can attain sufficient fairness by having relatively simple network devices send uniform congestion signals to all flows, and mandating that all protocols have equivalent responses to these congestion signals. To function appropriately in a shared environment, Relentless Congestion Control requires that the network allocates capacity through some technique such1/17/23 --> <xi:include href="https://datatracker.ietf.org/doc/bibxml3/reference.I-D.briscoe-iccrg-prague-congestion-control.xml"/> <!-- [I-D.mathis-iccrg-relentless-tcp] IESG state Expired asFair Queuing, Approximate Fair Dropping, etc. The salient features of these algorithms are that they segregate the traffic into distinct flows, and send different congestion signals to each flow. This alternative congestion control paradigm is described in a separate document, also under consideration by the ICCRG. The goalofthe document is to illustrate some new protocol features and properties might be possible if we relax the "TCP-friendly" mandate. A secondary goal of Relentless TCP is to make a distinction between the bottlenecks that belong to protocol itself, vs standard congestion control and the "TCP-friendly" paradigm.</t> </abstract> </front> <seriesInfo name="Internet-Draft" value="draft-mathis-iccrg-relentless-tcp-00"/> </reference> <!--{ToDo: DCttH ref will need to be updated, once stable}--> <reference anchor="DCttH19" target="https://bobbriscoe.net/pubs.html#DCttH_TR"> <front> <title>`Data Centre to the Home': Ultra-Low Latency for All</title> <author fullname="Koen De Schepper" initials="K." surname="De Schepper"> <organization>Nokia Bell Labs</organization> </author> <author fullname="Olga Bondarenko" initials="O." surname="Bondarenko"> <organization>Simula Research Lab</organization> </author> <author fullname="Olivier" initials="O." surname="Tilmans"> <organization>Nokia Bell Labs</organization> </author> <author fullname="Bob Briscoe" initials="B." surname="Briscoe"> <organization>Independent (bobbriscoe.net)</organization> </author> <date month="July" year="2019"/> </front> <seriesInfo name="Updated RITE project Technical Report" value=""/> <format target="https://bobbriscoe.net/projects/latency/dctth_journal_draft20190726.pdf" type="PDF"/> </reference>1/17/23 --> <xi:include href="https://datatracker.ietf.org/doc/bibxml3/reference.I-D.mathis-iccrg-relentless-tcp.xml"/> <reference anchor="PI2"target="https://riteproject.files.wordpress.com/2015/10/pi2_conext.pdf">target="https://dl.acm.org/doi/10.1145/2999572.2999578"> <front> <title>PI2: A Linearized AQM for both Classic and Scalable TCP</title> <author fullname="Koen De Schepper" initials="K." surname="De Schepper"> <organization>Nokia Bell Labs</organization> </author> <author fullname="Olga Bondarenko" initials="O." surname="Bondarenko"> <organization>Simula Research Lab</organization> </author> <author fullname="Bob Briscoe" initials="B." surname="Briscoe"> <organization>BT</organization> </author> <author fullname="Ing-jyh Tsang" initials="I." surname="Tsang"> <organization>Nokia Bell Labs</organization> </author> <date month="December" year="2016"/> </front> <seriesInfoname="ACM CoNEXT'16" value=""/>name="DOI" value="10.1145/2999572.2999578"/> <refcontent>ACM CoNEXT'16</refcontent> </reference> <reference anchor="L4Sdemo16"target="https//dl.acm.org/citation.cfm?doid=2910017.2910633 (videos of demos: https://riteproject.eu/dctth/#1511dispatchwg )">target="https://dl.acm.org/citation.cfm?doid=2910017.2910633"> <front> <title>Ultra-Low Delay for All: Live Experience, Live Analysis</title> <author fullname="Olga Bondarenko" initials="O." surname="Bondarenko"> <organization>Simula Research Lab</organization> </author> <author fullname="Koen De Schepper" initials="K." surname="De Schepper"> <organization>Bell Labs</organization> </author> <author fullname="Ing-jyh Tsang" initials="I." surname="Tsang"> <organization>Bell Labs</organization> </author> <author fullname="Bob Briscoe" initials="B." surname="Briscoe"> <organization>BT</organization> </author> <author fullname="Andreas Petlund" initials="A." surname="Petlund"> </author> <author fullname="Carsten Griwodz" initials="C." surname="Griwodz"> </author> <date month="May" year="2016"/> </front> <seriesInfoname="Proc. MMSYS'16" value="pp33:1--33:4"/>name="DOI" value="10.1145/2910017.2910633"/> <refcontent>Proceedings of the 7th International Conference on Multimedia Systems, Article No. 33, pp. 1-4</refcontent> <format target="https://dl.acm.org/citation.cfm?doid=2910017.2910633" type="PDF"/> </reference> <reference anchor="L4Seval22" target="https://arxiv.org/abs/2209.01078"> <front> <title>Dual Queue Coupled AQM: Deployable Very Low Queuing Delay for All</title> <author fullname="Koen De Schepper" initials="K." surname="De Schepper"> <organization>Nokia Bell Labs</organization> </author> <author fullname="Olga Albisser" initials="O." surname="Albisser"> <organization>Simula Research Lab</organization> </author> <author fullname="Olivier Tilmans" initials="O." surname="Tilmans"> <organization>Nokia Bell Labs</organization> </author> <author fullname="Bob Briscoe" initials="B." surname="Briscoe"> <organization>Independent (bobbriscoe.net)</organization> </author> <date month="September" year="2022"/> </front> <seriesInfo name="DOI" value="10.48550/arXiv.2209.01078"/> <refcontent>Preprint submitted to IEEE/ACM Transactions on Networking</refcontent> </reference> <reference anchor="Dukkipati06" target="https://dl.acm.org/doi/10.1145/1111322.1111336"> <front> <title>Why Flow-Completion Time is the Right Metric for Congestion Control</title> <author fullname="Nandita Dukkipati" initials="N." surname="Dukkipati"> <organization>StanfordUni</organization>University</organization> </author> <author fullname="Nick McKeown" initials="N." surname="McKeown"> <organization>StanfordUni</organization>University</organization> </author> <date month="January" year="2006"/> </front> <seriesInfoname="ACM CCR" value="36(1):59--62"/> <format target="http://yuba.stanford.edu/rcp/flowCompTime-dukkipati.pdf" type="PDF"/> </reference> <!-- <reference anchor="DCTCP_Pitfalls" target="http://blogs.usenix.org/conference/nsdi15/technical-sessions/presentation/judd"> <front> <title>Attaining the Promise and Avoiding the Pitfalls of TCP in the Datacenter</title> <author fullname="Glenn Judd" initials="G." surname="Judd"> <organization>Morgan Stanley</organization> </author> <date month="May" year="2015"/> </front> <seriesInfo name="12th USENIX Symposium on Networked Systems Design and Implementation (NSDI 15)" value="145-157"/> <format target="http://blogs.usenix.org/conference/nsdi15/technical-sessions/presentation/judd" type="PDF"/>name="DOI" value="10.1145/1111322.1111336"/> <refcontent>ACM SIGCOMM Computer Communication Review, Vol. 36, Issue 1, pp. 59-62</refcontent> </reference>--><reference anchor="CRED_Insights" target="https://arxiv.org/abs/1904.07339"> <front> <title>Insights from Curvy RED (Random Early Detection)</title> <author fullname="Bob Briscoe" initials="B." surname="Briscoe"> <organization>BT</organization> </author> <author fullname="Koen De Schepper" initials="K." surname="De Schepper"> <organization>BT</organization> </author> <dateday="" month="July"month="August" year="2015"/> </front> <seriesInfoname="BTname="DOI" value="10.48550/arXiv.1904.07339"/> <refcontent>BT TechnicalReport" value="TR-TUB8-2015-003 arXiv:1904.07339 [cs.NI]"/>Report, TR-TUB8-2015-003</refcontent> <format target="https://arxiv.org/pdf/1904.07339" type="PDF"/> </reference> <reference anchor="CoDel" target="https://queue.acm.org/issuedetail.cfm?issue=2208917"> <front> <title>Controlling Queue Delay</title> <author fullname="Kathleen Nichols" initials="K." surname="Nichols"> <organization>PARC</organization> </author> <author fullname="Van Jacobson" initials="V." surname="Jacobson"> <organization>Pollere Inc</organization> </author> <date month="May" year="2012"/> </front><seriesInfo name="ACM Queue" value="10(5)"/><refcontent>ACM Queue, Vol. 10, Issue 5</refcontent> </reference> <referenceanchor="MEDF" target="https://infocom2003.ieee-infocom.org/papers/27_04.PDF">anchor="MEDF"> <front> <title>MEDF -a simple scheduling algorithmA Simple Scheduling Algorithm fortwo real-time transport service classesTwo Real-Time Transport Service Classes withapplicationApplication in the UTRAN</title> <author fullname="Michael Menth " initials="M." surname="Menth"> <organization>University of Wuerzburg</organization> </author> <author fullname="Matthias Schmid " initials="M." surname="Schmid"> <organization>Infosim AG</organization> </author> <author fullname="Herbert Heiss" initials="H." surname="Heiss"> <organization>Siemens</organization> </author> <author fullname="Thomas Reim" initials="T." surname="Reim"> <organization>Siemens</organization> </author> <date month="March" year="2003"/> </front> <seriesInfoname="Proc.name="DOI" value="10.1109/INFCOM.2003.1208948"/> <refcontent>Proc. IEEE Conference on Computer Communications(INFOCOM'03)" value="Vol.2 pp.1116-1122"/>(INFOCOM'03), Vol. 2, pp. 1116-1122</refcontent> </reference> <referenceanchor="DualQ-Test" target="https://www.duo.uio.no/bitstream/handle/10852/57424/thesis-henrste.pdf?sequence=1">anchor="DualQ-Test"> <front> <title>Destruction Testing: Ultra-Low Delay using Dual Queue Coupled Active Queue Management</title> <author fullname="Henrik Steen" initials="H." surname="Steen"><organization>Uni<organization>University of Oslo</organization> </author> <date month="May" year="2017"/> </front><seriesInfo name="Master's<refcontent>Master's Thesis,DeptDepartment of Informatics,Uni Oslo" value=""/>University of Oslo</refcontent> </reference> <reference anchor="SigQ-Dyn" target="https://arxiv.org/abs/1904.07044"> <front> <title>Rapid Signalling of Queue Dynamics</title> <author fullname="Bob Briscoe" initials="B." surname="Briscoe"> <organization/> </author> <date month="September" year="2017"/> </front> <seriesInfoname="Technical Report" value="TR-BB-2017-001 arXiv:1904.07044 [cs.NI]"/>name="DOI" value="10.48550/arXiv.1904.07044"/> <refcontent>Technical Report, TR-BB-2017-001</refcontent> <format target="https://arxiv.org/pdf/1904.07044" type="PDF"/> </reference> <reference anchor="Alizadeh-stability" target="https://dl.acm.org/citation.cfm?id=1993753"> <front> <title>Analysis of DCTCP: Stability, Convergence, and Fairness</title> <author fullname="Mohamed Alizadeh" initials="M." surname="Alizadeh"/> <author fullname="Adel Javanmard" initials="A." surname="Javanmard"/> <author fullname="Balaji Prabhakar" initials="B." surname="Prabhakar"/> <date month="June" year="2011"/> </front> <seriesInfoname="ACMname="DOI" value="10.1145/1993744.1993753"/> <refcontent>SIGMETRICS '11: Proceedings of the ACM SIGMETRICS2011" value=""/>Joint International Conference on Measurement and Modeling of Computer Systems, pp. 73-84</refcontent> <format target="https://people.csail.mit.edu/alizadeh/papers/dctcp_analysis-sigmetrics11.pdf" type="PDF"/> </reference> <reference anchor="PragueLinux" target="https://www.netdevconf.org/0x13/session.html?talk-tcp-prague-l4s"> <front> <title>Implementing the`TCP'TCP Prague' Requirements forLow Latency Low Loss Scalable Throughput (L4S)</title>L4S</title> <author fullname="Bob Briscoe" initials="B." surname="Briscoe"> <organization>Independent</organization> </author> <author fullname="Koen De Schepper" initials="K." surname="De Schepper"> <organization>Nokia Bell Labs</organization> </author> <author fullname="Olga Albisser" initials="O." surname="Albisser"> <organization>Simula Research Lab</organization> </author> <author fullname="Joakim Misund" initials="J." surname="Misund"> <organization>Simula Research Lab</organization> </author> <author fullname="Olivier Tilmans" initials="O." surname="Tilmans"> <organization>Nokia Bell Labs</organization> </author> <author fullname="MirjaKühlewind"Kuehlewind" initials="M."surname="Kühlewind">surname="Kuehlewind"> <organization>ETH Zurich</organization> </author> <author fullname="Asad Sajjad Ahmed"initials="A.S."initials="A." surname="Ahmed"> <organization>Simula Research Lab</organization> </author> <date month="March" year="2019"/> </front><seriesInfo name="Proc.<refcontent>Proceedings of Linux Netdev0x13" value=""/> <format target="https://www.files.netdevconf.org/f/4d6939d5f1fb404fafd1/?dl=1" type="PDF"/>0x13</refcontent> </reference> <reference anchor="DualPI2Linux" target="https://www.netdevconf.org/0x13/session.html?talk-DUALPI2-AQM"> <front> <title>DUALPI2 - Low Latency, Low Loss and Scalable (L4S) AQM</title> <author fullname="Olga Albisser" initials="O." surname="Albisser"> <organization>Simula Research Lab</organization> </author> <author fullname="Koen De Schepper" initials="K." surname="De Schepper"> <organization>Nokia Bell Labs</organization> </author> <author fullname="Bob Briscoe" initials="B." surname="Briscoe"> <organization>Independent</organization> </author> <author fullname="Olivier Tilmans" initials="O." surname="Tilmans"> <organization>Nokia Bell Labs</organization> </author> <author fullname="Henrik Steen" initials="H." surname="Steen"> <organization>Simula Research Lab</organization> </author> <date month="March" year="2019"/> </front> <seriesInfoname="Proc.name="Proceedings of Linux Netdev 0x13" value=""/> <format target="https://www.files.netdevconf.org/f/febbe8c6a05b4ceab641/?dl=1" type="PDF"/> </reference> <reference anchor="DOCSIS3.1"target="https://specification-search.cablelabs.com/CM-SP-MULPIv3.1">target="https://specification-search.cablelabs.com/CM-SP-MULPIv3"> <front><title>MAC<title>DOCSIS 3.1 MAC and Upper Layer Protocols Interface(MULPI) Specification, CM-SP-MULPIv3.1</title>Specification</title> <author fullname="" surname=""> <organization>CableLabs</organization> </author> <dateday="21"month="January" year="2019"/> </front><seriesInfo name="Data-Over-Cable<refcontent>CM-SP-MULPIv3.1, Data-Over-Cable Service Interface SpecificationsDOCSIS® 3.1" value="Version i17DOCSIS 3.1 Version I17 orlater"/>later</refcontent> </reference> <reference anchor="LLD" target="https://cablela.bs/low-latency-docsis-technology-overview-february-2019"> <front> <title>Low Latency DOCSIS: Technology Overview</title> <author fullname="Greg White" initials="G." surname="White"> <organization>CableLabs</organization> </author> <author fullname="Karthik Sundaresan" initials="K." surname="Sundaresan"> <organization>CableLabs</organization> </author> <author fullname="Bob Briscoe" initials="B." surname="Briscoe"> <organization>CableLabs</organization> </author> <date day="" month="February" year="2019"/> </front><seriesInfo name="CableLabs<refcontent>CableLabs WhitePaper" value=""/>Paper</refcontent> </reference> <reference anchor="AQMmetrics" target="https://www.cs.purdue.edu/homes/fahmy/papers/ldc.pdf"> <front> <title>A Comparison of Load-based andQueue- basedQueue-based Active Queue Management Algorithms</title> <author fullname="Minseok Kwon" initials="M." surname="Kwon"> <organization>PurdueUni</organization>University</organization> </author> <author fullname="Sonia Fahmy" initials="S." surname="Fahmy"> <organization>PurdueUni</organization>University</organization> </author> <date year="2002"/> </front> <seriesInfoname="Proc.name="DOI" value="10.1117/12.473021"/> <refcontent>Proc. Int'l Soc. for Optical Engineering(SPIE)" value="4866:35--46 DOI: 10.1117/12.473021"/>(SPIE), Vol. 4866, pp. 35-46</refcontent> </reference> <reference anchor="CCcensus19" target="https://doi.org/10.1145/3366693"> <front> <title>The Great Internet TCP Congestion Control Census</title> <author fullname="Ayush Mishra" initials="A." surname="Mishra"> <organization/> </author> <author fullname="Xiangpeng Sun" initials="X." surname="Sun"><organization/> <address> <postal> <street/> <city/> <region/> <code/> <country/> </postal> <phone/> <email/> <uri/> </address></author> <author fullname="Atishya Jain" initials="A." surname="Jain"><organization/> <address> <postal> <street/> <city/> <region/> <code/> <country/> </postal> <phone/> <email/> <uri/> </address></author> <author fullname="Sameer Pande" initials="S." surname="Pande"><organization/> <address> <postal> <street/> <city/> <region/> <code/> <country/> </postal> <phone/> <email/> <uri/> </address></author> <author fullname="Raj Joshi" initials="R." surname="Joshi"><organization/> <address> <postal> <street/> <city/> <region/> <code/> <country/> </postal> <phone/> <email/> <uri/> </address></author> <author fullname="Ben Leong" initials="B." surname="Leong"><organization/> <address> <postal> <street/> <city/> <region/> <code/> <country/> </postal> <phone/> <email/> <uri/> </address></author> <date month="December" year="2019"/> </front> <seriesInfoname="Proc.name="DOI" value="10.1145/3366693"/> <refcontent>Proceedings of the ACM on Measurement and Analysis of ComputingSystems" value="3(3)"/>Systems, Vol. 3, Issue 3, Article No. 45, pp. 1-24</refcontent> <format target="https://dl.acm.org/doi/10.1145/3366693" type="PDF"/> </reference> <reference anchor="PI2param" target="https://arxiv.org/abs/2107.01003"> <front> <title>PI2 Parameters</title> <author fullname="Bob Briscoe" initials="B." surname="Briscoe"> <organization/> </author> <date month="July" year="2021"/> </front> <seriesInfoname="Technical Report" value="TR-BB-2021-001name="DOI" value="10.48550/arXiv.2107.01003"/> <refcontent>Technical Report, TR-BB-2021-001, arXiv:2107.01003[cs.NI]"/>[cs.NI]</refcontent> <format target="https://arxiv.org/pdf/2107.01003" type="PDF"/> </reference> <reference anchor="Labovitz10" target="https://doi.org/10.1145/1851275.1851194"> <front> <title>Internet Inter-Domain Traffic</title> <author fullname="Craig Labovitz" initials="C." surname="Labovitz"> <organization>Arbor Networks</organization> </author> <author fullname="Scott Iekel-Johnson" initials="S." surname="Iekel-Johnson"> <organization>Arbor Networks</organization> </author> <author fullname="Danny McPherson" initials="D." surname="McPherson"> <organization>Arbor Networks</organization> </author> <author fullname="Jon Oberheide" initials="J." surname="Oberheide"> <organization>Uni Michigan</organization> </author> <author fullname="Farnam Jahanian" initials="F." surname="Jahanian"> <organization>Uni Michigan</organization> </author> <date month="August" year="2010"/> </front> <seriesInfoname="Proc ACM SIGCOMM; ACM CCR" value="40(4):75--86"/>name="DOI" value="10.1145/1851275.1851194"/> <refcontent>ACM SIGCOMM Computer Communication Review, Vol. 40, Issue 4, pp. 75-86</refcontent> <format target="https://dl.acm.org/doi/pdf/10.1145/1851182.1851194" type="PDF"/> </reference> <referenceanchor="SCReAM" target="https://github.com/EricssonResearch/scream/blob/master/README.md">anchor="SCReAM-L4S" target="https://github.com/EricssonResearch/scream"> <front> <title>SCReAM</title><author fullname="Ingemar Johansson" initials="I" surname="Johansson"> <organization/> </author> <date/><author/> <date month="June" year="2022"/> </front><seriesInfo name="GitHub repository;" value=""/> <format target="https://github.com/google/bbr/blob/v2alpha/README.md" type="Source code"/><refcontent>commit fda6c53</refcontent> </reference> <reference anchor="L4S_5G" target="https://www.ericsson.com/en/reports-and-papers/white-papers/enabling-time-critical-applications-over-5g-with-rate-adaptation"> <front> <title>Enabling time-critical applications over 5G with rate adaptation</title> <author fullname="Per Willars" initials="P." surname="Willars"/> <author fullname="Emma Wittenmark" initials="E." surname="Wittenmark"/> <author fullname="Henrik Ronkainen" initials="H." surname="Ronkainen"/> <author fullname="Christer Östberg" initials="C." surname="Östberg"/> <author fullname="Ingemar Johansson" initials="I." surname="Johansson"/> <author fullname="Johan Strand" initials="J." surname="Strand"/> <author fullname="Petr Lédl" initials="P." surname="Lédl"/> <author fullname="Dominik Schnieders" initials="D." surname="Schnieders"/> <date month="May" year="2021"/> </front><seriesInfo name="Ericsson<refcontent>Ericsson - Deutsche Telekom WhitePaper" value="BNEW-21:025455 Uen"/>Paper, BNEW-21:025455</refcontent> <format target="https://www.ericsson.com/49bc82/assets/local/reports-papers/white-papers/26052021-enabling-time-critical-applications-over-5g-with-rate-adaptation-whitepaper.pdf" type="PDF"/> </reference> <reference anchor="BBRv2"target="https://github.com/google/bbr/blob/v2alpha/README.md">target="https://github.com/google/bbr"> <front><title>BRTCP<title>TCP BBR v2 Alpha/Preview Release</title><author fullname="Neal Cardwell" initials="N" surname="Cardwell"> <organization/> </author> <date/><author/> <date month="June" year="2022"/> </front><seriesInfo name="GitHub repository;" value="Linux congestion control module"/><refcontent>commit 17700ca</refcontent> </reference> <reference anchor="Heist21"target="https://github.com/heistp/l4s-tests/#underutilization-with-bursty-traffic">target="https://github.com/heistp/l4s-tests"> <front> <title>L4S Tests</title><author fullname="Pete Heist" initials="P." surname="Heist"> <organization/> </author> <author fullname="Jonathan Morton" initials="J." surname="Morton"> <organization/> </author><author/> <date month="August" year="2021"/> </front><seriesInfo name="GitHub" value="README"/><refcontent>commit e21cd91</refcontent> </reference> <reference anchor="Boru20" target="https://dl.acm.org/doi/abs/10.1145/3402413.3402419"> <front> <title>Validating the Sharing Behavior and Latency Characteristics of the L4S Architecture</title> <author fullname="Dejene Boru Oljira" initials="D." surname="Boru Oljira"> <organization>Karlstad Uni</organization> </author> <author fullname="Karl-Johan Grinnemo" initials="K-J." surname="Grinnemo"> <organization>Karlstad Uni</organization> </author> <author fullname="Anna Brunstrom" initials="A." surname="Brunstrom"> <organization>Karlstad Uni</organization> </author> <author fullname="Javid Taheri" initials="J." surname="Taheri"> <organization>Karlstad Uni</organization> </author> <date month="May" year="2020"/> </front> <seriesInfoname="ACM CCR" value="50(2):37--44"/>name="DOI" value="10.1145/3402413.3402419"/> <refcontent>ACM SIGCOMM Computer Communication Review, Vol. 50, Issue 2, pp. 37-44</refcontent> </reference> </references> </references> <section anchor="dualq_Ex_algo_pi2" numbered="true" toc="default"> <name>Example DualQ Coupled PI2 Algorithm</name> <t>As a first concrete example, the pseudocode below gives the DualPI2 algorithm. DualPI2 follows the structure of the DualQ Coupled AQM framework in <xref target="dualq_fig_structure" format="default"/>. A simple ramp function (configured in units of queuing time) with unsmoothed ECN marking is used for the Native L4S AQM. The ramp can also be configured as a step function. The PI2algorithm <xrefalgorithm <xref target="PI2" format="default"/> is used for the Classic AQM. PI2 is an improved variant of the PIEAQM <xrefAQM <xref target="RFC8033" format="default"/>.</t> <t>The pseudocode will be introduced in two passes. The first pass explains the core concepts, deferring handling of edge-cases like overload to the second pass. To aid comparison, line numbers are kept in step between the two passes by using letter suffixes where the longer code needs extra lines.</t> <t>All variables are assumed to be floating point in their basic units (size in bytes, time in seconds, rates in bytes/second, alpha and beta in Hz, and probabilities from 0 to1.1). Constants expressed in k (kilo), M (mega), G (giga), u (micro), m(milli) ,(milli), %,...and so forth, are assumed to be converted to their appropriate multiple or fraction to represent the basic units. A real implementation that wants to use integer values needs to handle appropriate scaling factors and allowaccordinglyappropriate resolution of its integer types (including temporary internal values during calculations).</t> <t>A full open source implementation for Linux is availableat: https://github.com/L4STeam/sch_dualpi2_upstreamat <eref target="https://github.com/L4STeam/sch_dualpi2_upstream" brackets="angle"/> and explained in <xref target="DualPI2Linux" format="default"/>. The specification of the DualQ Coupled AQM for DOCSIS cable modems andCMTSscable modem termination systems (CMTSs) is available in <xref target="DOCSIS3.1" format="default"/> and explained in <xref target="LLD" format="default"/>.</t> <section anchor="dualq_Ex_algo_pi2-1" numbered="true" toc="default"> <name>Pass #1: Core Concepts</name> <t>The pseudocode manipulates three main structures of variables: the packet (pkt), the L4S queue(lq)(lq), and the Classic queue (cq). The pseudocode consists of the following six functions:</t> <ul spacing="normal"> <li>The initialization function dualpi2_params_init(...) (<xref target="dualq_fig_Algo_pi2_core_header" format="default"/>) that sets parameter defaults (the API for setting non-default values is omitted forbrevity)</li>brevity).</li> <li>The enqueue function dualpi2_enqueue(lq, cq, pkt) (<xref target="dualq_fig_Algo_pi2_enqueue"format="default"/>)</li>format="default"/>).</li> <li>The dequeue function dualpi2_dequeue(lq, cq, pkt) (<xref target="dualq_fig_Algo_pi2_dequeue"format="default"/>)</li>format="default"/>).</li> <li>The recurrence function recur(q, likelihood) for de-randomized ECN marking (shown at the end of <xref target="dualq_fig_Algo_pi2_dequeue" format="default"/>).</li> <li>The L4S AQM function laqm(qdelay) (<xref target="dualq_fig_Algo_laqm_core" format="default"/>) used to calculate the ECN-marking probability for the L4Squeue</li>queue.</li> <li>ThebaseBase AQM function that implements the PI algorithm dualpi2_update(lq, cq) (<xref target="dualq_fig_Algo_pi2_core" format="default"/>) used to regularly update the base probability (p'), which is squared for the Classic AQM as well as being coupled across to the L4S queue.</li> </ul> <t>It also uses the following functions that are not shown in full here:</t> <ul spacing="normal"> <li>scheduler(), which selects between the head packets of the twoqueues; thequeues. The choice of scheduler technology is discussedlater;</li>later.</li> <li>cq.byt() or lq.byt() returns the current length(aka. backlog)(a.k.a. backlog) of the relevant queue inbytes;</li>bytes.</li> <li>cq.len() or lq.len() returns the current length of the relevant queue inpackets;</li>packets.</li> <li>cq.time() or lq.time() returns the current queuing delay of the relevant queue in units of time (seeNote a);</li><xref target="note_qdelay" format="none">Note a</xref> below).</li> <li>mark(pkt) and drop(pkt) forECN-markingECN marking and dropping apacket;</li>packet.</li> </ul> <t>In experiments so far (building on experiments with PIE) on broadband access links ranging from 4 Mb/s to 200 Mb/s with base RTTs from 5 ms to 100 ms, DualPI2 achieves good results with the default parameters in <xref target="dualq_fig_Algo_pi2_core_header" format="default"/>. The parameters are categorised by whether they relate to theBasePI2 AQM, the L4SAQMAQM, or the framework coupling them together. Constants and variables derived from these parameters are also included at the end of each category. Each parameter is explained as it is encountered in the walk-through of the pseudocode below, and the rationale for the chosen defaults are given so that sensible values can be used in scenarios other than the regular public Internet.</t> <figure anchor="dualq_fig_Algo_pi2_core_header"> <name>Example Header Pseudocode for DualQ Coupled PI2 AQM</name><artwork name="" type="" align="left" alt=""><![CDATA[1:<sourcecode><![CDATA[ 1: dualpi2_params_init(...) { % Set input parameter defaults 2: % DualQ Coupled framework parameters 5: limit = MAX_LINK_RATE * 250 ms % Dual buffer size 3: k = 2 % Coupling factor 4: % NOT SHOWN % scheduler-dependent weight or equival't parameter 6: 7: % PI2 Classic AQM parameters 8: target = 15 ms % Queue delay target 9: RTT_max = 100 ms % Worst case RTT expected 10: % PI2 constants derived from above PI2 parameters 11: p_Cmax = min(1/k^2, 1) % Max Classic drop/mark prob 12: Tupdate = min(target, RTT_max/3) % PI sampling interval 13: alpha = 0.1 * Tupdate / RTT_max^2 % PI integral gain in Hz 14: beta = 0.3 / RTT_max % PI proportional gain in Hz 15: 16: % L4S ramp AQM parameters 17: minTh = 800 us % L4S min marking threshold in time units 18: range = 400 us % Range of L4S ramp in time units 19: Th_len = 1 pkt % Min L4S marking threshold in packets 20: % L4S constants 21: p_Lmax = 1 % Max L4S marking prob 22:} ]]></artwork>}]]></sourcecode> </figure> <t>The overall goal of the code is to apply the marking and dropping probabilities for L4S and Classic traffic (p_L and p_C). These are derived from the underlying base probabilities p'_L and p'driven respectivelydriven, respectively, by the traffic in the L and C queues. The marking probability for the L queue (p_L) depends on both the base probability in its own queue (p'_L) and a probability called p_CL, which is coupled across from p' in the C queue (see <xref target="dualq_coupled_structure" format="default"/> for the derivation of the specific equations and dependencies).</t> <t>The probabilities p_CL and p_C are derived in lines 4 and 5 of the dualpi2_update() function (<xref target="dualq_fig_Algo_pi2_core" format="default"/>) then used in the dualpi2_dequeue() function where p_L is also derived from p_CL at line 6 (<xref target="dualq_fig_Algo_pi2_dequeue" format="default"/>). The code walk-through below builds up to explaining that part of the code eventually, but it starts from packet arrival.</t> <figure anchor="dualq_fig_Algo_pi2_enqueue"> <name>Example Enqueue Pseudocode for DualQ Coupled PI2 AQM</name><artwork name="" type="" align="left" alt=""><![CDATA[1:<sourcecode><![CDATA[ 1: dualpi2_enqueue(lq, cq, pkt) { % Test limit and classify lq or cq 2: if ( lq.byt() + cq.byt() + MTU > limit) 3: drop(pkt) % drop packet if buffer is full 4: timestamp(pkt) % only needed if using the sojourn technique 5: % Packet classifier 6: if ( ecn(pkt) modulo 2 == 1 ) % ECN bits = ECT(1) or CE 7: lq.enqueue(pkt) 8: else % ECN bits = not-ECT or ECT(0) 9: cq.enqueue(pkt) 10:} ]]></artwork>}]]></sourcecode> </figure> <figure anchor="dualq_fig_Algo_pi2_dequeue"> <name>Example Dequeue Pseudocode for DualQ Coupled PI2 AQM</name><artwork name="" type="" align="left" alt=""><![CDATA[1:<sourcecode><![CDATA[ 1: dualpi2_dequeue(lq, cq, pkt) { % Couples L4S & Classic queues 2: while ( lq.byt() + cq.byt() > 0 ) { 3: if ( scheduler() == lq ) { 4: lq.dequeue(pkt) % Scheduler chooses lq 5: p'_L = laqm(lq.time()) % Native LAQM 6: p_L = max(p'_L, p_CL) % Combining function 7: if ( recur(lq, p_L) ) % Linear marking 8: mark(pkt) 9: } else { 10: cq.dequeue(pkt) % Scheduler chooses cq 11: if ( recur(cq, p_C) ) { % probability p_C = p'^2 12: if ( ecn(pkt) == 0 ) { % if ECN field = not-ECT 13: drop(pkt) % squared drop 14: continue % continue to the top of the while loop 15: } 16: mark(pkt) % squared mark 17: } 18: } 19: return(pkt) % return the packet and stop 20: } 21: return(NULL) % no packet to dequeue 22: } 23: recur(q, likelihood) { % Returns TRUE with a certain likelihood 24: q.count += likelihood 25: if (q.count > 1) { 26: q.count -= 1 27: return TRUE 28: } 29: return FALSE 30:} ]]></artwork>}]]></sourcecode> </figure> <t>When packets arrive,firsta common queue limit is checked first as shown in line 2 of the enqueuing pseudocode in <xref target="dualq_fig_Algo_pi2_enqueue" format="default"/>. This assumes a shared buffer for the two queues(Note b(<xref target="note_separate_buffers" format="none">Note b</xref> discusses the merits of separate buffers). In order to avoid any bias against larger packets, 1 MTU of space is always allowed, and the limit is deliberately tested before enqueue.</t> <t>If limit is not exceeded, the packet is timestamped in line 4 (only if the sojourn time technique is being used to measure queue delay; seeNote a<xref target="note_qdelay" format="none">Note a</xref> below for alternatives).</t> <t>At lines 5-9, the packet is classified and enqueued to the Classic or L4S queue dependent on the least significant bit (LSB) of the ECN field in the IP header (line 6). Packets with a codepoint having an LSB of 0 (Not-ECT and ECT(0)) will be enqueued in the Classic queue. Otherwise, ECT(1) and CE packets will be enqueued in the L4S queue. Optional additional packet classification flexibility is omitted for brevity (see the L4S ECNprotocol <xref target="I-D.ietf-tsvwg-ecn-l4s-id"protocol <xref target="RFC9331" format="default"/>).</t> <t>The dequeue pseudocode (<xref target="dualq_fig_Algo_pi2_dequeue" format="default"/>) is repeatedly called whenever the lower layer is ready to forward a packet. It schedules one packet for dequeuing (or zero if the queue is empty) then returns control to thecaller,caller so that it does not block while that packet is being forwarded. While making this dequeue decision, it also makes the necessary AQM decisions on dropping or marking. The alternative of applying the AQMs at enqueue would shift some processing from the critical time when each packet is dequeued. However, it would also add a whole queue of delay to the control signals, making the control loop sloppier (for a typicalRTTRTT, it would double the Classic queue's feedback delay).</t> <t>All the dequeue code is contained within a large while loop so that if it decides to drop a packet, it will continue until it selects a packet to schedule. Line 3 of the dequeue pseudocode is where the scheduler chooses between the L4S queue (lq) and the Classic queue (cq). Detailed implementation of the scheduler is not shown (see discussion later). </t> <ul spacing="normal"> <li> <t>If an L4S packet is scheduled, in lines 7 and 8 the packet is ECN-marked with likelihood p_L. The recur() function at the end of <xref target="dualq_fig_Algo_pi2_dequeue" format="default"/> is used, which is preferred over random marking because it avoids delay due to randomization when interpreting congestion signals, but it still desynchronizes thesaw-teethsawteeth of the flows. Line 6 calculates p_L as the maximum of the coupled L4S probability p_CL and the probability from thenativeNative L4S AQM p'_L. This implements the max() function shown in <xref target="dualq_fig_structure" format="default"/> to couple the outputs of the two AQMs together. Of the two probabilities input to p_L in line 6:</t> <ul spacing="normal"> <li>p'_L is calculated per packet in line 5 by the laqm() function (see <xref target="dualq_fig_Algo_laqm_core"format="default"/>),</li> <li>Whereas p_CLformat="default"/>), whereas</li> <li>p_CL is maintained by the dualpi2_update()functionfunction, which runs every Tupdate (Tupdate is set in line 12 of <xref target="dualq_fig_Algo_pi2_core_header" format="default"/>).</li> </ul> </li> <li>If a Classic packet is scheduled, lines 10 to 17 drop or mark the packet with probability p_C.</li> </ul> <t>The Native L4S AQM algorithm (<xref target="dualq_fig_Algo_laqm_core" format="default"/>) is a ramp function, similar to the RED algorithm, but simplified as follows:</t> <ul spacing="normal"> <li>The extent of the ramp is defined in units of queuing delay, not bytes, so that configuration remains invariant as the queue departure rate varies.</li> <li>It uses instantaneousqueueingqueuing delay, which avoids the complexity of smoothing, but also avoids embedding a worst-case RTT of smoothing delay in the network (see <xref target="dualq_coupled" format="default"/>).</li> <li>The ramp rises linearly directly from 0 to 1, not to an intermediate value of p'_L as RED would, because there is no need to keepECN markingECN-marking probability low.</li> <li>Marking does not have to be randomized. Determinism is used instead ofrandomness;randomness to reduce the delay necessary to smooth out the noise of randomness from the signal.</li> </ul> <t>The ramp function requires two configuration parameters, the minimum threshold (minTh) and the width of the ramp (range), both in units of queuing time, as shown in lines 17&and 18 of the initialization function in <xref target="dualq_fig_Algo_pi2_core_header" format="default"/>. The ramp function can be configured as a step (seeNote c).</t><xref target="note_ramp" format="none">Note c</xref>).</t> <t>Although the DCTCPpaper <xrefpaper <xref target="Alizadeh-stability" format="default"/> recommends anECN markingECN-marking threshold of 0.17*RTT_typ, it also shows that the threshold can be much shallower with hardly any worseunder-utilizationunderutilization of the link (because the amplitude of DCTCP's sawteeth is so small). Based on extensive experiments, for the public Internet the default minimumECN markingECN-marking threshold (target) in <xref target="dualq_fig_Algo_pi2_core_header" format="default"/> is considered a good compromise, even though it is a significantly smaller fraction of RTT_typ.</t> <figure anchor="dualq_fig_Algo_laqm_core"> <name>Example Pseudocode for the Native L4S AQM</name><artwork name="" type="" align="left" alt=""><![CDATA[1:<sourcecode><![CDATA[ 1: laqm(qdelay) { % ReturnsnativeNative L4S AQM probability 2: if (qdelay >= maxTh) 3: return 1 4: else if (qdelay > minTh) 5: return (qdelay - minTh)/range % Divide could use a bit-shift 6: else 7: return 0 8:} ]]></artwork>}]]></sourcecode> </figure> <t/> <figure anchor="dualq_fig_Algo_pi2_core"> <name>ExamplePI-UpdatePI-update Pseudocode for DualQ Coupled PI2 AQM</name><artwork name="" type="" align="left" alt=""><![CDATA[1:<sourcecode><![CDATA[ 1: dualpi2_update(lq, cq) { % Update p' every Tupdate 2: curq = cq.time() % use queuing time of first-in Classic packet 3: p' = p' + alpha * (curq - target) + beta * (curq - prevq) 4: p_CL = k * p' % Coupled L4S prob = base prob * coupling factor 5: p_C = p'^2 % Classic prob = (base prob)^2 6: prevq = curq 7:} ]]></artwork>}]]></sourcecode> </figure> <tkeepWithPrevious="true">(ClampingkeepWithPrevious="true" indent='3'>(Note: Clamping p' within the range [0,1] omitted for clarity--- seetext)</t>below.)</t> <t>The coupled markingprobability,probability p_CL depends on the base probability (p'), which is kept up to date by executing the core PI algorithm in <xref target="dualq_fig_Algo_pi2_core" format="default"/>executedevery Tupdate.</t> <t>Note that p' solely depends on the queuing time in the Classic queue. In line 2, the current queuing delay (curq) is evaluated from how long the head packet was in the Classic queue (cq). The function cq.time() (not shown) subtracts the time stamped at enqueue from the current time (seeNote a)<xref target="note_qdelay" format="none">Note a</xref> below) and implicitly takes the current queuing delay as 0 if the queue is empty.</t> <t>The algorithm centres on line 3, which is a classicalProportional-Integral (PI)PI controller that alters p' dependent on: a) the error between the current queuing delay (curq) and the target queuingdelay, 'target';delay (target) and b) the change in queuing delay since the last sample. The name 'PI' represents the fact that the second factor (how fast the queue is growing) is<em>P</em>roportionalProportional to load while the first is the<em>I</em>ntegralIntegral of the load (so it removes any standing queue in excess of the target).</t> <t>The target parameter can be set based on local knowledge, but the aim is for the default to be a good compromise for anywhere in the intended deployment environment -- the public Internet. According to <xref target="PI2param" format="default"/>, the target queuing delay on line98 of <xref target="dualq_fig_Algo_pi2_core_header" format="default"/> is related to the typical base RTT worldwide, RTT_typ, by two factors: target = RTT_typ * g * f.BelowBelow, we summarize the rationale behind these factors and introduce a further adjustment. The two factors ensure that, in a large proportion of cases (say 90%), the sawtooth variations in RTT of a single flow will fit within the buffer without underutilizing the link. Frankly, these factors are educated guesses, but with the emphasis closer to 'educated' than to 'guess' (see <xref target="PI2param" format="default"/> for the full background):</t> <ul spacing="normal"> <li>RTT_typ is taken as 25 ms. This is based on an average CDN latency measured in each country weighted by the number of Internet users in that country to produce an overall weighted average for theInternet <xrefInternet <xref target="PI2param" format="default"/>. Countries were ranked by number of Internet users, and once 90% of Internet users were covered, smaller countries were excluded to avoidunrepresentativelysmall samplesizes.sizes that would be less representative. Also, importantly, the data for the average CDN latency in China (with the largest number of Internet users) has been removed, because the CDN latency was a significant outlier and, on reflection, the experimental technique seemed inappropriate to the CDN market in China.</li> <li>g is taken as 0.38. The factor g is a geometry factor that characterizes the shape of the sawteeth of prevalent Classic congestion controllers. The geometry factor is the fraction of the amplitude of the sawtooth variability in queue delay that lies below the AQM's target. For instance, at lowbit rate,bitrates, the geometry factor of standard Reno is 0.5, but at higherratesrates, it tendstotowards just under 1. According to the census of congestion controllers conducted by Mishra etal. inal. in Jul-Oct2019 <xref2019 <xref target="CCcensus19" format="default"/>, most Classic TCP traffic usesCubic.CUBIC. And, according to the analysis in <xref target="PI2param" format="default"/>, if running over a PI2 AQM, a large proportion of thisCubicCUBIC traffic would be in itsReno-FriendlyReno-friendly mode, which has a geometry factor of ~0.39(all(for all known implementations). The rest of theCubicCUBIC traffic would be in trueCubicCUBIC mode, which has a geometry factor of ~0.36. Without modelling the sawtooth profiles from all the other less prevalent congestion controllers, we estimate a 7:3 weighted average of these two, resulting in an average geometry factor of 0.38.</li> <li>f is taken as 2. The factor f is a safety factor that increases the target queue to allow for the distribution of RTT_typ around its mean. Otherwise, the target queue would only avoid underutilization for those users below the mean. It also provides a safety margin for the proportion of paths in use that span beyond the distance between a user and their local CDN. Currently, no data is available on the variance of queue delay around the mean in each region, so there is plenty of room for this guess to become more educated.</li> <li> <xref target="PI2param" format="default"/> recommends target = RTT_typ * g * f =25ms25 ms * 0.38 * 2 = 19 ms. However, a further adjustment is warranted, because target is moving year-on-year. The paper is based on data collected in 2019, and it mentions evidence fromspeedtest.netthe Speedtest Global Index that suggests RTT_typ reduced by 17% (fixed) or 12% (mobile) between 2020 and 2021. Therefore, we recommend a default of target = 15 ms at the time of writing (2021).</li> </ul> <t>Operators can always use the data and discussion in <xref target="PI2param" format="default"/> to configure a more appropriate target for their environment. For instance, an operator might wish to question the assumptions called out in that paper, such as the goal of no underutilization for a large majority of single flow transfers (given many large transfers use multiple flows to avoid the scaling limitations of Classic flows).</t> <t>The two 'gain factors' in line 3 of <xref target="dualq_fig_Algo_pi2_core" format="default"/>, alpha and beta, respectively weight how strongly each of the two elements (Integral and Proportional) alters p'. They are in units of 'per second of delay' or Hz, because they transform differences inqueueingqueuing delay into changes in probability (assuming probability has a value from 0 to 1).</t> <t>Alpha and beta determine how much p' ought to change after each update interval (Tupdate). For a smaller Tupdate, p' should change by the same amount persecond,second but in finer more frequent steps. So alpha depends on Tupdate (see line 13 of the initialization function in <xref target="dualq_fig_Algo_pi2_core_header" format="default"/>). It is best to update p' as frequently as possible, but Tupdate will probably be constrained by hardware performance. As shown in line13,12, the update interval should be frequent enough to update at least once in the time taken for the target queue to drain ('target') as long as it updates at least three times per maximum RTT. Tupdate defaults to 16 ms in the reference Linux implementation because it has to be rounded to a multiple of 4 ms. For link rates from 4 to 200 Mb/s and a maximum RTT of100ms,100 ms, it has been verified through extensive testing thatTupdate=16msTupdate = 16 ms (as also recommended in the PIEspec <xrefspec <xref target="RFC8033" format="default"/>) is sufficient.</t> <t>The choice of alpha and beta also determines the AQM's stable operating range. The AQM ought to change p' as fast as possible in response to changes in load withoutover-compensatingovercompensating and therefore causing oscillations in the queue. Therefore, the values of alpha and beta also depend on the RTT of the expected worst-case flow (RTT_max).</t> <t>The maximum RTT of a PI controller (RTT_max in line109 of <xref target="dualq_fig_Algo_pi2_core_header" format="default"/>) is not an absolute maximum, but more instability (more queue variability) sets in for long-running flows with an RTT above this value. The propagation delay halfway round the planet and back in glass fibre is 200 ms. However, hardly any traffic traverses such extreme paths and, since the significant consolidation of Internet traffic between 2007 and2009 <xref2009 <xref target="Labovitz10" format="default"/>, a high and growing proportion of all Internet traffic (roughly two-thirds at the time of writing) has been served fromcontent distribution networks (CDNs)CDNs or 'cloud' services distributed close toend-users.end users. The Internet might change again, but for now, designing for a maximum RTT of100ms100 ms is a good compromise between faster queue control at low RTT and some instability on the occasions when a longer path is necessary.</t> <t>Recommended derivations of the gain constants alpha and beta can be approximated for Reno over a PI2 AQM as: alpha = 0.1 * Tupdate / RTT_max^2; beta = 0.3 / RTT_max, as shown in lines 13 and 14& 15of <xref target="dualq_fig_Algo_pi2_core_header" format="default"/>. These are derived from the stability analysis in <xref target="PI2" format="default"/>. For the default values ofTupdate=16Tupdate = 16 ms and RTT_max = 100 ms, they result in alpha = 0.16; beta = 3.2 (discrepancies are due to rounding). These defaults have been verified with a wide range of link rates, targetdelaysdelays, anda range oftraffic models with mixed and similar RTTs, short and long flows, etc.</t> <t>In corner cases, p' can overflow the range [0,1] so the resulting value of p' has to be bounded (omitted from the pseudocode). Then, as already explained, the coupled and Classic probabilities are derived from the new p' in lines 4 and 5 of <xref target="dualq_fig_Algo_pi2_core" format="default"/> as p_CL = k*p' and p_C = p'^2.</t> <t>Because the coupled L4S marking probability (p_CL) is factored up by k, the dynamic gain parameters alpha and beta are also inherently factored up by k for the L4S queue. So, the effective gain factor for the L4S queue is k*alpha (with defaults alpha = 0.16 Hz andk=2,k = 2, effective L4S alpha = 0.32 Hz).</t> <t>Unlike inPIE <xrefPIE <xref target="RFC8033" format="default"/>, alpha and beta do not need to be tuned every Tupdate dependent on p'. Instead, in PI2, alpha and beta are independent of p' because the squaring applied to Classic traffic tunes them inherently. This is explained in <xref target="PI2" format="default"/>, which also explains why this more principled approach removes the need for most of the heuristics that had to be added to PIE.</t> <t>Nonetheless, an implementer might wish to add selected details to either AQM. Forinstanceinstance, the Linux reference DualPI2 implementation includes the following (not shown in the pseudocode above):</t> <ul spacing="normal"> <li>Classic and coupled marking or dropping(i.e. based(i.e., based on p_C and p_CL from the PI controller) is not applied to a packet if the aggregate queue length in bytes is < 2 MTU (prior to enqueuing the packet or dequeuing it, depending on whether the AQM is configured to be applied at enqueue ordequeue);</li> <li>Indequeue); and</li> <li>in the WRR scheduler, the 'credit' indicating which queue should transmit is only changed if there are packets in both queues(i.e. if(i.e., if there is actual resource contention). This means that a properly paced L flow might never be delayed by the WRR. The WRR credit is reset in favour of the L queue when the link is idle.</li> </ul> <t>An implementer might also wish to add other heuristics,e.g. burst protection <xrefe.g., burst protection <xref target="RFC8033" format="default"/> or enhanced burstprotection <xrefprotection <xref target="RFC8034" format="default"/>.</t> <t>Notes:</t> <ol spacing="normal"type="a"><li anchor="dualq_note_qdelay">type="a"> <li anchor="note_qdelay"> <t>The drain rate of the queue can vary if it is scheduled relative to otherqueues,queues orto cater forif it accommodates fluctuations in a wireless medium. To auto-adjust to changes in drain rate, the queue needs to be measured in time, not bytes orpackets <xrefpackets <xref target="AQMmetrics"format="default"/>,format="default"/> <xref target="CoDel" format="default"/>. Queuing delay could be measured directly as the sojourn time(aka.(a.k.a. service time) of thequeue,queue by storing a per-packettime-stamptimestamp as each packet isenqueued,enqueued and subtractingthisit from the system time when the packet is dequeued. Iftime-stampingtimestamping is not easy to introduce with certain hardware, queuing delay could be predicted indirectly by dividing the size of the queue by the predicted departure rate, which might be known precisely for some link technologies(see(see, forexample inexample, DOCSIS PIE[RFC8034]).<xref target="RFC8034"/>). </t> <t>However, sojourn time is slow to detect bursts. For instance, if a burst arrives at an empty queue, the sojourn time only fully measures the burst's delay when its last packet is dequeued, even though the queue has known the size of the burst since its last packet was enqueued--- so it could have signalled congestion earlier. To remedy this, each head packet can be marked when it is dequeued based on the expected delay of the tail packet behind it, as explained below, rather than based on the head packet's own delay due to the packets in front ofit. <xrefit. "Underutilization with Bursty Traffic" in <xref target="Heist21" format="default"/> identifies a specific scenario where bursty traffic significantly hits utilization of the L queue. If this effect proves to be more widely applicable, using the delay behind the head could improve performance.</t> <t>The delay behind the head can be implemented by dividing the backlog at dequeue by the link rate or equivalently multiplying the backlog by the delay per unit of backlog. The implementation details will depend on whether the link rate is known; if it is not, a moving average of the delay per unit backlog can be maintained. This delay consists of serialization as well as media acquisition for shared media. So the details will depend strongly on the specific linktechnology,technology. This approach should be less sensitive to timing errors and cost less in operations and memory than the otherwise equivalent 'scaled sojourn time' metric, which is the sojourn time of a packet scaled by the ratio of the queue sizes when the packet departed andarrived <xrefarrived <xref target="SigQ-Dyn" format="default"/>.</t> </li><li>Line<li anchor="note_separate_buffers">Line 2 of the dualpi2_enqueue() function (<xref target="dualq_fig_Algo_pi2_enqueue" format="default"/>) assumes an implementation where lq and cq share common buffer memory. An alternative implementation could use separate buffers for each queue, in which case the arriving packet would have to be classified first to determine which buffer to check for available space. The choice is a trade-off; a shared buffer can use less memory whereas separate buffers isolate the L4S queue fromtail-droptail drop due to large bursts of Classic traffic(e.g. a(e.g., a Classic Reno TCP during slow-start over a long RTT).</li><li><li anchor="note_ramp"> <t>There has been some concern that using the step function of DCTCP for the Native L4S AQM requiresend-systemsend systems to smooth the signal for an unnecessarily large number of round trips to ensure sufficient fidelity. A ramp is no worse than a step in initial experiments with existing DCTCP. Therefore, it is recommended that a ramp is configured in place of a step, which will allow congestion control algorithms to investigate faster smoothing algorithms.</t> <t>A ramp is more generalthatthan a step, because an operator can effectively turn the ramp into a step function, as used by DCTCP, by setting the range to zero. There will not be a divide by zero problem at line 5 of <xref target="dualq_fig_Algo_laqm_core" format="default"/> because, if minTh is equal to maxTh, the condition for this ramp calculation cannot arise.</t> </li> </ol> </section> <section anchor="dualq_Ex_algo_pi2-2" numbered="true" toc="default"> <name>Pass #2: Edge-Case Details</name> <t>This section takes a second pass through the pseudocodeaddingto add details of two edge-cases: low link rate and overload. <xref target="dualq_fig_Algo_pi2_full_dequeue" format="default"/> repeats the dequeue function of <xref target="dualq_fig_Algo_pi2_dequeue" format="default"/>, but with details of both edge-cases added. Similarly, <xref target="dualq_fig_Algo_pi2_full_core" format="default"/> repeats the core PI algorithm of <xref target="dualq_fig_Algo_pi2_core" format="default"/>, but with overload details added. The initialization, enqueue, L4SAQMAQM, and recur functions are unchanged.</t> <t>The link rate can be so low that it takes a single packet queue longer to serialize than the threshold delay at which ECN marking starts to be applied in the L queue. Therefore, a minimum marking threshold parameter in units of packets rather than time is necessary (Th_len, default 1 packet in line 19 of <xref target="dualq_fig_Algo_pi2_core_header" format="default"/>) to ensure that the ramp does not trigger excessive marking on slow links. Where an implementation knows the link rate, it can set up this minimum at the time it is configured. For instance, it would divide 1 MTU by the link rate to convert it into a serialization time, then if the lower threshold of the Native L AQM ramp was lower than this serialization time, it could increase the thresholds to shift the bottom of the ramp to 2 MTU. This is the approach used inDOCSIS <xrefDOCSIS <xref target="DOCSIS3.1" format="default"/>, because the configured link rate is dedicated to the DualQ.</t> <t>The pseudocode given here applies where the link rate is unknown, which is more common for software implementations that might be deployed in scenarios where the link is shared with other queues. In lines 5a to 5d in <xref target="dualq_fig_Algo_pi2_full_dequeue"format="default"/>format="default"/>, the native L4S marking probability, p'_L, is zeroed if the queue is only 1 packet (in the default configuration).</t><t>Linux<aside><t>Linux implementationnote:</t> <ul spacing="normal"> <li>Innote: In Linux, the check that the queue exceeds Th_len before marking with thenativeNative L4S AQM is actually at enqueue, notdequeue, otherwisedequeue; otherwise, it would exempt the last packet of a burst from being marked. The result of the check is conveyed from enqueue to the dequeue function via a boolean in the packetmetadata.</li> </ul>metadata.</t> </aside> <t>Persistent overload is deemed to have occurred when Classic drop/marking probability reaches p_Cmax. Above this point, the Classic drop probability is applied to both the L and C queues, irrespective of whether any packet is ECN-capable. ECT packets that are not dropped can still be ECN-marked.</t> <t>In line1011 of the initialization function (<xref target="dualq_fig_Algo_pi2_core_header" format="default"/>), the maximum Classic drop probability p_Cmax = min(1/k^2, 1) or 1/4 for the default coupling factork=2.k = 2. In practice, 25% has been found to be a good threshold to preserve fairness betweenECN capableECN-capable andnon ECN capablenon-ECN-capable traffic. This protects the queues against both temporary overload from responsive flows and more persistent overload from any unresponsive traffic that falsely claims to be responsive to ECN.</t> <t>When the ClassicECN markingECN-marking probability reaches the p_Cmax threshold (1/k^2), the marking probability that is coupled to the L4S queue,p_CLp_CL, will always be 100% for any k (by equation (1) in <xreftarget="dualq_algo"target="dualq_coupled" format="default"/>). So, for readability, the constant p_Lmax is defined as 1 in line2221 of the initialization function (<xref target="dualq_fig_Algo_pi2_core_header" format="default"/>). This is intended to ensure that the L4S queue starts to introduce dropping onceECN-markingECN marking saturates at 100% and can rise no further. The 'PragueL4S' requirements <xref target="I-D.ietf-tsvwg-ecn-l4s-id"L4S requirements' <xref target="RFC9331" format="default"/> statethat,that when an L4S congestion control detects a drop, it falls back to a response that coexists with 'Classic' Reno congestion control.SoSo, it is correctthat,that when the L4S queue drops packets, it drops them proportional to p'^2, as if they are Classic packets.</t> <t>The two queues each test for overload in lines 4b and 12b of the dequeue function (<xref target="dualq_fig_Algo_pi2_full_dequeue" format="default"/>). Lines 8c to 8g drop L4S packets with probability p'^2. Lines 8h to 8i mark the remaining packets with probability p_CL. Given p_Lmax = 1, all remaining packets will be marked because, to have reached the else block at line 8b, p_CL >= 1.</t> <t>Line 2a in the core PI algorithm (<xref target="dualq_fig_Algo_pi2_full_core" format="default"/>) deals with overload of the L4S queue when there is little or no Classic traffic. This is necessary, because the core PI algorithm maintains the appropriate drop probability to regulate overload, but it depends on the length of the Classic queue. If there is little or no Classicqueuequeue, the naivePI updatePI-update functionin <xref(<xref target="dualq_fig_Algo_pi2_core"format="default"/>format="default"/>) would drop nothing, even if the L4S queue were overloaded--- so tail drop would have to take over (lines 2 and 3 of <xref target="dualq_fig_Algo_pi2_enqueue" format="default"/>).</t> <t>Instead, line 2a of the fullPI updatePI-update functionin <xref(<xref target="dualq_fig_Algo_pi2_full_core"format="default"/>format="default"/>) ensures that thebaseBase PI AQM in line 3 is driven by whichever of the two queue delays is greater, but line 3 still always uses the same Classic target (default 15 ms). If L queue delay is greater just because there is little or no Classic traffic, normally it will still be well below thebaseBase AQM target. This is because L4S traffic is also governed by the shallow threshold of its ownnativeNative AQM (lines5 and5a to 6 of the dequeue algorithm in <xref target="dualq_fig_Algo_pi2_full_dequeue" format="default"/>). So thebaseBase AQM will be driven to zero and not contribute. However, if the L queue is overloaded by traffic that is unresponsive to its marking, the max() in line22a of <xref target="dualq_fig_Algo_pi2_full_core" format="default"/> enables the L queue to smoothly take over driving thebaseBase AQM into overload mode even if there is little or no Classic traffic. Then thebaseBase AQM will keep the L queue to the Classic target (default 15 ms) by shedding L packets.</t> <figure anchor="dualq_fig_Algo_pi2_full_dequeue"> <name>Example Dequeue Pseudocode for DualQ Coupled PI2 AQM (Including Code for Edge-Cases)</name><artwork name="" type="" align="left" alt=""><![CDATA[1:<sourcecode><![CDATA[ 1: dualpi2_dequeue(lq, cq, pkt) { % Couples L4S & Classic queues 2: while ( lq.byt() + cq.byt() > 0 ) { 3: if ( scheduler() == lq ) { 4a: lq.dequeue(pkt) % L4S scheduled 4b: if ( p_CL < p_Lmax ) { % Check for overload saturation 5a: if (lq.len()>Th_len) % >1 packet queued 5b: p'_L = laqm(lq.time()) % Native LAQM 5c: else 5d: p'_L = 0 % Suppress marking 1 pkt queue 6: p_L = max(p'_L, p_CL) % Combining function 7: if ( recur(lq, p_L) %Linear marking 8a: mark(pkt) 8b: } else { % overload saturation 8c: if ( recur(lq, p_C) ) { % probability p_C = p'^2 8e: drop(pkt) % revert to Classic drop due to overload 8f: continue % continue to the top of the while loop 8g: } 8h: if ( recur(lq, p_CL) ) % probability p_CL = k * p' 8i: mark(pkt) % linear marking of remaining packets 8j: } 9: } else { 10: cq.dequeue(pkt) % Classic scheduled 11: if ( recur(cq, p_C) ) { % probability p_C = p'^2 12a: if ( (ecn(pkt) == 0) % ECN field = not-ECT 12b: OR (p_C >= p_Cmax) ) { % Overload disables ECN 13: drop(pkt) % squared drop, redo loop 14: continue % continue to the top of the while loop 15: } 16: mark(pkt) % squared mark 17: } 18: } 19: return(pkt) % return the packet and stop 20: } 21: return(NULL) % no packet to dequeue 22: }]]></artwork>]]></sourcecode> </figure> <figure anchor="dualq_fig_Algo_pi2_full_core"> <name>ExamplePI-UpdatePI-update Pseudocode for DualQ Coupled PI2 AQM (Including Overload Code)</name><artwork name="" type="" align="left" alt=""><![CDATA[1:<sourcecode><![CDATA[ 1: dualpi2_update(lq, cq) { % Update p' every Tupdate 2a: curq = max(cq.time(), lq.time()) % use greatest queuing time 3: p' = p' + alpha * (curq - target) + beta * (curq - prevq) 4: p_CL = p' * k % Coupled L4S prob = base prob * coupling factor 5: p_C = p'^2 % Classic prob = (base prob)^2 6: prevq = curq 7: }]]></artwork>]]></sourcecode> </figure> <t/> <t>The choice of scheduler technology is critical to overload protection (see <xref target="dualq_Overload_Starvation" format="default"/>). </t> <ul spacing="normal"> <li>A well-understood weighted scheduler such asweighted round-robin (WRR)WRR is recommended. As long as the scheduler weight for Classic is small(e.g. 1/16),(e.g., 1/16), its exact value isunimportantunimportant, because it does not normally determine capacity shares. The weight is only important to prevent unresponsive L4S traffic starving Classic traffic in the short term (see <xref target="dualq_Overload_Starvation" format="default"/>). This is because capacity sharing between the queues is normally determined by the coupled congestion signal, which overrides the scheduler, by making L4S sources leave roughly equal per-flow capacity available for Classic flows.</li> <li> <t>Alternatively, a time-shifted FIFO (TS-FIFO) could be used. It works by selecting the head packet that has waited the longest, biased against the Classic traffic by a time-shift of tshift. To implementtime-shifted FIFO,TS-FIFO, the scheduler() function in line 3 of the dequeue code would simply be implemented as the scheduler() function at the bottom of <xref target="dualq_fig_Algo_Real" format="default"/> in <xref target="dualq_Ex_algo" format="default"/>. For the publicInternetInternet, a good value for tshift is50ms.50 ms. For private networks with smaller diameter, about 4*target would be reasonable. TS-FIFO is a very simple scheduler, but complexity might need to be added to address some deficiencies (which is why it is not recommended over WRR):</t> <ul spacing="normal"> <li>TS-FIFO does not fully isolate latency in the L4S queue from uncontrolled bursts in the Classic queue;</li><li>Using<li>using sojourn time for TS-FIFO is only appropriate iftime-stampingtimestamping of packets isfeasible;</li> <li>Evenfeasible; and</li> <li>even iftime-stampingtimestamping is supported, the sojourn time of the head packet is always stale, so a more instantaneous measure of queue delay could be used (seeNote a<xref target="note_qdelay" format="none">Note a</xref> in <xref target="dualq_Ex_algo_pi2-1" format="default"/>).</li> </ul> </li> <li>A strict priority scheduler would be inappropriate as discussed in <xref target="dualq_Overload_Starvation" format="default"/>.</li> </ul> </section> </section> <section anchor="dualq_Ex_algo" numbered="true" toc="default"> <name>Example DualQ Coupled Curvy RED Algorithm</name> <t>As another example of a DualQ Coupled AQM algorithm, the pseudocode below gives theCurvy RED basedCurvy-RED-based algorithm. Although the AQM was designed to be efficient in integer arithmetic, to aid understanding it is first given using floating point arithmetic (<xref target="dualq_fig_Algo_Real" format="default"/>). Then, one possible optimization for integer arithmetic is given, also in pseudocode (<xref target="dualq_fig_Algo_Int" format="default"/>). To aid comparison, the line numbers are kept in step between the two by using letter suffixes where the longer code needs extra lines.</t> <section anchor="dualq_Ex_algo_float" numbered="true" toc="default"> <name>Curvy RED in Pseudocode</name> <t>The pseudocode manipulates three main structures of variables: the packet (pkt), the L4S queue(lq)(lq), and the Classic queue(cq)(cq). It is defined andconsists ofdescribed below in the followingfivethree functions:</t> <ul spacing="normal"><li>The<li>the initialization function cred_params_init(...) (<xref target="dualq_fig_Algo_pi2_core_header" format="default"/>) that sets parameter defaults (the API for setting non-default values is omitted for brevity);</li><li>The<li>the dequeue function cred_dequeue(lq, cq, pkt) (<xref target="dualq_fig_Algo_pi2_dequeue"format="default"/>);</li> <li>Theformat="default"/>); and</li> <li>the scheduling function scheduler(), which selects between the head packets of the two queues.</li> </ul> <t>It also uses the following functions that are either shownelsewhere,elsewhere or not shown in full here:</t> <ul spacing="normal"><li>The<li>the enqueue function, which is identical to that used for DualPI2, dualpi2_enqueue(lq, cq, pkt) in <xref target="dualq_fig_Algo_pi2_enqueue" format="default"/>;</li> <li>mark(pkt) and drop(pkt) forECN-markingECN marking and dropping a packet;</li> <li>cq.byt() or lq.byt() returns the current length(aka. backlog)(a.k.a. backlog) of the relevant queue inbytes;</li>bytes; and</li> <li>cq.time() or lq.time() returns the current queuing delay of the relevant queue in units of time (seeNote a<xref target="note_qdelay" format="none">Note a</xref> in <xref target="dualq_Ex_algo_pi2-1" format="default"/>).</li> </ul> <t>Because Curvy RED was evaluated before DualPI2, certain improvements introduced for DualPI2 were not evaluated for Curvy RED. In the pseudocode below, the straightforward improvements have been added on the assumption they will provide similar benefits, but that has not been proven experimentally. They are: i) a conditional priority scheduler instead of strictprioritypriority; ii) a time-based threshold for thenativeNative L4S AQM; and iii) ECN support for the Classic AQM. A recent evaluation has proved that a minimum ECN-marking threshold (minTh) greatly improves performance, so this is also included in the pseudocode.</t> <t>Overload protection has not been added to the Curvy RED pseudocode below so as not to detract from the main features. It would be added in exactly the same way as in <xref target="dualq_Ex_algo_pi2-2" format="default"/> for the DualPI2 pseudocode. ThenativeNative L4S AQM uses a step threshold, but a ramp like that described for DualPI2 could be used instead. The scheduler uses the simple TS-FIFO algorithm, but it could be replaced with WRR.</t> <t>The Curvy RED algorithm has not been maintained or evaluated to the same degree as the DualPI2 algorithm. In initial experiments on broadband access links ranging from 4 Mb/s to 200 Mb/s with base RTTs from 5 ms to 100 ms, Curvy RED achieved good results with the default parameters in <xref target="dualq_fig_Algo_cred_core_header" format="default"/>.</t> <t>The parameters arecategorisedcategorized by whether they relate to the Classic AQM, the L4SAQMAQM, or the framework coupling them together. Constants and variables derived from these parameters are also included at the end of each category. These are the raw input parameters for the algorithm. A configuration front-end could accept more meaningful parameters(e.g. RTT_max(e.g., RTT_max and RTT_typ) and convert them into these raw parameters, as has been done for DualPI2 in <xref target="dualq_Ex_algo_pi2" format="default"/>. Where necessary, parameters are explained further in the walk-through of the pseudocode below.</t> <figure anchor="dualq_fig_Algo_cred_core_header"> <name>Example Header Pseudocode for DualQ Coupled Curvy RED AQM</name><artwork name="" type="" align="left" alt=""><![CDATA[1:<sourcecode><![CDATA[ 1: cred_params_init(...) { % Set input parameter defaults 2: % DualQ Coupled framework parameters 3: limit = MAX_LINK_RATE * 250 ms % Dual buffer size 4: k' = 1 % Coupling factor as a power of 2 5: tshift = 50 ms %Time shiftTime-shift of TS-FIFO scheduler 6: % Constants derived from Classic AQM parameters 7: k = 2^k' % Coupling factor fromEquationequation (1) 6: 7: % Classic AQM parameters 8: g_C = 5 % EWMA smoothing parameter as a power of 1/2 9: S_C = -1 % Classic ramp scaling factor as a power of 2 10: minTh = 500 ms % No Classic drop/mark below this queue delay 11: % Constants derived from Classic AQM parameters 12: gamma = 2^(-g_C) % EWMA smoothing parameter 13: range_C = 2^S_C % Range of Classic ramp 14: 15: % L4S AQM parameters 16: T = 1 ms % Queue delay threshold fornativeNative L4S AQM 17: % Constants derived from above parameters 18: S_L = S_C - k' % L4S ramp scaling factor as a power of 2 19: range_L = 2^S_L % Range of L4S ramp 20: }]]></artwork>]]></sourcecode> </figure> <figure anchor="dualq_fig_Algo_Real"> <name>Example Dequeue Pseudocode for DualQ Coupled Curvy RED AQM</name><artwork name="" type="" align="left" alt=""><![CDATA[1:<sourcecode><![CDATA[ 1: cred_dequeue(lq, cq, pkt) { % Couples L4S & Classic queues 2: while ( lq.byt() + cq.byt() > 0 ) { 3: if ( scheduler() == lq ) { 4: lq.dequeue(pkt) % L4S scheduled 5a: p_CL = (Q_C - minTh) / range_L 5b: if ( ( lq.time() > T ) 5c: OR ( p_CL > maxrand(U) ) ) 6: mark(pkt) 7: } else { 8: cq.dequeue(pkt) % Classic scheduled 9a: Q_C = gamma * cq.time() + (1-gamma) * Q_C % Classic Q EWMA 10a: sqrt_p_C = (Q_C - minTh) / range_C 10b: if ( sqrt_p_C > maxrand(2*U) ) { 11: if ( (ecn(pkt) == 0) { % ECN field = not-ECT 12: drop(pkt) % Squared drop, redo loop 13: continue % continue to the top of the while loop 14: } 15: mark(pkt) 16: } 17: } 18: return(pkt) % return the packet and stop here 19: } 20: return(NULL) % no packet to dequeue 21: } 22: maxrand(u) { % return the max of u random numbers 23: maxr=0 24: while (u-- > 0) 25: maxr = max(maxr, rand()) % 0 <= rand() < 1 26: return(maxr) 27: } 28: scheduler() { 29: if ( lq.time() + tshift >= cq.time() ) 30: return lq; 31: else 32: return cq; 33: }]]></artwork>]]></sourcecode> </figure> <t>The dequeue pseudocode (<xref target="dualq_fig_Algo_Real" format="default"/>) is repeatedly called whenever the lower layer is ready to forward a packet. It schedules one packet for dequeuing (or zero if the queue is empty) then returns control to thecaller,caller so that it does not block while that packet is being forwarded. While making this dequeue decision, it also makes the necessary AQM decisions on dropping or marking. The alternative of applying the AQMs at enqueue would shift some processing from the critical time when each packet is dequeued. However, it would also add a whole queue of delay to the control signals, making the control loop very sloppy.</t> <t>The code is written assuming the AQMs are applied on dequeue(Note <xref format="counter" target="dualq_note_dequeue"/>).(<xref format="none" target="dualq_note_dequeue">Note 1</xref>). All the dequeue code is contained within a large while loop so that if it decides to drop a packet, it will continue until it selects a packet to schedule. If both queues are empty, the routine returns NULL at line 20. Line 3 of the dequeue pseudocode is where the conditional priority scheduler chooses between the L4S queue (lq) and the Classic queue (cq). Thetime-shifted FIFOTS-FIFO scheduler is shown at lines 28-33, which would be suitable if simplicity is paramount (seeNote<xrefformat="counter" target="dualq_note_conditional_priority"/>).</t>format="none" target="dualq_note_conditional_priority">Note 2</xref>).</t> <t>Within each queue, the decision whether to forward,dropdrop, or mark is taken as follows (to simplify the explanation, it is assumed thatU=1):</t>U = 1):</t> <dlnewline="false"newline="true" spacing="normal"> <dt>L4S:</dt> <dd> <t>If the test at line 3 determines there is an L4S packet to dequeue, the tests at lines 5b and 5c determine whether to mark it. The first is a simple test of whether the L4S queue delay (lq.time()) is greater than a step threshold T(Note <xref format="counter" target="dualq_note_step"/>).(<xref target="dualq_note_step" format="none">Note 3</xref>). The second test is similar to the random ECN marking inRED,RED but with the following differences: i) marking depends on queuing time, not bytes, in order to scale for any link rate without being reconfigured; ii) marking of the L4S queue depends on a logical OR of twotests;tests: one against its own queuing time and one against the queuing time of the <em>other</em> (Classic) queue; iii) the tests are against the instantaneous queuing time of the L4Squeue,queue but against a smoothed average of the other (Classic) queue; and iv) the queue is compared with the maximum of U random numbers (but ifU=1,U = 1, this is the same as the single random number used in RED).</t> <t>Specifically, in line5a5a, the coupled marking probability p_CL is set to the amount by which the averaged Classicqueueingqueuing delay Q_C exceeds the minimum queuing delay threshold(minTh)(minTh), all divided by the L4S scaling parameter range_L. range_L represents the queuing delay (in seconds) added to minTh at which marking probability would hit 100%.ThenThen, in line 5c (ifU=1)U = 1), the result is compared with a uniformly distributed random number between 0 and 1, which ensures that, over range_L, marking probability will linearly increase withqueueingqueuing time.</t> </dd> <dt>Classic:</dt> <dd> <t>If the scheduler at line 3 chooses to dequeue a Classic packet and jumps to line 7, the test at line 10b determines whether to drop or mark it. But before that, line 9a updates Q_C, which is an exponentially weighted moving average (Note <xref format="counter" target="dualq_note_non-EWMA"/>) of the queuing time of the Classic queue, where cq.time() is the current instantaneousqueueingqueuing time of the packet at the head of the Classic queue (zero ifempty)empty), and gamma is theEWMAexponentially weighted moving average (EWMA) constant (default1/32,1/32; see line 12 of the initialization function). </t> <t>Lines 10a and 10b implement the Classic AQM. In line10a10a, the averaged queuing time Q_C is divided by the Classic scaling parameter range_C, in the same way that queuing time was scaled for L4S marking. This scaled queuing time will be squared to compute Classic dropprobability so,probability. So, before it is squared, it is effectively the square root of the dropprobability, henceprobability; hence, it is given the variable name sqrt_p_C. The squaring is done by comparing it with the maximum out of two random numbers (assumingU=1).U = 1). Comparing it with the maximum out of two is the same as the logical`AND''AND' of two tests, which ensures drop probability rises with the square of queuing time.</t> </dd> </dl> <t>The AQM functions in each queue (lines 5c&and 10b) are two cases of a new generalization of RED calledCurvy RED,'Curvy RED', motivated as follows. When the performance of this AQM was compared with FQ-CoDel and PIE, their goal of holding queuing delay to a fixed target seemedmisguided <xrefmisguided <xref target="CRED_Insights" format="default"/>. As the number of flows increases, if the AQM does not allow host congestion controllers to increase queuing delay, it has to introduce abnormally high levels of loss. Then loss rather than queuing becomes the dominant cause of delay for short flows, due to timeouts and tail losses.</t> <t>Curvy RED constrains delay with a softened target that allows some increase in delay as load increases. This is achieved by increasing drop probability on a convex curve relative to queue growth (the square curve in the Classic queue, ifU=1).U = 1). Like RED, the curve hugs the zero axis while the queue is shallow. Then, as load increases, it introduces a growing barrier to higher delay. But, unlike RED, it requires only two parameters, not three. The disadvantage of Curvy RED (compared to a PIcontrollercontroller, for example) is that it is not adapted to a wide range of RTTs. Curvy RED can be used as is when the RTT range to be supported islimited, otherwiselimited; otherwise, an adaptation mechanism is needed.</t> <t>From our limited experiments with Curvy RED so far, recommended values of these parameters are: S_C = -1; g_C = 5; T = 5 * MTU at the link rate (about1ms1 ms at60Mb/s)60 Mb/s) for the range of base RTTs typical on the public Internet. <xref target="CRED_Insights" format="default"/> explains why these parameters are applicable whatever rate link this AQM implementation is deployed on and how the parameters would need to be adjusted for a scenario with a different range of RTTs(e.g. a(e.g., a data centre). The setting of k depends on policy (see <xref target="dualq_norm_reqs" format="default"/> and <xref target="dualq_Choosing_k"format="default"/> respectivelyformat="default"/>, respectively, for its recommended setting and guidance on alternatives).</t> <t>There is also a cUrviness parameter, U, which is a small positive integer. It is likely to take the same hard-coded value for all implementations, once experiments have determined a good value. OnlyU=1U = 1 has been used in experiments so far, but results might be even better withU=2U = 2 or higher.</t> <t>Notes:</t> <ol spacing="normal"type="1"><litype="1"> <li anchor="dualq_note_dequeue">The alternative of applying the AQMs at enqueue would shift some processing from the critical time when each packet is dequeued. However, it would also add a whole queue of delay to the control signals, making the control loop sloppier (for a typicalRTTRTT, it would double the Classic queue's feedback delay). On a platform where packet timestamping is feasible,e.g. Linux,e.g., Linux, it is also easiest to apply the AQMs atdequeuedequeue, because that is where queuing time is also measured.</li> <li anchor="dualq_note_conditional_priority">WRR better isolates the L4S queue from large delay bursts in the Classic queue, but it is slightly less simple than TS-FIFO. If WRR were used, a low default Classic weight(e.g. 1/16)(e.g., 1/16) would need to be configured in place of thetime shifttime-shift in line 5 of the initialization function (<xref target="dualq_fig_Algo_cred_core_header" format="default"/>).</li> <li anchor="dualq_note_step">A step function is shown for simplicity. A ramp function (see <xref target="dualq_fig_Algo_laqm_core" format="default"/> and the discussion around it in <xref target="dualq_Ex_algo_pi2-1" format="default"/>) is recommended, because it is more general than a step and has the potential to enable L4S congestion controls to converge more rapidly.</li> <li anchor="dualq_note_non-EWMA">An EWMA is only one possible way to filter bursts; other more adaptive smoothing methods could bevalidvalid, and it might be appropriate to decrease the EWMA faster than it increases,e.g. bye.g., by using the minimum of the smoothed and instantaneous queue delays, min(Q_C, qc.time()).</li> </ol> </section> <section numbered="true" toc="default"> <name>Efficient Implementation of Curvy RED</name> <t>Although code optimization depends on the platform, the following notes explain where the design of Curvy RED was particularly motivated by efficient implementation.</t> <t>The Classic AQM at line 10b in <xref target="dualq_fig_Algo_Real" format="default"/> calls maxrand(2*U), which gives twice as much curviness as the call to maxrand(U) in the marking function at line 5c. This is the trick that implements the square rule in equation (1) (<xref target="dualq_coupled" format="default"/>). This is based on the fact that, given a number X from 1 to 6, the probability that two dice throws will both be less than X is the square of the probability that one throw will be less than X. So, whenU=1,U = 1, the L4S marking function is linear and the Classic dropping function is squared. IfU=2,U = 2, L4S would be a square function and Classic would be quartic. And so on.</t> <t>The maxrand(u) function in lines16-2122-27 simply generates u random numbers and returns the maximum. Typically, maxrand(u) could be run in parallel out of band. For instance, ifU=1,U = 1, the Classic queue would require the maximum of two random numbers. So, instead of calling maxrand(2*U) in-band, the maximum of every pair of values from a pseudorandom number generator could be generatedout-of-band,out of band and held in a buffer ready for the Classic queue to consume.</t> <figure anchor="dualq_fig_Algo_Int"> <name>Optimised Example Dequeue Pseudocode for DualQ Coupled AQM using Integer Arithmetic</name><artwork name="" type="" align="left" alt=""><![CDATA[1:<sourcecode><![CDATA[ 1: cred_dequeue(lq, cq, pkt) { % Couples L4S & Classic queues 2: while ( lq.byt() + cq.byt() > 0 ) { 3: if ( scheduler() == lq ) { 4: lq.dequeue(pkt) % L4S scheduled 5: if ((lq.time() > T) OR (Q_C >> (S_L-2) > maxrand(U))) 6: mark(pkt) 7: } else { 8: cq.dequeue(pkt) % Classic scheduled 9: Q_C += (qc.ns() - Q_C) >> g_C % Classic Q EWMA 10: if ( (Q_C >> (S_C-2) ) > maxrand(2*U) ) { 11: if ( (ecn(pkt) == 0) { % ECN field = not-ECT 12: drop(pkt) % Squared drop, redo loop 13: continue % continue to the top of the while loop 14: } 15: mark(pkt) 16: } 17: } 18: return(pkt) % return the packet and stop here 19: } 20: return(NULL) % no packet to dequeue 21: }]]></artwork>]]></sourcecode> </figure> <t>The two ranges, range_L andrange_Crange_C, are expressed as powers of 2 so that division can be implemented as a right bit-shift (>>) in lines 5 and 10 of the integer variant of the pseudocode (<xref target="dualq_fig_Algo_Int" format="default"/>).</t> <t>For the integer variant of the pseudocode, an integer version of the rand() function used at line 25 of themaxrand(function)maxrand() function in <xref target="dualq_fig_Algo_Real" format="default"/> would be arranged to return an integer in the range 0 <= maxrand() < 2^32 (not shown). This would scale up all the floating point probabilities in the range [0,1] by 2^32.</t> <t>Queuing delays are also scaled up by 2^32, but in two stages: i)Inin line99, queuing time qc.ns() is returned in integer nanoseconds, making the value about 2^30 times larger than when the units were seconds,ii)and then ii) in lines 5 and1010, an adjustment of -2 to the right bit-shift multiplies the result by 2^2, to complete the scaling by 2^32.</t> <t>In line 8 of the initialization function, the EWMA constant gamma is represented as an integer power of 2, g_C, so that in line 9 of the integer code (<xref target="dualq_fig_Algo_Int" format="default"/>), the division needed to weight the moving average can be implemented by a right bit-shift (>> g_C).</t> </section> </section> <section numbered="true" toc="default"> <name>Choice of Coupling Factor, k</name> <t/> <section anchor="dualq_rtt-dependence" numbered="true" toc="default"> <name>RTT-Dependence</name> <t>Where Classic flows compete for the same capacity, their relative flow rates depend not only on the congestionprobability,probability but also on their end-to-end RTT (= base RTT + queue delay). The rates ofReno <xrefReno <xref target="RFC5681" format="default"/> flows competing over an AQM are roughly inversely proportional to their RTTs.CubicCUBIC exhibits similar RTT-dependence when inReno-compatibilityReno-friendly mode, but it is less RTT-dependent otherwise.</t> <t>Until the early experiments with the DualQ Coupled AQM, the importance of the reasonably large Classic queue in mitigating RTT-dependence when the base RTT is low had not been appreciated. AppendixA.1.6<xref target="RFC9331" sectionFormat="bare" section="A.1.6"/> of the L4S ECNprotocol <xref target="I-D.ietf-tsvwg-ecn-l4s-id"Protocol <xref target="RFC9331" format="default"/> uses numerical examples to explain why bloated buffers had concealed the RTT-dependence of Classic congestion controls before that time.ThenThen, it explains why, the more that queuing delays have reduced, the more that RTT-dependence has surfaced as a potential starvation problem for long RTT flows, when competing against very short RTT flows.</t> <t>Given that congestion control onend-systemsend systems is voluntary, there is no reason why it has to be voluntarily RTT-dependent. The RTT-dependence of existing Classic traffic cannot be 'undeployed'. Therefore, <xreftarget="I-D.ietf-tsvwg-ecn-l4s-id"target="RFC9331" format="default"/> requires L4S congestion controls to be significantly less RTT-dependent than the standard Reno congestioncontrol <xrefcontrol <xref target="RFC5681" format="default"/>, at least at low RTT. Then RTT-dependence ought to be no worse than it is with appropriately sized Classic buffers. Following this approach means there is no need for network devices to address RTT-dependence, although there would be no harm if they did, which per-flow queuing inherently does.</t> </section> <section anchor="dualq_Choosing_k" numbered="true" toc="default"> <name>Guidance on Controlling Throughput Equivalence</name> <t>The coupling factor, k, determines the balance between L4S and Classic flow rates (see <xref target="dualq_config" format="default"/> and equation(1)).</t>(1) in <xref target="dualq_coupled" format="default"/>).</t> <t>For the public Internet, a coupling factor ofk=2k = 2 isrecommended,recommended and justified below. For scenarios other than the public Internet, a good coupling factor can be derived by plugging the appropriate numbers into the same working.</t> <t>To summarize the maths below, from equation (7) it can be seen that choosingk=1.64k = 1.64 would theoretically make L4S throughput roughly the same as Classic, <em>if their actual end-to-end RTTs were the same</em>. However, even if the base RTTs are the same, the actual RTTs are unlikely to be the same, because Classic traffic needs a fairly large queue to avoidunder-utilizationunderutilization and excessdrop. Whereasdrop, whereas L4S does not.</t> <t>Therefore, to determine the appropriate coupling factor policy, the operator needs to decide at what base RTT it wants L4S and Classic flows to have roughly equal throughput, once the effect of the additional Classic queue on Classic throughput has been taken into account. With this approach, a network operator can determine a good coupling factor without knowing the precise L4S algorithm for reducing RTT-dependence--- or even in the absence of any algorithm.</t> <t>The following additional terminology will be used, with appropriate subscripts:</t> <dl newline="false" spacing="normal"> <dt>r:</dt> <dd>Packet rate [pkt/s]</dd> <dt>R:</dt> <dd>RTT [s/round]</dd> <dt>p:</dt><dd>ECN marking<dd>ECN-marking probability []</dd> </dl> <t>On the Classic side, we consider Reno as the most sensitive and therefore worst-case Classic congestion control. We will also considerCubicCUBIC in its Reno-friendly mode('CReno'),('CReno') as the most prevalent congestion control, according to the references and analysis in <xref target="PI2param" format="default"/>. In either case, the Classic packet rate in steady state is given by the well-known square root formula for Reno congestion control:</t><artwork name="" type="" align="left" alt=""><![CDATA[<sourcecode><![CDATA[ r_C = 1.22 / (R_C * p_C^0.5)(5)]]></artwork>(5)]]></sourcecode> <t>On the L4S side, we consider the Prague congestioncontrol <xrefcontrol <xref target="I-D.briscoe-iccrg-prague-congestion-control" format="default"/> as the reference for steady-state dependence on congestion. Prague conforms to the same equation as DCTCP, but we do not use the equation derived in the DCTCP paper, which is only appropriate for step marking. The coupled marking, p_CL, is the appropriate one when considering throughput equivalence with Classic flows. Unlike step marking, coupled markings are inherently spaced out, so we use the formula for DCTCP packet rate with probabilistic marking derived in Appendix A of <xref target="PI2" format="default"/>. We use the equation without RTT-independence enabled, which will be explained later.</t><artwork name="" type="" align="left" alt=""><![CDATA[<sourcecode><![CDATA[ r_L = 2 / (R_L * p_CL)(6)]]></artwork>(6)]]></sourcecode> <t>For packet rate equivalence, we equate the two packet rates and rearrange the equation into the same form asEquation (1),equation (1) (copied from <xref target="dualq_coupled" format="default"/>) so the two can be equated and simplified to produce a formula for a theoretical coupling factor, which we shall call k*:</t><artwork name="" type="" align="left" alt=""><![CDATA[<sourcecode><![CDATA[ r_c = r_L => p_C = (p_CL/1.64 *R_L/R_C)^2R_L/R_C)^2. p_C = ( p_CL / k)^2)^2. (1) k* = 1.64 * (R_C /R_L)R_L). (7)]]></artwork>]]></sourcecode> <t>We say that this coupling factor is theoretical, because it is in terms of two RTTs, which raises two practical questions: i) for multiple flows with different RTTs, the RTT for each traffic class would have to be derived from the RTTs of all the flows in that class (actually the harmonic mean would beneeded);needed) and ii) a network node cannot easily know the RTT of the flows anyway.</t> <t>RTT-dependence is caused by window-based congestion control, so it ought to be reversed there, not in the network. Therefore, we use a fixed coupling factor in thenetwork,network and reduce RTT-dependence in L4S senders. We cannot expect Classic senders to all be updated to reduce their RTT-dependence. But solely addressing the problem in L4S senders at least makes RTT-dependence no worse--- not just between L4S senders, but also between L4S and Classic senders.</t><t>Traditionally, throughput<t>Throughput equivalencehas beenis defined for flows under comparable conditions, including with the same baseRTT <xrefRTT <xref target="RFC2914" format="default"/>. So if we assume the same base RTT, R_b, for comparable flows, we can put both R_C and R_L in terms of R_b.</t> <t>We can approximate the L4S RTT to be hardly greater than the base RTT,i.e. R_Li.e., R_L ~= R_b. And we can replace R_C with (R_b + q_C), where the Classic queue, q_C, depends on the target queue delay that the operator has configured for the Classic AQM.</t> <t>Taking PI2 as an example Classic AQM, it seems that we could just take R_C = R_b + target (recommended 15 ms by default in <xref target="dualq_Ex_algo_pi2-1" format="default"/>). However, target is roughly the queue depth reached by the tips of the sawteeth of a congestion control, not the average <xref target="PI2param" format="default"/>. That is R_max = R_b + target.</t> <t>The position of the average in relation to the max depends on the amplitude and geometry of the sawteeth. We consider two examples:Reno <xrefReno <xref target="RFC5681" format="default"/>, as the most sensitiveworst-case,worst case, andCubic <xrefCUBIC <xref target="RFC8312" format="default"/> in its Reno-friendly mode ('CReno') as the most prevalent congestion control algorithm on the Internet according to the references in <xref target="PI2param" format="default"/>. Both areAIMD,Additive Increase Multiplicative Decrease (AIMD), so we will generalize using b as the multiplicative decrease factor (b_r = 0.5 for Reno, b_c = 0.7 for CReno).Then:</t> <artwork name="" type="" align="left" alt=""><![CDATA[Then</t> <sourcecode><![CDATA[ R_C = (R_max + b*R_max) / 2 = R_max *(1+b)/2(1+b)/2. R_reno = 0.75 * (R_b + target); R_creno = 0.85 * (R_b + target). (8)]]></artwork>]]></sourcecode> <t>Plugging all this into equation(7)(7), at any particular base RTT, R_b, we get a fixed coupling factor for each:</t><artwork name="" type="" align="left" alt=""><![CDATA[k_reno<sourcecode><![CDATA[ k_reno = 1.64*0.75*(R_b+target)/R_b = 1.23*(1 + target/R_b); k_creno = 1.39 * (1 +target/R_b) ]]></artwork>target/R_b). ]]></sourcecode> <t>An operator can then choose the base RTT at which it wants throughput to be equivalent. For instance, if we recommend that the operator chooses R_b = 25 ms, as a typical base RTT between Internet users andCDNs <xrefCDNs <xref target="PI2param" format="default"/>, then these coupling factors become:</t><artwork name="" type="" align="left" alt=""><![CDATA[k_reno<sourcecode><![CDATA[ k_reno = 1.23 * (1 + 15/25) k_creno = 1.39 * (1 + 15/25) = 1.97 = 2.22 ~=22. ~=22. (9)]]></artwork>]]></sourcecode> <t>The approximation is relevant to any of the above example DualQ Coupled algorithms, which use a coupling factor that is an integer power of 2 to aid efficient implementation. It also fits besttofor the worst case (Reno).</t> <t>To check the outcome of this coupling factor, we can express the ratio of L4S to Classic throughput by substituting from their rate equations (5) and (6), then also substituting for p_C in terms ofp_CL,p_CL using equation (1) withk=2k = 2 as just determined for the Internet:</t><artwork name="" type="" align="left" alt=""><![CDATA[r_L<sourcecode><![CDATA[ r_L / r_C = 2 (R_C * p_C^0.5) / 1.22 (R_L * p_CL) = (R_C * p_CL) / (1.22 * R_L * p_CL) = R_C / (1.22 *R_L)R_L). (10)]]></artwork>]]></sourcecode> <t>As an example, we can then consider single competing CReno and Prague flows, by expressing both their RTTs in (10) in terms of their base RTTs, R_bC and R_bL. So R_C is replaced by equation (8) for CReno. And R_L is replaced by the max() function below, which represents the effective RTT of the current Prague congestioncontrol <xrefcontrol <xref target="I-D.briscoe-iccrg-prague-congestion-control" format="default"/> in its (default) RTT-independent mode, because it sets a floor to the effective RTT that it uses for additive increase:</t><artwork name="" type="" align="left" alt=""><![CDATA[<sourcecode><![CDATA[ r_L / r_C ~= 0.85 * (R_bC + target) / (1.22 * max(R_bL, R_typ)) ~= (R_bC + target) / (1.4 * max(R_bL,R_typ)) ]]></artwork>R_typ)). ]]></sourcecode> <t>It can be seen that, for base RTTs below target (15 ms), both the numerator and the denominator plateau, which has the desired effect of limiting RTT-dependence.</t> <t>At the start of the above derivations, an explanation was promised for why the L4S throughput equation in equation (6) did not need to model RTT-independence. This is because we only use one point--- at the typical base RTT where the operator chooses to calculate the coupling factor.Then,Then throughput equivalence will at least hold at that chosen point. Nonetheless, assuming Prague senders implement RTT-independence over a range of RTTs below this, the throughput equivalence will then extend over that range as well.</t> <t>Congestion control designers can choose different ways to reduce RTT-dependence. And each operator can make a policy choice to decide on a different base RTT, and therefore a different k, at which it wants throughput equivalence. Nonetheless, for the Internet, it makes sense to choose what is believed to be the typical RTT most users experience, because a Classic AQM's target queuing delay is also derived from a typical RTT for the Internet.</t> <t>As a non-Internet example, for localized traffic from a particular ISP's data centre, using the measured RTTs, it was calculated that a value of k = 8 would achieve throughput equivalence, and experiments verified the formula very closely.</t> <t>But, for a typical mix of RTTs across the general Internet, a value ofk=2k = 2 is recommended as a good workable compromise.</t> </section> </section><!-- <section title="Open Issues"> <t>Minor open issues are tagged '{ToDo}' at the appropriate point in the document. Major open issues are listed below:<list> <t>None</t> </list></t> </section> <section title="Change Log (to be Deleted before Publication)"> <t>A detailed version history can be accessed at <http://datatracker.ietf.org/doc/draft-briscoe-aqm-ecn-roadmap/history/></t> <t><list style="hanging"> <t hangText="From briscoe-...-00 to briscoe-...-01:">Technical changes:<list style="symbols"> <t/> </list>Editorial changes:<list style="symbols"> <t/> </list></t> </list></t> </section> --><section numbered="false" toc="default"> <name>Acknowledgements</name> <t>Thanks toAnil Agarwal, Sowmini Varadhan, Gabi Bracha, Nicolas Kuhn, Greg Skinner, Tom Henderson, David Pullen, Mirja Kuehlewind, Gorry Fairhurst, Pete Heist, Ermin Sakic and Martin Duke<contact fullname="Anil Agarwal"/>, <contact fullname="Sowmini Varadhan"/>, <contact fullname="Gabi Bracha"/>, <contact fullname="Nicolas Kuhn"/>, <contact fullname="Greg Skinner"/>, <contact fullname="Tom Henderson"/>, <contact fullname="David Pullen"/>, <contact fullname="Mirja Kühlewind"/>, <contact fullname="Gorry Fairhurst"/>, <contact fullname="Pete Heist"/>, <contact fullname="Ermin Sakic"/>, and <contact fullname="Martin Duke"/> for detailed reviewcommentscomments, particularly of theappendicesappendices, and suggestions on how to make the explanations clearer. Thanks also toTom Henderson<contact fullname="Tom Henderson"/> forinsightsinsight on the choice of schedulers and queue delay measurement techniques. And thanks to the area reviewersChrister Holmberg, Lars Eggert and Roman Danyliw.</t><contact fullname="Christer Holmberg"/>, <contact fullname="Lars Eggert"/>, and <contact fullname="Roman Danyliw"/>.</t> <t>The early contributions ofKoen<contact fullname="Koen DeSchepper, Bob Briscoe, Olga BondarenkoSchepper"/>, <contact fullname="Bob Briscoe"/>, <contact fullname="Olga Bondarenko"/>, andInton Tsang<contact fullname="Inton Tsang"/> werepart-fundedpartly funded by the European Community under its Seventh Framework Programme through the Reducing Internet Transport Latency (RITE) project (ICT-317700). Contributions ofKoen<contact fullname="Koen DeSchepperSchepper"/> andOlivier Tilmans<contact fullname="Olivier Tilmans"/> were alsopart-fundedpartly funded by the 5Growth and DAEMON EU H2020 projects.Bob Briscoe's<contact fullname="Bob Briscoe"/>'s contribution was alsopart-fundedpartly funded by the Comcast Innovation Fund and the Research Council of Norway through the TimeIn project. The views expressed here are solely those of the authors.</t> </section> <section numbered="false" toc="default"> <name>Contributors</name> <t>The following contributed implementations and evaluations that validated and helped to improve this specification:</t><ul empty="true" spacing="normal"> <li>Olga Albisser<t><contact fullname="Olga Albisser"/> <olga@albisser.org> of Simula Research Lab, Norway (Olga Bondarenko during earlydrafts)draft versions) implemented the prototype DualPI2 AQM for Linux with Koen De Schepper and conducted extensive evaluations as well as implementing the live performance visualizationGUI <xrefGUI <xref target="L4Sdemo16"format="default"/>.</li> <li>Olivier Tilmansformat="default"/>.</t> <t><contact fullname="Olivier Tilmans"/> <olivier.tilmans@nokia-bell-labs.com> of Nokia Bell Labs, Belgium prepared and maintains the Linux implementation of DualPI2 forupstreaming.</li> <li>Shravya K.S.upstreaming.</t> <t><contact fullname="Shravya K.S."/> wrote a model for the ns-3 simulator based onthe -01draft-ietf-tsvwg-aqm-dualq-coupled-01 (a draft version of thisInternet-Draft.document). Based on this initial work,Tom Henderson<contact fullname="Tom Henderson"/> <tomh@tomh.org> updated that earlier model and created a model for the DualQ variant specified as part of the Low Latency DOCSIS specification, as well as conducting extensiveevaluations.</li> <li>Ingevaluations.</t> <t><contact fullname="Ing Jyh (Inton)TsangTsang"/> of Nokia, Belgium built the End-to-End Data Centre to the Home broadband testbed on which DualQ Coupled AQM implementations weretested.</li> </ul>tested.</t> </section> </back> </rfc>