idnits 2.17.1 draft-ietf-tsvwg-ecn-l4s-id-24.txt: Checking boilerplate required by RFC 5378 and the IETF Trust (see https://trustee.ietf.org/license-info): ---------------------------------------------------------------------------- No issues found here. Checking nits according to https://www.ietf.org/id-info/1id-guidelines.txt: ---------------------------------------------------------------------------- == There are 2 instances of lines with non-ascii characters in the document. Checking nits according to https://www.ietf.org/id-info/checklist : ---------------------------------------------------------------------------- == There are 1 instance of lines with non-RFC6890-compliant IPv4 addresses in the document. If these are example addresses, they should be changed. Miscellaneous warnings: ---------------------------------------------------------------------------- == The copyright year in the IETF Trust and authors Copyright Line does not match the current year -- The document date (1 February 2022) is 815 days in the past. Is this intentional? -- Found something which looks like a code comment -- if you have code sections in the document, please surround them with '' and '' lines. Checking references for intended status: Experimental ---------------------------------------------------------------------------- -- Looks like a reference, but probably isn't: '1' on line 1510 == Missing Reference: 'RFCXXXX' is mentioned on line 1510, but not defined == Outdated reference: A later version (-07) exists of draft-briscoe-docsis-q-protection-02 == Outdated reference: A later version (-03) exists of draft-briscoe-iccrg-prague-congestion-control-00 == Outdated reference: A later version (-02) exists of draft-cardwell-iccrg-bbr-congestion-control-01 == Outdated reference: A later version (-28) exists of draft-ietf-tcpm-accurate-ecn-15 == Outdated reference: A later version (-15) exists of draft-ietf-tcpm-generalized-ecn-09 == Outdated reference: A later version (-25) exists of draft-ietf-tsvwg-aqm-dualq-coupled-20 == Outdated reference: A later version (-22) exists of draft-ietf-tsvwg-ecn-encap-guidelines-16 == Outdated reference: A later version (-20) exists of draft-ietf-tsvwg-l4s-arch-15 == Outdated reference: A later version (-06) exists of draft-ietf-tsvwg-l4sops-02 == Outdated reference: A later version (-22) exists of draft-ietf-tsvwg-nqb-08 == Outdated reference: A later version (-23) exists of draft-ietf-tsvwg-rfc6040update-shim-14 == Outdated reference: A later version (-07) exists of draft-stewart-tsvwg-sctpecn-05 -- Obsolete informational reference (is this intentional?): RFC 2309 (Obsoleted by RFC 7567) -- Obsolete informational reference (is this intentional?): RFC 4960 (Obsoleted by RFC 9260) -- Obsolete informational reference (is this intentional?): RFC 6347 (Obsoleted by RFC 9147) -- Obsolete informational reference (is this intentional?): RFC 8312 (Obsoleted by RFC 9438) Summary: 0 errors (**), 0 flaws (~~), 16 warnings (==), 7 comments (--). Run idnits with the --verbose option for more detailed information about the items above. -------------------------------------------------------------------------------- 2 Transport Services (tsv) K. De Schepper 3 Internet-Draft Nokia Bell Labs 4 Intended status: Experimental B. Briscoe, Ed. 5 Expires: 5 August 2022 Independent 6 1 February 2022 8 Explicit Congestion Notification (ECN) Protocol for Very Low Queuing 9 Delay (L4S) 10 draft-ietf-tsvwg-ecn-l4s-id-24 12 Abstract 14 This specification defines the protocol to be used for a new network 15 service called low latency, low loss and scalable throughput (L4S). 16 L4S uses an Explicit Congestion Notification (ECN) scheme at the IP 17 layer that is similar to the original (or 'Classic') ECN approach, 18 except as specified within. L4S uses 'scalable' congestion control, 19 which induces much more frequent control signals from the network and 20 it responds to them with much more fine-grained adjustments, so that 21 very low (typically sub-millisecond on average) and consistently low 22 queuing delay becomes possible for L4S traffic without compromising 23 link utilization. Thus even capacity-seeking (TCP-like) traffic can 24 have high bandwidth and very low delay at the same time, even during 25 periods of high traffic load. 27 The L4S identifier defined in this document distinguishes L4S from 28 'Classic' (e.g. TCP-Reno-friendly) traffic. It gives an incremental 29 migration path so that suitably modified network bottlenecks can 30 distinguish and isolate existing traffic that still follows the 31 Classic behaviour, to prevent it degrading the low queuing delay and 32 low loss of L4S traffic. This specification defines the rules that 33 L4S transports and network elements need to follow with the intention 34 that L4S flows neither harm each other's performance nor that of 35 Classic traffic. Examples of new active queue management (AQM) 36 marking algorithms and examples of new transports (whether TCP-like 37 or real-time) are specified separately. 39 Status of This Memo 41 This Internet-Draft is submitted in full conformance with the 42 provisions of BCP 78 and BCP 79. 44 Internet-Drafts are working documents of the Internet Engineering 45 Task Force (IETF). Note that other groups may also distribute 46 working documents as Internet-Drafts. The list of current Internet- 47 Drafts is at https://datatracker.ietf.org/drafts/current/. 49 Internet-Drafts are draft documents valid for a maximum of six months 50 and may be updated, replaced, or obsoleted by other documents at any 51 time. It is inappropriate to use Internet-Drafts as reference 52 material or to cite them other than as "work in progress." 54 This Internet-Draft will expire on 5 August 2022. 56 Copyright Notice 58 Copyright (c) 2022 IETF Trust and the persons identified as the 59 document authors. All rights reserved. 61 This document is subject to BCP 78 and the IETF Trust's Legal 62 Provisions Relating to IETF Documents (https://trustee.ietf.org/ 63 license-info) in effect on the date of publication of this document. 64 Please review these documents carefully, as they describe your rights 65 and restrictions with respect to this document. Code Components 66 extracted from this document must include Revised BSD License text as 67 described in Section 4.e of the Trust Legal Provisions and are 68 provided without warranty as described in the Revised BSD License. 70 Table of Contents 72 1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . 4 73 1.1. Latency, Loss and Scaling Problems . . . . . . . . . . . 5 74 1.2. Terminology . . . . . . . . . . . . . . . . . . . . . . . 7 75 1.3. Scope . . . . . . . . . . . . . . . . . . . . . . . . . . 9 76 2. Choice of L4S Packet Identifier: Requirements . . . . . . . . 10 77 3. L4S Packet Identification . . . . . . . . . . . . . . . . . . 11 78 4. Transport Layer Behaviour (the 'Prague Requirements') . . . . 11 79 4.1. Codepoint Setting . . . . . . . . . . . . . . . . . . . . 12 80 4.2. Prerequisite Transport Feedback . . . . . . . . . . . . . 12 81 4.3. Prerequisite Congestion Response . . . . . . . . . . . . 13 82 4.3.1. Guidance on Congestion Response in the RFC Series . . 16 83 4.4. Filtering or Smoothing of ECN Feedback . . . . . . . . . 19 84 5. Network Node Behaviour . . . . . . . . . . . . . . . . . . . 19 85 5.1. Classification and Re-Marking Behaviour . . . . . . . . . 19 86 5.2. The Strength of L4S CE Marking Relative to Drop . . . . . 21 87 5.3. Exception for L4S Packet Identification by Network Nodes 88 with Transport-Layer Awareness . . . . . . . . . . . . . 22 89 5.4. Interaction of the L4S Identifier with other 90 Identifiers . . . . . . . . . . . . . . . . . . . . . . . 22 91 5.4.1. DualQ Examples of Other Identifiers Complementing L4S 92 Identifiers . . . . . . . . . . . . . . . . . . . . . 22 93 5.4.1.1. Inclusion of Additional Traffic with L4S . . . . 22 94 5.4.1.2. Exclusion of Traffic From L4S Treatment . . . . . 24 95 5.4.1.3. Generalized Combination of L4S and Other 96 Identifiers . . . . . . . . . . . . . . . . . . . . 25 98 5.4.2. Per-Flow Queuing Examples of Other Identifiers 99 Complementing L4S Identifiers . . . . . . . . . . . . 26 100 5.5. Limiting Packet Bursts from Links . . . . . . . . . . . . 27 101 5.5.1. Limiting Packet Bursts from Links Fed by an L4S 102 AQM . . . . . . . . . . . . . . . . . . . . . . . . . 27 103 5.5.2. Limiting Packet Bursts from Links Upstream of an L4S 104 AQM . . . . . . . . . . . . . . . . . . . . . . . . . 28 105 6. Behaviour of Tunnels and Encapsulations . . . . . . . . . . . 28 106 6.1. No Change to ECN Tunnels and Encapsulations in General . 28 107 6.2. VPN Behaviour to Avoid Limitations of Anti-Replay . . . . 29 108 7. L4S Experiments . . . . . . . . . . . . . . . . . . . . . . . 30 109 7.1. Open Questions . . . . . . . . . . . . . . . . . . . . . 30 110 7.2. Open Issues . . . . . . . . . . . . . . . . . . . . . . . 32 111 7.3. Future Potential . . . . . . . . . . . . . . . . . . . . 32 112 8. IANA Considerations . . . . . . . . . . . . . . . . . . . . . 33 113 9. Security Considerations . . . . . . . . . . . . . . . . . . . 33 114 10. Acknowledgements . . . . . . . . . . . . . . . . . . . . . . 34 115 11. References . . . . . . . . . . . . . . . . . . . . . . . . . 34 116 11.1. Normative References . . . . . . . . . . . . . . . . . . 34 117 11.2. Informative References . . . . . . . . . . . . . . . . . 35 118 Appendix A. Rationale for the 'Prague L4S Requirements' . . . . 45 119 A.1. Rationale for the Requirements for Scalable Transport 120 Protocols . . . . . . . . . . . . . . . . . . . . . . . . 45 121 A.1.1. Use of L4S Packet Identifier . . . . . . . . . . . . 46 122 A.1.2. Accurate ECN Feedback . . . . . . . . . . . . . . . . 46 123 A.1.3. Capable of Replacement by Classic Congestion 124 Control . . . . . . . . . . . . . . . . . . . . . . . 46 125 A.1.4. Fall back to Classic Congestion Control on Packet 126 Loss . . . . . . . . . . . . . . . . . . . . . . . . 46 127 A.1.5. Coexistence with Classic Congestion Control at Classic 128 ECN bottlenecks . . . . . . . . . . . . . . . . . . . 47 129 A.1.6. Reduce RTT dependence . . . . . . . . . . . . . . . . 51 130 A.1.7. Scaling down to fractional congestion windows . . . . 51 131 A.1.8. Measuring Reordering Tolerance in Time Units . . . . 53 132 A.2. Scalable Transport Protocol Optimizations . . . . . . . . 55 133 A.2.1. Setting ECT in Control Packets and Retransmissions . 55 134 A.2.2. Faster than Additive Increase . . . . . . . . . . . . 56 135 A.2.3. Faster Convergence at Flow Start . . . . . . . . . . 57 136 Appendix B. Compromises in the Choice of L4S Identifier . . . . 57 137 Appendix C. Potential Competing Uses for the ECT(1) Codepoint . 62 138 C.1. Integrity of Congestion Feedback . . . . . . . . . . . . 62 139 C.2. Notification of Less Severe Congestion than CE . . . . . 63 140 Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . 63 142 1. Introduction 144 This specification defines the protocol to be used for a new network 145 service called low latency, low loss and scalable throughput (L4S). 146 L4S uses an Explicit Congestion Notification (ECN) scheme at the IP 147 layer with the same set of codepoint transitions as the original (or 148 'Classic') Explicit Congestion Notification (ECN [RFC3168]). RFC 149 3168 required an ECN mark to be equivalent to a drop, both when 150 applied in the network and when responded to by a transport. Unlike 151 Classic ECN marking, the network applies L4S marking more immediately 152 and more aggressively than drop, and the transport response to each 153 mark is reduced and smoothed relative to that for drop. The two 154 changes counterbalance each other so that the throughput of an L4S 155 flow will be roughly the same as a comparable non-L4S flow under the 156 same conditions. Nonetheless, the much more frequent ECN control 157 signals and the finer responses to these signals result in very low 158 queuing delay without compromising link utilization, and this low 159 delay can be maintained during high load. For instance, queuing 160 delay under heavy and highly varying load with the example DCTCP/ 161 DualQ solution cited below on a DSL or Ethernet link is sub- 162 millisecond on average and roughly 1 to 2 milliseconds at the 99th 163 percentile without losing link utilization [DualPI2Linux], [DCttH19]. 164 Note that the inherent queuing delay while waiting to acquire a 165 discontinuous medium such as WiFi has to be minimized in its own 166 right, so it would be additional to the above (see section 6.3 of 167 [I-D.ietf-tsvwg-l4s-arch]). 169 L4S relies on 'scalable' congestion controls for these delay 170 properties and for preserving low delay as flow rate scales, hence 171 the name. The congestion control used in Data Center TCP (DCTCP) is 172 an example of a scalable congestion control, but DCTCP is applicable 173 solely to controlled environments like data centres [RFC8257], 174 because it is too aggressive to co-exist with existing TCP-Reno- 175 friendly traffic. The DualQ Coupled AQM, which is defined in a 176 complementary experimental specification 177 [I-D.ietf-tsvwg-aqm-dualq-coupled], is an AQM framework that enables 178 scalable congestion controls derived from DCTCP to co-exist with 179 existing traffic, each getting roughly the same flow rate when they 180 compete under similar conditions. Note that a scalable congestion 181 control is still not safe to deploy on the Internet unless it 182 satisfies the requirements listed in Section 4. 184 L4S is not only for elastic (TCP-like) traffic - there are scalable 185 congestion controls for real-time media, such as the L4S variant of 186 the SCReAM [RFC8298] real-time media congestion avoidance technique 187 (RMCAT). The factor that distinguishes L4S from Classic traffic is 188 its behaviour in response to congestion. The transport wire 189 protocol, e.g. TCP, QUIC, SCTP, DCCP, RTP/RTCP, is orthogonal (and 190 therefore not suitable for distinguishing L4S from Classic packets). 192 The L4S identifier defined in this document is the key piece that 193 distinguishes L4S from 'Classic' (e.g. Reno-friendly) traffic. It 194 gives an incremental migration path so that suitably modified network 195 bottlenecks can distinguish and isolate existing Classic traffic from 196 L4S traffic to prevent the former from degrading the very low delay 197 and loss of the new scalable transports, without harming Classic 198 performance at these bottlenecks. Initial implementation of the 199 separate parts of the system has been motivated by the performance 200 benefits. 202 1.1. Latency, Loss and Scaling Problems 204 Latency is becoming the critical performance factor for many (most?) 205 applications on the public Internet, e.g. interactive Web, Web 206 services, voice, conversational video, interactive video, interactive 207 remote presence, instant messaging, online gaming, remote desktop, 208 cloud-based applications, and video-assisted remote control of 209 machinery and industrial processes. In the 'developed' world, 210 further increases in access network bit-rate offer diminishing 211 returns, whereas latency is still a multi-faceted problem. In the 212 last decade or so, much has been done to reduce propagation time by 213 placing caches or servers closer to users. However, queuing remains 214 a major intermittent component of latency. 216 The Diffserv architecture provides Expedited Forwarding [RFC3246], so 217 that low latency traffic can jump the queue of other traffic. If 218 growth in high-throughput latency-sensitive applications continues, 219 periods with solely latency-sensitive traffic will become 220 increasingly common on links where traffic aggregation is low. For 221 instance, on the access links dedicated to individual sites (homes, 222 small enterprises or mobile devices). These links also tend to 223 become the path bottleneck under load. During these periods, if all 224 the traffic were marked for the same treatment, at these bottlenecks 225 Diffserv would make no difference. Instead, it becomes imperative to 226 remove the underlying causes of any unnecessary delay. 228 The bufferbloat project has shown that excessively-large buffering 229 ('bufferbloat') has been introducing significantly more delay than 230 the underlying propagation time. These delays appear only 231 intermittently--only when a capacity-seeking (e.g. TCP) flow is long 232 enough for the queue to fill the buffer, making every packet in other 233 flows sharing the buffer sit through the queue. 235 Active queue management (AQM) was originally developed to solve this 236 problem (and others). Unlike Diffserv, which gives low latency to 237 some traffic at the expense of others, AQM controls latency for _all_ 238 traffic in a class. In general, AQM methods introduce an increasing 239 level of discard from the buffer the longer the queue persists above 240 a shallow threshold. This gives sufficient signals to capacity- 241 seeking (aka. greedy) flows to keep the buffer empty for its intended 242 purpose: absorbing bursts. However, RED [RFC2309] and other 243 algorithms from the 1990s were sensitive to their configuration and 244 hard to set correctly. So, this form of AQM was not widely deployed. 246 More recent state-of-the-art AQM methods, e.g. FQ-CoDel [RFC8290], 247 PIE [RFC8033], Adaptive RED [ARED01], are easier to configure, 248 because they define the queuing threshold in time not bytes, so it is 249 invariant for different link rates. However, no matter how good the 250 AQM, the sawtoothing sending window of a Classic congestion control 251 will either cause queuing delay to vary or cause the link to be 252 underutilized. Even with a perfectly tuned AQM, the additional 253 queuing delay will be of the same order as the underlying speed-of- 254 light delay across the network, thereby roughly doubling the total 255 round-trip time. 257 If a sender's own behaviour is introducing queuing delay variation, 258 no AQM in the network can 'un-vary' the delay without significantly 259 compromising link utilization. Even flow-queuing (e.g. [RFC8290]), 260 which isolates one flow from another, cannot isolate a flow from the 261 delay variations it inflicts on itself. Therefore those applications 262 that need to seek out high bandwidth but also need low latency will 263 have to migrate to scalable congestion control. 265 Altering host behaviour is not enough on its own though. Even if 266 hosts adopt low latency behaviour (scalable congestion controls), 267 they need to be isolated from the behaviour of existing Classic 268 congestion controls that induce large queue variations. L4S enables 269 that migration by providing latency isolation in the network and 270 distinguishing the two types of packets that need to be isolated: L4S 271 and Classic. L4S isolation can be achieved with a queue per flow 272 (e.g. [RFC8290]) but a DualQ [I-D.ietf-tsvwg-aqm-dualq-coupled] is 273 sufficient, and actually gives better tail latency. Both approaches 274 are addressed in this document. 276 The DualQ solution was developed to make very low latency available 277 without requiring per-flow queues at every bottleneck. This was 278 because per-flow-queuing (FQ) has well-known downsides - not least 279 the need to inspect transport layer headers in the network, which 280 makes it incompatible with privacy approaches such as IPSec VPN 281 tunnels, and incompatible with link layer queue management, where 282 transport layer headers can be hidden, e.g. 5G. 284 Latency is not the only concern addressed by L4S: It was known when 285 TCP congestion avoidance was first developed that it would not scale 286 to high bandwidth-delay products (footnote 6 of Jacobson and Karels 287 [TCP-CA]). Given regular broadband bit-rates over WAN distances are 288 already [RFC3649] beyond the scaling range of Reno congestion 289 control, 'less unscalable' Cubic [RFC8312] and 290 Compound [I-D.sridharan-tcpm-ctcp] variants of TCP have been 291 successfully deployed. However, these are now approaching their 292 scaling limits. Unfortunately, fully scalable congestion controls 293 such as DCTCP [RFC8257] outcompete Classic ECN congestion controls 294 sharing the same queue, which is why they have been confined to 295 private data centres or research testbeds. 297 It turns out that these scalable congestion control algorithms that 298 solve the latency problem can also solve the scalability problem of 299 Classic congestion controls. The finer sawteeth in the congestion 300 window have low amplitude, so they cause very little queuing delay 301 variation and the average time to recover from one congestion signal 302 to the next (the average duration of each sawtooth) remains 303 invariant, which maintains constant tight control as flow-rate 304 scales. A background paper [DCttH19] gives the full explanation of 305 why the design solves both the latency and the scaling problems, both 306 in plain English and in more precise mathematical form. The 307 explanation is summarised without the maths in Section 4 of the L4S 308 architecture document [I-D.ietf-tsvwg-l4s-arch]. 310 1.2. Terminology 312 The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", 313 "SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and 314 "OPTIONAL" in this document are to be interpreted as described in 315 [RFC2119]. In this document, these words will appear with that 316 interpretation only when in ALL CAPS. Lower case uses of these words 317 are not to be interpreted as carrying RFC-2119 significance. 319 Note: [I-D.ietf-tsvwg-l4s-arch] repeats the following definitions, 320 but if there are accidental differences those below take precedence. 322 Classic Congestion Control: A congestion control behaviour that can 323 co-exist with standard Reno [RFC5681] without causing 324 significantly negative impact on its flow rate [RFC5033]. With 325 Classic congestion controls, such as Reno or Cubic, because flow 326 rate has scaled since TCP congestion control was first designed in 327 1988, it now takes hundreds of round trips (and growing) to 328 recover after a congestion signal (whether a loss or an ECN mark) 329 as shown in the examples in section 5.1 of 330 [I-D.ietf-tsvwg-l4s-arch] and in [RFC3649]. Therefore control of 331 queuing and utilization becomes very slack, and the slightest 332 disturbances (e.g. from new flows starting) prevent a high rate 333 from being attained. 335 Scalable Congestion Control: A congestion control where the average 336 time from one congestion signal to the next (the recovery time) 337 remains invariant as the flow rate scales, all other factors being 338 equal. This maintains the same degree of control over queueing 339 and utilization whatever the flow rate, as well as ensuring that 340 high throughput is robust to disturbances. For instance, DCTCP 341 averages 2 congestion signals per round-trip whatever the flow 342 rate, as do other recently developed scalable congestion controls, 343 e.g. Relentless TCP [Mathis09], TCP Prague 344 [I-D.briscoe-iccrg-prague-congestion-control], [PragueLinux], 345 BBRv2 [I-D.cardwell-iccrg-bbr-congestion-control] and the L4S 346 variant of SCREAM for real-time media [SCReAM], [RFC8298]). See 347 Section 4.3 for more explanation. 349 Classic service: The Classic service is intended for all the 350 congestion control behaviours that co-exist with Reno [RFC5681] 351 (e.g. Reno itself, Cubic [RFC8312], Compound 352 [I-D.sridharan-tcpm-ctcp], TFRC [RFC5348]). The term 'Classic 353 queue' means a queue providing the Classic service. 355 Low-Latency, Low-Loss Scalable throughput (L4S) service: The 'L4S' 356 service is intended for traffic from scalable congestion control 357 algorithms, such as TCP Prague 358 [I-D.briscoe-iccrg-prague-congestion-control], which was derived 359 from DCTCP [RFC8257]. The L4S service is for more general traffic 360 than just TCP Prague--it allows the set of congestion controls 361 with similar scaling properties to Prague to evolve, such as the 362 examples listed above (Relentless, SCReAM). The term 'L4S queue' 363 means a queue providing the L4S service. 365 The terms Classic or L4S can also qualify other nouns, such as 366 'queue', 'codepoint', 'identifier', 'classification', 'packet', 367 'flow'. For example: an L4S packet means a packet with an L4S 368 identifier sent from an L4S congestion control. 370 Both Classic and L4S services can cope with a proportion of 371 unresponsive or less-responsive traffic as well, but in the L4S 372 case its rate has to be smooth enough or low enough not to build a 373 queue (e.g. DNS, VoIP, game sync datagrams, etc). 375 Reno-friendly: The subset of Classic traffic that is friendly to the 376 standard Reno congestion control defined for TCP in [RFC5681]. 377 The TFRC spec. [RFC5348] indirectly implies that 'friendly' is 378 defined as "generally within a factor of two of the sending rate 379 of a TCP flow under the same conditions". Reno-friendly is used 380 here in place of 'TCP-friendly', given the latter has become 381 imprecise, because the TCP protocol is now used with so many 382 different congestion control behaviours, and Reno is used in non- 383 TCP transports such as QUIC [RFC9000]. 385 Classic ECN: The original Explicit Congestion Notification (ECN) 386 protocol [RFC3168], which requires ECN signals to be treated the 387 same as drops, both when generated in the network and when 388 responded to by the sender. For L4S, the names used for the four 389 codepoints of the 2-bit IP-ECN field are unchanged from those 390 defined in [RFC3168]: Not ECT, ECT(0), ECT(1) and CE, where ECT 391 stands for ECN-Capable Transport and CE stands for Congestion 392 Experienced. A packet marked with the CE codepoint is termed 393 'ECN-marked' or sometimes just 'marked' where the context makes 394 ECN obvious. 396 Site: A home, mobile device, small enterprise or campus, where the 397 network bottleneck is typically the access link to the site. Not 398 all network arrangements fit this model but it is a useful, widely 399 applicable generalization. 401 1.3. Scope 403 The new L4S identifier defined in this specification is applicable 404 for IPv4 and IPv6 packets (as for Classic ECN [RFC3168]). It is 405 applicable for the unicast, multicast and anycast forwarding modes. 407 The L4S identifier is an orthogonal packet classification to the 408 Differentiated Services Code Point (DSCP) [RFC2474]. Section 5.4 409 explains what this means in practice. 411 This document is intended for experimental status, so it does not 412 update any standards track RFCs. Therefore it depends on [RFC8311], 413 which is a standards track specification that: 415 * updates the ECN proposed standard [RFC3168] to allow experimental 416 track RFCs to relax the requirement that an ECN mark must be 417 equivalent to a drop (when the network applies markings and/or 418 when the sender responds to them). For instance, in the ABE 419 experiment [RFC8511] this permits a sender to respond less to ECN 420 marks than to drops; 422 * changes the status of the experimental ECN nonce [RFC3540] to 423 historic; 425 * makes consequent updates to the following additional proposed 426 standard RFCs to reflect the above two bullets: 428 - ECN for RTP [RFC6679]; 430 - the congestion control specifications of various DCCP 431 congestion control identifier (CCID) profiles [RFC4341], 432 [RFC4342], [RFC5622]. 434 This document is about identifiers that are used for interoperation 435 between hosts and networks. So the audience is broad, covering 436 developers of host transports and network AQMs, as well as covering 437 how operators might wish to combine various identifiers, which would 438 require flexibility from equipment developers. 440 2. Choice of L4S Packet Identifier: Requirements 442 This subsection briefly records the process that led to the chosen 443 L4S identifier. 445 The identifier for packets using the Low Latency, Low Loss, Scalable 446 throughput (L4S) service needs to meet the following requirements: 448 * it SHOULD survive end-to-end between source and destination end- 449 points: across the boundary between host and network, between 450 interconnected networks, and through middleboxes; 452 * it SHOULD be visible at the IP layer; 454 * it SHOULD be common to IPv4 and IPv6 and transport-agnostic; 456 * it SHOULD be incrementally deployable; 458 * it SHOULD enable an AQM to classify packets encapsulated by outer 459 IP or lower-layer headers; 461 * it SHOULD consume minimal extra codepoints; 463 * it SHOULD be consistent on all the packets of a transport layer 464 flow, so that some packets of a flow are not served by a different 465 queue to others. 467 Whether the identifier would be recoverable if the experiment failed 468 is a factor that could be taken into account. However, this has not 469 been made a requirement, because that would favour schemes that would 470 be easier to fail, rather than those more likely to succeed. 472 It is recognised that any choice of identifier is unlikely to satisfy 473 all these requirements, particularly given the limited space left in 474 the IP header. Therefore a compromise will always be necessary, 475 which is why all the above requirements are expressed with the word 476 'SHOULD' not 'MUST'. 478 After extensive assessment of alternative schemes, "ECT(1) and CE 479 codepoints" was chosen as the best compromise. Therefore this scheme 480 is defined in detail in the following sections, while Appendix B 481 records its pros and cons against the above requirements. 483 3. L4S Packet Identification 485 The L4S treatment is an experimental track alternative packet marking 486 treatment to the Classic ECN treatment in [RFC3168], which has been 487 updated by [RFC8311] to allow experiments such as the one defined in 488 the present specification. [RFC4774] discusses some of the issues 489 and evaluation criteria when defining alternative ECN semantics. 490 Like Classic ECN, L4S ECN identifies both network and host behaviour: 491 it identifies the marking treatment that network nodes are expected 492 to apply to L4S packets, and it identifies packets that have been 493 sent from hosts that are expected to comply with a broad type of 494 sending behaviour. 496 For a packet to receive L4S treatment as it is forwarded, the sender 497 sets the ECN field in the IP header to the ECT(1) codepoint. See 498 Section 4 for full transport layer behaviour requirements, including 499 feedback and congestion response. 501 A network node that implements the L4S service always classifies 502 arriving ECT(1) packets for L4S treatment and by default classifies 503 CE packets for L4S treatment unless the heuristics described in 504 Section 5.3 are employed. See Section 5 for full network element 505 behaviour requirements, including classification, ECN-marking and 506 interaction of the L4S identifier with other identifiers and per-hop 507 behaviours. 509 4. Transport Layer Behaviour (the 'Prague Requirements') 510 4.1. Codepoint Setting 512 A sender that wishes a packet to receive L4S treatment as it is 513 forwarded, MUST set the ECN field in the IP header (v4 or v6) to the 514 ECT(1) codepoint. 516 4.2. Prerequisite Transport Feedback 518 For a transport protocol to provide scalable congestion control 519 (Section 4.3) it MUST provide feedback of the extent of CE marking on 520 the forward path. When ECN was added to TCP [RFC3168], the feedback 521 method reported no more than one CE mark per round trip. Some 522 transport protocols derived from TCP mimic this behaviour while 523 others report the accurate extent of ECN marking. This means that 524 some transport protocols will need to be updated as a prerequisite 525 for scalable congestion control. The position for a few well-known 526 transport protocols is given below. 528 TCP: Support for the accurate ECN feedback requirements [RFC7560] 529 (such as that provided by AccECN [I-D.ietf-tcpm-accurate-ecn]) by 530 both ends is a prerequisite for scalable congestion control in 531 TCP. Therefore, the presence of ECT(1) in the IP headers even in 532 one direction of a TCP connection will imply that both ends 533 support accurate ECN feedback. However, the converse does not 534 apply. So even if both ends support AccECN, either of the two 535 ends can choose not to use a scalable congestion control, whatever 536 the other end's choice. 538 SCTP: A suitable ECN feedback mechanism for SCTP could add a chunk 539 to report the number of received CE marks 540 (e.g. [I-D.stewart-tsvwg-sctpecn]), and update the ECN feedback 541 protocol sketched out in Appendix A of the standards track 542 specification of SCTP [RFC4960]. 544 RTP over UDP: A prerequisite for scalable congestion control is for 545 both (all) ends of one media-level hop to signal ECN support 546 [RFC6679] and use the new generic RTCP feedback format of 547 [RFC8888]. The presence of ECT(1) implies that both (all) ends of 548 that media-level hop support ECN. However, the converse does not 549 apply. So each end of a media-level hop can independently choose 550 not to use a scalable congestion control, even if both ends 551 support ECN. 553 QUIC: Support for sufficiently fine-grained ECN feedback is provided 554 by the v1 IETF QUIC transport [RFC9000]. 556 DCCP: The ACK vector in DCCP [RFC4340] is already sufficient to 557 report the extent of CE marking as needed by a scalable congestion 558 control. 560 4.3. Prerequisite Congestion Response 562 As a condition for a host to send packets with the L4S identifier 563 (ECT(1)), it SHOULD implement a congestion control behaviour that 564 ensures that, in steady state, the average duration between induced 565 ECN marks does not increase as flow rate scales up, all other factors 566 being equal. This is termed a scalable congestion control. This 567 invariant duration ensures that, as flow rate scales, the average 568 period with no feedback information about capacity does not become 569 excessive. It also ensures that queue variations remain small, 570 without having to sacrifice utilization. 572 With a congestion control that sawtooths to probe capacity, this 573 duration is called the recovery time, because each time the sawtooth 574 yields, on average it take this time to recover to its previous high 575 point. A scalable congestion control does not have to sawtooth, but 576 it has to coexist with scalable congestion controls that do. 578 For instance, for DCTCP [RFC8257], TCP Prague 579 [I-D.briscoe-iccrg-prague-congestion-control], [PragueLinux] and the 580 L4S variant of SCReAM [RFC8298], the average recovery time is always 581 half a round trip (or half a reference round trip), whatever the flow 582 rate. 584 As with all transport behaviours, a detailed specification (probably 585 an experimental RFC) is expected for each congestion control, 586 following the guidelines for specifying new congestion control 587 algorithms in [RFC5033]. In addition it is expected to document 588 these L4S-specific matters, specifically the timescale over which the 589 proportionality is averaged, and control of burstiness. The recovery 590 time requirement above is worded as a 'SHOULD' rather than a 'MUST' 591 to allow reasonable flexibility for such implementations. 593 The condition 'all other factors being equal', allows the recovery 594 time to be different for different round trip times, as long as it 595 does not increase with flow rate for any particular RTT. 597 Saying that the recovery time remains roughly invariant is equivalent 598 to saying that the number of ECN CE marks per round trip remains 599 invariant as flow rate scales, all other factors being equal. For 600 instance, an average recovery time of half of 1 RTT is equivalent to 601 2 ECN marks per round trip. For those familiar with steady-state 602 congestion response functions, it is also equivalent to say that the 603 congestion window is inversely proportional to the proportion of 604 bytes in packets marked with the CE codepoint (see section 2 of 605 [PI2]). 607 In order to coexist safely with other Internet traffic, a scalable 608 congestion control MUST NOT tag its packets with the ECT(1) codepoint 609 unless it complies with the following bulleted requirements: 611 1. A scalable congestion control MUST be capable of being replaced 612 by a Classic congestion control (by application and/or by 613 administrative control). If a Classic congestion control is 614 activated, it will not tag its packets with the ECT(1) codepoint 615 (see Appendix A.1.3 for rationale). 617 2. As well as responding to ECN markings, a scalable congestion 618 control MUST react to packet loss in a way that will coexist 619 safely with Classic congestion controls such as standard Reno 620 [RFC5681], as required by [RFC5033] (see Appendix A.1.4 for 621 rationale). 623 3. In uncontrolled environments, monitoring MUST be implemented to 624 support detection of problems with an ECN-capable AQM at the path 625 bottleneck that appears not to support L4S and might be in a 626 shared queue. Such monitoring SHOULD be applied to live traffic 627 that is using Scalable congestion control. Alternatively, 628 monitoring need not be applied to live traffic, if monitoring has 629 been arranged to cover the paths that live traffic takes through 630 uncontrolled environments. 632 A function to detect the above problems with an ECN-capable AQM 633 MUST also be implemented and used. The detection function SHOULD 634 be capable of making the congestion control adapt its ECN-marking 635 response in real-time to coexist safely with Classic congestion 636 controls such as standard Reno [RFC5681], as required by 637 [RFC5033]. This could be complemented by more detailed offline 638 detection of potential problems. If only offline detection is 639 used and potential problems with such an AQM are detected on 640 certain paths, the scalable congestion control MUST be replaced 641 by a Classic congestion control, at least for the problem paths. 643 See Section 4.3.1, Appendix A.1.5 and [I-D.ietf-tsvwg-l4sops] for 644 rationale. 646 Note that a scalable congestion control is not expected to change 647 to setting ECT(0) while it transiently adapts to coexist with 648 Classic congestion controls, whereas a replacement congestion 649 control that solely behaves in the Classic way will set ECT(0). 651 4. In the range between the minimum likely RTT and typical RTTs 652 expected in the intended deployment scenario, a scalable 653 congestion control MUST converge towards a rate that is as 654 independent of RTT as is possible without compromising stability 655 or efficiency (see Appendix A.1.6 for rationale). 657 5. A scalable congestion control SHOULD remain responsive to 658 congestion when typical RTTs over the public Internet are 659 significantly smaller because they are no longer inflated by 660 queuing delay. It would be preferable for the minimum window of 661 a scalable congestion control to be lower than 1 segment rather 662 than use the timeout approach described for TCP in S.6.1.2 of 663 [RFC3168] (or an equivalent for other transports). However, a 664 lower minimum is not set as a formal requirement for L4S 665 experiments (see Appendix A.1.7 for rationale). 667 6. A scalable congestion control's loss detection SHOULD be 668 resilient to reordering over an adaptive time interval that 669 scales with throughput and adapts to reordering (as in 670 [RFC8985]), as opposed to counting only in fixed units of packets 671 (as in the 3 DupACK rule of [RFC5681] and [RFC6675], which is not 672 scalable). As data rates increase (e.g., due to new and/or 673 improved technology), congestion controls that detect loss by 674 counting in units of packets become more likely to incorrectly 675 treat reordering events as congestion-caused loss events (see 676 Appendix A.1.8 for further rationale). This requirement does not 677 apply to congestion controls that are solely used in controlled 678 environments where the network introduces hardly any reordering. 680 7. A scalable congestion control is expected to limit the queue 681 caused by bursts of packets. It would not seem necessary to set 682 the limit any lower than 10% of the minimum RTT expected in a 683 typical deployment (e.g. additional queuing of roughly 250 us for 684 the public Internet). This would be converted to a number of 685 packets under the worst-case assumption that the bottleneck link 686 capacity equals the current flow rate. No normative requirement 687 to limit bursts is given here and, until there is more industry 688 experience from the L4S experiment, it is not even known whether 689 one is needed - it seems to be in an L4S sender's self-interest 690 to limit bursts. 692 Each sender in a session can use a scalable congestion control 693 independently of the congestion control used by the receiver(s) when 694 they send data. Therefore there might be ECT(1) packets in one 695 direction and ECT(0) or Not-ECT in the other. 697 Later (Section 5.4.1.1) this document discusses the conditions for 698 mixing other "'Safe' Unresponsive Traffic" (e.g. DNS, LDAP, NTP, 699 voice, game sync packets) with L4S traffic. To be clear, although 700 such traffic can share the same queue as L4S traffic, it is not 701 appropriate for the sender to tag it as ECT(1), except in the 702 (unlikely) case that it satisfies the above conditions. 704 4.3.1. Guidance on Congestion Response in the RFC Series 706 RFC 3168 requires the congestion responses to a CE-marked packet and 707 a dropped packet to be the same. RFC 8311 is a standards-track 708 update to RFC 3168 intended to enable experimentation with ECN, 709 including the L4S experiment. RFC 8311 allows an experimental 710 congestion control's response to a CE-marked packet to differ from 711 the response to a dropped packet, provided that the differences are 712 documented in an experimental RFC, such as the present document. 714 BCP 124 [RFC4774] gives guidance to protocol designers, when 715 specifying alternative semantics for the ECN field. RFC 8311 716 explained that it did not need to update the best current practice in 717 BCP 124 in order to relax the 'equivalence with drop' requirement 718 because, although BCP 124 quotes the same requirement from RFC 3168, 719 the BCP does not impose requirements based on it. BCP124 describes 720 three options for incremental deployment, with Option 3 (in 721 Section 4.3 of BCP 124) best matching the L4S case. Option 3's 722 requirement for end-nodes is that they respond to CE marks "in a way 723 that is friendly to flows using IETF-conformant congestion control." 724 This echoes other general congestion control requirements in the RFC 725 series, for example [RFC5033], which says "...congestion controllers 726 that have a significantly negative impact on traffic using standard 727 congestion control may be suspect", or [RFC8085] concerning UDP 728 congestion control says "Bulk-transfer applications that choose not 729 to implement TFRC or TCP-like windowing SHOULD implement a congestion 730 control scheme that results in bandwidth (capacity) use that competes 731 fairly with TCP within an order of magnitude." 733 The third normative bullet in Section 4.3 above (which concerns L4S 734 response to congestion from a Classic ECN AQM) aims to ensure that 735 these 'coexistence' requirements are satisfied, but it makes some 736 compromises. This subsection highlights and justifies those 737 compromises and Appendix A.1.5 and [I-D.ietf-tsvwg-l4sops] give 738 detailed analysis, examples and references (the normative text in 739 that bullet takes precedence if any informative elaboration leads to 740 ambiguity). The approach is based on an assessment of the risk of 741 harm, which is a combination of the prevalence of the conditions 742 necessary for harm to occur, and the potential severity of the harm 743 if they do. 745 Prevalence: There are three cases: 747 * Drop Tail: Coexistence between L4S and Classic flows is not in 748 doubt where the bottleneck does not support any form of ECN, 749 which has remained by far the most prevalent case since the ECN 750 RFC was published in 2001. 752 * L4S: Coexistence is not in doubt if the bottleneck supports 753 L4S. 755 * Classic ECN [RFC3168]: The compromises centre around cases 756 where the bottleneck supports Classic ECN but not L4S. But it 757 depends on which sub-case: 759 - Shared Queue with Classic ECN: The members of the Transport 760 Working group are not aware of any current deployments of 761 single-queue Classic ECN bottlenecks in the Internet. 762 Nonetheless, at the scale of the Internet, rarity need not 763 imply small numbers, nor that there will be rarity in 764 future. 766 - Per-Flow-queues with Classic ECN: Most AQMs with per-flow- 767 queuing (FQ) deployed from 2012 onwards had Classic ECN 768 enabled by default, specifically FQ-CoDel [RFC8290] and 769 COBALT [COBALT]. But the compromises only apply to the 770 second of two further sub-cases: 772 o With per-flow-queuing, co-existence between Classic and 773 L4S flows is not normally a problem, because different 774 flows are not meant to coexist within the same queue, 776 o However, the isolation between L4S and Classic flows is 777 not perfect in cases where the hashes of flow IDs collide 778 or where multiple flows within a layer-3 VPN are 779 encapsulated within one flow ID. 781 To summarize, the coexistence problem is confined to cases of 782 imperfect flow isolation in an FQ, or in potential cases where a 783 Classic ECN AQM has been deployed in a shared queue (see 784 [I-D.ietf-tsvwg-l4sops] for further details including recent 785 surveys attempting to quantify prevalence). Further, if one of 786 these cases does occur, the coexistence problem does not arise 787 unless sources of Classic and L4S flows are simultaneously sharing 788 the same bottleneck queue (e.g. different applications in the same 789 household) and flows of each type have to be large enough to 790 coincide for long enough for any throughput imbalance to have 791 developed. 793 Severity: Where long-running L4S and Classic flows coincide in a 794 shared queue, testing of one L4S congestion control (TCP Prague) 795 has found that the imbalance in average throughput between an L4S 796 and a Classic flow can reach 25:1 in favour of L4S in the worst 797 case [ecn-fallback]. However, when capacity is most scarce, the 798 Classic flow gets a higher proportion of the link, for instance 799 over a 4 Mb/s link the throughput ratio is below ~10:1 over paths 800 with a base RTT below 100 ms, and falls below ~5:1 for base RTTs 801 below 20ms. 803 These throughput ratios can clearly fall well outside current RFC 804 guidance on coexistence. However, the tendency towards leaving a 805 greater share for Classic flows at lower link rate and the very 806 limited prevalence of the conditions necessary for harm to occur led 807 to the possibility of allowing the RFC requirements to be 808 compromised, albeit briefly:: 810 * The recommended approach is still to detect and adapt to a Classic 811 ECN AQM in real-time, which is fully consistent with all the RFCs 812 on coexistence. In other words, the "SHOULD"s in the third bullet 813 of Section 4.3 above expect the sender to implement something 814 similar to the proof of concept code that detects the presence of 815 a Classic ECN AQM and falls back to a Classic congestion response 816 within a few round trips [ecn-fallback]. However, although this 817 code reliably detects a Classic ECN AQM, the current code can also 818 wrongly categorize an L4S AQM as Classic, most often in cases when 819 link rate is low or RTT is high. Although this is the safe way 820 round, and although implementers are expected to be able to 821 improve on this proof of concept, concerns have been raised that 822 implementers might lose faith in such detection and disable it. 824 * Therefore the third bullet in Section 4.3 above allows a 825 compromise where coexistence could diverge from the requirements 826 in the RFC Series briefly, but mandatory monitoring is required, 827 in order to detect such cases and trigger remedial action. This 828 approach tolerates a brief divergence from the RFCs given the 829 likely low prevalence and given harm here means a flow progresses 830 more slowly than otherwise, but it does progress. 831 [I-D.ietf-tsvwg-l4sops] outlines a range of example remedial 832 actions that include alterations either to the sender or to the 833 network. However, the final normative requirement in the third 834 bullet of Section 4.3 above places ultimate responsibility for 835 remedial action on the sender. If coexistence problems with a 836 Classic ECN AQM are detected (implying they have not been resolved 837 by the network), it says the sender "MUST" revert to a Classic 838 congestion control." 840 [I-D.ietf-tsvwg-l4sops] also gives example ways in which L4S 841 congestion controls can be rolled out initially in lower risk 842 scenarios. 844 4.4. Filtering or Smoothing of ECN Feedback 846 Section 5.2 below specifies that an L4S AQM is expected to signal L4S 847 ECN immediately, to avoid introducing delay due to filtering or 848 smoothing. This contrasts with a Classic AQM, which filters out 849 variations in the queue before signalling ECN marking or drop. In 850 the L4S architecture [I-D.ietf-tsvwg-l4s-arch], responsibility for 851 smoothing out these variations shifts to the sender's congestion 852 control. 854 This shift of responsibility has the advantage that each sender can 855 smooth variations over a timescale proportionate to its own RTT. 856 Whereas, in the Classic approach, the network doesn't know the RTTs 857 of any of the flows, so it has to smooth out variations for a worst- 858 case RTT to ensure stability. For all the typical flows with shorter 859 RTT than the worst-case, this makes congestion control unnecessarily 860 sluggish. 862 This also gives an L4S sender the choice not to smooth, depending on 863 its context (start-up, congestion avoidance, etc). Therefore, this 864 document places no requirement on an L4S congestion control to smooth 865 out variations in any particular way. Implementers are encouraged to 866 openly publish the approach they take to smoothing, and the results 867 and experience they gain during the L4S experiment. 869 5. Network Node Behaviour 871 5.1. Classification and Re-Marking Behaviour 873 A network node that implements the L4S service: 875 * MUST classify arriving ECT(1) packets for L4S treatment, unless 876 overridden by another classifier (e.g., see Section 5.4.1.2); 878 * MUST classify arriving CE packets for L4S treatment as well, 879 unless overridden by a another classifier or unless the exception 880 referred to next applies; 881 CE packets might have originated as ECT(1) or ECT(0), but the 882 above rule to classify them as if they originated as ECT(1) is the 883 safe choice (see Appendix B for rationale). The exception is 884 where some flow-aware in-network mechanism happens to be available 885 for distinguishing CE packets that originated as ECT(0), as 886 described in Section 5.3, but there is no implication that such a 887 mechanism is necessary. 889 An L4S AQM treatment follows similar codepoint transition rules to 890 those in RFC 3168. Specifically, the ECT(1) codepoint MUST NOT be 891 changed to any other codepoint than CE, and CE MUST NOT be changed to 892 any other codepoint. An ECT(1) packet is classified as ECN-capable 893 and, if congestion increases, an L4S AQM algorithm will increasingly 894 mark the ECN field as CE, otherwise forwarding packets unchanged as 895 ECT(1). Necessary conditions for an L4S marking treatment are 896 defined in Section 5.2. 898 Under persistent overload an L4S marking treatment MUST begin 899 applying drop to L4S traffic until the overload episode has subsided, 900 as recommended for all AQM methods in [RFC7567] (Section 4.2.1), 901 which follows the similar advice in RFC 3168 (Section 7). During 902 overload, it MUST apply the same drop probability to L4S traffic as 903 it would to Classic traffic. 905 Where an L4S AQM is transport-aware, this requirement could be 906 satisfied by using drop in only the most overloaded individual per- 907 flow AQMs. In a DualQ with flow-aware queue protection (e.g. 908 [I-D.briscoe-docsis-q-protection]), this could be achieved by 909 redirecting packets in those flows contributing most to the overload 910 out of the L4S queue so that they are subjected to drop in the 911 Classic queue. 913 For backward compatibility in uncontrolled environments, a network 914 node that implements the L4S treatment MUST also implement an AQM 915 treatment for the Classic service as defined in Section 1.2. This 916 Classic AQM treatment need not mark ECT(0) packets, but if it does, 917 see Section 5.2 for the strengths of the markings relative to drop. 918 It MUST classify arriving ECT(0) and Not-ECT packets for treatment by 919 this Classic AQM (for the DualQ Coupled AQM, see the extensive 920 discussion on classification in Sections 2.3 and 2.5.1.1 of 921 [I-D.ietf-tsvwg-aqm-dualq-coupled]). 923 In case unforeseen problems arise with the L4S experiment, it MUST be 924 possible to configure an L4S implementation to disable the L4S 925 treatment. Once disabled, all packets of all ECN codepoints will 926 receive Classic treatment and ECT(1) packets MUST be treated as if 927 they were Not-ECT. 929 5.2. The Strength of L4S CE Marking Relative to Drop 931 The relative strengths of L4S CE and drop are irrelevant where AQMs 932 are implemented in separate queues per-application-flow, which are 933 then explicitly scheduled (e.g. with an FQ scheduler as in 934 [RFC8290]). Nonetheless, the relationship between them needs to be 935 defined for the coupling between L4S and Classic congestion signals 936 in a DualQ Coupled AQM [I-D.ietf-tsvwg-aqm-dualq-coupled], as below. 938 Unless an AQM node schedules application flows explicitly, the 939 likelihood that the AQM drops a Not-ECT Classic packet (p_C) MUST be 940 roughly proportional to the square of the likelihood that it would 941 have marked it if it had been an L4S packet (p_L). That is 943 p_C ~= (p_L / k)^2 945 The constant of proportionality (k) does not have to be standardised 946 for interoperability, but a value of 2 is RECOMMENDED. The term 947 'likelihood' is used above to allow for marking and dropping to be 948 either probabilistic or deterministic. 950 This formula ensures that Scalable and Classic flows will converge to 951 roughly equal congestion windows, for the worst case of Reno 952 congestion control. This is because the congestion windows of 953 Scalable and Classic congestion controls are inversely proportional 954 to p_L and sqrt(p_C) respectively. So squaring p_C in the above 955 formula counterbalances the square root that characterizes Reno- 956 friendly flows. 958 Note that, contrary to RFC 3168, an AQM implementing the L4S and 959 Classic treatments does not mark an ECT(1) packet under the same 960 conditions that it would have dropped a Not-ECT packet, as allowed by 961 [RFC8311], which updates RFC 3168. However, if it marks ECT(0) 962 packets, it does so under the same conditions that it would have 963 dropped a Not-ECT packet [RFC3168]. 965 Also, In the L4S architecture [I-D.ietf-tsvwg-l4s-arch], the sender, 966 not the network, is responsible for smoothing out variations in the 967 queue. So, an L4S AQM MUST signal congestion as soon as possible. 968 Then, an L4S sender generally interprets CE marking as an unsmoothed 969 signal. 971 This requirement does not prevent an L4S AQM from mixing in 972 additional congestion signals that are smoothed, such as the signals 973 from a Classic smoothed AQM that are coupled with unsmoothed L4S 974 signals in the coupled DualQ [I-D.ietf-tsvwg-aqm-dualq-coupled]. But 975 only as long as the onset of congestion can be signalled immediately, 976 and can be interpreted by the sender as if it has been signalled 977 immediately, which is important for interoperability 979 5.3. Exception for L4S Packet Identification by Network Nodes with 980 Transport-Layer Awareness 982 To implement L4S packet classification, a network node does not need 983 to identify transport-layer flows. Nonetheless, if an L4S network 984 node classifies packets by their transport-layer flow ID and their 985 ECN field, and if all the ECT packets in a flow have been ECT(0), the 986 node MAY classify any CE packets in the same flow as if they were 987 Classic ECT(0) packets. In all other cases, a network node MUST 988 classify all CE packets as if they were ECT(1) packets. Examples of 989 such other cases are: i) if no ECT packets have yet been identified 990 in a flow; ii) if it is not desirable for a network node to identify 991 transport-layer flows; or iii) if some ECT packets in a flow have 992 been ECT(1) (this advice will need to be verified as part of L4S 993 experiments). 995 5.4. Interaction of the L4S Identifier with other Identifiers 997 The examples in this section concern how additional identifiers might 998 complement the L4S identifier to classify packets between class-based 999 queues. Firstly Section 5.4.1 considers two queues, L4S and Classic, 1000 as in the Coupled DualQ AQM [I-D.ietf-tsvwg-aqm-dualq-coupled], 1001 either alone (Section 5.4.1.1) or within a larger queuing hierarchy 1002 (Section 5.4.1.2). Then Section 5.4.2 considers schemes that might 1003 combine per-flow 5-tuples with other identifiers. 1005 5.4.1. DualQ Examples of Other Identifiers Complementing L4S 1006 Identifiers 1008 5.4.1.1. Inclusion of Additional Traffic with L4S 1010 In a typical case for the public Internet a network element that 1011 implements L4S in a shared queue might want to classify some low-rate 1012 but unresponsive traffic (e.g. DNS, LDAP, NTP, voice, game sync 1013 packets) into the low latency queue to mix with L4S traffic. In this 1014 case it would not be appropriate to call the queue an L4S queue, 1015 because it is shared by L4S and non-L4S traffic. Instead it will be 1016 called the low latency or L queue. The L queue then offers two 1017 different treatments: 1019 * The L4S treatment, which is a combination of the L4S AQM treatment 1020 and a priority scheduling treatment; 1022 * The low latency treatment, which is solely the priority scheduling 1023 treatment, without ECN-marking by the AQM. 1025 To identify packets for just the scheduling treatment, it would be 1026 inappropriate to use the L4S ECT(1) identifier, because such traffic 1027 is unresponsive to ECN marking. Examples of relevant non-ECN 1028 identifiers are: 1030 * address ranges of specific applications or hosts configured to be, 1031 or known to be, safe, e.g. hard-coded IoT devices sending low 1032 intensity traffic; 1034 * certain low data-volume applications or protocols (e.g. ARP, DNS); 1036 * specific Diffserv codepoints that indicate traffic with limited 1037 burstiness such as the EF (Expedited Forwarding [RFC3246]), Voice- 1038 Admit [RFC5865] or proposed NQB (Non-Queue-Building 1039 [I-D.ietf-tsvwg-nqb]) service classes or equivalent local-use 1040 DSCPs (see [I-D.briscoe-tsvwg-l4s-diffserv]). 1042 In summary, a network element that implements L4S in a shared queue 1043 MAY classify additional types of packets into the L queue based on 1044 identifiers other than the ECN field, but the types SHOULD be 'safe' 1045 to mix with L4S traffic, where 'safe' is explained in 1046 Section 5.4.1.1.1. 1048 A packet that carries one of these non-ECN identifiers to classify it 1049 into the L queue would not be subject to the L4S ECN marking 1050 treatment, unless it also carried an ECT(1) or CE codepoint. The 1051 specification of an L4S AQM MUST define the behaviour for packets 1052 with unexpected combinations of codepoints, e.g. a non-ECN-based 1053 classifier for the L queue, but ECT(0) in the ECN field (for examples 1054 see section 2.5.1.1 of [I-D.ietf-tsvwg-aqm-dualq-coupled]). 1056 For clarity, non-ECN identifiers, such as the examples itemized 1057 above, might be used by some network operators who believe they 1058 identify non-L4S traffic that would be safe to mix with L4S traffic. 1059 They are not alternative ways for a host to indicate that it is 1060 sending L4S packets. Only the ECT(1) ECN codepoint indicates to a 1061 network element that a host is sending L4S packets (and CE indicates 1062 that it could have originated as ECT(1)). Specifically ECT(1) 1063 indicates that the host claims its behaviour satisfies the 1064 prerequisite transport requirements in Section 4. 1066 In order to include non-L4S packets in the L queue, a network node 1067 MUST NOT alter Not-ECT or ECT(0) in the IP-ECN field to an L4S 1068 identifier. This ensures that these codepoints survive for any 1069 potential use later on the network path. 1071 5.4.1.1.1. 'Safe' Unresponsive Traffic 1073 The above section requires unresponsive traffic to be 'safe' to mix 1074 with L4S traffic. Ideally this means that the sender never sends any 1075 sequence of packets at a rate that exceeds the available capacity of 1076 the bottleneck link. However, typically an unresponsive transport 1077 does not even know the bottleneck capacity of the path, let alone its 1078 available capacity. Nonetheless, an application can be considered 1079 safe enough if it paces packets out (not necessarily completely 1080 regularly) such that its maximum instantaneous rate from packet to 1081 packet stays well below a typical broadband access rate. 1083 This is a vague but useful definition, because many low latency 1084 applications of interest, such as DNS, voice, game sync packets, RPC, 1085 ACKs, keep-alives, could match this description. 1087 Low rate streams such as voice and game sync packets, might not use 1088 continuously adapting ECN-based congestion control, but they ought to 1089 at least use a 'circuit-breaker' style of congestion response 1090 [RFC8083]. If the volume of traffic from unresponsive applications 1091 is high enough to overload the link, this will at least protect the 1092 capacity available to responsive applications. However, queuing 1093 delay in the L queue will probably rise to that controlled by the 1094 Classic (drop-based) AQM. If a network operator considers that such 1095 self-restraint is not enough, it might want to police the L queue 1096 (see Section 8.2 of [I-D.ietf-tsvwg-l4s-arch]). 1098 5.4.1.2. Exclusion of Traffic From L4S Treatment 1100 To extend the above example, an operator might want to exclude some 1101 traffic from the L4S treatment for a policy reason, e.g. security 1102 (traffic from malicious sources) or commercial (e.g. initially the 1103 operator may wish to confine the benefits of L4S to business 1104 customers). 1106 In this exclusion case, the classifier MUST classify on the relevant 1107 locally-used identifiers (e.g. source addresses) before classifying 1108 the non-matching traffic on the end-to-end L4S ECN identifier. 1110 A network node MUST NOT alter the end-to-end L4S ECN identifier from 1111 L4S to Classic, because an operator decision to exclude certain 1112 traffic from L4S treatment is local-only. The end-to-end L4S 1113 identifier then survives for other operators to use, or indeed, they 1114 can apply their own policy, independently based on their own choice 1115 of locally-used identifiers. This approach also allows any operator 1116 to remove its locally-applied exclusions in future, e.g. if it wishes 1117 to widen the benefit of the L4S treatment to all its customers. 1119 A network node that supports L4S but excludes certain packets 1120 carrying the L4S identifier from L4S treatment MUST still apply 1121 marking or dropping that is compatible with an L4S congestion 1122 response. For instance, it could either drop such packets with the 1123 same likelihood as Classic packets or it could ECN-mark them with a 1124 likelihood appropriate to L4S traffic (e.g. the coupled probability 1125 in a DualQ coupled AQM) but aiming for the Classic delay target. It 1126 MUST NOT ECN-mark such packets with a Classic marking probability, 1127 which could confuse the sender. 1129 5.4.1.3. Generalized Combination of L4S and Other Identifiers 1131 L4S concerns low latency, which it can provide for all traffic 1132 without differentiation and without _necessarily_ affecting bandwidth 1133 allocation. Diffserv provides for differentiation of both bandwidth 1134 and low latency, but its control of latency depends on its control of 1135 bandwidth. The two can be combined if a network operator wants to 1136 control bandwidth allocation but it also wants to provide low latency 1137 - for any amount of traffic within one of these allocations of 1138 bandwidth (rather than only providing low latency by limiting 1139 bandwidth) [I-D.briscoe-tsvwg-l4s-diffserv]. 1141 The DualQ examples so far have been framed in the context of 1142 providing the default Best Efforts Per-Hop Behaviour (PHB) using two 1143 queues - a Low Latency (L) queue and a Classic (C) Queue. This 1144 single DualQ structure is expected to be the most common and useful 1145 arrangement. But, more generally, an operator might choose to 1146 control bandwidth allocation through a hierarchy of Diffserv PHBs at 1147 a node, and to offer one (or more) of these PHBs using a pair of 1148 queues for a low latency and a Classic variant of the PHB. 1150 In the first case, if we assume that a network element provides no 1151 PHBs except the DualQ, if a packet carries ECT(1) or CE, the network 1152 element would classify it for the L4S treatment irrespective of its 1153 DSCP. And, if a packet carried (say) the EF DSCP, the network 1154 element could classify it into the L queue irrespective of its ECN 1155 codepoint. However, where the DualQ is in a hierarchy of other PHBs, 1156 the classifier would classify some traffic into other PHBs based on 1157 DSCP before classifying between the low latency and Classic queues 1158 (based on ECT(1), CE and perhaps also the EF DSCP or other 1159 identifiers as in the above example). 1160 [I-D.briscoe-tsvwg-l4s-diffserv] gives a number of examples of such 1161 arrangements to address various requirements. 1163 [I-D.briscoe-tsvwg-l4s-diffserv] describes how an operator might use 1164 L4S to offer low latency as well as using Diffserv for bandwidth 1165 differentiation. It identifies two main types of approach, which can 1166 be combined: the operator might split certain Diffserv PHBs between 1167 L4S and a corresponding Classic service. Or it might split the L4S 1168 and/or the Classic service into multiple Diffserv PHBs. In either of 1169 these cases, a packet would have to be classified on its Diffserv and 1170 ECN codepoints. 1172 In summary, there are numerous ways in which the L4S ECN identifier 1173 (ECT(1) and CE) could be combined with other identifiers to achieve 1174 particular objectives. The following categorization articulates 1175 those that are valid, but it is not necessarily exhaustive. Those 1176 tagged 'Recommended-standard-use' could be set by the sending host or 1177 a network. Those tagged 'Local-use' would only be set by a network: 1179 1. Identifiers Complementing the L4S Identifier 1181 a. Including More Traffic in the L Queue 1183 (Could use Recommended-standard-use or Local-use identifiers) 1185 b. Excluding Certain Traffic from the L Queue 1187 (Local-use only) 1189 2. Identifiers to place L4S classification in a PHB Hierarchy 1191 (Could use Recommended-standard-use or Local-use identifiers) 1193 a. PHBs Before L4S ECN Classification 1195 b. PHBs After L4S ECN Classification 1197 5.4.2. Per-Flow Queuing Examples of Other Identifiers Complementing L4S 1198 Identifiers 1200 At a node with per-flow queueing (e.g. FQ-CoDel [RFC8290]), the L4S 1201 identifier could complement the Layer-4 flow ID as a further level of 1202 flow granularity (i.e. Not-ECT and ECT(0) queued separately from 1203 ECT(1) and CE packets). "Risk of reordering Classic CE packets" in 1204 Appendix B discusses the resulting ambiguity if packets originally 1205 marked ECT(0) are marked CE by an upstream AQM before they arrive at 1206 a node that classifies CE as L4S. It argues that the risk of 1207 reordering is vanishingly small and the consequence of such a low 1208 level of reordering is minimal. 1210 Alternatively, it could be assumed that it is not in a flow's own 1211 interest to mix Classic and L4S identifiers. Then the AQM could use 1212 the ECN field to switch itself between a Classic and an L4S AQM 1213 behaviour within one per-flow queue. For instance, for ECN-capable 1214 packets, the AQM might consist of a simple marking threshold and an 1215 L4S ECN identifier might simply select a shallower threshold than a 1216 Classic ECN identifier would. 1218 5.5. Limiting Packet Bursts from Links 1220 As well as senders needing to limit packet bursts (Section 4.3), 1221 links need to limit the degree of burstiness they introduce. In both 1222 cases (senders and links) this is a tradeoff, because batch-handling 1223 of packets is done for good reason, e.g. processing efficiency or to 1224 make efficient use of medium acquisition delay. Some take the 1225 attitude that there is no point reducing burst delay at the sender 1226 below that introduced by links (or vice versa). However, delay 1227 reduction proceeds by cutting down 'the longest pole in the tent', 1228 which turns the spotlight on the next longest, and so on. 1230 This document does not set any quantified requirements for links to 1231 limit burst delay, primarily because link technologies are outside 1232 the remit of L4S specifications. Nonetheless, the following two 1233 subsections outline opportunities for addressing bursty links in the 1234 process of L4S implementation and deployment. 1236 5.5.1. Limiting Packet Bursts from Links Fed by an L4S AQM 1238 It would not make sense to implement an L4S AQM that feeds into a 1239 particular link technology without also reviewing opportunities to 1240 reduce any form of burst delay introduced by that link technology. 1241 This would at least limit the bursts that the link would otherwise 1242 introduce into the onward traffic, which would cause jumpy feedback 1243 to the sender as well as potential extra queuing delay downstream. 1244 This document does not presume to even give guidance on an 1245 appropriate target for such burst delay until there is more industry 1246 experience of L4S. However, as suggested in Section 4.3 it would not 1247 seem necessary to limit bursts lower than roughly 10% of the minimum 1248 base RTT expected in the typical deployment scenario (e.g. 250 us 1249 burst duration for links within the public Internet). 1251 5.5.2. Limiting Packet Bursts from Links Upstream of an L4S AQM 1253 The initial scope of the L4S experiment is to deploy L4S AQMs at 1254 bottlenecks and L4S congestion controls at senders. This is expected 1255 to highlight interactions with the most bursty upstream links and 1256 lead operators to tune down the burstiness of those links in their 1257 network that are configurable, or failing that, to have to compromise 1258 on the delay target of some L4S AQMs. It might also require specific 1259 redesign work relevant to the most problematic link types. Such 1260 knock-on effects of initial L4S deployment would all be part of the 1261 learning from the L4S experiment. 1263 The details of such link changes are beyond the scope of the present 1264 document. Nonetheless, where L4S technology is being implemented on 1265 an outgoing interface of a device, it would make sense to consider 1266 opportunities for reducing bursts arriving at other incoming 1267 interface(s). For instance, where an L4S AQM is implemented to feed 1268 into the upstream WAN interface of a home gateway, there would be 1269 opportunities to alter the WiFi profiles sent out of any WiFi 1270 interfaces from the same device, in order to mitigate incoming bursts 1271 of aggregated WiFi frames from other WiFi stations. 1273 6. Behaviour of Tunnels and Encapsulations 1275 6.1. No Change to ECN Tunnels and Encapsulations in General 1277 The L4S identifier is expected to work through and within any tunnel 1278 without modification, as long as the tunnel propagates the ECN field 1279 in any of the ways that have been defined since the first variant in 1280 the year 2001 [RFC3168]. L4S will also work with (but does not rely 1281 on) any of the more recent updates to ECN propagation in [RFC4301], 1282 [RFC6040] or [I-D.ietf-tsvwg-rfc6040update-shim]. However, it is 1283 likely that some tunnels still do not implement ECN propagation at 1284 all. In these cases, L4S will work through such tunnels, but within 1285 them the outer header of L4S traffic will appear as Classic. 1287 AQMs are typically implemented where an IP-layer buffer feeds into a 1288 lower layer, so they are agnostic to link layer encapsulations. 1289 Where a bottleneck link is not IP-aware, the L4S identifier is still 1290 expected to work within any lower layer encapsulation without 1291 modification, as long it propagates the ECN field as defined for the 1292 link technology, for example for MPLS [RFC5129] or TRILL 1293 [I-D.ietf-trill-ecn-support]. In some of these cases, e.g. layer-3 1294 Ethernet switches, the AQM accesses the IP layer header within the 1295 outer encapsulation, so again the L4S identifier is expected to work 1296 without modification. Nonetheless, the programme to define ECN for 1297 other lower layers is still in progress 1298 [I-D.ietf-tsvwg-ecn-encap-guidelines]. 1300 6.2. VPN Behaviour to Avoid Limitations of Anti-Replay 1302 If a mix of L4S and Classic packets is sent into the same security 1303 association (SA) of a virtual private network (VPN), and if the VPN 1304 egress is employing the optional anti-replay feature, it could 1305 inappropriately discard Classic packets (or discard the records in 1306 Classic packets) by mistaking their greater queuing delay for a 1307 replay attack (see "Dropped Packets for Tunnels with Replay 1308 Protection Enabled" in [Heist21] for the potential performance 1309 impact). This known problem is common to both IPsec [RFC4301] and 1310 DTLS [RFC6347] VPNs, given they use similar anti-replay window 1311 mechanisms. The mechanism used can only check for replay within its 1312 window, so if the window is smaller than the degree of reordering, it 1313 can only assume there might be a replay attack and discard all the 1314 packets behind the trailing edge of the window. The specifications 1315 of IPsec AH [RFC4302] and ESP [RFC4303] suggest that an implementer 1316 scales the size of the anti-replay window with interface speed, and 1317 the current draft of DTLS 1.3 [I-D.ietf-tls-dtls13] says "The 1318 receiver SHOULD pick a window large enough to handle any plausible 1319 reordering, which depends on the data rate." However, in practice, 1320 the size of a VPN's anti-replay window is not always scaled 1321 appropriately. 1323 If a VPN carrying traffic participating in the L4S experiment 1324 experiences inappropriate replay detection, the foremost remedy would 1325 be to ensure that the egress is configured to comply with the above 1326 window-sizing requirements. 1328 If an implementation of a VPN egress does not support a sufficiently 1329 large anti-replay window, e.g. due to hardware limitations, one of 1330 the temporary alternatives listed in order of preference below might 1331 be feasible instead: 1333 * If the VPN can be configured to classify packets into different 1334 SAs indexed by DSCP, apply the appropriate locally defined DSCPs 1335 to Classic and L4S packets. The DSCPs could be applied by the 1336 network (based on the least significant bit of the ECN field), or 1337 by the sending host. Such DSCPs would only need to survive as far 1338 as the VPN ingress. 1340 * If the above is not possible and it is necessary to use L4S, 1341 either of the following might be appropriate as a last resort: 1343 - disable anti-replay protection at the VPN egress, after 1344 considering the security implications (optional anti-replay is 1345 mandatory in both IPsec and DTLS); 1347 - configure the tunnel ingress not to propagate ECN to the outer, 1348 which would lose the benefits of L4S and Classic ECN over the 1349 VPN. 1351 Modification to VPN implementations is outside the present scope, 1352 which is why this section has so far focused on reconfiguration. 1353 Although this document does not define any requirements for VPN 1354 implementations, determining whether there is a need for such 1355 requirements could be one aspect of L4S experimentation. 1357 7. L4S Experiments 1359 This section describes open questions that L4S Experiments ought to 1360 focus on. This section also documents outstanding open issues that 1361 will need to be investigated as part of L4S experimentation, given 1362 they could not be fully resolved during the WG phase. It also lists 1363 metrics that will need to be monitored during experiments 1364 (summarizing text elsewhere in L4S documents) and finally lists some 1365 potential future directions that researchers might wish to 1366 investigate. 1368 In addition to this section, [I-D.ietf-tsvwg-aqm-dualq-coupled] sets 1369 operational and management requirements for experiments with DualQ 1370 Coupled AQMs; and General operational and management requirements for 1371 experiments with L4S congestion controls are given in Section 4 and 1372 Section 5 above, e.g. co-existence and scaling requirements, 1373 incremental deployment arrangements. 1375 The specification of each scalable congestion control will need to 1376 include protocol-specific requirements for configuration and 1377 monitoring performance during experiments. Appendix A of [RFC5706] 1378 provides a helpful checklist. 1380 7.1. Open Questions 1382 L4S experiments would be expected to answer the following questions: 1384 * Have all the parts of L4S been deployed, and if so, what 1385 proportion of paths support it? 1387 - What types of L4S AQMs were deployed, e.g. FQ, coupled DualQ, 1388 uncoupled DualQ, other? And how prevalent was each? 1390 - Are the signalling patterns emitted by the deployed AQMs in any 1391 way different from those expected when the Prague requirements 1392 for endpoints were written? 1394 * Does use of L4S over the Internet result in significantly improved 1395 user experience? 1397 * Has L4S enabled novel interactive applications? 1399 * Did use of L4S over the Internet result in improvements to the 1400 following metrics: 1402 - queue delay (mean and 99th percentile) under various loads; 1404 - utilization; 1406 - starvation / fairness; 1408 - scaling range of flow rates and RTTs? 1410 * How dependent was the performance of L4S service on the bottleneck 1411 bandwidth or the path RTT? 1413 * How much do bursty links in the Internet affect L4S performance 1414 (see "Underutilization with Bursty Links" in [Heist21]) and how 1415 prevalent are they? How much limitation of burstiness from 1416 upstream links was needed and/or was realized - both at senders 1417 and at links, especially radio links or how much did L4S target 1418 delay have to be increased to accommodate the bursts (see bullet 1419 #7 in Section 4.3 and Section 5.5.2)? 1421 * Is the initial experiment with mis-marked bursty traffic at high 1422 RTT (see "Underutilization with Bursty Traffic" in [Heist21]) 1423 indicative of similar problems at lower RTTs and, if so, how 1424 effective is the suggested remedy in Appendix A.1 of 1425 [I-D.ietf-tsvwg-aqm-dualq-coupled] (or possible other remedies)? 1427 * Was per-flow queue protection typically (un)necessary? 1429 - How well did overload protection or queue protection work? 1431 * How well did L4S flows coexist with Classic flows when sharing a 1432 bottleneck? 1434 - How frequently did problems arise? 1436 - What caused any coexistence problems, and were any problems due 1437 to single-queue Classic ECN AQMs (this assumes single-queue 1438 Classic ECN AQMs can be distinguished from FQ ones)? 1440 * How prevalent were problems with the L4S service due to tunnels / 1441 encapsulations that do not support ECN decapsulation? 1443 * How easy was it to implement a fully compliant L4S congestion 1444 control, over various different transport protocols (TCP, QUIC, 1445 RMCAT, etc)? 1447 Monitoring for harm to other traffic, specifically bandwidth 1448 starvation or excess queuing delay, will need to be conducted 1449 alongside all early L4S experiments. It is hard, if not impossible, 1450 for an individual flow to measure its impact on other traffic. So 1451 such monitoring will need to be conducted using bespoke monitoring 1452 across flows and/or across classes of traffic. 1454 7.2. Open Issues 1456 * What is the best way forward to deal with L4S over single-queue 1457 Classic ECN AQM bottlenecks, given current problems with 1458 misdetecting L4S AQMs as Classic ECN AQMs? See 1459 [I-D.ietf-tsvwg-l4sops]. 1461 * Fixing the poor Interaction between current L4S congestion 1462 controls and CoDel with only Classic ECN support during flow 1463 startup. Originally, this was due to a bug in the initialization 1464 of the congestion EWMA in the Linux implementation of TCP Prague. 1465 That was quickly fixed, which removed the main performance impact, 1466 but further improvement would be useful (either by modifying 1467 CoDel, Scalable congestion controls, or both). 1469 7.3. Future Potential 1471 Researchers might find that L4S opens up the following interesting 1472 areas for investigation: 1474 * Potential for faster convergence time and tracking of available 1475 capacity; 1477 * Potential for improvements to particular link technologies, and 1478 cross-layer interactions with them; 1480 * Potential for using virtual queues, e.g. to further reduce latency 1481 jitter, or to leave headroom for capacity variation in radio 1482 networks; 1484 * Development and specification of reverse path congestion control 1485 using L4S building bocks (e.g. AccECN, QUIC); 1487 * Once queuing delay is cut down, what becomes the 'second longest 1488 pole in the tent' (other than the speed of light)? 1490 * Novel alternatives to the existing set of L4S AQMs; 1491 * Novel applications enabled by L4S. 1493 8. IANA Considerations 1495 The 01 codepoint of the ECN Field of the IP header is specified by 1496 the present Experimental RFC. The process for an experimental RFC to 1497 assign this codepoint in the IP header (v4 and v6) is documented in 1498 Proposed Standard [RFC8311], which updates the Proposed Standard 1499 [RFC3168]. 1501 When the present document is published as an RFC, IANA is asked to 1502 update the 01 entry in the registry, "ECN Field (Bits 6-7)" to the 1503 following (see https://www.iana.org/assignments/dscp-registry/dscp- 1504 registry.xhtml#ecn-field ): 1506 +========+=====================+=============================+ 1507 | Binary | Keyword | References | 1508 +========+=====================+=============================+ 1509 | 01 | ECT(1) (ECN-Capable | [RFC8311] [RFC Errata 5399] | 1510 | | Transport(1))[1] | [RFCXXXX] | 1511 +--------+---------------------+-----------------------------+ 1513 Table 1 1515 [XXXX is the number that the RFC Editor assigns to the present 1516 document (this sentence to be removed by the RFC Editor)]. 1518 9. Security Considerations 1520 Approaches to assure the integrity of signals using the new 1521 identifier are introduced in Appendix C.1. See the security 1522 considerations in the L4S architecture [I-D.ietf-tsvwg-l4s-arch] for 1523 further discussion of mis-use of the identifier, as well as extensive 1524 discussion of policing rate and latency in regard to L4S. 1526 If the anti-replay window of a VPN egress is too small, it will 1527 mistake deliberate delay differences as a replay attack, and discard 1528 higher delay packets (e.g. Classic) carried within the same security 1529 association (SA) as low delay packets (e.g. L4S). Section 6.2 1530 recommends that VPNs used in L4S experiments are configured with a 1531 sufficiently large anti-replay window, as required by the relevant 1532 specifications. It also discusses other alternatives. 1534 If a user taking part in the L4S experiment sets up a VPN without 1535 being aware of the above advice, and if the user allows anyone to 1536 send traffic into their VPN, they would open up a DoS vulnerability 1537 in which an attacker could induce the VPN's anti-replay mechanism to 1538 discard enough of the user's Classic (C) traffic (if they are 1539 receiving any) to cause a significant rate reduction. While the user 1540 is actively downloading C traffic, the attacker sends C traffic into 1541 the VPN to fill the remainder of the bottleneck link, then sends 1542 intermittent L4S packets to maximize the chance of exceeding the 1543 VPN's replay window. The user can prevent this attack by following 1544 the recommendations in Section 6.2. 1546 The recommendation to detect loss in time units prevents the ACK- 1547 splitting attacks described in [Savage-TCP]. 1549 10. Acknowledgements 1551 Thanks to Richard Scheffenegger, John Leslie, David Taeht, Jonathan 1552 Morton, Gorry Fairhurst, Michael Welzl, Mikael Abrahamsson and Andrew 1553 McGregor for the discussions that led to this specification. Ing-jyh 1554 (Inton) Tsang was a contributor to the early drafts of this document. 1555 And thanks to Mikael Abrahamsson, Lloyd Wood, Nicolas Kuhn, Greg 1556 White, Tom Henderson, David Black, Gorry Fairhurst, Brian Carpenter, 1557 Jake Holland, Rod Grimes, Richard Scheffenegger, Sebastian Moeller, 1558 Neal Cardwell, Praveen Balasubramanian, Reza Marandian Hagh, Pete 1559 Heist, Stuart Cheshire, Vidhi Goel, Mirja Kuehlewind and Ermin Sakic 1560 for providing help and reviewing this draft and thanks to Ingemar 1561 Johansson for reviewing and providing substantial text. Thanks to 1562 Sebastian Moeller for identifying the interaction with VPN anti- 1563 replay and to Jonathan Morton for identifying the attack based on 1564 this. Particular thanks to tsvwg chairs Gorry Fairhurst, David Black 1565 and Wes Eddy for patiently helping this and the other L4S drafts 1566 through the IETF process. Appendix A listing the Prague L4S 1567 Requirements is based on text authored by Marcelo Bagnulo Braun that 1568 was originally an appendix to [I-D.ietf-tsvwg-l4s-arch]. That text 1569 was in turn based on the collective output of the attendees listed in 1570 the minutes of a 'bar BoF' on DCTCP Evolution during IETF-94 1571 [TCPPrague]. 1573 The authors' contributions were part-funded by the European Community 1574 under its Seventh Framework Programme through the Reducing Internet 1575 Transport Latency (RITE) project (ICT-317700). Bob Briscoe was also 1576 funded partly by the Research Council of Norway through the TimeIn 1577 project, partly by CableLabs and partly by the Comcast Innovation 1578 Fund. The views expressed here are solely those of the authors. 1580 11. References 1582 11.1. Normative References 1584 [RFC2119] Bradner, S., "Key words for use in RFCs to Indicate 1585 Requirement Levels", BCP 14, RFC 2119, 1586 DOI 10.17487/RFC2119, March 1997, 1587 . 1589 [RFC3168] Ramakrishnan, K., Floyd, S., and D. Black, "The Addition 1590 of Explicit Congestion Notification (ECN) to IP", 1591 RFC 3168, DOI 10.17487/RFC3168, September 2001, 1592 . 1594 [RFC4774] Floyd, S., "Specifying Alternate Semantics for the 1595 Explicit Congestion Notification (ECN) Field", BCP 124, 1596 RFC 4774, DOI 10.17487/RFC4774, November 2006, 1597 . 1599 [RFC6679] Westerlund, M., Johansson, I., Perkins, C., O'Hanlon, P., 1600 and K. Carlberg, "Explicit Congestion Notification (ECN) 1601 for RTP over UDP", RFC 6679, DOI 10.17487/RFC6679, August 1602 2012, . 1604 11.2. Informative References 1606 [A2DTCP] Zhang, T., Wang, J., Huang, J., Huang, Y., Chen, J., and 1607 Y. Pan, "Adaptive-Acceleration Data Center TCP", IEEE 1608 Transactions on Computers 64(6):1522-1533, June 2015, 1609 . 1612 [Ahmed19] Ahmed, A.S., "Extending TCP for Low Round Trip Delay", 1613 Masters Thesis, Uni Oslo , August 2019, 1614 . 1616 [Alizadeh-stability] 1617 Alizadeh, M., Javanmard, A., and B. Prabhakar, "Analysis 1618 of DCTCP: Stability, Convergence, and Fairness", ACM 1619 SIGMETRICS 2011 , June 2011, 1620 . 1623 [ARED01] Floyd, S., Gummadi, R., and S. Shenker, "Adaptive RED: An 1624 Algorithm for Increasing the Robustness of RED's Active 1625 Queue Management", ACIRI Technical Report , August 2001, 1626 . 1628 [COBALT] Palmei, J., Gupta, S., Imputato, P., Morton, J., 1629 Tahiliani, M., Avallone, S., and D. Taht, "Design and 1630 Evaluation of COBALT Queue Discipline", In Proc. IEEE 1631 Int'l Symp. on Local and Metropolitan Area Networks 2019, 1632 pp1--6, 2019, 1633 . 1635 [DCttH19] De Schepper, K., Bondarenko, O., Tilmans, O., and B. 1636 Briscoe, "`Data Centre to the Home': Ultra-Low Latency for 1637 All", Updated RITE project Technical Report , July 2019, 1638 . 1640 [DualPI2Linux] 1641 Albisser, O., De Schepper, K., Briscoe, B., Tilmans, O., 1642 and H. Steen, "DUALPI2 - Low Latency, Low Loss and 1643 Scalable (L4S) AQM", Proc. Linux Netdev 0x13 , March 2019, 1644 . 1647 [ecn-fallback] 1648 Briscoe, B. and A.S. Ahmed, "TCP Prague Fall-back on 1649 Detection of a Classic ECN AQM", bobbriscoe.net Technical 1650 Report TR-BB-2019-002, April 2020, 1651 . 1653 [Heist21] Heist, P. and J. Morton, "L4S Tests", github README, May 1654 2021, . 1656 [I-D.briscoe-docsis-q-protection] 1657 Briscoe, B. and G. White, "The DOCSIS(r) Queue Protection 1658 Algorithm to Preserve Low Latency", Work in Progress, 1659 Internet-Draft, draft-briscoe-docsis-q-protection-02, 31 1660 January 2022, . 1663 [I-D.briscoe-iccrg-prague-congestion-control] 1664 Schepper, K. D., Tilmans, O., and B. Briscoe, "Prague 1665 Congestion Control", Work in Progress, Internet-Draft, 1666 draft-briscoe-iccrg-prague-congestion-control-00, 9 March 1667 2021, . 1670 [I-D.briscoe-tsvwg-l4s-diffserv] 1671 Briscoe, B., "Interactions between Low Latency, Low Loss, 1672 Scalable Throughput (L4S) and Differentiated Services", 1673 Work in Progress, Internet-Draft, draft-briscoe-tsvwg-l4s- 1674 diffserv-02, 4 November 2018, 1675 . 1678 [I-D.cardwell-iccrg-bbr-congestion-control] 1679 Cardwell, N., Cheng, Y., Yeganeh, S. H., Swett, I., and V. 1680 Jacobson, "BBR Congestion Control", Work in Progress, 1681 Internet-Draft, draft-cardwell-iccrg-bbr-congestion- 1682 control-01, 7 November 2021, 1683 . 1686 [I-D.ietf-tcpm-accurate-ecn] 1687 Briscoe, B., Kühlewind, M., and R. Scheffenegger, "More 1688 Accurate ECN Feedback in TCP", Work in Progress, Internet- 1689 Draft, draft-ietf-tcpm-accurate-ecn-15, 12 July 2021, 1690 . 1693 [I-D.ietf-tcpm-generalized-ecn] 1694 Bagnulo, M. and B. Briscoe, "ECN++: Adding Explicit 1695 Congestion Notification (ECN) to TCP Control Packets", 1696 Work in Progress, Internet-Draft, draft-ietf-tcpm- 1697 generalized-ecn-09, 31 January 2022, 1698 . 1701 [I-D.ietf-tls-dtls13] 1702 Rescorla, E., Tschofenig, H., and N. Modadugu, "The 1703 Datagram Transport Layer Security (DTLS) Protocol Version 1704 1.3", Work in Progress, Internet-Draft, draft-ietf-tls- 1705 dtls13-43, 30 April 2021, 1706 . 1709 [I-D.ietf-trill-ecn-support] 1710 Eastlake, D. E. and B. Briscoe, "TRILL (TRansparent 1711 Interconnection of Lots of Links): ECN (Explicit 1712 Congestion Notification) Support", Work in Progress, 1713 Internet-Draft, draft-ietf-trill-ecn-support-07, 25 1714 February 2018, . 1717 [I-D.ietf-tsvwg-aqm-dualq-coupled] 1718 Schepper, K. D., Briscoe, B., and G. White, "DualQ Coupled 1719 AQMs for Low Latency, Low Loss and Scalable Throughput 1720 (L4S)", Work in Progress, Internet-Draft, draft-ietf- 1721 tsvwg-aqm-dualq-coupled-20, 24 December 2021, 1722 . 1725 [I-D.ietf-tsvwg-ecn-encap-guidelines] 1726 Briscoe, B. and J. Kaippallimalil, "Guidelines for Adding 1727 Congestion Notification to Protocols that Encapsulate IP", 1728 Work in Progress, Internet-Draft, draft-ietf-tsvwg-ecn- 1729 encap-guidelines-16, 25 May 2021, 1730 . 1733 [I-D.ietf-tsvwg-l4s-arch] 1734 Briscoe, B., Schepper, K. D., Bagnulo, M., and G. White, 1735 "Low Latency, Low Loss, Scalable Throughput (L4S) Internet 1736 Service: Architecture", Work in Progress, Internet-Draft, 1737 draft-ietf-tsvwg-l4s-arch-15, 24 December 2021, 1738 . 1741 [I-D.ietf-tsvwg-l4sops] 1742 White, G., "Operational Guidance for Deployment of L4S in 1743 the Internet", Work in Progress, Internet-Draft, draft- 1744 ietf-tsvwg-l4sops-02, 25 October 2021, 1745 . 1748 [I-D.ietf-tsvwg-nqb] 1749 White, G. and T. Fossati, "A Non-Queue-Building Per-Hop 1750 Behavior (NQB PHB) for Differentiated Services", Work in 1751 Progress, Internet-Draft, draft-ietf-tsvwg-nqb-08, 25 1752 October 2021, . 1755 [I-D.ietf-tsvwg-rfc6040update-shim] 1756 Briscoe, B., "Propagating Explicit Congestion Notification 1757 Across IP Tunnel Headers Separated by a Shim", Work in 1758 Progress, Internet-Draft, draft-ietf-tsvwg-rfc6040update- 1759 shim-14, 25 May 2021, 1760 . 1763 [I-D.sridharan-tcpm-ctcp] 1764 Sridharan, M., Tan, K., Bansal, D., and D. Thaler, 1765 "Compound TCP: A New TCP Congestion Control for High-Speed 1766 and Long Distance Networks", Work in Progress, Internet- 1767 Draft, draft-sridharan-tcpm-ctcp-02, 11 November 2008, 1768 . 1771 [I-D.stewart-tsvwg-sctpecn] 1772 Stewart, R. R., Tuexen, M., and X. Dong, "ECN for Stream 1773 Control Transmission Protocol (SCTP)", Work in Progress, 1774 Internet-Draft, draft-stewart-tsvwg-sctpecn-05, 15 January 1775 2014, . 1778 [LinuxPacedChirping] 1779 Misund, J. and B. Briscoe, "Paced Chirping - Rethinking 1780 TCP start-up", Proc. Linux Netdev 0x13 , March 2019, 1781 . 1783 [Mathis09] Mathis, M., "Relentless Congestion Control", PFLDNeT'09 , 1784 May 2009, . 1787 [Paced-Chirping] 1788 Misund, J., "Rapid Acceleration in TCP Prague", Masters 1789 Thesis , May 2018, 1790 . 1793 [PI2] De Schepper, K., Bondarenko, O., Tsang, I., and B. 1794 Briscoe, "PI^2 : A Linearized AQM for both Classic and 1795 Scalable TCP", Proc. ACM CoNEXT 2016 pp.105-119, December 1796 2016, 1797 . 1799 [PragueLinux] 1800 Briscoe, B., De Schepper, K., Albisser, O., Misund, J., 1801 Tilmans, O., Kühlewind, M., and A.S. Ahmed, "Implementing 1802 the `TCP Prague' Requirements for Low Latency Low Loss 1803 Scalable Throughput (L4S)", Proc. Linux Netdev 0x13 , 1804 March 2019, . 1807 [QV] Briscoe, B. and P. Hurtig, "Up to Speed with Queue View", 1808 RITE Technical Report D2.3; Appendix C.2, August 2015, 1809 . 1812 [RFC2309] Braden, B., Clark, D., Crowcroft, J., Davie, B., Deering, 1813 S., Estrin, D., Floyd, S., Jacobson, V., Minshall, G., 1814 Partridge, C., Peterson, L., Ramakrishnan, K., Shenker, 1815 S., Wroclawski, J., and L. Zhang, "Recommendations on 1816 Queue Management and Congestion Avoidance in the 1817 Internet", RFC 2309, DOI 10.17487/RFC2309, April 1998, 1818 . 1820 [RFC2474] Nichols, K., Blake, S., Baker, F., and D. Black, 1821 "Definition of the Differentiated Services Field (DS 1822 Field) in the IPv4 and IPv6 Headers", RFC 2474, 1823 DOI 10.17487/RFC2474, December 1998, 1824 . 1826 [RFC3246] Davie, B., Charny, A., Bennet, J.C.R., Benson, K., Le 1827 Boudec, J.Y., Courtney, W., Davari, S., Firoiu, V., and D. 1828 Stiliadis, "An Expedited Forwarding PHB (Per-Hop 1829 Behavior)", RFC 3246, DOI 10.17487/RFC3246, March 2002, 1830 . 1832 [RFC3540] Spring, N., Wetherall, D., and D. Ely, "Robust Explicit 1833 Congestion Notification (ECN) Signaling with Nonces", 1834 RFC 3540, DOI 10.17487/RFC3540, June 2003, 1835 . 1837 [RFC3649] Floyd, S., "HighSpeed TCP for Large Congestion Windows", 1838 RFC 3649, DOI 10.17487/RFC3649, December 2003, 1839 . 1841 [RFC4301] Kent, S. and K. Seo, "Security Architecture for the 1842 Internet Protocol", RFC 4301, DOI 10.17487/RFC4301, 1843 December 2005, . 1845 [RFC4302] Kent, S., "IP Authentication Header", RFC 4302, 1846 DOI 10.17487/RFC4302, December 2005, 1847 . 1849 [RFC4303] Kent, S., "IP Encapsulating Security Payload (ESP)", 1850 RFC 4303, DOI 10.17487/RFC4303, December 2005, 1851 . 1853 [RFC4340] Kohler, E., Handley, M., and S. Floyd, "Datagram 1854 Congestion Control Protocol (DCCP)", RFC 4340, 1855 DOI 10.17487/RFC4340, March 2006, 1856 . 1858 [RFC4341] Floyd, S. and E. Kohler, "Profile for Datagram Congestion 1859 Control Protocol (DCCP) Congestion Control ID 2: TCP-like 1860 Congestion Control", RFC 4341, DOI 10.17487/RFC4341, March 1861 2006, . 1863 [RFC4342] Floyd, S., Kohler, E., and J. Padhye, "Profile for 1864 Datagram Congestion Control Protocol (DCCP) Congestion 1865 Control ID 3: TCP-Friendly Rate Control (TFRC)", RFC 4342, 1866 DOI 10.17487/RFC4342, March 2006, 1867 . 1869 [RFC4960] Stewart, R., Ed., "Stream Control Transmission Protocol", 1870 RFC 4960, DOI 10.17487/RFC4960, September 2007, 1871 . 1873 [RFC5033] Floyd, S. and M. Allman, "Specifying New Congestion 1874 Control Algorithms", BCP 133, RFC 5033, 1875 DOI 10.17487/RFC5033, August 2007, 1876 . 1878 [RFC5129] Davie, B., Briscoe, B., and J. Tay, "Explicit Congestion 1879 Marking in MPLS", RFC 5129, DOI 10.17487/RFC5129, January 1880 2008, . 1882 [RFC5348] Floyd, S., Handley, M., Padhye, J., and J. Widmer, "TCP 1883 Friendly Rate Control (TFRC): Protocol Specification", 1884 RFC 5348, DOI 10.17487/RFC5348, September 2008, 1885 . 1887 [RFC5562] Kuzmanovic, A., Mondal, A., Floyd, S., and K. 1888 Ramakrishnan, "Adding Explicit Congestion Notification 1889 (ECN) Capability to TCP's SYN/ACK Packets", RFC 5562, 1890 DOI 10.17487/RFC5562, June 2009, 1891 . 1893 [RFC5622] Floyd, S. and E. Kohler, "Profile for Datagram Congestion 1894 Control Protocol (DCCP) Congestion ID 4: TCP-Friendly Rate 1895 Control for Small Packets (TFRC-SP)", RFC 5622, 1896 DOI 10.17487/RFC5622, August 2009, 1897 . 1899 [RFC5681] Allman, M., Paxson, V., and E. Blanton, "TCP Congestion 1900 Control", RFC 5681, DOI 10.17487/RFC5681, September 2009, 1901 . 1903 [RFC5706] Harrington, D., "Guidelines for Considering Operations and 1904 Management of New Protocols and Protocol Extensions", 1905 RFC 5706, DOI 10.17487/RFC5706, November 2009, 1906 . 1908 [RFC5865] Baker, F., Polk, J., and M. Dolly, "A Differentiated 1909 Services Code Point (DSCP) for Capacity-Admitted Traffic", 1910 RFC 5865, DOI 10.17487/RFC5865, May 2010, 1911 . 1913 [RFC5925] Touch, J., Mankin, A., and R. Bonica, "The TCP 1914 Authentication Option", RFC 5925, DOI 10.17487/RFC5925, 1915 June 2010, . 1917 [RFC6040] Briscoe, B., "Tunnelling of Explicit Congestion 1918 Notification", RFC 6040, DOI 10.17487/RFC6040, November 1919 2010, . 1921 [RFC6077] Papadimitriou, D., Ed., Welzl, M., Scharf, M., and B. 1922 Briscoe, "Open Research Issues in Internet Congestion 1923 Control", RFC 6077, DOI 10.17487/RFC6077, February 2011, 1924 . 1926 [RFC6347] Rescorla, E. and N. Modadugu, "Datagram Transport Layer 1927 Security Version 1.2", RFC 6347, DOI 10.17487/RFC6347, 1928 January 2012, . 1930 [RFC6660] Briscoe, B., Moncaster, T., and M. Menth, "Encoding Three 1931 Pre-Congestion Notification (PCN) States in the IP Header 1932 Using a Single Diffserv Codepoint (DSCP)", RFC 6660, 1933 DOI 10.17487/RFC6660, July 2012, 1934 . 1936 [RFC6675] Blanton, E., Allman, M., Wang, L., Jarvinen, I., Kojo, M., 1937 and Y. Nishida, "A Conservative Loss Recovery Algorithm 1938 Based on Selective Acknowledgment (SACK) for TCP", 1939 RFC 6675, DOI 10.17487/RFC6675, August 2012, 1940 . 1942 [RFC7560] Kuehlewind, M., Ed., Scheffenegger, R., and B. Briscoe, 1943 "Problem Statement and Requirements for Increased Accuracy 1944 in Explicit Congestion Notification (ECN) Feedback", 1945 RFC 7560, DOI 10.17487/RFC7560, August 2015, 1946 . 1948 [RFC7567] Baker, F., Ed. and G. Fairhurst, Ed., "IETF 1949 Recommendations Regarding Active Queue Management", 1950 BCP 197, RFC 7567, DOI 10.17487/RFC7567, July 2015, 1951 . 1953 [RFC7713] Mathis, M. and B. Briscoe, "Congestion Exposure (ConEx) 1954 Concepts, Abstract Mechanism, and Requirements", RFC 7713, 1955 DOI 10.17487/RFC7713, December 2015, 1956 . 1958 [RFC8033] Pan, R., Natarajan, P., Baker, F., and G. White, 1959 "Proportional Integral Controller Enhanced (PIE): A 1960 Lightweight Control Scheme to Address the Bufferbloat 1961 Problem", RFC 8033, DOI 10.17487/RFC8033, February 2017, 1962 . 1964 [RFC8083] Perkins, C. and V. Singh, "Multimedia Congestion Control: 1965 Circuit Breakers for Unicast RTP Sessions", RFC 8083, 1966 DOI 10.17487/RFC8083, March 2017, 1967 . 1969 [RFC8085] Eggert, L., Fairhurst, G., and G. Shepherd, "UDP Usage 1970 Guidelines", BCP 145, RFC 8085, DOI 10.17487/RFC8085, 1971 March 2017, . 1973 [RFC8257] Bensley, S., Thaler, D., Balasubramanian, P., Eggert, L., 1974 and G. Judd, "Data Center TCP (DCTCP): TCP Congestion 1975 Control for Data Centers", RFC 8257, DOI 10.17487/RFC8257, 1976 October 2017, . 1978 [RFC8290] Hoeiland-Joergensen, T., McKenney, P., Taht, D., Gettys, 1979 J., and E. Dumazet, "The Flow Queue CoDel Packet Scheduler 1980 and Active Queue Management Algorithm", RFC 8290, 1981 DOI 10.17487/RFC8290, January 2018, 1982 . 1984 [RFC8298] Johansson, I. and Z. Sarker, "Self-Clocked Rate Adaptation 1985 for Multimedia", RFC 8298, DOI 10.17487/RFC8298, December 1986 2017, . 1988 [RFC8311] Black, D., "Relaxing Restrictions on Explicit Congestion 1989 Notification (ECN) Experimentation", RFC 8311, 1990 DOI 10.17487/RFC8311, January 2018, 1991 . 1993 [RFC8312] Rhee, I., Xu, L., Ha, S., Zimmermann, A., Eggert, L., and 1994 R. Scheffenegger, "CUBIC for Fast Long-Distance Networks", 1995 RFC 8312, DOI 10.17487/RFC8312, February 2018, 1996 . 1998 [RFC8511] Khademi, N., Welzl, M., Armitage, G., and G. Fairhurst, 1999 "TCP Alternative Backoff with ECN (ABE)", RFC 8511, 2000 DOI 10.17487/RFC8511, December 2018, 2001 . 2003 [RFC8888] Sarker, Z., Perkins, C., Singh, V., and M. Ramalho, "RTP 2004 Control Protocol (RTCP) Feedback for Congestion Control", 2005 RFC 8888, DOI 10.17487/RFC8888, January 2021, 2006 . 2008 [RFC8985] Cheng, Y., Cardwell, N., Dukkipati, N., and P. Jha, "The 2009 RACK-TLP Loss Detection Algorithm for TCP", RFC 8985, 2010 DOI 10.17487/RFC8985, February 2021, 2011 . 2013 [RFC9000] Iyengar, J., Ed. and M. Thomson, Ed., "QUIC: A UDP-Based 2014 Multiplexed and Secure Transport", RFC 9000, 2015 DOI 10.17487/RFC9000, May 2021, 2016 . 2018 [Savage-TCP] 2019 Savage, S., Cardwell, N., Wetherall, D., and T. Anderson, 2020 "TCP Congestion Control with a Misbehaving Receiver", ACM 2021 SIGCOMM Computer Communication Review 29(5):71--78, 2022 October 1999. 2024 [SCReAM] Johansson, I., "SCReAM", github repository; , 2025 . 2028 [sub-mss-prob] 2029 Briscoe, B. and K. De Schepper, "Scaling TCP's Congestion 2030 Window for Small Round Trip Times", BT Technical Report 2031 TR-TUB8-2015-002, May 2015, 2032 . 2034 [TCP-CA] Jacobson, V. and M.J. Karels, "Congestion Avoidance and 2035 Control", Laurence Berkeley Labs Technical Report , 2036 November 1988, . 2038 [TCPPrague] 2039 Briscoe, B., "Notes: DCTCP evolution 'bar BoF': Tue 21 Jul 2040 2015, 17:40, Prague", tcpprague mailing list archive , 2041 July 2015, . 2044 [VCP] Xia, Y., Subramanian, L., Stoica, I., and S. Kalyanaraman, 2045 "One more bit is enough", Proc. SIGCOMM'05, ACM CCR 2046 35(4)37--48, 2005, 2047 . 2049 Appendix A. Rationale for the 'Prague L4S Requirements' 2051 This appendix is informative, not normative. It gives a list of 2052 modifications to current scalable congestion controls so that they 2053 can be deployed over the public Internet and coexist safely with 2054 existing traffic. The list complements the normative requirements in 2055 Section 4 that a sender has to comply with before it can set the L4S 2056 identifier in packets it sends into the Internet. As well as 2057 rationale for safety improvements (the requirements in Section 4) 2058 this appendix also includes preferable performance improvements 2059 (optimizations). 2061 The requirements and recommendations in Section 4) have become know 2062 as the Prague L4S Requirements, because they were originally 2063 identified at an ad hoc meeting during IETF-94 in Prague [TCPPrague]. 2064 They were originally called the 'TCP Prague Requirements', but they 2065 are not solely applicable to TCP, so the name and wording has been 2066 generalized for all transport protocols, and the name 'TCP Prague' is 2067 now used for a specific implementation of the requirements. 2069 At the time of writing, DCTCP [RFC8257] is the most widely used 2070 scalable transport protocol. In its current form, DCTCP is specified 2071 to be deployable only in controlled environments. Deploying it in 2072 the public Internet would lead to a number of issues, both from the 2073 safety and the performance perspective. The modifications and 2074 additional mechanisms listed in this section will be necessary for 2075 its deployment over the global Internet. Where an example is needed, 2076 DCTCP is used as a base, but the requirements in Section 4 apply 2077 equally to other scalable congestion controls, covering adaptive 2078 real-time media, etc., not just capacity-seeking behaviours. 2080 A.1. Rationale for the Requirements for Scalable Transport Protocols 2081 A.1.1. Use of L4S Packet Identifier 2083 Description: A scalable congestion control needs to distinguish the 2084 packets it sends from those sent by Classic congestion controls (see 2085 the precise normative requirement wording in Section 4.1). 2087 Motivation: It needs to be possible for a network node to classify 2088 L4S packets without flow state into a queue that applies an L4S ECN 2089 marking behaviour and isolates L4S packets from the queuing delay of 2090 Classic packets. 2092 A.1.2. Accurate ECN Feedback 2094 Description: The transport protocol for a scalable congestion control 2095 needs to provide timely, accurate feedback about the extent of ECN 2096 marking experienced by all packets (see the precise normative 2097 requirement wording in Section 4.2). 2099 Motivation: Classic congestion controls only need feedback about the 2100 existence of a congestion episode within a round trip, not precisely 2101 how many packets were marked with ECN or dropped. Therefore, in 2102 2001, when ECN feedback was added to TCP [RFC3168], it could not 2103 inform the sender of more than one ECN mark per RTT. Since then, 2104 requirements for more accurate ECN feedback in TCP have been defined 2105 in [RFC7560] and [I-D.ietf-tcpm-accurate-ecn] specifies a change to 2106 the TCP protocol to satisfy these requirements. Most other transport 2107 protocols already satisfy this requirement (see Section 4.2). 2109 A.1.3. Capable of Replacement by Classic Congestion Control 2111 Description: It needs to be possible to replace the implementation of 2112 a scalable congestion control with a Classic control (see the precise 2113 normative requirement wording in Section 4.3). 2115 Motivation: L4S is an experimental protocol, therefore it seems 2116 prudent to be able to disable it at source in case of insurmountable 2117 problems, perhaps due to some unexpected interaction on a particular 2118 sender; over a particular path or network; with a particular receiver 2119 or even ultimately an insurmountable problem with the experiment as a 2120 whole. 2122 A.1.4. Fall back to Classic Congestion Control on Packet Loss 2124 Description: As well as responding to ECN markings in a scalable way, 2125 a scalable congestion control needs to react to packet loss in a way 2126 that will coexist safely with a Reno congestion control [RFC5681] 2127 (see the precise normative requirement wording in Section 4.3). 2129 Motivation: Part of the safety conditions for deploying a scalable 2130 congestion control on the public Internet is to make sure that it 2131 behaves properly when it builds a queue at a network bottleneck that 2132 has not been upgraded to support L4S. Packet loss can have many 2133 causes, but it usually has to be conservatively assumed that it is a 2134 sign of congestion. Therefore, on detecting packet loss, a scalable 2135 congestion control will need to fall back to Classic congestion 2136 control behaviour. If it does not comply, it could starve Classic 2137 traffic. 2139 A scalable congestion control can be used for different types of 2140 transport, e.g. for real-time media or for reliable transport like 2141 TCP. Therefore, the particular Classic congestion control behaviour 2142 to fall back on will need to be dependent on the specific congestion 2143 control implementation. In the particular case of DCTCP, the DCTCP 2144 specification [RFC8257] states that "It is RECOMMENDED that an 2145 implementation deal with loss episodes in the same way as 2146 conventional TCP." For safe deployment, Section 4.3 requires any 2147 specification of a scalable congestion control for the public 2148 Internet to define the above requirement as a "MUST". 2150 Even though a bottleneck is L4S capable, it might still become 2151 overloaded and have to drop packets. In this case, the sender may 2152 receive a high proportion of packets marked with the CE bit set and 2153 also experience loss. Current DCTCP implementations each react 2154 differently to this situation. At least one implementation reacts 2155 only to the drop signal (e.g. by halving the CWND) and at least 2156 another DCTCP implementation reacts to both signals (e.g. by halving 2157 the CWND due to the drop and also further reducing the CWND based on 2158 the proportion of marked packet). A third approach for the public 2159 Internet has been proposed that adjusts the loss response to result 2160 in a halving when combined with the ECN response. We believe that 2161 further experimentation is needed to understand what is the best 2162 behaviour for the public Internet, which may or not be one of these 2163 existing approaches. 2165 A.1.5. Coexistence with Classic Congestion Control at Classic ECN 2166 bottlenecks 2168 Description: Monitoring has to be in place so that a non-L4S but ECN- 2169 capable AQM can be detected at path bottlenecks. This is in case 2170 such an AQM has been implemented in a shared queue, in which case any 2171 long-running scalable flow would predominate over any simultaneous 2172 long-running Classic flow sharing the queue. The precise requirement 2173 wording in Section 4.3 is written so that such a problem could either 2174 be resolved in real-time, or via administrative intervention. 2176 Motivation: Similarly to the discussion in Appendix A.1.4, this 2177 requirement in Section 4.3 is a safety condition to ensure an L4S 2178 congestion control coexists well with Classic flows when it builds a 2179 queue at a shared network bottleneck that has not been upgraded to 2180 support L4S. Nonetheless, if necessary, it is considered reasonable 2181 to resolve such problems over management timescales (possibly 2182 involving human intervention) because: 2184 * although a Classic flow can considerably reduce its throughput in 2185 the face of a competing scalable flow, it still makes progress and 2186 does not starve; 2188 * implementations of a Classic ECN AQM in a queue that is intended 2189 to be shared are believed to be rare; 2191 * detection of such AQMs is not always clear-cut; so focused out-of- 2192 band testing (or even contacting the relevant network operator) 2193 would improve certainty. 2195 Therefore, the relevant normative requirement (Section 4.3) is 2196 divided into three stages: monitoring, detection and action: 2198 Monitoring: Monitoring involves collection of the measurement data 2199 to be analysed. Monitoring is expressed as a 'MUST' for 2200 uncontrolled environments, although the placement of the 2201 monitoring function is left open. Whether monitoring has to be 2202 applied in real-time is expressed as a 'SHOULD'. This allows for 2203 the possibility that the operator of an L4S sender (e.g. a CDN) 2204 might prefer to test out-of-band for signs of Classic ECN AQMs, 2205 perhaps to avoid continually consuming resources to monitor live 2206 traffic. 2208 Detection: Detection involves analysis of the monitored data to 2209 detect the likelihood of a Classic ECN AQM. Detection can either 2210 directly detect actual coexistence problems between flows, or it 2211 can aim to identify AQM technologies that are likely to present 2212 coexistence problems, based on knowledge of AQMs deployed at the 2213 time. The requirements recommend that detection occurs live in 2214 real-time. However, detection is allowed to be deferred (e.g. it 2215 might involve further testing targeted at candidate AQMs); 2217 Action: This involves the act of switching the sender to a Classic 2218 congestion control. This might occur in real-time within the 2219 congestion control for the subsequent duration of a flow, or it 2220 might involve administrative action to switch to Classic 2221 congestion control for a specific interface or for a certain set 2222 of destination addresses. 2224 Instead of the sender taking action itself, the operator of the 2225 sender (e.g. a CDN) might prefer to ask the network operator to 2226 modify the Classic AQM's treatment of L4S packets; or to ensure 2227 L4S packets bypass the AQM; or to upgrade the AQM to support L4S 2228 (see [I-D.ietf-tsvwg-l4sops]). Once L4S flows no longer shared 2229 the Classic ECN AQM they would obviously no longer detect it, and 2230 the requirement to act on it would no longer apply. 2232 The whole set of normative requirements concerning Classic ECN AQMs 2233 in Section 4.3 is worded so that it does not apply in controlled 2234 environments, such as private networks or data centre networks. CDN 2235 servers placed within an access ISP's network can be considered as a 2236 single controlled environment, but any onward networks served by the 2237 access network, including all the attached customer networks, would 2238 be unlikely to fall under the same degree of coordinated control. 2239 Monitoring is expressed as a 'MUST' for these uncontrolled segments 2240 of paths (e.g. beyond the access ISP in a home network), because 2241 there is a possibility that there might be a shared queue Classic ECN 2242 AQM in that segment. Nonetheless, the intent of the wording is to 2243 only require occasional monitoring of these uncontrolled regions, and 2244 not to burden CDN operators if monitoring never uncovers any 2245 potential problems. 2247 More detailed discussion of all the above options and alternatives 2248 can be found in [I-D.ietf-tsvwg-l4sops]. 2250 Having said all the above, the approach recommended in Section 4.3 is 2251 to monitor, detect and act in real-time on live traffic. A passive 2252 monitoring algorithm to detect a Classic ECN AQM at the bottleneck 2253 and fall back to Classic congestion control is described in an 2254 extensive technical report [ecn-fallback], which also provides a link 2255 to Linux source code, and a large online visualization of its 2256 evaluation results. Very briefly, the algorithm primarily monitors 2257 RTT variation using the same algorithm that maintains the mean 2258 deviation of TCP's smoothed RTT, but it smooths over a duration of 2259 the order of a Classic sawtooth. The outcome is also conditioned on 2260 other metrics such as the presence of CE marking and congestion 2261 avoidance phase having stabilized. The report also identifies 2262 further work to improve the approach, for instance improvements with 2263 low capacity links and combining the measurements with a cache of 2264 what had been learned about a path in previous connections. The 2265 report also suggests alternative approaches. 2267 Although using passive measurements within live traffic (as above) 2268 can detect a Classic ECN AQM, it is much harder (perhaps impossible) 2269 to determine whether or not the AQM is in a shared queue. 2270 Nonetheless, this is much easier using active test traffic out-of- 2271 band, because two flows can be used. Section 4 of the same report 2272 [ecn-fallback] describes a simple technique to detect a Classic ECN 2273 AQM and determine whether it is in a shared queue, summarized here. 2275 An L4S-enabled test server could be set up so that, when a test 2276 client accesses it, it serves a script that gets the client to open 2277 two parallel long-running flows. It could serve one with a Classic 2278 congestion control (C, that sets ECT(0)) and one with a scalable CC 2279 (L, that sets ECT(1)). If neither flow induces any ECN marks, it can 2280 be presumed the path does not contain a Classic ECN AQM. If either 2281 flow induces some ECN marks, the server could measure the relative 2282 flow rates and round trip times of the two flows. Table 2 shows the 2283 AQM that can be inferred for various cases (presuming the AQM 2284 behaviours known at the time of writing). 2286 +========+=======+========================+ 2287 | Rate | RTT | Inferred AQM | 2288 +========+=======+========================+ 2289 | L > C | L = C | Classic ECN AQM (FIFO) | 2290 +--------+-------+------------------------+ 2291 | L = C | L = C | Classic ECN AQM (FQ) | 2292 +--------+-------+------------------------+ 2293 | L = C | L < C | FQ-L4S AQM | 2294 +--------+-------+------------------------+ 2295 | L ~= C | L < C | Coupled DualQ AQM | 2296 +--------+-------+------------------------+ 2298 Table 2: Out-of-band testing with two 2299 parallel flows. L:=L4S, C:=Classic. 2301 Finally, we motivate the recommendation in Section 4.3 that a 2302 scalable congestion control is not expected to change to setting 2303 ECT(0) while it adapts its behaviour to coexist with Classic flows. 2304 This is because the sender needs to continue to check whether it made 2305 the right decision - and switch back if it was wrong, or if a 2306 different link becomes the bottleneck: 2308 * If, as recommended, the sender changes only its behaviour but not 2309 its codepoint to Classic, its codepoint will still be compatible 2310 with either an L4S or a Classic AQM. If the bottleneck does 2311 actually support both, it will still classify ECT(1) into the same 2312 L4S queue, where the sender can measure that switching to Classic 2313 behaviour was wrong, so that it can switch back. 2315 * In contrast, if the sender changes both its behaviour and its 2316 codepoint to Classic, even if the bottleneck supports both, it 2317 will classify ECT(0) into the Classic queue, reinforcing the 2318 sender's incorrect decision so that it never switches back. 2320 * Also, not changing codepoint avoids the risk of being flipped to a 2321 different path by a load balancer or multipath routing that hashes 2322 on the whole of the ex-ToS byte (unfortunately still a common 2323 pathology). 2325 Note that if a flow is configured to _only_ use a Classic congestion 2326 control, it is then entirely appropriate not to use ECT(1). 2328 A.1.6. Reduce RTT dependence 2330 Description: A scalable congestion control needs to reduce RTT bias 2331 as much as possible at least over the low to typical range of RTTs 2332 that will interact in the intended deployment scenario (see the 2333 precise normative requirement wording in Section 4.3). 2335 Motivation: The throughput of Classic congestion controls is known to 2336 be inversely proportional to RTT, so one would expect flows over very 2337 low RTT paths to nearly starve flows over larger RTTs. However, 2338 Classic congestion controls have never allowed a very low RTT path to 2339 exist because they induce a large queue. For instance, consider two 2340 paths with base RTT 1 ms and 100 ms. If a Classic congestion control 2341 induces a 100 ms queue, it turns these RTTs into 101 ms and 200 ms 2342 leading to a throughput ratio of about 2:1. Whereas if a scalable 2343 congestion control induces only a 1 ms queue, the ratio is 2:101, 2344 leading to a throughput ratio of about 50:1. 2346 Therefore, with very small queues, long RTT flows will essentially 2347 starve, unless scalable congestion controls comply with this 2348 requirement in Section 4.3. 2350 The RTT bias in current Classic congestion controls works 2351 satisfactorily when the RTT is higher than typical, and L4S does not 2352 change that. So, there is no additional requirement in Section 4.3 2353 for high RTT L4S flows to remove RTT bias - they can but they don't 2354 have to. 2356 A.1.7. Scaling down to fractional congestion windows 2358 Description: A scalable congestion control needs to remain responsive 2359 to congestion when typical RTTs over the public Internet are 2360 significantly smaller because they are no longer inflated by queuing 2361 delay (see the precise normative requirement wording in Section 4.3). 2363 Motivation: As currently specified, the minimum congestion window of 2364 ECN-capable TCP (and its derivatives) is expected to be 2 sender 2365 maximum segment sizes (SMSS), or 1 SMSS after a retransmission 2366 timeout. Once the congestion window reaches this minimum, if there 2367 is further ECN-marking, TCP is meant to wait for a retransmission 2368 timeout before sending another segment (see section 6.1.2 of 2369 [RFC3168]). In practice, most known window-based congestion control 2370 algorithms become unresponsive to ECN congestion signals at this 2371 point. No matter how much ECN marking, the congestion window no 2372 longer reduces. Instead, the sender's lack of any further congestion 2373 response forces the queue to grow, overriding any AQM and increasing 2374 queuing delay (making the window large enough to become responsive 2375 again). This can result in a stable but deeper queue, or it might 2376 drive the queue to loss, then the retransmission timeout mechanism 2377 acts as a backstop. 2379 Most window-based congestion controls for other transport protocols 2380 have a similar minimum window, albeit when measured in bytes for 2381 those that use smaller packets. 2383 L4S mechanisms significantly reduce queueing delay so, over the same 2384 path, the RTT becomes lower. Then this problem becomes surprisingly 2385 common [sub-mss-prob]. This is because, for the same link capacity, 2386 smaller RTT implies a smaller window. For instance, consider a 2387 residential setting with an upstream broadband Internet access of 8 2388 Mb/s, assuming a max segment size of 1500 B. Two upstream flows will 2389 each have the minimum window of 2 SMSS if the RTT is 6 ms or less, 2390 which is quite common when accessing a nearby data centre. So, any 2391 more than two such parallel TCP flows will become unresponsive to ECN 2392 and increase queuing delay. 2394 Unless scalable congestion controls address the requirement in 2395 Section 4.3 from the start, they will frequently become unresponsive 2396 to ECN, negating the low latency benefit of L4S, for themselves and 2397 for others. 2399 That would seem to imply that scalable congestion controllers ought 2400 to be required to be able work with a congestion window less than 2401 1 SMSS. For instance, if an ECN-capable TCP gets an ECN-mark when it 2402 is already sitting at a window of 1 SMSS, RFC 3168 requires it to 2403 defer sending for a retransmission timeout. A less drastic but more 2404 complex mechanism can maintain a congestion window less than 1 SMSS 2405 (significantly less if necessary), as described in [Ahmed19]. Other 2406 approaches are likely to be feasible. 2408 However, the requirement in Section 4.3 is worded as a "SHOULD" 2409 because it is believed that the existence of a minimum window is not 2410 all bad. When competing with an unresponsive flow, a minimum window 2411 naturally protects the flow from starvation by at least keeping some 2412 data flowing. 2414 By stating the requirement to go lower than 1 SMSS as a "SHOULD", 2415 while the requirement in RFC 3168 still stands as well, we shall be 2416 able to watch the choices of minimum window evolve in different 2417 scalable congestion controllers. 2419 A.1.8. Measuring Reordering Tolerance in Time Units 2421 Description: When detecting loss, a scalable congestion control needs 2422 to be tolerant to reordering over an adaptive time interval, which 2423 scales with throughput, rather than counting only in fixed units of 2424 packets, which does not scale (see the precise normative requirement 2425 wording in Section 4.3). 2427 Motivation: A primary purpose of L4S is scalable throughput (it's in 2428 the name). Scalability in all dimensions is, of course, also a goal 2429 of all IETF technology. The inverse linear congestion response in 2430 Section 4.3 is necessary, but not sufficient, to solve the congestion 2431 control scalability problem identified in [RFC3649]. As well as 2432 maintaining frequent ECN signals as rate scales, it is also important 2433 to ensure that a potentially false perception of loss does not limit 2434 throughput scaling. 2436 End-systems cannot know whether a missing packet is due to loss or 2437 reordering, except in hindsight - if it appears later. So they can 2438 only deem that there has been a loss if a gap in the sequence space 2439 has not been filled, either after a certain number of subsequent 2440 packets has arrived (e.g. the 3 DupACK rule of standard TCP 2441 congestion control [RFC5681]) or after a certain amount of time 2442 (e.g. the RACK approach [RFC8985]). 2444 As we attempt to scale packet rate over the years: 2446 * Even if only _some_ sending hosts still deem that loss has 2447 occurred by counting reordered packets, _all_ networks will have 2448 to keep reducing the time over which they keep packets in order. 2449 If some link technologies keep the time within which reordering 2450 occurs roughly unchanged, then loss over these links, as perceived 2451 by these hosts, will appear to continually rise over the years. 2453 * In contrast, if all senders detect loss in units of time, the time 2454 over which the network has to keep packets in order stays roughly 2455 invariant. 2457 Therefore hosts have an incentive to detect loss in time units (so as 2458 not to fool themselves too often into detecting losses when there are 2459 none). And for hosts that are changing their congestion control 2460 implementation to L4S, there is no downside to including time-based 2461 loss detection code in the change (loss recovery implemented in 2462 hardware is an exception, covered later). Therefore requiring L4S 2463 hosts to detect loss in time-based units would not be a burden. 2465 If the requirement in Section 4.3 were not placed on L4S hosts, even 2466 though it would be no burden on hosts to comply, all networks would 2467 face unnecessary uncertainty over whether some L4S hosts might be 2468 detecting loss by counting packets. Then _all_ link technologies 2469 will have to unnecessarily keep reducing the time within which 2470 reordering occurs. That is not a problem for some link technologies, 2471 but it becomes increasingly challenging for other link technologies 2472 to continue to scale, particularly those relying on channel bonding 2473 for scaling, such as LTE, 5G and DOCSIS. 2475 Given Internet paths traverse many link technologies, any scaling 2476 limit for these more challenging access link technologies would 2477 become a scaling limit for the Internet as a whole. 2479 It might be asked how it helps to place this loss detection 2480 requirement only on L4S hosts, because networks will still face 2481 uncertainty over whether non-L4S flows are detecting loss by counting 2482 DupACKs. The answer is that those link technologies for which it is 2483 challenging to keep squeezing the reordering time will only need to 2484 do so for non-L4S traffic (which they can do because the L4S 2485 identifier is visible at the IP layer). Therefore, they can focus 2486 their processing and memory resources into scaling non-L4S (Classic) 2487 traffic. Then, the higher the proportion of L4S traffic, the less of 2488 a scaling challenge they will have. 2490 To summarize, there is no reason for L4S hosts not to be part of the 2491 solution instead of part of the problem. 2493 Requirement ("MUST") or recommendation ("SHOULD")? As explained 2494 above, this is a subtle interoperability issue between hosts and 2495 networks, which seems to need a "MUST". Unless networks can be 2496 certain that all L4S hosts follow the time-based approach, they still 2497 have to cater for the worst case - continually squeeze reordering 2498 into a smaller and smaller duration - just for hosts that might be 2499 using the counting approach. However, it was decided to express this 2500 as a recommendation, using "SHOULD". The main justification was that 2501 networks can still be fairly certain that L4S hosts will follow this 2502 recommendation, because following it offers only gain and no pain. 2504 Details: 2506 The speed of loss recovery is much more significant for short flows 2507 than long, therefore a good compromise is to adapt the reordering 2508 window; from a small fraction of the RTT at the start of a flow, to a 2509 larger fraction of the RTT for flows that continue for many round 2510 trips. 2512 This is broadly the approach adopted by TCP RACK (Recent 2513 ACKnowledgements) [RFC8985]. However, RACK starts with the 3 DupACK 2514 approach, because the RTT estimate is not necessarily stable. As 2515 long as the initial window is paced, such initial use of 3 DupACK 2516 counting would amount to time-based loss detection and therefore 2517 would satisfy the time-based loss detection recommendation of 2518 Section 4.3. This is because pacing of the initial window would 2519 ensure that 3 DupACKs early in the connection would be spread over a 2520 small fraction of the round trip. 2522 As mentioned above, hardware implementations of loss recovery using 2523 DupACK counting exist (e.g. some implementations of RoCEv2 for RDMA). 2524 For low latency, these implementations can change their congestion 2525 control to implement L4S, because the congestion control (as distinct 2526 from loss recovery) is implemented in software. But they cannot 2527 easily satisfy this loss recovery requirement. However, it is 2528 believed they do not need to, because such implementations are 2529 believed to solely exist in controlled environments, where the 2530 network technology keeps reordering extremely low anyway. This is 2531 why controlled environments with hardly any reordering are excluded 2532 from the scope of the normative recommendation in Section 4.3. 2534 Detecting loss in time units also prevents the ACK-splitting attacks 2535 described in [Savage-TCP]. 2537 A.2. Scalable Transport Protocol Optimizations 2539 A.2.1. Setting ECT in Control Packets and Retransmissions 2541 Description: This item concerns TCP and its derivatives (e.g. SCTP) 2542 as well as RTP/RTCP [RFC6679]. The original specification of ECN for 2543 TCP precluded the use of ECN on control packets and retransmissions. 2544 Similarly [RFC6679] precludes the use of ECT on RTCP datagrams, in 2545 case the path changes after it has been checked for ECN traversal. 2546 To improve performance, scalable transport protocols ought to enable 2547 ECN at the IP layer in TCP control packets (SYN, SYN-ACK, pure ACKs, 2548 etc.) and in retransmitted packets. The same is true for other 2549 transports, e.g. SCTP, RTCP. 2551 Motivation (TCP): RFC 3168 prohibits the use of ECN on these types of 2552 TCP packet, based on a number of arguments. This means these packets 2553 are not protected from congestion loss by ECN, which considerably 2554 harms performance, particularly for short flows. 2555 [I-D.ietf-tcpm-generalized-ecn] proposes experimental use of ECN on 2556 all types of TCP packet as long as AccECN feedback 2557 [I-D.ietf-tcpm-accurate-ecn] is available (which itself satisfies the 2558 accurate feedback requirement in Section 4.2 for using a scalable 2559 congestion control). 2561 Motivation (RTCP): L4S experiments in general will need to observe 2562 the rule in [RFC6679] that precludes ECT on RTCP datagrams. 2563 Nonetheless, as ECN usage becomes more widespread, it would be useful 2564 to conduct specific experiments with ECN-capable RTCP to gather data 2565 on whether such caution is necessary. 2567 A.2.2. Faster than Additive Increase 2569 Description: It would improve performance if scalable congestion 2570 controls did not limit their congestion window increase to the 2571 standard additive increase of 1 SMSS per round trip [RFC5681] during 2572 congestion avoidance. The same is true for derivatives of TCP 2573 congestion control, including similar approaches used for real-time 2574 media. 2576 Motivation: As currently defined [RFC8257], DCTCP uses the 2577 traditional Reno additive increase in congestion avoidance phase. 2578 When the available capacity suddenly increases (e.g. when another 2579 flow finishes, or if radio capacity increases) it can take very many 2580 round trips to take advantage of the new capacity. TCP Cubic 2581 [RFC8312] was designed to solve this problem, but as flow rates have 2582 continued to increase, the delay accelerating into available capacity 2583 has become prohibitive. See, for instance, the examples in 2584 Section 5.1 of [I-D.ietf-tsvwg-l4s-arch]. Even when out of its Reno- 2585 compatibility mode, every 8x scaling of Cubic's flow rate leads to 2x 2586 more acceleration delay. 2588 In the steady state, DCTCP induces about 2 ECN marks per round trip, 2589 so it is possible to quickly detect when these signals have 2590 disappeared and seek available capacity more rapidly, while 2591 minimizing the impact on other flows (Classic and scalable) 2592 [LinuxPacedChirping]. Alternatively, approaches such as Adaptive 2593 Acceleration (A2DTCP [A2DTCP]) have been proposed to address this 2594 problem in data centres, which might be deployable over the public 2595 Internet. 2597 A.2.3. Faster Convergence at Flow Start 2599 Description: It would improve performance if scalable congestion 2600 controls converged (reached their steady-state share of the capacity) 2601 faster than Classic congestion controls or at least no slower. This 2602 affects the flow start behaviour of any L4S congestion control 2603 derived from a Classic transport that uses TCP slow start, including 2604 those for real-time media. 2606 Motivation: As an example, a new DCTCP flow takes longer than a 2607 Classic congestion control to obtain its share of the capacity of the 2608 bottleneck when there are already ongoing flows using the bottleneck 2609 capacity. In a data centre environment DCTCP takes about a factor of 2610 1.5 to 2 longer to converge due to the much higher typical level of 2611 ECN marking that DCTCP background traffic induces, which causes new 2612 flows to exit slow start early [Alizadeh-stability]. In testing for 2613 use over the public Internet the convergence time of DCTCP relative 2614 to a regular loss-based TCP slow start is even less favourable 2615 [Paced-Chirping] due to the shallow ECN marking threshold needed for 2616 L4S. It is exacerbated by the typically greater mismatch between the 2617 link rate of the sending host and typical Internet access 2618 bottlenecks. This problem is detrimental in general, but would 2619 particularly harm the performance of short flows relative to Classic 2620 congestion controls. 2622 Appendix B. Compromises in the Choice of L4S Identifier 2624 This appendix is informative, not normative. As explained in 2625 Section 2, there is insufficient space in the IP header (v4 or v6) to 2626 fully accommodate every requirement. So the choice of L4S identifier 2627 involves tradeoffs. This appendix records the pros and cons of the 2628 choice that was made. 2630 Non-normative recap of the chosen codepoint scheme: 2632 Packets with ECT(1) and conditionally packets with CE signify L4S 2633 semantics as an alternative to the semantics of Classic ECN 2634 [RFC3168], specifically: 2636 - The ECT(1) codepoint signifies that the packet was sent by an 2637 L4S-capable sender. 2639 - Given shortage of codepoints, both L4S and Classic ECN sides of 2640 an AQM have to use the same CE codepoint to indicate that a 2641 packet has experienced congestion. If a packet that had 2642 already been marked CE in an upstream buffer arrived at a 2643 subsequent AQM, this AQM would then have to guess whether to 2644 classify CE packets as L4S or Classic ECN. Choosing the L4S 2645 treatment is a safer choice, because then a few Classic packets 2646 might arrive early, rather than a few L4S packets arriving 2647 late. 2649 - Additional information might be available if the classifier 2650 were transport-aware. Then it could classify a CE packet for 2651 Classic ECN treatment if the most recent ECT packet in the same 2652 flow had been marked ECT(0). However, the L4S service ought 2653 not to need transport-layer awareness. 2655 Cons: 2657 Consumes the last ECN codepoint: The L4S service could potentially 2658 supersede the service provided by Classic ECN, therefore using 2659 ECT(1) to identify L4S packets could ultimately mean that the 2660 ECT(0) codepoint was 'wasted' purely to distinguish one form of 2661 ECN from its successor. 2663 ECN hard in some lower layers: It is not always possible to support 2664 the equivalent of an IP-ECN field in an AQM acting in a buffer 2665 below the IP layer [I-D.ietf-tsvwg-ecn-encap-guidelines]. Then, 2666 depending on the lower layer scheme, the L4S service might have to 2667 drop rather than mark frames even though they might encapsulate an 2668 ECN-capable packet. 2670 Risk of reordering Classic CE packets within a flow: Classifying all 2671 CE packets into the L4S queue risks any CE packets that were 2672 originally ECT(0) being incorrectly classified as L4S. If there 2673 were delay in the Classic queue, these incorrectly classified CE 2674 packets would arrive early, which is a form of reordering. 2675 Reordering within a microflow can cause TCP senders (and senders 2676 of similar transports) to retransmit spuriously. However, the 2677 risk of spurious retransmissions would be extremely low for the 2678 following reasons: 2680 1. It is quite unusual to experience queuing at more than one 2681 bottleneck on the same path (the available capacities have to 2682 be identical). 2684 2. In only a subset of these unusual cases would the first 2685 bottleneck support Classic ECN marking while the second 2686 supported L4S ECN marking, which would be the only scenario 2687 where some ECT(0) packets could be CE marked by an AQM 2688 supporting Classic ECN then the remainder experienced further 2689 delay through the Classic side of a subsequent L4S DualQ AQM. 2691 3. Even then, when a few packets are delivered early, it takes 2692 very unusual conditions to cause a spurious retransmission, in 2693 contrast to when some packets are delivered late. The first 2694 bottleneck has to apply CE-marks to at least N contiguous 2695 packets and the second bottleneck has to inject an 2696 uninterrupted sequence of at least N of these packets between 2697 two packets earlier in the stream (where N is the reordering 2698 window that the transport protocol allows before it considers 2699 a packet is lost). 2701 For example consider N=3, and consider the sequence of 2702 packets 100, 101, 102, 103,... and imagine that packets 2703 150,151,152 from later in the flow are injected as follows: 2704 100, 150, 151, 101, 152, 102, 103... If this were late 2705 reordering, even one packet arriving out of sequence would 2706 trigger a spurious retransmission, but there is no spurious 2707 retransmission here with early reordering, because packet 2708 101 moves the cumulative ACK counter forward before 3 2709 packets have arrived out of order. Later, when packets 2710 148, 149, 153... arrive, even though there is a 3-packet 2711 hole, there will be no problem, because the packets to fill 2712 the hole are already in the receive buffer. 2714 4. Even with the current TCP recommendation of N=3 [RFC5681] 2715 spurious retransmissions will be unlikely for all the above 2716 reasons. As RACK [RFC8985] is becoming widely deployed, it 2717 tends to adapt its reordering window to a larger value of N, 2718 which will make the chance of a contiguous sequence of N early 2719 arrivals vanishingly small. 2721 5. Even a run of 2 CE marks within a Classic ECN flow is 2722 unlikely, given FQ-CoDel is the only known widely deployed AQM 2723 that supports Classic ECN marking and it takes great care to 2724 separate out flows and to space any markings evenly along each 2725 flow. 2727 It is extremely unlikely that the above set of 5 eventualities 2728 that are each unusual in themselves would all happen 2729 simultaneously. But, even if they did, the consequences would 2730 hardly be dire: the odd spurious fast retransmission. Whenever 2731 the traffic source (a Classic congestion control) mistakes the 2732 reordering of a string of CE marks for a loss, one might think 2733 that it will reduce its congestion window as well as emitting a 2734 spurious retransmission. However, it would have already reduced 2735 its congestion window when the CE markings arrived early. If it 2736 is using ABE [RFC8511], it might reduce cwnd a little more for a 2737 loss than for a CE mark. But it will revert that reduction once 2738 it detects that the retransmission was spurious. 2740 In conclusion, the impact of early reordering on spurious 2741 retransmissions due to CE being ambiguous will generally be 2742 vanishingly small. 2744 Insufficient anti-replay window in some pre-existing VPNs: If delay 2745 is reduced for a subset of the flows within a VPN, the anti-replay 2746 feature of some VPNs is known to potentially mistake the 2747 difference in delay for a replay attack. Section 6.2 recommends 2748 that the anti-replay window at the VPN egress is sufficiently 2749 sized, as required by the relevant specifications. However, in 2750 some VPN implementations the maximum anti-replay window is 2751 insufficient to cater for a large delay difference at prevailing 2752 packet rates. Section 6.2 suggests alternative work-rounds for 2753 such cases, but end-users using L4S over a VPN will need to be 2754 able to recognize the symptoms of this problem, in order to seek 2755 out these work-rounds. 2757 Hard to distinguish Classic ECN AQM: With this scheme, when a source 2758 receives ECN feedback, it is not explicitly clear which type of 2759 AQM generated the CE markings. This is not a problem for Classic 2760 ECN sources that send ECT(0) packets, because an L4S AQM will 2761 recognize the ECT(0) packets as Classic and apply the appropriate 2762 Classic ECN marking behaviour. 2764 However, in the absence of explicit disambiguation of the CE 2765 markings, an L4S source needs to use heuristic techniques to work 2766 out which type of congestion response to apply (see 2767 Appendix A.1.5). Otherwise, if long-running Classic flow(s) are 2768 sharing a Classic ECN AQM bottleneck with long-running L4S 2769 flow(s), which then apply an L4S response to Classic CE signals, 2770 the L4S flows would outcompete the Classic flow(s). Experiments 2771 have shown that L4S flows can take about 20 times more capacity 2772 share than equivalent Classic flows. Nonetheless, as link 2773 capacity reduces (e.g. to 4 Mb/s), the inequality reduces. So 2774 Classic flows always make progress and are not starved. 2776 When L4S was first proposed (in 2015, 14 years after [RFC3168] was 2777 published), it was believed that Classic ECN AQMs had failed to be 2778 deployed, because research measurements had found little or no 2779 evidence of CE marking. In subsequent years Classic ECN was 2780 included in per-flow-queuing (FQ) deployments, however an FQ 2781 scheduler stops an L4S flow outcompeting Classic, because it 2782 enforces equality between flow rates. It is not known whether 2783 there have been any non-FQ deployments of Classic ECN AQMs in the 2784 subsequent years, or whether there will be in future. 2786 An algorithm for detecting a Classic ECN AQM as soon as a flow 2787 stabilizes after start-up has been proposed [ecn-fallback] (see 2788 Appendix A.1.5 for a brief summary). Testbed evaluations of v2 of 2789 the algorithm have shown detection is reasonably good for Classic 2790 ECN AQMs, in a wide range of circumstances. However, although it 2791 can correctly detect an L4S ECN AQM in many circumstances, its is 2792 often incorrect at low link capacities and/or high RTTs. Although 2793 this is the safe way round, there is a danger that it will 2794 discourage use of the algorithm. 2796 Non-L4S service for control packets: Solely for the case of TCP, the 2797 Classic ECN RFCs [RFC3168] and [RFC5562] require a sender to clear 2798 the ECN field to Not-ECT on retransmissions and on certain control 2799 packets specifically pure ACKs, window probes and SYNs. When L4S 2800 packets are classified by the ECN field, these TCP control packets 2801 would not be classified into an L4S queue, and could therefore be 2802 delayed relative to the other packets in the flow. This would not 2803 cause reordering (because retransmissions are already out of 2804 order, and these control packets typically carry no data). 2805 However, it would make critical TCP control packets more 2806 vulnerable to loss and delay. To address this problem, 2807 [I-D.ietf-tcpm-generalized-ecn] proposes an experiment in which 2808 all TCP control packets and retransmissions are ECN-capable as 2809 long as appropriate ECN feedback is available in each case. 2811 Pros: 2813 Should work e2e: The ECN field generally propagates end-to-end 2814 across the Internet without being wiped or mangled, at least over 2815 fixed networks. Unlike the DSCP, the setting of the ECN field is 2816 at least meant to be forwarded unchanged by networks that do not 2817 support ECN. 2819 Should work in tunnels: The L4S identifiers work across and within 2820 any tunnel that propagates the ECN field in any of the variant 2821 ways it has been defined since ECN-tunneling was first specified 2822 in the year 2001 [RFC3168]. However, it is likely that some 2823 tunnels still do not implement ECN propagation at all. 2825 Should work for many link technologies: At most, but not all, path 2826 bottlenecks there is IP-awareness, so that L4S AQMs can be located 2827 where the IP-ECN field can be manipulated. Bottlenecks at lower 2828 layer nodes without IP-awareness either have to use drop to signal 2829 congestion or a specific congestion notification facility has to 2830 be defined for that link technology, including propagation to and 2831 from IP-ECN. The programme to define these is progressing and in 2832 each case so far the scheme already defined for ECN inherently 2833 supports L4S as well (see Section 6.1). 2835 Could migrate to one codepoint: If all Classic ECN senders 2836 eventually evolve to use the L4S service, the ECT(0) codepoint 2837 could be reused for some future purpose, but only once use of 2838 ECT(0) packets had reduced to zero, or near-zero, which might 2839 never happen. 2841 L4 not required: Being based on the ECN field, this scheme does not 2842 need the network to access transport layer flow identifiers. 2843 Nonetheless, it does not preclude solutions that do. 2845 Appendix C. Potential Competing Uses for the ECT(1) Codepoint 2847 The ECT(1) codepoint of the ECN field has already been assigned once 2848 for the ECN nonce [RFC3540], which has now been categorized as 2849 historic [RFC8311]. ECN is probably the only remaining field in the 2850 Internet Protocol that is common to IPv4 and IPv6 and still has 2851 potential to work end-to-end, with tunnels and with lower layers. 2852 Therefore, ECT(1) should not be reassigned to a different 2853 experimental use (L4S) without carefully assessing competing 2854 potential uses. These fall into the following categories: 2856 C.1. Integrity of Congestion Feedback 2858 Receiving hosts can fool a sender into downloading faster by 2859 suppressing feedback of ECN marks (or of losses if retransmissions 2860 are not necessary or available otherwise). 2862 The historic ECN nonce protocol [RFC3540] proposed that a TCP sender 2863 could set either of ECT(0) or ECT(1) in each packet of a flow and 2864 remember the sequence it had set. If any packet was lost or 2865 congestion marked, the receiver would miss that bit of the sequence. 2866 An ECN Nonce receiver had to feed back the least significant bit of 2867 the sum, so it could not suppress feedback of a loss or mark without 2868 a 50-50 chance of guessing the sum incorrectly. 2870 It is highly unlikely that ECT(1) will be needed for integrity 2871 protection in future. The ECN Nonce RFC [RFC3540] as been 2872 reclassified as historic, partly because other ways have been 2873 developed to protect feedback integrity of TCP and other transports 2874 [RFC8311] that do not consume a codepoint in the IP header. For 2875 instance: 2877 * the sender can test the integrity of the receiver's feedback by 2878 occasionally setting the IP-ECN field to a value normally only set 2879 by the network. Then it can test whether the receiver's feedback 2880 faithfully reports what it expects (see para 2 of Section 20.2 of 2881 [RFC3168]. This works for loss and it will work for the accurate 2882 ECN feedback [RFC7560] intended for L4S. 2884 * A network can enforce a congestion response to its ECN markings 2885 (or packet losses) by auditing congestion exposure (ConEx) 2886 [RFC7713]. Whether the receiver or a downstream network is 2887 suppressing congestion feedback or the sender is unresponsive to 2888 the feedback, or both, ConEx audit can neutralise any advantage 2889 that any of these three parties would otherwise gain. 2891 * The TCP authentication option (TCP-AO [RFC5925]) can be used to 2892 detect any tampering with TCP congestion feedback (whether 2893 malicious or accidental). TCP's congestion feedback fields are 2894 immutable end-to-end, so they are amenable to TCP-AO protection, 2895 which covers the main TCP header and TCP options by default. 2896 However, TCP-AO is often too brittle to use on many end-to-end 2897 paths, where middleboxes can make verification fail in their 2898 attempts to improve performance or security, e.g. by 2899 resegmentation or shifting the sequence space. 2901 C.2. Notification of Less Severe Congestion than CE 2903 Various researchers have proposed to use ECT(1) as a less severe 2904 congestion notification than CE, particularly to enable flows to fill 2905 available capacity more quickly after an idle period, when another 2906 flow departs or when a flow starts, e.g. VCP [VCP], Queue View (QV) 2907 [QV]. 2909 Before assigning ECT(1) as an identifier for L4S, we must carefully 2910 consider whether it might be better to hold ECT(1) in reserve for 2911 future standardisation of rapid flow acceleration, which is an 2912 important and enduring problem [RFC6077]. 2914 Pre-Congestion Notification (PCN) is another scheme that assigns 2915 alternative semantics to the ECN field. It uses ECT(1) to signify a 2916 less severe level of pre-congestion notification than CE [RFC6660]. 2917 However, the ECN field only takes on the PCN semantics if packets 2918 carry a Diffserv codepoint defined to indicate PCN marking within a 2919 controlled environment. PCN is required to be applied solely to the 2920 outer header of a tunnel across the controlled region in order not to 2921 interfere with any end-to-end use of the ECN field. Therefore a PCN 2922 region on the path would not interfere with the L4S service 2923 identifier defined in Section 3. 2925 Authors' Addresses 2927 Koen De Schepper 2928 Nokia Bell Labs 2929 Antwerp 2930 Belgium 2931 Email: koen.de_schepper@nokia.com 2932 URI: https://www.bell-labs.com/usr/koen.de_schepper 2934 Bob Briscoe (editor) 2935 Independent 2936 United Kingdom 2938 Email: ietf@bobbriscoe.net 2939 URI: http://bobbriscoe.net/