idnits 2.17.1 draft-ietf-tsvwg-l4s-arch-02.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: ---------------------------------------------------------------------------- No issues found here. Checking nits according to https://www.ietf.org/id-info/checklist : ---------------------------------------------------------------------------- No issues found here. Miscellaneous warnings: ---------------------------------------------------------------------------- == The copyright year in the IETF Trust and authors Copyright Line does not match the current year == The document doesn't use any RFC 2119 keywords, yet seems to have RFC 2119 boilerplate text. -- The document date (March 22, 2018) is 2228 days in the past. Is this intentional? Checking references for intended status: Informational ---------------------------------------------------------------------------- == Outdated reference: A later version (-02) exists of draft-briscoe-tsvwg-l4s-diffserv-00 == Outdated reference: A later version (-28) exists of draft-ietf-tcpm-accurate-ecn-06 == Outdated reference: A later version (-12) exists of draft-ietf-tcpm-alternativebackoff-ecn-07 == Outdated reference: A later version (-15) exists of draft-ietf-tcpm-generalized-ecn-02 == Outdated reference: A later version (-25) exists of draft-ietf-tsvwg-aqm-dualq-coupled-04 == Outdated reference: A later version (-22) exists of draft-ietf-tsvwg-ecn-encap-guidelines-10 == Outdated reference: A later version (-29) exists of draft-ietf-tsvwg-ecn-l4s-id-02 == Outdated reference: A later version (-07) exists of draft-stewart-tsvwg-sctpecn-05 -- Obsolete informational reference (is this intentional?): RFC 4960 (Obsoleted by RFC 9260) -- Obsolete informational reference (is this intentional?): RFC 7540 (Obsoleted by RFC 9113) -- Obsolete informational reference (is this intentional?): RFC 8312 (Obsoleted by RFC 9438) Summary: 0 errors (**), 0 flaws (~~), 10 warnings (==), 4 comments (--). Run idnits with the --verbose option for more detailed information about the items above. -------------------------------------------------------------------------------- 2 Transport Area Working Group B. Briscoe, Ed. 3 Internet-Draft CableLabs 4 Intended status: Informational K. De Schepper 5 Expires: September 23, 2018 Nokia Bell Labs 6 M. Bagnulo Braun 7 Universidad Carlos III de Madrid 8 March 22, 2018 10 Low Latency, Low Loss, Scalable Throughput (L4S) Internet Service: 11 Architecture 12 draft-ietf-tsvwg-l4s-arch-02 14 Abstract 16 This document describes the L4S architecture for the provision of a 17 new Internet service that could eventually replace best efforts for 18 all traffic: Low Latency, Low Loss, Scalable throughput (L4S). It is 19 becoming common for _all_ (or most) applications being run by a user 20 at any one time to require low latency. However, the only solution 21 the IETF can offer for ultra-low queuing delay is Diffserv, which 22 only favours a minority of packets at the expense of others. In 23 extensive testing the new L4S service keeps average queuing delay 24 under a millisecond for _all_ applications even under very heavy 25 load, without sacrificing utilization; and it keeps congestion loss 26 to zero. It is becoming widely recognized that adding more access 27 capacity gives diminishing returns, because latency is becoming the 28 critical problem. Even with a high capacity broadband access, the 29 reduced latency of L4S remarkably and consistently improves 30 performance under load for applications such as interactive video, 31 conversational video, voice, Web, gaming, instant messaging, remote 32 desktop and cloud-based apps (even when all being used at once over 33 the same access link). The insight is that the root cause of queuing 34 delay is in TCP, not in the queue. By fixing the sending TCP (and 35 other transports) queuing latency becomes so much better than today 36 that operators will want to deploy the network part of L4S to enable 37 new products and services. Further, the network part is simple to 38 deploy - incrementally with zero-config. Both parts, sender and 39 network, ensure coexistence with other legacy traffic. At the same 40 time L4S solves the long-recognized problem with the future 41 scalability of TCP throughput. 43 This document describes the L4S architecture, briefly describing the 44 different components and how the work together to provide the 45 aforementioned enhanced Internet service. 47 Status of This Memo 49 This Internet-Draft is submitted in full conformance with the 50 provisions of BCP 78 and BCP 79. 52 Internet-Drafts are working documents of the Internet Engineering 53 Task Force (IETF). Note that other groups may also distribute 54 working documents as Internet-Drafts. The list of current Internet- 55 Drafts is at https://datatracker.ietf.org/drafts/current/. 57 Internet-Drafts are draft documents valid for a maximum of six months 58 and may be updated, replaced, or obsoleted by other documents at any 59 time. It is inappropriate to use Internet-Drafts as reference 60 material or to cite them other than as "work in progress." 62 This Internet-Draft will expire on September 23, 2018. 64 Copyright Notice 66 Copyright (c) 2018 IETF Trust and the persons identified as the 67 document authors. All rights reserved. 69 This document is subject to BCP 78 and the IETF Trust's Legal 70 Provisions Relating to IETF Documents 71 (https://trustee.ietf.org/license-info) in effect on the date of 72 publication of this document. Please review these documents 73 carefully, as they describe your rights and restrictions with respect 74 to this document. Code Components extracted from this document must 75 include Simplified BSD License text as described in Section 4.e of 76 the Trust Legal Provisions and are provided without warranty as 77 described in the Simplified BSD License. 79 Table of Contents 81 1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . 3 82 2. L4S Architecture Overview . . . . . . . . . . . . . . . . . . 4 83 3. Terminology . . . . . . . . . . . . . . . . . . . . . . . . . 6 84 4. L4S Architecture Components . . . . . . . . . . . . . . . . . 7 85 5. Rationale . . . . . . . . . . . . . . . . . . . . . . . . . . 9 86 5.1. Why These Primary Components? . . . . . . . . . . . . . . 9 87 5.2. Why Not Alternative Approaches? . . . . . . . . . . . . . 10 88 6. Applicability . . . . . . . . . . . . . . . . . . . . . . . . 13 89 6.1. Applications . . . . . . . . . . . . . . . . . . . . . . 13 90 6.2. Use Cases . . . . . . . . . . . . . . . . . . . . . . . . 14 91 6.3. Deployment Considerations . . . . . . . . . . . . . . . . 15 92 6.3.1. Deployment Topology . . . . . . . . . . . . . . . . . 16 93 6.3.2. Deployment Sequences . . . . . . . . . . . . . . . . 17 94 6.3.3. L4S Flow but Non-L4S Bottleneck . . . . . . . . . . . 19 95 6.3.4. Other Potential Deployment Issues . . . . . . . . . . 20 96 7. IANA Considerations . . . . . . . . . . . . . . . . . . . . . 21 97 8. Security Considerations . . . . . . . . . . . . . . . . . . . 21 98 8.1. Traffic (Non-)Policing . . . . . . . . . . . . . . . . . 21 99 8.2. 'Latency Friendliness' . . . . . . . . . . . . . . . . . 22 100 8.3. Interaction between Rate Policing and L4S . . . . . . . . 22 101 8.4. ECN Integrity . . . . . . . . . . . . . . . . . . . . . . 23 102 9. Acknowledgements . . . . . . . . . . . . . . . . . . . . . . 23 103 10. References . . . . . . . . . . . . . . . . . . . . . . . . . 24 104 10.1. Normative References . . . . . . . . . . . . . . . . . . 24 105 10.2. Informative References . . . . . . . . . . . . . . . . . 24 106 Appendix A. Standardization items . . . . . . . . . . . . . . . 28 107 Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . 31 109 1. Introduction 111 It is increasingly common for _all_ of a user's applications at any 112 one time to require low delay: interactive Web, Web services, voice, 113 conversational video, interactive video, interactive remote presence, 114 instant messaging, online gaming, remote desktop, cloud-based 115 applications and video-assisted remote control of machinery and 116 industrial processes. In the last decade or so, much has been done 117 to reduce propagation delay by placing caches or servers closer to 118 users. However, queuing remains a major, albeit intermittent, 119 component of latency. For instance spikes of hundreds of 120 milliseconds are common. During a long-running flow, even with 121 state-of-the-art active queue management (AQM), the base speed-of- 122 light path delay roughly doubles. Low loss is also important 123 because, for interactive applications, losses translate into even 124 longer retransmission delays. 126 It has been demonstrated that, once access network bit rates reach 127 levels now common in the developed world, increasing capacity offers 128 diminishing returns if latency (delay) is not addressed. 129 Differentiated services (Diffserv) offers Expedited Forwarding 130 [RFC3246] for some packets at the expense of others, but this is not 131 sufficient when all (or most) of a user's applications require low 132 latency. 134 Therefore, the goal is an Internet service with ultra-Low queueing 135 Latency, ultra-Low Loss and Scalable throughput (L4S) - for _all_ 136 traffic. A service for all traffic will need none of the 137 configuration or management baggage (traffic policing, traffic 138 contracts) associated with favouring some packets over others. This 139 document describes the L4S architecture for achieving that goal. 141 It must be said that queuing delay only degrades performance 142 infrequently [Hohlfeld14]. It only occurs when a large enough 143 capacity-seeking (e.g. TCP) flow is running alongside the user's 144 traffic in the bottleneck link, which is typically in the access 145 network. Or when the low latency application is itself a large 146 capacity-seeking flow (e.g. interactive video). At these times, the 147 performance improvement from L4S must be so remarkable that network 148 operators will be motivated to deploy it. 150 Active Queue Management (AQM) is part of the solution to queuing 151 under load. AQM improves performance for all traffic, but there is a 152 limit to how much queuing delay can be reduced by solely changing the 153 network; without addressing the root of the problem. 155 The root of the problem is the presence of standard TCP congestion 156 control (Reno [RFC5681]) or compatible variants (e.g. TCP Cubic 157 [RFC8312]). We shall call this family of congestion controls 158 'Classic' TCP. It has been demonstrated that if the sending host 159 replaces Classic TCP with a 'Scalable' alternative, when a suitable 160 AQM is deployed in the network the performance under load of all the 161 above interactive applications can be stunningly improved. For 162 instance, queuing delay under heavy load with the example DCTCP/DualQ 163 solution cited below is roughly 1 millisecond (1 ms) at the 99th 164 percentile without losing link utilization. This compares with 5 to 165 20 ms on _average_ with a Classic TCP and current state-of-the-art 166 AQMs such as fq_CoDel [RFC8290] or PIE [RFC8033]. Also, with a 167 Classic TCP, 5 ms of queuing is usually only possible by losing some 168 utilization. 170 It has been convincingly demonstrated [DCttH15] that it is possible 171 to deploy such an L4S service alongside the existing best efforts 172 service so that all of a user's applications can shift to it when 173 their stack is updated. Access networks are typically designed with 174 one link as the bottleneck for each site (which might be a home, 175 small enterprise or mobile device), so deployment at a single node 176 should give nearly all the benefit. The L4S approach requires 177 component mechanisms in different parts of an Internet path to 178 fulfill its goal. This document presents the L4S architecture, by 179 describing the different components and how they interact to provide 180 the scalable low-latency, low-loss, Internet service. 182 2. L4S Architecture Overview 184 There are three main components to the L4S architecture (illustrated 185 in Figure 1): 187 1) Network: The L4S service traffic needs to be isolated from the 188 queuing latency of the Classic service traffic. However, the two 189 should be able to freely share a common pool of capacity. This is 190 because there is no way to predict how many flows at any one time 191 might use each service and capacity in access networks is too 192 scarce to partition into two. So a 'semi-permeable' membrane is 193 needed that partitions latency but not bandwidth. The Dual Queue 194 Coupled AQM [I-D.ietf-tsvwg-aqm-dualq-coupled] is an example of 195 such a semi-permeable membrane. 197 Per-flow queuing such as in [RFC8290] could be used, but it 198 partitions both latency and bandwidth between every end-to-end 199 flow. So it is rather overkill, which brings disadvantages (see 200 Section 5.2), not least that thousands of queues are needed when 201 two are sufficient. 203 2) Protocol: A host needs to distinguish L4S and Classic packets 204 with an identifier so that the network can classify them into 205 their separate treatments. [I-D.ietf-tsvwg-ecn-l4s-id] considers 206 various alternative identifiers, and concludes that all 207 alternatives involve compromises, but the ECT(1) codepoint of the 208 ECN field is a workable solution. 210 3) Host: Scalable congestion controls already exist. They solve the 211 scaling problem with TCP first pointed out in [RFC3649]. The one 212 used most widely (in controlled environments) is Data Centre TCP 213 (DCTCP [RFC8257]), which has been implemented and deployed in 214 Windows Server Editions (since 2012), in Linux and in FreeBSD. 215 Although DCTCP as-is 'works' well over the public Internet, most 216 implementations lack certain safety features that will be 217 necessary once it is used outside controlled environments like 218 data centres (see later). A similar scalable congestion control 219 will also need to be transplanted into protocols other than TCP 220 (SCTP, RTP/RTCP, RMCAT, etc.) 221 (2) (1) 222 .-------^------. .--------------^-------------------. 223 ,-(3)-----. ______ 224 ; ________ : L4S --------. | | 225 :|Scalable| : _\ ||___\_| mark | 226 :| sender | : __________ / / || / |______|\ _________ 227 :|________|\; | |/ --------' ^ \1| | 228 `---------'\_| IP-ECN | Coupling : \|priority |_\ 229 ________ / |Classifier| : /|scheduler| / 230 |Classic |/ |__________|\ --------. ___:__ / |_________| 231 | sender | \_\ || | |||___\_| mark/|/ 232 |________| / || | ||| / | drop | 233 Classic --------' |______| 235 Figure 1: Components of an L4S Solution: 1) Isolation in separate 236 network queues; 2) Packet Identification Protocol; and 3) Scalable 237 Sending Host 239 3. Terminology 241 The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", 242 "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this 243 document are to be interpreted as described in [RFC2119]. In this 244 document, these words will appear with that interpretation only when 245 in ALL CAPS. Lower case uses of these words are not to be 246 interpreted as carrying RFC-2119 significance. COMMENT: Since this 247 will be an information document, This should be removed. 249 Classic service: The 'Classic' service is intended for all the 250 congestion control behaviours that currently co-exist with TCP 251 Reno (e.g. TCP Cubic, Compound, SCTP, etc). 253 Low-Latency, Low-Loss and Scalable (L4S) service: The 'L4S' service 254 is intended for traffic from scalable TCP algorithms such as Data 255 Centre TCP. But it is also more general--it will allow a set of 256 congestion controls with similar scaling properties to DCTCP (e.g. 257 Relentless [Mathis09]) to evolve. 259 Both Classic and L4S services can cope with a proportion of 260 unresponsive or less-responsive traffic as well (e.g. DNS, VoIP, 261 etc). 263 Scalable Congestion Control: A congestion control where the packet 264 flow rate per round trip (the window) is inversely proportional to 265 the level (probability) of congestion signals. Then, as flow rate 266 scales, the number of congestion signals per round trip remains 267 invariant, maintaining the same degree of control. For instance, 268 DCTCP averages 2 congestion signals per round-trip whatever the 269 flow rate. 271 Classic Congestion Control: A congestion control with a flow rate 272 compatible with standard TCP Reno [RFC5681]. With Classic 273 congestion controls, as capacity increases enabling higher flow 274 rates, the number of round trips between congestion signals 275 (losses or ECN marks) rises in proportion to the flow rate. So 276 control of queuing and/or utilization becomes very slack. For 277 instance, with 1500 B packets and an RTT of 18 ms, as TCP Reno 278 flow rate increases from 2 to 100 Mb/s the number of round trips 279 between congestion signals rises proportionately, from 2 to 100. 281 The default congestion control in Linux (TCP Cubic) is Reno- 282 compatible for most Internet access scenarios expected for some 283 years. For instance, with a typical domestic round-trip time 284 (RTT) of 18ms, TCP Cubic only switches out of Reno-compatibility 285 mode once the flow rate approaches 1 Gb/s. For a typical data 286 centre RTT of 1 ms, the switch-over point is theoretically 1.3 Tb/ 287 s. However, with a less common transcontinental RTT of 100 ms, it 288 only remains Reno-compatible up to 13 Mb/s. All examples assume 289 1,500 B packets. 291 Classic ECN: The original proposed standard Explicit Congestion 292 Notification (ECN) protocol [RFC3168], which requires ECN signals 293 to be treated the same as drops, both when generated in the 294 network and when responded to by the sender. 296 Site: A home, mobile device, small enterprise or campus, where the 297 network bottleneck is typically the access link to the site. Not 298 all network arrangements fit this model but it is a useful, widely 299 applicable generalisation. 301 4. L4S Architecture Components 303 The L4S architecture is composed of the following elements. 305 Protocols:The L4S architecture encompasses the two protocol changes 306 (an unassignment and an assignment) that we describe next: 308 a. An essential aspect of a scalable congestion control is the use 309 of explicit congestion signals rather than losses, because the 310 signals need to be sent immediately and frequently--too often to 311 use drops. 'Classic' ECN [RFC3168] requires an ECN signal to be 312 treated the same as a drop, both when it is generated in the 313 network and when it is responded to by hosts. L4S needs networks 314 and hosts to support two separate meanings for ECN. So the 315 standards track [RFC3168] needs to be updated to allow L4S 316 packets to depart from the 'same as drop' constraint. 318 [RFC8311] is a standards track update to relax specific 319 requirements in RFC 3168 (and certain other standards track 320 RFCs), which clears the way for the experimental changes proposed 321 for L4S. [RFC8311] also reclassifies the original experimental 322 assignment of the ECT(1) codepoint as an ECN nonce [RFC3540] as 323 historic. 325 b. [I-D.ietf-tsvwg-ecn-l4s-id] recommends ECT(1) is used as the 326 identifier to classify L4S packets into a separate treatment from 327 Classic packets. This satisfies the requirements for identifying 328 an alternative ECN treatment in [RFC4774]. 330 Network components:The Dual Queue Coupled AQM has been specified as 331 generically as possible [I-D.ietf-tsvwg-aqm-dualq-coupled] as a 332 'semi-permeable' membrane without specifying the particular AQMs to 333 use in the two queues. An informational appendix of the draft is 334 provided for pseudocode examples of different possible AQM 335 approaches. Initially a zero-config variant of RED called Curvy RED 336 was implemented, tested and documented. The aim is for designers to 337 be free to implement diverse ideas. So the brief normative body of 338 the draft only specifies the minimum constraints an AQM needs to 339 comply with to ensure that the L4S and Classic services will coexist. 340 For instance, a variant of PIE called Dual PI Squared [PI2] has been 341 implemented and found to perform better than Curvy RED over a wide 342 range of conditions, so it has been documented in a second appendix 343 of [I-D.ietf-tsvwg-aqm-dualq-coupled]. 345 Host mechanisms: The L4S architecture includes a number of mechanisms 346 in the end host that we enumerate next: 348 a. Data Centre TCP is the most widely used example of a scalable 349 congestion control. It has been documented as an informational 350 record of the protocol currently in use [RFC8257]. It will be 351 necessary to define a number of safety features for a variant 352 usable on the public Internet. A draft list of these, known as 353 the TCP Prague requirements, has been drawn up (see Appendix A of 354 [I-D.ietf-tsvwg-ecn-l4s-id]). The list also includes some 355 optional performance improvements. 357 b. Transport protocols other than TCP use various congestion 358 controls designed to be friendly with Classic TCP. Before they 359 can use the L4S service, it will be necessary to implement 360 scalable variants of each of these congestion control behaviours. 361 The following standards track RFCs currently define these 362 protocols: ECN in TCP [RFC3168], in SCTP [RFC4960], in RTP 364 [RFC6679], and in DCCP [RFC4340]. Not all are in widespread use, 365 but those that are will eventually need to be updated to allow a 366 different congestion response, which they will have to indicate 367 by using the ECT(1) codepoint. Scalable variants are under 368 consideration for some new transport protocols that are 369 themselves under development, e.g. QUIC [I-D.johansson-quic-ecn] 370 and certain real-time media congestion avoidance techniques 371 (RMCAT) protocols. 373 c. ECN feedback is sufficient for L4S in some transport protocols 374 (RTCP, DCCP) but not others: 376 * For the case of TCP, the feedback protocol for ECN embeds the 377 assumption from Classic ECN that an ECN mark is the same as a 378 drop, making it unusable for a scalable TCP. Therefore, the 379 implementation of TCP receivers will have to be upgraded 380 [RFC7560]. Work to standardize more accurate ECN feedback for 381 TCP (AccECN [I-D.ietf-tcpm-accurate-ecn]) is in progress. 383 * ECN feedback is only roughly sketched in an appendix of the 384 SCTP specification. A fuller specification has been proposed 385 [I-D.stewart-tsvwg-sctpecn], which would need to be 386 implemented and deployed before SCTCP could support L4S. 388 5. Rationale 390 5.1. Why These Primary Components? 392 Explicit congestion signalling (protocol): Explicit congestion 393 signalling is a key part of the L4S approach. In contrast, use of 394 drop as a congestion signal creates a tension because drop is both 395 a useful signal (more would reduce delay) and an impairment (less 396 would reduce delay). Explicit congestion signals can be used many 397 times per round trip, to keep tight control, without any 398 impairment. Under heavy load, even more explicit signals can be 399 applied so the queue can be kept short whatever the load. Whereas 400 state-of-the-art AQMs have to introduce very high packet drop at 401 high load to keep the queue short. Further, when using ECN TCP's 402 sawtooth reduction can be smaller, and therefore return to the 403 operating point more often, without worrying that this causes more 404 signals (one at the top of each smaller sawtooth). The consequent 405 smaller amplitude sawteeth fit between a very shallow marking 406 threshold and an empty queue, so delay variation can be very low, 407 without risk of under-utilization. 409 All the above makes it clear that explicit congestion signalling 410 is only advantageous for latency if it does not have to be 411 considered 'the same as' drop (as required with Classic ECN 413 [RFC3168]). Therefore, in a DualQ AQM, the L4S queue uses a new 414 L4S variant of ECN that is not equivalent to drop 415 [I-D.ietf-tsvwg-ecn-l4s-id], while the Classic queue uses either 416 classic ECN [RFC3168] or drop, which are equivalent. 418 Before Classic ECN was standardized, there were various proposals 419 to give an ECN mark a different meaning from drop. However, there 420 was no particular reason to agree on any one of the alternative 421 meanings, so 'the same as drop' was the only compromise that could 422 be reached. RFC 3168 contains a statement that: 424 "An environment where all end nodes were ECN-Capable could 425 allow new criteria to be developed for setting the CE 426 codepoint, and new congestion control mechanisms for end-node 427 reaction to CE packets. However, this is a research issue, and 428 as such is not addressed in this document." 430 Latency isolation with coupled congestion notification (network): 431 Using just two queues is not essential to L4S (more would be 432 possible), but it is the simplest way to isolate all the L4S 433 traffic that keeps latency low from all the legacy Classic traffic 434 that does not. 436 Similarly, coupling the congestion notification between the queues 437 is not necessarily essential, but it is a clever and simple way to 438 allow senders to determine their rate, packet-by-packet, rather 439 than be overridden by a network scheduler. Because otherwise a 440 network scheduler would have to inspect at least transport layer 441 headers, and it would have to continually assign a rate to each 442 flow without any easy way to understand application intent. 444 L4S packet identifier (protocol): Once there are at least two 445 separate treatments in the network, hosts need an identifier at 446 the IP layer to distinguish which treatment they intend to use. 448 Scalable congestion notification (host): A scalable congestion 449 control keeps the signalling frequency high so that rate 450 variations can be small when signalling is stable, and rate can 451 track variations in available capacity as rapidly as possible 452 otherwise. 454 5.2. Why Not Alternative Approaches? 456 All the following approaches address some part of the same problem 457 space as L4S. In each case, it is shown that L4S complements them or 458 improves on them, rather than being a mutually exclusive alternative: 460 Diffserv: Diffserv addresses the problem of bandwidth apportionment 461 for important traffic as well as queuing latency for delay- 462 sensitive traffic. L4S solely addresses the problem of queuing 463 latency (as well as loss and throughput scaling). Diffserv will 464 still be necessary where important traffic requires priority (e.g. 465 for commercial reasons, or for protection of critical 466 infrastructure traffic) - see [I-D.briscoe-tsvwg-l4s-diffserv]. 467 Nonetheless, if there are Diffserv classes for important traffic, 468 the L4S approach can provide low latency for _all_ traffic within 469 each Diffserv class (including the case where there is only one 470 Diffserv class). 472 Also, as already explained, Diffserv only works for a small subset 473 of the traffic on a link. It is not applicable when all the 474 applications in use at one time at a single site (home, small 475 business or mobile device) require low latency. Also, because L4S 476 is for all traffic, it needs none of the management baggage 477 (traffic policing, traffic contracts) associated with favouring 478 some packets over others. This baggage has held Diffserv back 479 from widespread end-to-end deployment. 481 State-of-the-art AQMs: AQMs such as PIE and fq_CoDel give a 482 significant reduction in queuing delay relative to no AQM at all. 483 The L4S work is intended to complement these AQMs, and we 484 definitely do not want to distract from the need to deploy them as 485 widely as possible. Nonetheless, without addressing the large 486 saw-toothing rate variations of Classic congestion controls, AQMs 487 alone cannot reduce queuing delay too far without significantly 488 reducing link utilization. The L4S approach resolves this tension 489 by ensuring hosts can minimize the size of their sawteeth without 490 appearing so aggressive to legacy flows that they starve them. 492 Per-flow queuing: Similarly per-flow queuing is not incompatible 493 with the L4S approach. However, one queue for every flow can be 494 thought of as overkill compared to the minimum of two queues for 495 all traffic needed for the L4S approach. The overkill of per-flow 496 queuing has side-effects: 498 A. fq makes high performance networking equipment costly 499 (processing and memory) - in contrast dual queue code can be 500 very simple; 502 B. fq requires packet inspection into the end-to-end transport 503 layer, which doesn't sit well alongside encryption for privacy 504 - in contrast the use of ECN as the classifier for L4S 505 requires no deeper inspection than the IP layer; 507 C. fq isolates the queuing of each flow from the others but not 508 from itself so, unlike L4S, it does not support applications 509 that need both capacity-seeking behaviour and very low 510 latency. 512 It might seem that self-inflicted queuing delay should not 513 count, because if the delay wasn't in the network it would 514 just shift to the sender. However, modern adaptive 515 applications, e.g. HTTP/2 [RFC7540] or the interactive media 516 applications described in Section 6, can keep low latency 517 objects at the front of their local send queue by shuffling 518 priorities of other objects dependent on the progress of other 519 transfers. They cannot shuffle packets once they have 520 released them into the network. 522 D. fq prevents any one flow from consuming more than 1/N of the 523 capacity at any instant, where N is the number of flows. This 524 is fine if all flows are elastic, but it does not sit well 525 with a variable bit rate real-time multimedia flow, which 526 requires wriggle room to sometimes take more and other times 527 less than a 1/N share. 529 It might seem that an fq scheduler offers the benefit that it 530 prevents individual flows from hogging all the bandwidth. 531 However, L4S has been deliberately designed so that policing 532 of individual flows can be added as a policy choice, rather 533 than requiring one specific policy choice as the mechanism 534 itself. A scheduler (like fq) has to decide packet-by-packet 535 which flow to schedule without knowing application intent. 536 Whereas a separate policing function can be configured less 537 strictly, so that senders can still control the instantaneous 538 rate of each flow dependent on the needs of each application 539 (e.g. variable rate video), giving more wriggle-room before a 540 flow is deemed non-compliant. Also policing of queuing and of 541 flow-rates can be applied independently. 543 Alternative Back-off ECN (ABE): Yet again, L4S is not an alternative 544 to ABE but a complement that introduces much lower queuing delay. 545 ABE [I-D.ietf-tcpm-alternativebackoff-ecn] alters the host 546 behaviour in response to ECN marking to utilize a link better and 547 give ECN flows a faster throughput, but it assumes the network 548 still treats ECN and drop the same. Therefore ABE exploits any 549 lower queuing delay that AQMs can provide. But as explained 550 above, AQMs still cannot reduce queuing delay too far without 551 losing link utilization (to allow for other, non-ABE, flows). 553 6. Applicability 555 6.1. Applications 557 A transport layer that solves the current latency issues will provide 558 new service, product and application opportunities. 560 With the L4S approach, the following existing applications will 561 immediately experience significantly better quality of experience 562 under load in the best effort class: 564 o Gaming; 566 o VoIP; 568 o Video conferencing; 570 o Web browsing; 572 o (Adaptive) video streaming; 574 o Instant messaging. 576 The significantly lower queuing latency also enables some interactive 577 application functions to be offloaded to the cloud that would hardly 578 even be usable today: 580 o Cloud based interactive video; 582 o Cloud based virtual and augmented reality. 584 The above two applications have been successfully demonstrated with 585 L4S, both running together over a 40 Mb/s broadband access link 586 loaded up with the numerous other latency sensitive applications in 587 the previous list as well as numerous downloads - all sharing the 588 same bottleneck queue simultaneously [L4Sdemo16]. For the former, a 589 panoramic video of a football stadium could be swiped and pinched so 590 that, on the fly, a proxy in the cloud could generate a sub-window of 591 the match video under the finger-gesture control of each user. For 592 the latter, a virtual reality headset displayed a viewport taken from 593 a 360 degree camera in a racing car. The user's head movements 594 controlled the viewport extracted by a cloud-based proxy. In both 595 cases, with 7 ms end-to-end base delay, the additional queuing delay 596 of roughly 1 ms was so low that it seemed the video was generated 597 locally. 599 Using a swiping finger gesture or head movement to pan a video are 600 extremely latency-demanding actions--far more demanding than VoIP. 602 Because human vision can detect extremely low delays of the order of 603 single milliseconds when delay is translated into a visual lag 604 between a video and a reference point (the finger or the orientation 605 of the head sensed by the balance system in the inner ear (the 606 vestibular system). 608 Without the low queuing delay of L4S, cloud-based applications like 609 these would not be credible without significantly more access 610 bandwidth (to deliver all possible video that might be viewed) and 611 more local processing, which would increase the weight and power 612 consumption of head-mounted displays. When all interactive 613 processing can be done in the cloud, only the data to be rendered for 614 the end user needs to be sent. 616 Other low latency high bandwidth applications such as: 618 o Interactive remote presence; 620 o Video-assisted remote control of machinery or industrial 621 processes. 623 are not credible at all without very low queuing delay. No amount of 624 extra access bandwidth or local processing can make up for lost time. 626 6.2. Use Cases 628 The following use-cases for L4S are being considered by various 629 interested parties: 631 o Where the bottleneck is one of various types of access network: 632 DSL, cable, mobile, satellite 634 * Radio links (cellular, WiFi, satellite) that are distant from 635 the source are particularly challenging. The radio link 636 capacity can vary rapidly by orders of magnitude, so it is 637 often desirable to hold a buffer to utilise sudden increases of 638 capacity; 640 * cellular networks are further complicated by a perceived need 641 to buffer in order to make hand-overs imperceptible; 643 * Satellite networks generally have a very large base RTT, so 644 even with minimal queuing, overall delay can never be extremely 645 low; 647 * Nonetheless, it is certainly desirable not to hold a buffer 648 purely because of the sawteeth of Classic TCP, when it is more 649 than is needed for all the above reasons. 651 o Private networks of heterogeneous data centres, where there is no 652 single administrator that can arrange for all the simultaneous 653 changes to senders, receivers and network needed to deploy DCTCP: 655 * a set of private data centres interconnected over a wide area 656 with separate administrations, but within the same company 658 * a set of data centres operated by separate companies 659 interconnected by a community of interest network (e.g. for the 660 finance sector) 662 * multi-tenant (cloud) data centres where tenants choose their 663 operating system stack (Infrastructure as a Service - IaaS) 665 o Different types of transport (or application) congestion control: 667 * elastic (TCP/SCTP); 669 * real-time (RTP, RMCAT); 671 * query (DNS/LDAP). 673 o Where low delay quality of service is required, but without 674 inspecting or intervening above the IP layer 675 [I-D.smith-encrypted-traffic-management]: 677 * mobile and other networks have tended to inspect higher layers 678 in order to guess application QoS requirements. However, with 679 growing demand for support of privacy and encryption, L4S 680 offers an alternative. There is no need to select which 681 traffic to favour for queuing, when L4S gives favourable 682 queuing to all traffic. 684 o If queuing delay is minimized, applications with a fixed delay 685 budget can communicate over longer distances, or via a longer 686 chain of service functions [RFC7665] or onion routers. 688 6.3. Deployment Considerations 690 The DualQ is, in itself, an incremental deployment framework for L4S 691 AQMs so that L4S traffic can coexist with existing Classic "TCP- 692 friendly" traffic. Section 6.3.1 explains why only deploying a DualQ 693 AQM [I-D.ietf-tsvwg-aqm-dualq-coupled] in one node at each end of the 694 access link will realize nearly all the benefit of L4S. 696 L4S involves both end systems and the network, so Section 6.3.2 697 suggests some typical sequences to deploy each part, and why there 698 will be an immediate and significant benefit after deploying just one 699 part. 701 If an ECN-enabled DualQ AQM has not been deployed at a bottleneck, an 702 L4S flow is required to include a fall-back strategy to Classic 703 behaviour. Section 6.3.3 describes how an L4S flow detects this, and 704 how to minimize the effect of false negative detection. 706 6.3.1. Deployment Topology 708 DualQ AQMs will not have to be deployed throughout the Internet 709 before L4S will work for anyone. Operators of public Internet access 710 networks typically design their networks so that the bottleneck will 711 nearly always occur at one known (logical) link. This confines the 712 cost of queue management technology to one place. 714 The case of mesh networks is different and will be discussed later. 715 But the known bottleneck case is generally true for Internet access 716 to all sorts of different 'sites', where the word 'site' includes 717 home networks, small-to-medium sized campus or enterprise networks 718 and even cellular devices (Figure 2). Also, this known-bottleneck 719 case tends to be true whatever the access link technology; whether 720 xDSL, cable, cellular, line-of-sight wireless or satellite. 722 Therefore, the full benefit of the L4S service should be available in 723 the downstream direction when the DualQ AQM is deployed at the 724 ingress to this bottleneck link (or links for multihomed sites). And 725 similarly, the full upstream service will be available once the DualQ 726 is deployed at the upstream ingress. 728 ______ 729 ( ) 730 __ __ ( ) 731 |DQ\________/DQ|( enterprise ) 732 ___ |__/ \__| ( /campus ) 733 ( ) (______) 734 ( ) ___||_ 735 +----+ ( ) __ __ / \ 736 | DC |-----( Core )|DQ\_______________/DQ|| home | 737 +----+ ( ) |__/ \__||______| 738 (_____) __ 739 |DQ\__/\ __ ,===. 740 |__/ \ ____/DQ||| ||mobile 741 \/ \__|||_||device 742 | o | 743 `---' 745 Figure 2: Likely location of DualQ (DQ) Deployments in common access 746 topologies 748 Deployment in mesh topologies depends on how over-booked the core is. 749 If the core is non-blocking, or at least generously provisioned so 750 that the edges are nearly always the bottlenecks, it would only be 751 necessary to deploy the DualQ AQM at the edge bottlenecks. For 752 example, some datacentre networks are designed with the bottleneck in 753 the hypervisor or host NICs, while others bottleneck at the top-of- 754 rack switch (both the output ports facing hosts and those facing the 755 core). 757 The DualQ would eventually also need to be deployed at any other 758 persistent bottlenecks such as network interconnections, e.g. some 759 public Internet exchange points and the ingress and egress to WAN 760 links interconnecting datacentres. 762 6.3.2. Deployment Sequences 764 For any one L4S flow to work, it requires 3 parts to have been 765 deployed. This was the same deployment problem that ECN faced 766 [RFC8170] so we have learned from this. 768 Firstly, L4S deployment exploits the fact that DCTCP already exists 769 on many Internet hosts (Windows, FreeBSD and Linux); both servers and 770 clients. Therefore, just deploying DualQ AQM at a network bottleneck 771 immediately gives a working deployment of all the L4S parts. DCTCP 772 needs some safety concerns to be fixed for general use over the 773 public Internet (see Section 2.3 of [I-D.ietf-tsvwg-ecn-l4s-id]), but 774 DCTCP is not on by default, so these issues can be managed within 775 controlled deployments or controlled trials. 777 Secondly, the performance improvement with L4S is so significant that 778 it enables new interactive services and products that were not 779 previously possible. It is much easier for companies to initiate new 780 work on deployment if there is budget for a new product trial. If, 781 in contrast, there were only an incremental performance improvement 782 (as with Classic ECN), spending on deployment tends to be much harder 783 to justify. 785 Thirdly, the L4S identifier is defined so that intially network 786 operators can enable L4S exclusively for certain customers or certain 787 applications. But this is carefully defined so that it does not 788 compromise future evolution towards L4S as an Internet-wide service. 789 This is because the L4S identifier is defined not only as the end-to- 790 end ECN field, but it can also optionally be combined with any other 791 packet header or some status of a customer or their access link 792 [I-D.ietf-tsvwg-ecn-l4s-id]. Operators could do this anyway, even if 793 it were not blessed by the IETF. However, it is best for the IETF to 794 specify that they must use their own local identifier in combination 795 with the IETF's identifier. Then, if an operator enables the 796 optional local-use approach, they only have to remove this extra rule 797 to make the service work Internet-wide - it will already traverse 798 middleboxes, peerings, etc. 800 +-+--------------------+----------------------+---------------------+ 801 | | Servers or proxies | Access link | Clients | 802 +-+--------------------+----------------------+---------------------+ 803 |1| DCTCP (existing) | | DCTCP (existing) | 804 | | | DualQ AQM downstream | | 805 | | WORKS DOWNSTREAM FOR CONTROLLED DEPLOYMENTS/TRIALS | 806 +-+--------------------+----------------------+---------------------+ 807 |2| TCP Prague | | AccECN (already in | 808 | | | | progress:DCTCP/BBR) | 809 | | FULLY WORKS DOWNSTREAM | 810 +-+--------------------+----------------------+---------------------+ 811 |3| | DualQ AQM upstream | TCP Prague | 812 | | | | | 813 | | FULLY WORKS UPSTREAM AND DOWNSTREAM | 814 +-+--------------------+----------------------+---------------------+ 816 Figure 3: Example L4S Deployment Sequences 818 Figure 3 illustrates some example sequences in which the parts of L4S 819 might be deployed. It consists of the following stages: 821 1. Here, the immediate benefit of a single AQM deployment can be 822 seen, but limited to a controlled trial or controlled deployment. 823 In this example downstream deployment is first, but in other 824 scenarios the upstream might be deployed first. If no AQM at all 825 was previously deployed for the downstream access, the DualQ AQM 826 greatly improves the Classic service (as well as adding the L4S 827 service). If an AQM was already deployed, the Classic service 828 will be unchanged (and L4S will still be added). 830 2. In this stage, the name 'TCP Prague' is used to represent a 831 variant of DCTCP that is safe to use in a production environment. 832 If the application is primarily unidirectional, 'TCP Prague' at 833 one end will provide all the benefit needed. Accurate ECN 834 feedback (AccECN) [I-D.ietf-tcpm-accurate-ecn] is needed at the 835 other end, but it is a generic ECN feedback facility that is 836 already planned to be deployed for other purposes, e.g. DCTCP, 837 BBR [BBR]. The two ends can be deployed in either order, because 838 TCP Prague only enables itself if it has negotiated the use of 839 AccECN feedback with the other end during the connection 840 handshake. Thus, deployment of TCP Prague on a server enables 841 L4S trials to move to a production service in one direction, 842 wherever AccECN is deployed at the other end. This stage might 843 be further motivated by performance improvements between DCTCP 844 and TCP Prague (see Appendix A.2 of [I-D.ietf-tsvwg-ecn-l4s-id]). 846 3. This is a two-move stage to enable L4S upstream. The DualQ or 847 TCP Prague can be deployed in either order as already explained. 848 To motivate the first of two independent moves, the deferred 849 benefit of enabling new services after the second move has to be 850 worth it to cover the first mover's investment risk. As 851 explained already, the potential for new interactive services 852 provides this motivation. The DualQ AQM also greatly improves 853 the upstream Classic service, assuming no other AQM has already 854 been deployed. 856 Note that other deployment sequences might occur. For instance: the 857 upstream might be deployed first; a non-TCP protocol might be used 858 end-to-end, e.g. QUIC, RMCAT; a body such as the 3GPP might require 859 L4S to be implemented in 5G user equipment, or other random acts of 860 kindness. 862 6.3.3. L4S Flow but Non-L4S Bottleneck 864 If L4S is enabled between two hosts but there is no L4S AQM at the 865 bottleneck, any drop from the bottleneck will trigger the L4S sender 866 to fall back to a classic ('TCP-Friendly') behaviour (see 867 Appendix A.1.3 of [I-D.ietf-tsvwg-ecn-l4s-id]). 869 Unfortunately, as well as protecting legacy traffic, this rule 870 degrades the L4S service whenever there is a loss, even if the loss 871 was not from a non-DualQ bottleneck (false negative). And 872 unfortunately, prevalent drop can be due to other causes, e.g.: 874 o congestion loss at other transient bottlenecks, e.g. due to bursts 875 in shallower queues; 877 o transmission errors, e.g. due to electrical interference; 879 o rate policing. 881 Three complementary approaches are in progress to address this issue, 882 but they are all currently research: 884 o In TCP Prague, ignore certain losses deemed unlikely to be due to 885 congestion (using some ideas from BBR [BBR] but with no need to 886 ignore nearly all losses). This could mask any of the above types 887 of loss (requires consensus on how to safely interoperate with 888 drop-based congestion controls). 890 o A combination of RACK, reconfigured link retransmission and L4S 891 could address transmission errors (no reference yet); 893 o Hybrid ECN/drop policers (see Section 8.3). 895 L4S deployment scenarios that minimize these issues (e.g. over 896 wireline networks) can proceed in parallel to this research, in the 897 expectation that research success will continually widen L4S 898 applicability. 900 Classic ECN support is starting to materialize (in the upstream of 901 some home routers as of early 2017), so an L4S sender will have to 902 fall back to a classic ('TCP-Friendly') behaviour if it detects that 903 ECN marking is accompanied by greater queuing delay or greater delay 904 variation than would be expected with L4S (see Appendix A.1.4 of 905 [I-D.ietf-tsvwg-ecn-l4s-id]). 907 6.3.4. Other Potential Deployment Issues 909 An L4S AQM uses the ECN field to signal congestion. So, in common 910 with Classic ECN, if the AQM is within a tunnel or at a lower layer, 911 correct functioning of ECN signalling requires correct propagation of 912 the ECN field up the layers [I-D.ietf-tsvwg-ecn-encap-guidelines]. 914 7. IANA Considerations 916 This specification contains no IANA considerations. 918 8. Security Considerations 920 8.1. Traffic (Non-)Policing 922 Because the L4S service can serve all traffic that is using the 923 capacity of a link, it should not be necessary to police access to 924 the L4S service. In contrast, Diffserv only works if some packets 925 get less favourable treatment than others. So Diffserv has to use 926 traffic policers to limit how much traffic can be favoured, In turn, 927 traffic policers require traffic contracts between users and networks 928 as well as pairwise between networks. Because L4S will lack all this 929 management complexity, it is more likely to work end-to-end. 931 During early deployment (and perhaps always), some networks will not 932 offer the L4S service. These networks do not need to police or re- 933 mark L4S traffic - they just forward it unchanged as best efforts 934 traffic, as they already forward traffic with ECT(1) today. At a 935 bottleneck, such networks will introduce some queuing and dropping. 936 When a scalable congestion control detects a drop it will have to 937 respond as if it is a Classic congestion control (as required in 938 Section 2.3 of [I-D.ietf-tsvwg-ecn-l4s-id]). This will ensure safe 939 interworking with other traffic at the 'legacy' bottleneck, but it 940 will degrade the L4S service to no better (but never worse) than 941 classic best efforts, whenever a legacy (non-L4S) bottleneck is 942 encountered on a path. 944 Certain network operators might choose to restrict access to the L4S 945 class, perhaps only to selected premium customers as a value-added 946 service. Their packet classifier (item 2 in Figure 1) could identify 947 such customers against some other field (e.g. source address range) 948 as well as ECN. If only the ECN L4S identifier matched, but not the 949 source address (say), the classifier could direct these packets (from 950 non-premium customers) into the Classic queue. Allowing operators to 951 use an additional local classifier is intended to remove any 952 incentive to bleach the L4S identifier. Then at least the L4S ECN 953 identifier will be more likely to survive end-to-end even though the 954 service may not be supported at every hop. Such arrangements would 955 only require simple registered/not-registered packet classification, 956 rather than the managed application-specific traffic policing against 957 customer-specific traffic contracts that Diffserv requires. 959 8.2. 'Latency Friendliness' 961 The L4S service does rely on self-constraint - not in terms of 962 limiting rate, but in terms of limiting latency. It is hoped that 963 standardisation of dynamic behaviour (cf. TCP slow-start) and self- 964 interest will be sufficient to prevent transports from sending 965 excessive bursts of L4S traffic, given the application's own latency 966 will suffer most from such behaviour. 968 Whether burst policing becomes necessary remains to be seen. Without 969 it, there will be potential for attacks on the low latency of the L4S 970 service. However it may only be necessary to apply such policing 971 reactively, e.g. punitively targeted at any deployments of new bursty 972 malware. 974 8.3. Interaction between Rate Policing and L4S 976 As mentioned in Section 5.2, L4S should remove the need for low 977 latency Diffserv classes. However, those Diffserv classes that give 978 certain applications or users priority over capacity, would still be 979 applicable. Then, within such Diffserv classes, L4S would often be 980 applicable to give traffic low latency and low loss as well. Within 981 such a Diffserv class, the bandwidth available to a user or 982 application is often limited by a rate policer. Similarly, in the 983 default Diffserv class, rate policers are used to partition shared 984 capacity. 986 A classic rate policer drops any packets exceeding a set rate, 987 usually also giving a burst allowance (variants exist where the 988 policer re-marks non-compliant traffic to a discard-eligible Diffserv 989 codepoint, so they may be dropped elsewhere during contention). 990 Whenever L4S traffic encounters one of these rate policers, it will 991 experience drops and the source has to fall back to a Classic 992 congestion control, thus losing the benefits of L4S. So, in networks 993 that already use rate policers and plan to deploy L4S, it will be 994 preferable to redesign these rate policers to be more friendly to the 995 L4S service. 997 This is currently a research area. It might be achieved by setting a 998 threshold where ECN marking is introduced, such that it is just under 999 the policed rate or just under the burst allowance where drop is 1000 introduced. This could be applied to various types of policer, e.g. 1001 [RFC2697], [RFC2698] or the 'local' (non-ConEx) variant of the ConEx 1002 congestion policer [I-D.briscoe-conex-policing]. It might also be 1003 possible to design scalable congestion controls to respond less 1004 catastrophically to loss that has not been preceded by a period of 1005 increasing delay. 1007 The design of L4S-friendly rate policers will require a separate 1008 dedicated document. For further discussion of the interaction 1009 between L4S and Diffserv, see [I-D.briscoe-tsvwg-l4s-diffserv]. 1011 8.4. ECN Integrity 1013 Receiving hosts can fool a sender into downloading faster by 1014 suppressing feedback of ECN marks (or of losses if retransmissions 1015 are not necessary or available otherwise). Various ways to protect 1016 TCP feedback integrity have been developed. For instance: 1018 o The sender can test the integrity of the receiver's feedback by 1019 occasionally setting the IP-ECN field to the congestion 1020 experienced (CE) codepoint, which is normally only set by a 1021 congested link. Then the sender can test whether the receiver's 1022 feedback faithfully reports what it expects (see 2nd para of 1023 [RFC3168]). 1025 o A network can enforce a congestion response to its ECN markings 1026 (or packet losses) by auditing congestion exposure (ConEx) 1027 [RFC7713]. 1029 o The TCP authentication option (TCP-AO [RFC5925]) can be used to 1030 detect tampering with TCP congestion feedback. 1032 o The ECN Nonce [RFC3540] was proposed to detect tampering with 1033 congestion feedback, but it has been reclassified as historic 1034 [RFC8311]. 1036 Appendix C.1 of [I-D.ietf-tsvwg-ecn-l4s-id] gives more details of 1037 these techniques including their applicability and pros and cons. 1039 9. Acknowledgements 1041 Thanks to Wes Eddy, Karen Nielsen and David Black for their useful 1042 review comments. 1044 Bob Briscoe and Koen De Schepper were part-funded by the European 1045 Community under its Seventh Framework Programme through the Reducing 1046 Internet Transport Latency (RITE) project (ICT-317700). Bob Briscoe 1047 was also part-funded by the Research Council of Norway through the 1048 TimeIn project. The views expressed here are solely those of the 1049 authors. 1051 10. References 1053 10.1. Normative References 1055 [RFC2119] Bradner, S., "Key words for use in RFCs to Indicate 1056 Requirement Levels", BCP 14, RFC 2119, 1057 DOI 10.17487/RFC2119, March 1997, 1058 . 1060 10.2. Informative References 1062 [BBR] Cardwell, N., Cheng, Y., Gunn, C., Yeganeh, S., and V. 1063 Jacobson, "BBR: Congestion-Based Congestion Control; 1064 Measuring bottleneck bandwidth and round-trip propagation 1065 time", ACM Queue (14)5, December 2016. 1067 [DCttH15] De Schepper, K., Bondarenko, O., Tsang, I., and B. 1068 Briscoe, "'Data Centre to the Home': Ultra-Low Latency for 1069 All", 2015, . 1072 (Under submission) 1074 [Hohlfeld14] 1075 Hohlfeld , O., Pujol, E., Ciucu, F., Feldmann, A., and P. 1076 Barford, "A QoE Perspective on Sizing Network Buffers", 1077 Proc. ACM Internet Measurement Conf (IMC'14) hmm, November 1078 2014. 1080 [I-D.briscoe-conex-policing] 1081 Briscoe, B., "Network Performance Isolation using 1082 Congestion Policing", draft-briscoe-conex-policing-01 1083 (work in progress), February 2014. 1085 [I-D.briscoe-tsvwg-l4s-diffserv] 1086 Briscoe, B., "Interactions between Low Latency, Low Loss, 1087 Scalable Throughput (L4S) and Differentiated Services", 1088 draft-briscoe-tsvwg-l4s-diffserv-00 (work in progress), 1089 March 2018. 1091 [I-D.ietf-tcpm-accurate-ecn] 1092 Briscoe, B., Kuehlewind, M., and R. Scheffenegger, "More 1093 Accurate ECN Feedback in TCP", draft-ietf-tcpm-accurate- 1094 ecn-06 (work in progress), March 2018. 1096 [I-D.ietf-tcpm-alternativebackoff-ecn] 1097 Khademi, N., Welzl, M., Armitage, G., and G. Fairhurst, 1098 "TCP Alternative Backoff with ECN (ABE)", draft-ietf-tcpm- 1099 alternativebackoff-ecn-07 (work in progress), March 2018. 1101 [I-D.ietf-tcpm-generalized-ecn] 1102 Bagnulo, M. and B. Briscoe, "ECN++: Adding Explicit 1103 Congestion Notification (ECN) to TCP Control Packets", 1104 draft-ietf-tcpm-generalized-ecn-02 (work in progress), 1105 October 2017. 1107 [I-D.ietf-tsvwg-aqm-dualq-coupled] 1108 Schepper, K., Briscoe, B., Bondarenko, O., and I. Tsang, 1109 "DualQ Coupled AQMs for Low Latency, Low Loss and Scalable 1110 Throughput (L4S)", draft-ietf-tsvwg-aqm-dualq-coupled-04 1111 (work in progress), March 2018. 1113 [I-D.ietf-tsvwg-ecn-encap-guidelines] 1114 Briscoe, B., Kaippallimalil, J., and P. Thaler, 1115 "Guidelines for Adding Congestion Notification to 1116 Protocols that Encapsulate IP", draft-ietf-tsvwg-ecn- 1117 encap-guidelines-10 (work in progress), March 2018. 1119 [I-D.ietf-tsvwg-ecn-l4s-id] 1120 Schepper, K. and B. Briscoe, "Identifying Modified 1121 Explicit Congestion Notification (ECN) Semantics for 1122 Ultra-Low Queuing Delay (L4S)", draft-ietf-tsvwg-ecn-l4s- 1123 id-02 (work in progress), March 2018. 1125 [I-D.johansson-quic-ecn] 1126 Johansson, I., "ECN support in QUIC", draft-johansson- 1127 quic-ecn-03 (work in progress), May 2017. 1129 [I-D.smith-encrypted-traffic-management] 1130 Smith, K., "Network management of encrypted traffic", 1131 draft-smith-encrypted-traffic-management-05 (work in 1132 progress), May 2016. 1134 [I-D.stewart-tsvwg-sctpecn] 1135 Stewart, R., Tuexen, M., and X. Dong, "ECN for Stream 1136 Control Transmission Protocol (SCTP)", draft-stewart- 1137 tsvwg-sctpecn-05 (work in progress), January 2014. 1139 [L4Sdemo16] 1140 Bondarenko, O., De Schepper, K., Tsang, I., and B. 1141 Briscoe, "Ultra-Low Delay for All: Live Experience, Live 1142 Analysis", Proc. MMSYS'16 pp33:1--33:4, May 2016, 1143 . 1147 [Mathis09] 1148 Mathis, M., "Relentless Congestion Control", PFLDNeT'09 , 1149 May 2009, . 1152 [NewCC_Proc] 1153 Eggert, L., "Experimental Specification of New Congestion 1154 Control Algorithms", IETF Operational Note ion-tsv-alt-cc, 1155 July 2007. 1157 [PI2] De Schepper, K., Bondarenko, O., Tsang, I., and B. 1158 Briscoe, "PI^2 : A Linearized AQM for both Classic and 1159 Scalable TCP", Proc. ACM CoNEXT 2016 pp.105-119, December 1160 2016, 1161 . 1163 [RFC2697] Heinanen, J. and R. Guerin, "A Single Rate Three Color 1164 Marker", RFC 2697, DOI 10.17487/RFC2697, September 1999, 1165 . 1167 [RFC2698] Heinanen, J. and R. Guerin, "A Two Rate Three Color 1168 Marker", RFC 2698, DOI 10.17487/RFC2698, September 1999, 1169 . 1171 [RFC3168] Ramakrishnan, K., Floyd, S., and D. Black, "The Addition 1172 of Explicit Congestion Notification (ECN) to IP", 1173 RFC 3168, DOI 10.17487/RFC3168, September 2001, 1174 . 1176 [RFC3246] Davie, B., Charny, A., Bennet, J., Benson, K., Le Boudec, 1177 J., Courtney, W., Davari, S., Firoiu, V., and D. 1178 Stiliadis, "An Expedited Forwarding PHB (Per-Hop 1179 Behavior)", RFC 3246, DOI 10.17487/RFC3246, March 2002, 1180 . 1182 [RFC3540] Spring, N., Wetherall, D., and D. Ely, "Robust Explicit 1183 Congestion Notification (ECN) Signaling with Nonces", 1184 RFC 3540, DOI 10.17487/RFC3540, June 2003, 1185 . 1187 [RFC3649] Floyd, S., "HighSpeed TCP for Large Congestion Windows", 1188 RFC 3649, DOI 10.17487/RFC3649, December 2003, 1189 . 1191 [RFC4340] Kohler, E., Handley, M., and S. Floyd, "Datagram 1192 Congestion Control Protocol (DCCP)", RFC 4340, 1193 DOI 10.17487/RFC4340, March 2006, 1194 . 1196 [RFC4774] Floyd, S., "Specifying Alternate Semantics for the 1197 Explicit Congestion Notification (ECN) Field", BCP 124, 1198 RFC 4774, DOI 10.17487/RFC4774, November 2006, 1199 . 1201 [RFC4960] Stewart, R., Ed., "Stream Control Transmission Protocol", 1202 RFC 4960, DOI 10.17487/RFC4960, September 2007, 1203 . 1205 [RFC5681] Allman, M., Paxson, V., and E. Blanton, "TCP Congestion 1206 Control", RFC 5681, DOI 10.17487/RFC5681, September 2009, 1207 . 1209 [RFC5925] Touch, J., Mankin, A., and R. Bonica, "The TCP 1210 Authentication Option", RFC 5925, DOI 10.17487/RFC5925, 1211 June 2010, . 1213 [RFC6679] Westerlund, M., Johansson, I., Perkins, C., O'Hanlon, P., 1214 and K. Carlberg, "Explicit Congestion Notification (ECN) 1215 for RTP over UDP", RFC 6679, DOI 10.17487/RFC6679, August 1216 2012, . 1218 [RFC7540] Belshe, M., Peon, R., and M. Thomson, Ed., "Hypertext 1219 Transfer Protocol Version 2 (HTTP/2)", RFC 7540, 1220 DOI 10.17487/RFC7540, May 2015, 1221 . 1223 [RFC7560] Kuehlewind, M., Ed., Scheffenegger, R., and B. Briscoe, 1224 "Problem Statement and Requirements for Increased Accuracy 1225 in Explicit Congestion Notification (ECN) Feedback", 1226 RFC 7560, DOI 10.17487/RFC7560, August 2015, 1227 . 1229 [RFC7665] Halpern, J., Ed. and C. Pignataro, Ed., "Service Function 1230 Chaining (SFC) Architecture", RFC 7665, 1231 DOI 10.17487/RFC7665, October 2015, 1232 . 1234 [RFC7713] Mathis, M. and B. Briscoe, "Congestion Exposure (ConEx) 1235 Concepts, Abstract Mechanism, and Requirements", RFC 7713, 1236 DOI 10.17487/RFC7713, December 2015, 1237 . 1239 [RFC8033] Pan, R., Natarajan, P., Baker, F., and G. White, 1240 "Proportional Integral Controller Enhanced (PIE): A 1241 Lightweight Control Scheme to Address the Bufferbloat 1242 Problem", RFC 8033, DOI 10.17487/RFC8033, February 2017, 1243 . 1245 [RFC8170] Thaler, D., Ed., "Planning for Protocol Adoption and 1246 Subsequent Transitions", RFC 8170, DOI 10.17487/RFC8170, 1247 May 2017, . 1249 [RFC8257] Bensley, S., Thaler, D., Balasubramanian, P., Eggert, L., 1250 and G. Judd, "Data Center TCP (DCTCP): TCP Congestion 1251 Control for Data Centers", RFC 8257, DOI 10.17487/RFC8257, 1252 October 2017, . 1254 [RFC8290] Hoeiland-Joergensen, T., McKenney, P., Taht, D., Gettys, 1255 J., and E. Dumazet, "The Flow Queue CoDel Packet Scheduler 1256 and Active Queue Management Algorithm", RFC 8290, 1257 DOI 10.17487/RFC8290, January 2018, 1258 . 1260 [RFC8311] Black, D., "Relaxing Restrictions on Explicit Congestion 1261 Notification (ECN) Experimentation", RFC 8311, 1262 DOI 10.17487/RFC8311, January 2018, 1263 . 1265 [RFC8312] Rhee, I., Xu, L., Ha, S., Zimmermann, A., Eggert, L., and 1266 R. Scheffenegger, "CUBIC for Fast Long-Distance Networks", 1267 RFC 8312, DOI 10.17487/RFC8312, February 2018, 1268 . 1270 [TCP-sub-mss-w] 1271 Briscoe, B. and K. De Schepper, "Scaling TCP's Congestion 1272 Window for Small Round Trip Times", BT Technical Report 1273 TR-TUB8-2015-002, May 2015, 1274 . 1277 Appendix A. Standardization items 1279 The following table includes all the items that will need to be 1280 standardized to provide a full L4S architecture. 1282 The table is too wide for the ASCII draft format, so it has been 1283 split into two, with a common column of row index numbers on the 1284 left. 1286 The columns in the second part of the table have the following 1287 meanings: 1289 WG: The IETF WG most relevant to this requirement. The "tcpm/iccrg" 1290 combination refers to the procedure typically used for congestion 1291 control changes, where tcpm owns the approval decision, but uses 1292 the iccrg for expert review [NewCC_Proc]; 1294 TCP: Applicable to all forms of TCP congestion control; 1296 DCTCP: Applicable to Data Centre TCP as currently used (in 1297 controlled environments); 1299 DCTCP bis: Applicable to an future Data Centre TCP congestion 1300 control intended for controlled environments; 1302 XXX Prague: Applicable to a Scalable variant of XXX (TCP/SCTP/RMCAT) 1303 congestion control. 1305 +-----+------------------------+------------------------------------+ 1306 | Req | Requirement | Reference | 1307 | # | | | 1308 +-----+------------------------+------------------------------------+ 1309 | 0 | ARCHITECTURE | | 1310 | 1 | L4S IDENTIFIER | [I-D.ietf-tsvwg-ecn-l4s-id] | 1311 | 2 | DUAL QUEUE AQM | [I-D.ietf-tsvwg-aqm-dualq-coupled] | 1312 | 3 | Suitable ECN Feedback | [I-D.ietf-tcpm-accurate-ecn], | 1313 | | | [I-D.stewart-tsvwg-sctpecn]. | 1314 | | | | 1315 | | SCALABLE TRANSPORT - | | 1316 | | SAFETY ADDITIONS | | 1317 | 4-1 | Fall back to | [I-D.ietf-tsvwg-ecn-l4s-id] S.2.3, | 1318 | | Reno/Cubic on loss | [RFC8257] | 1319 | 4-2 | Fall back to | [I-D.ietf-tsvwg-ecn-l4s-id] S.2.3 | 1320 | | Reno/Cubic if classic | | 1321 | | ECN bottleneck | | 1322 | | detected | | 1323 | | | | 1324 | 4-3 | Reduce RTT-dependence | [I-D.ietf-tsvwg-ecn-l4s-id] S.2.3 | 1325 | | | | 1326 | 4-4 | Scaling TCP's | [I-D.ietf-tsvwg-ecn-l4s-id] S.2.3, | 1327 | | Congestion Window for | [TCP-sub-mss-w] | 1328 | | Small Round Trip Times | | 1329 | | SCALABLE TRANSPORT - | | 1330 | | PERFORMANCE | | 1331 | | ENHANCEMENTS | | 1332 | 5-1 | Setting ECT in TCP | [I-D.ietf-tcpm-generalized-ecn] | 1333 | | Control Packets and | | 1334 | | Retransmissions | | 1335 | 5-2 | Faster-than-additive | [I-D.ietf-tsvwg-ecn-l4s-id] (Appx | 1336 | | increase | A.2.2) | 1337 | 5-3 | Faster Convergence at | [I-D.ietf-tsvwg-ecn-l4s-id] (Appx | 1338 | | Flow Start | A.2.2) | 1339 +-----+------------------------+------------------------------------+ 1340 +-----+--------+-----+-------+-----------+--------+--------+--------+ 1341 | # | WG | TCP | DCTCP | DCTCP-bis | TCP | SCTP | RMCAT | 1342 | | | | | | Prague | Prague | Prague | 1343 +-----+--------+-----+-------+-----------+--------+--------+--------+ 1344 | 0 | tsvwg | Y | Y | Y | Y | Y | Y | 1345 | 1 | tsvwg | | | Y | Y | Y | Y | 1346 | 2 | tsvwg | n/a | n/a | n/a | n/a | n/a | n/a | 1347 | | | | | | | | | 1348 | | | | | | | | | 1349 | | | | | | | | | 1350 | 3 | tcpm | Y | Y | Y | Y | n/a | n/a | 1351 | | | | | | | | | 1352 | 4-1 | tcpm | | Y | Y | Y | Y | Y | 1353 | | | | | | | | | 1354 | 4-2 | tcpm/ | | | | Y | Y | ? | 1355 | | iccrg? | | | | | | | 1356 | | | | | | | | | 1357 | | | | | | | | | 1358 | | | | | | | | | 1359 | | | | | | | | | 1360 | 4-3 | tcpm/ | | | Y | Y | Y | ? | 1361 | | iccrg? | | | | | | | 1362 | 4-4 | tcpm | Y | Y | Y | Y | Y | ? | 1363 | | | | | | | | | 1364 | | | | | | | | | 1365 | 5-1 | tcpm | Y | Y | Y | Y | n/a | n/a | 1366 | | | | | | | | | 1367 | 5-2 | tcpm/ | | | Y | Y | Y | ? | 1368 | | iccrg? | | | | | | | 1369 | 5-3 | tcpm/ | | | Y | Y | Y | ? | 1370 | | iccrg? | | | | | | | 1371 +-----+--------+-----+-------+-----------+--------+--------+--------+ 1373 Authors' Addresses 1375 Bob Briscoe (editor) 1376 CableLabs 1377 UK 1379 Email: ietf@bobbriscoe.net 1380 URI: http://bobbriscoe.net/ 1381 Koen De Schepper 1382 Nokia Bell Labs 1383 Antwerp 1384 Belgium 1386 Email: koen.de_schepper@nokia.com 1387 URI: https://www.bell-labs.com/usr/koen.de_schepper 1389 Marcelo Bagnulo 1390 Universidad Carlos III de Madrid 1391 Av. Universidad 30 1392 Leganes, Madrid 28911 1393 Spain 1395 Phone: 34 91 6249500 1396 Email: marcelo@it.uc3m.es 1397 URI: http://www.it.uc3m.es