idnits 2.17.1 draft-ietf-tcpm-alternativebackoff-ecn-08.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 date (August 7, 2018) is 2088 days in the past. Is this intentional? Checking references for intended status: Experimental ---------------------------------------------------------------------------- == Unused Reference: 'RFC7713' is defined on line 525, but no explicit reference was found in the text == Outdated reference: A later version (-28) exists of draft-ietf-tcpm-accurate-ecn-06 -- Obsolete informational reference (is this intentional?): RFC 8312 (Obsoleted by RFC 9438) Summary: 0 errors (**), 0 flaws (~~), 3 warnings (==), 2 comments (--). Run idnits with the --verbose option for more detailed information about the items above. -------------------------------------------------------------------------------- 2 Network Working Group N. Khademi 3 Internet-Draft M. Welzl 4 Intended status: Experimental University of Oslo 5 Expires: February 8, 2019 G. Armitage 6 Netflix 7 G. Fairhurst 8 University of Aberdeen 9 August 7, 2018 11 TCP Alternative Backoff with ECN (ABE) 12 draft-ietf-tcpm-alternativebackoff-ecn-08 14 Abstract 16 Active Queue Management (AQM) mechanisms allow for burst tolerance 17 while enforcing short queues to minimise the time that packets spend 18 enqueued at a bottleneck. This can cause noticeable performance 19 degradation for TCP connections traversing such a bottleneck, 20 especially if there are only a few flows or their bandwidth-delay- 21 product is large. The reception of a Congestion Experienced (CE) ECN 22 mark indicates that an AQM mechanism is used at the bottleneck, and 23 therefore the bottleneck network queue is likely to be short. 24 Feedback of this signal allows the TCP sender-side ECN reaction in 25 congestion avoidance to reduce the Congestion Window (cwnd) by a 26 smaller amount than the congestion control algorithm's reaction to 27 inferred packet loss. This specification therefore defines an 28 experimental change to the TCP reaction specified in RFC3168, as 29 permitted by RFC 8311. 31 Status of This Memo 33 This Internet-Draft is submitted in full conformance with the 34 provisions of BCP 78 and BCP 79. 36 Internet-Drafts are working documents of the Internet Engineering 37 Task Force (IETF). Note that other groups may also distribute 38 working documents as Internet-Drafts. The list of current Internet- 39 Drafts is at https://datatracker.ietf.org/drafts/current/. 41 Internet-Drafts are draft documents valid for a maximum of six months 42 and may be updated, replaced, or obsoleted by other documents at any 43 time. It is inappropriate to use Internet-Drafts as reference 44 material or to cite them other than as "work in progress." 46 This Internet-Draft will expire on February 8, 2019. 48 Copyright Notice 50 Copyright (c) 2018 IETF Trust and the persons identified as the 51 document authors. All rights reserved. 53 This document is subject to BCP 78 and the IETF Trust's Legal 54 Provisions Relating to IETF Documents 55 (https://trustee.ietf.org/license-info) in effect on the date of 56 publication of this document. Please review these documents 57 carefully, as they describe your rights and restrictions with respect 58 to this document. Code Components extracted from this document must 59 include Simplified BSD License text as described in Section 4.e of 60 the Trust Legal Provisions and are provided without warranty as 61 described in the Simplified BSD License. 63 Table of Contents 65 1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . 2 66 2. Definitions . . . . . . . . . . . . . . . . . . . . . . . . . 3 67 3. Specification . . . . . . . . . . . . . . . . . . . . . . . . 4 68 3.1. Choice of ABE Multiplier . . . . . . . . . . . . . . . . 4 69 4. Discussion . . . . . . . . . . . . . . . . . . . . . . . . . 6 70 4.1. Why Use ECN to Vary the Degree of Backoff? . . . . . . . 6 71 4.2. An RTT-based response to indicated congestion . . . . . . 7 72 5. ABE Deployment Requirements . . . . . . . . . . . . . . . . . 7 73 6. ABE Experiment Goals . . . . . . . . . . . . . . . . . . . . 8 74 7. Acknowledgements . . . . . . . . . . . . . . . . . . . . . . 8 75 8. IANA Considerations . . . . . . . . . . . . . . . . . . . . . 9 76 9. Implementation Status . . . . . . . . . . . . . . . . . . . . 9 77 10. Security Considerations . . . . . . . . . . . . . . . . . . . 9 78 11. Revision Information . . . . . . . . . . . . . . . . . . . . 9 79 12. References . . . . . . . . . . . . . . . . . . . . . . . . . 11 80 12.1. Normative References . . . . . . . . . . . . . . . . . . 11 81 12.2. Informative References . . . . . . . . . . . . . . . . . 11 82 Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . 12 84 1. Introduction 86 Explicit Congestion Notification (ECN) [RFC3168] makes it possible 87 for an Active Queue Management (AQM) mechanism to signal the presence 88 of incipient congestion without necessarily incurring packet loss. 89 This lets the network deliver some packets to an application that 90 would have been dropped if the application or transport did not 91 support ECN. This packet loss reduction is the most obvious benefit 92 of ECN, but it is often relatively modest. Other benefits of 93 deploying ECN have been documented in RFC8087 [RFC8087]. 95 The rules for ECN were originally written to be very conservative, 96 and required the congestion control algorithms of ECN-Capable 97 transport protocols to treat indications of congestion signalled by 98 ECN exactly the same as they would treat an inferred packet loss 99 [RFC3168]. Research has demonstrated the benefits of reducing 100 network delays that are caused by interaction of loss-based TCP 101 congestion control and excessive buffering [BUFFERBLOAT]. This has 102 led to the creation of AQM mechanisms like PIE [RFC8033] and CoDel 103 [CODEL2012][RFC8289], which prevent bloated queues that are common 104 with unmanaged and excessively large buffers deployed across the 105 Internet [BUFFERBLOAT]. 107 The AQM mechanisms mentioned above aim to keep a sustained queue 108 short while tolerating transient (short-term) packet bursts. 109 However, currently used loss-based congestion control mechanisms 110 cannot always utilise a bottleneck link well where there are short 111 queues. For example, a TCP sender using the Reno congestion control 112 needs to be able to store at least an end-to-end bandwidth-delay 113 product (BDP) worth of data at the bottleneck buffer if it is to 114 maintain full path utilisation in the face of loss-induced reduction 115 of the congestion window (cwnd) [RFC5681], which effectively doubles 116 the amount of data that can be in flight, the maximum round-trip time 117 (RTT) experience, and the path's effective RTT using the network 118 path. 120 Modern AQM mechanisms can use ECN to signal the early signs of 121 impending queue buildup long before a tail-drop queue would be forced 122 to resort to dropping packets. It is therefore appropriate for the 123 transport protocol congestion control algorithm to have a more 124 measured response when it receives an indication with an early- 125 warning of congestion after the remote endpoint receives an ECN CE- 126 marked packet. Recognizing these changes in modern AQM practices, 127 the strict requirement that ECN CE signals be treated identically to 128 inferred packet loss have been relaxed [RFC8311]. This document 129 therefore defines a new sender-side-only congestion control response, 130 called "ABE" (Alternative Backoff with ECN). ABE improves TCP's 131 average throughput when routers use AQM controlled buffers that allow 132 for short queues only. 134 2. Definitions 136 The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", 137 "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this 138 document are to be interpreted as described in RFC 2119 [RFC2119]. 140 3. Specification 142 This specification changes the congestion control algorithm of an 143 ECN-Capable TCP transport protocol by changing the TCP sender 144 response to feedback from the TCP receiver that indicates reception 145 of a CE-marked packet, i.e., receipt of a packet with the ECN-Echo 146 flag (defined in [RFC3168]) set, following the process defined in 147 [RFC8311]. 149 The TCP sender response is currently specified in section 6.1.2 of 150 the ECN specification [RFC3168], updated by [RFC8311]: 152 The indication of congestion should be treated just as a 153 congestion loss in non-ECN-Capable TCP. That is, the TCP source 154 halves the congestion window "cwnd" and reduces the slow start 155 threshold "ssthresh", unless otherwise specified by an 156 Experimental RFC in the IETF document stream. 158 This is replaced with: 160 Receipt of a packet with the ECN-Echo flag SHOULD trigger the TCP 161 source to set the slow start threshold (ssthresh) to 0.8 times the 162 FlightSize, with a lower bound of 2 * SMSS applied to the result. 163 As in [RFC5681], the TCP sender also reduces the cwnd value to no 164 more than the new ssthresh value. RFC 3168 section 6.1.2 provides 165 guidance on setting a cwnd less than 2 * SMSS. 167 3.1. Choice of ABE Multiplier 169 ABE decouples the reaction of a TCP sender to inferred packet loss 170 and indication of ECN-signalled congestion in the congestion 171 avoidance phase. To achieve this, ABE uses a different scaling 172 factor in Equation 4 in Section 3.1 of [RFC5681]. The description 173 respectively uses beta_{loss} and beta_{ecn} to refer to the 174 multiplicative decrease factors applied in response to inferred 175 packet loss, and in response to a receiver indicating ECN-signalled 176 congestion. For non-ECN-enabled TCP connections, only beta_{loss} 177 applies. 179 In other words, in response to inferred packet loss: 181 ssthresh = max (FlightSize * beta_{loss}, 2 * SMSS) 183 and in response to an indication of an ECN-signalled congestion: 185 ssthresh = max (FlightSize * beta_{ecn}, 2 * SMSS) 187 and 188 cwnd = ssthresh 190 (If ssthresh == 2 * SMSS, RFC 3168 section 6.1.2 provides guidance 191 on setting a cwnd lower than 2 * SMSS.) 193 where FlightSize is the amount of outstanding data in the network, 194 upper-bounded by the smaller of the sender's cwnd and the receiver's 195 advertised window (rwnd) [RFC5681]. The higher the values of 196 beta_{loss} and beta_{ecn}, the less aggressive the response of any 197 individual backoff event. 199 The appropriate choice for beta_{loss} and beta_{ecn} values is a 200 balancing act between path utilisation and draining the bottleneck 201 queue. More aggressive backoff (smaller beta_*) risks underutilising 202 the path, while less aggressive backoff (larger beta_*) can result in 203 slower draining of the bottleneck queue. 205 The Internet has already been running with at least two different 206 beta_{loss} values for several years: the standard value is 0.5 207 [RFC5681], and the Linux implementation of CUBIC [RFC8312] has used a 208 multiplier of 0.7 since kernel version 2.6.25 released in 2008. ABE 209 does not change the value of beta_{loss} used by current TCP 210 implementations. 212 The recommendation in this document specifies a value of 213 beta_{ecn}=0.8. This recommended beta_{ecn} value is only applicable 214 for the standard TCP congestion control [RFC5681]. The selection of 215 beta_{ecn} enables tuning the response of a TCP connection to shallow 216 AQM marking thresholds. beta_{loss} characterizes the response of a 217 congestion control algorithm to packet loss, i.e., exhaustion of 218 buffers (of unknown depth). Different values for beta_{loss} have 219 been suggested for TCP congestion control algorithms. Consequently, 220 beta_{ecn} is likely to be an algorithm-specific parameter rather 221 than a constant multiple of the algorithm's existing beta_{loss}. 223 A range of tests (section IV, [ABE2017]) with NewReno and CUBIC over 224 CoDel and PIE in lightly-multiplexed scenarios have explored this 225 choice of parameter. The results of these tests indicate that CUBIC 226 connections benefit from beta_{ecn} of 0.85 (cf. beta_{loss} = 0.7), 227 and NewReno connections see improvements with beta_{ecn} in the range 228 0.7 to 0.85 (cf. beta_{loss} = 0.5). 230 4. Discussion 232 Much of the technical background to ABE can be found in a research 233 paper [ABE2017]. This paper used a mix of experiments, theory and 234 simulations with NewReno [RFC5681] and CUBIC [RFC8312] to evaluate 235 the technique. The technique was shown to present "...significant 236 performance gains in lightly-multiplexed [few concurrent flows] 237 scenarios, without losing the delay-reduction benefits of deploying 238 CoDel or PIE". The performance improvement is achieved when reacting 239 to ECN-Echo in congestion avoidance (when ssthresh > cwnd) by 240 multiplying cwnd and ssthresh with a value in the range [0.7,0.85]. 241 Applying ABE when cwnd <= ssthresh is not currently recommended, but 242 may benefit from additional attention, experimentation and 243 specification. 245 4.1. Why Use ECN to Vary the Degree of Backoff? 247 AQM mechanisms such as CoDel [RFC8289] and PIE [RFC8033] set a delay 248 target in routers and use congestion notifications to constrain the 249 queuing delays experienced by packets, rather than in response to 250 impending or actual bottleneck buffer exhaustion. With current 251 default delay targets, CoDel and PIE both effectively emulate a 252 bottleneck with a short queue (section II, [ABE2017]) while also 253 allowing short traffic bursts into the queue. This provides 254 acceptable performance for TCP connections over a path with a low 255 BDP, or in highly multiplexed scenarios (many concurrent transport 256 flows). However, in a lightly-multiplexed case over a path with a 257 large BDP, conventional TCP backoff leads to gaps in packet 258 transmission and under-utilisation of the path. 260 Instead of discarding packets, an AQM mechanism is allowed to mark 261 ECN-Capable packets with an ECN CE-mark. The reception of a CE-mark 262 feedback not only indicates congestion on the network path, it also 263 indicates that an AQM mechanism exists at the bottleneck along the 264 path, and hence the CE-mark likely came from a bottleneck with a 265 controlled short queue. Reacting differently to an ECN-signalled 266 congestion than to an inferred packet loss can then yield the benefit 267 of a reduced back-off when queues are short. Using ECN can also be 268 advantageous for several other reasons [RFC8087]. 270 The idea of reacting differently to inferred packet loss and 271 detection of an ECN-signalled congestion pre-dates this 272 specification. For example, previous research proposed using ECN CE- 273 marked feedback to modify TCP congestion control behaviour via a 274 larger multiplicative decrease factor in conjunction with a smaller 275 additive increase factor [ICC2002]. The goal of this former work was 276 to operate across AQM bottlenecks using Random Early Detection (RED) 277 that were not necessarily configured to emulate a short queue (The 278 current usage of RED as an Internet AQM method is limited [RFC7567]). 280 4.2. An RTT-based response to indicated congestion 282 This specification applies to the use of ECN feedback as defined in 283 [RFC3168], which specifies a response to indicated congestion that is 284 no more frequent that once per path round trip time. Since ABE 285 responds to indicated congestion once per RTT, it therefore does not 286 respond to any further loss within the same RTT, because an ABE 287 sender has already reduced the congestion window. If congestion 288 persists after such reduction, ABE continues to reduce the congestion 289 window in each consecutive RTT. This consecutive reduction can 290 protect the network against long-standing unfairness in the case of 291 AQM algorithms that do not keep a small average queue length. The 292 mechanism does not rely on Accurate ECN 293 ([I-D.ietf-tcpm-accurate-ecn]). 295 In contrast, transport protocol mechanisms can also be designed to 296 utilise more frequent and detailed ECN feedback (e.g., Accurate ECN 297 [I-D.ietf-tcpm-accurate-ecn]), which then permit a congestion control 298 response that adjusts the sending rate more frequently. Datacenter 299 TCP (DCTCP) [RFC8257] is an example of this approach. 301 5. ABE Deployment Requirements 303 This update is a sender-side only change. Like other changes to 304 congestion control algorithms, it does not require any change to the 305 TCP receiver or to network devices. It does not require any ABE- 306 specific changes in routers or the use of Accurate ECN feedback 307 [I-D.ietf-tcpm-accurate-ecn] by a receiver. 309 If the method is only deployed by some senders, and not by others, 310 the senders that use this method can gain some advantage, possibly at 311 the expense of other flows that do not use this updated method. 312 Because this advantage applies only to ECN-marked packets and not to 313 packet loss indications, an ECN-Capable bottleneck will still fall 314 back to dropping packets if an TCP sender using ABE is too 315 aggressive, and the result is no different than if the TCP sender was 316 using traditional loss-based congestion control. 318 When used with bottlenecks that do not support ECN-marking the 319 specification does not modify the transport protocol. 321 6. ABE Experiment Goals 323 RFC3168 states that the congestion control response following an 324 indication of ECN-signalled congestion is the same as the response to 325 a dropped packet [RFC3168]. [RFC8311] updates this specification to 326 allow systems to provide a different behaviour when they experience 327 ECN-signalled congestion rather than packet loss. The present 328 specification defines such an experiment and has thus been assigned 329 an Experimental status before being proposed as a Standards-Track 330 update. 332 The purpose of the Internet experiment is to collect experience with 333 deployment of ABE, and confirm acceptable safety in deployed networks 334 that use this update to TCP congestion control. To evaluate ABE, 335 this experiment therefore requires support in AQM routers for ECN- 336 marking of packets carrying the ECN-Capable Transport, ECT(0), 337 codepoint [RFC3168]. 339 The result of this Internet experiment ought to include an 340 investigation of the implications of experiencing an ECN-CE mark 341 followed by loss within the same RTT. At the end of the experiment, 342 this will be reported to the TCPM WG or IESG. 344 7. Acknowledgements 346 Authors N. Khademi, M. Welzl and G. Fairhurst were part-funded by 347 the European Community under its Seventh Framework Programme through 348 the Reducing Internet Transport Latency (RITE) project (ICT-317700). 349 The views expressed are solely those of the authors. 351 Author G. Armitage performed most of his work on this document while 352 employed by Swinburne University of Technology, Melbourne, Australia. 354 The authors would like to thank Stuart Cheshire for many suggestions 355 when revising the draft, and the following people for their 356 contributions to [ABE2017]: Chamil Kulatunga, David Ros, Stein 357 Gjessing, Sebastian Zander. Thanks also to (in alphabetical order) 358 Roland Bless, Bob Briscoe, David Black, Markku Kojo, John Leslie, 359 Lawrence Stewart, Dave Taht and the TCPM working group for providing 360 valuable feedback on this document. 362 The authors would finally like to thank everyone who provided 363 feedback on the congestion control behaviour specified in this update 364 received from the IRTF Internet Congestion Control Research Group 365 (ICCRG). 367 8. IANA Considerations 369 XX RFC ED - PLEASE REMOVE THIS SECTION XXX 371 This document includes no request to IANA. 373 9. Implementation Status 375 ABE is implemented as a patch for Linux and FreeBSD. It is meant for 376 research and available for download from 377 http://heim.ifi.uio.no/naeemk/research/ABE/. This code was used to 378 produce the test results that are reported in [ABE2017]. The FreeBSD 379 code has been committed to the mainline kernel on March 19, 2018 380 [ABE-FreeBSD]. 382 10. Security Considerations 384 The described method is a sender-side only transport change, and does 385 not change the protocol messages exchanged. The security 386 considerations for ECN [RFC3168] therefore still apply. 388 This is a change to TCP congestion control with ECN that will 389 typically lead to a change in the capacity achieved when flows share 390 a network bottleneck. This could result in some flows receiving more 391 than their fair share of capacity. Similar unfairness in the way 392 that capacity is shared is also exhibited by other congestion control 393 mechanisms that have been in use in the Internet for many years 394 (e.g., CUBIC [RFC8312]). Unfairness may also be a result of other 395 factors, including the round trip time experienced by a flow. ABE 396 applies only when ECN-marked packets are received, not when packets 397 are lost, hence use of ABE cannot lead to congestion collapse. 399 11. Revision Information 401 XX RFC ED - PLEASE REMOVE THIS SECTION XXX 403 -08. Addressed comments from AD review on the document structure, 404 and relationship to existing RFCs. 406 -07. Addressed comments following WGLC. 408 o Updated Reference citations. 410 o Removed paragraph containing a wrong statement related to timeout 411 in section 4.1. 413 o Discuss what happens when cwnd <= ssthresh. 415 o Added text on Concern about lower bound of 2*SMSS. 417 -06. Addressed Michael Scharf's comments. 419 -05. Refined the description of the experiment based on feedback at 420 IETF-100. Incorporated comments from David Black. 422 -04. Incorporates review comments from Lawrence Stewart and the 423 remaining comments from Roland Bless. References are updated. 425 -03. Several review comments from Roland Bless are addressed. 426 Consistent terminology and equations. Clarification on the scope of 427 recommended beta_{ecn} value. 429 -02. Corrected the equations in Section 3.1. Updated the 430 affiliations. Lower bound for cwnd is defined. A recommendation for 431 window-based transport protocols is changed to cover all transport 432 protocols that implement a congestion control reduction to an ECN 433 congestion signal. Added text about ABE's FreeBSD mainline kernel 434 status including a reference to the FreeBSD code review page. 435 References are updated. 437 -01. Text improved, mainly incorporating comments from Stuart 438 Cheshire. The reference to a technical report has been updated to a 439 published version of the tests [ABE2017]. Used "AQM Mechanism" 440 throughout in place of other alternatives, and more consistent use of 441 technical language and clarification on the intended purpose of the 442 experiments required by EXP status. There was no change to the 443 technical content. 445 -00. draft-ietf-tcpm-alternativebackoff-ecn-00 replaces draft- 446 khademi-tcpm-alternativebackoff-ecn-01. Text describing the nature 447 of the experiment was added. 449 Individual draft -01. This I-D now refers to draft-black-tsvwg-ecn- 450 experimentation-02, which replaces draft-khademi-tsvwg-ecn- 451 response-00 to make a broader update to RFC3168 for the sake of 452 allowing experiments. As a result, some of the motivating and 453 discussing text that was moved from draft-khademi-alternativebackoff- 454 ecn-03 to draft-khademi-tsvwg-ecn-response-00 has now been re- 455 inserted here. 457 Individual draft -00. draft-khademi-tsvwg-ecn-response-00 and draft- 458 khademi-tcpm-alternativebackoff-ecn-00 replace draft-khademi- 459 alternativebackoff-ecn-03, following discussion in the TSVWG and TCPM 460 working groups. 462 12. References 464 12.1. Normative References 466 [RFC2119] Bradner, S., "Key words for use in RFCs to Indicate 467 Requirement Levels", BCP 14, RFC 2119, 468 DOI 10.17487/RFC2119, March 1997, 469 . 471 [RFC3168] Ramakrishnan, K., Floyd, S., and D. Black, "The Addition 472 of Explicit Congestion Notification (ECN) to IP", 473 RFC 3168, DOI 10.17487/RFC3168, September 2001, 474 . 476 [RFC5681] Allman, M., Paxson, V., and E. Blanton, "TCP Congestion 477 Control", RFC 5681, DOI 10.17487/RFC5681, September 2009, 478 . 480 [RFC7567] Baker, F., Ed. and G. Fairhurst, Ed., "IETF 481 Recommendations Regarding Active Queue Management", 482 BCP 197, RFC 7567, DOI 10.17487/RFC7567, July 2015, 483 . 485 [RFC8257] Bensley, S., Thaler, D., Balasubramanian, P., Eggert, L., 486 and G. Judd, "Data Center TCP (DCTCP): TCP Congestion 487 Control for Data Centers", RFC 8257, DOI 10.17487/RFC8257, 488 October 2017, . 490 [RFC8311] Black, D., "Relaxing Restrictions on Explicit Congestion 491 Notification (ECN) Experimentation", RFC 8311, 492 DOI 10.17487/RFC8311, January 2018, 493 . 495 12.2. Informative References 497 [ABE-FreeBSD] 498 "ABE patch review in FreeBSD", 499 . 502 [ABE2017] Khademi, N., Armitage, G., Welzl, M., Fairhurst, G., 503 Zander, S., and D. Ros, "Alternative Backoff: Achieving 504 Low Latency and High Throughput with ECN and AQM", IFIP 505 NETWORKING 2017, Stockholm, Sweden, June 2017. 507 [BUFFERBLOAT] 508 Gettys, J. and K. Nichols, "Bufferbloat: Dark Buffers in 509 the Internet", November 2011. 511 [CODEL2012] 512 Nichols, K. and V. Jacobson, "Controlling Queue Delay", 513 July 2012, . 515 [I-D.ietf-tcpm-accurate-ecn] 516 Briscoe, B., Kuehlewind, M., and R. Scheffenegger, "More 517 Accurate ECN Feedback in TCP", draft-ietf-tcpm-accurate- 518 ecn-06 (work in progress), March 2018. 520 [ICC2002] Kwon, M. and S. Fahmy, "TCP Increase/Decrease Behavior 521 with Explicit Congestion Notification (ECN)", IEEE 522 ICC 2002, New York, New York, USA, May 2002, 523 . 525 [RFC7713] Mathis, M. and B. Briscoe, "Congestion Exposure (ConEx) 526 Concepts, Abstract Mechanism, and Requirements", RFC 7713, 527 DOI 10.17487/RFC7713, December 2015, 528 . 530 [RFC8033] Pan, R., Natarajan, P., Baker, F., and G. White, 531 "Proportional Integral Controller Enhanced (PIE): A 532 Lightweight Control Scheme to Address the Bufferbloat 533 Problem", RFC 8033, DOI 10.17487/RFC8033, February 2017, 534 . 536 [RFC8087] Fairhurst, G. and M. Welzl, "The Benefits of Using 537 Explicit Congestion Notification (ECN)", RFC 8087, 538 DOI 10.17487/RFC8087, March 2017, 539 . 541 [RFC8289] Nichols, K., Jacobson, V., McGregor, A., Ed., and J. 542 Iyengar, Ed., "Controlled Delay Active Queue Management", 543 RFC 8289, DOI 10.17487/RFC8289, January 2018, 544 . 546 [RFC8312] Rhee, I., Xu, L., Ha, S., Zimmermann, A., Eggert, L., and 547 R. Scheffenegger, "CUBIC for Fast Long-Distance Networks", 548 RFC 8312, DOI 10.17487/RFC8312, February 2018, 549 . 551 Authors' Addresses 552 Naeem Khademi 553 University of Oslo 554 PO Box 1080 Blindern 555 Oslo N-0316 556 Norway 558 Email: naeemk@ifi.uio.no 560 Michael Welzl 561 University of Oslo 562 PO Box 1080 Blindern 563 Oslo N-0316 564 Norway 566 Email: michawe@ifi.uio.no 568 Grenville Armitage 569 Netflix Inc. 571 Email: garmitage@netflix.com 573 Godred Fairhurst 574 University of Aberdeen 575 School of Engineering, Fraser Noble Building 576 Aberdeen AB24 3UE 577 UK 579 Email: gorry@erg.abdn.ac.uk