idnits 2.17.1 draft-fairhurst-quic-ack-scaling-03.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 (September 14, 2020) is 1319 days in the past. Is this intentional? Checking references for intended status: Informational ---------------------------------------------------------------------------- == Outdated reference: A later version (-34) exists of draft-ietf-quic-recovery-30 == Outdated reference: A later version (-34) exists of draft-ietf-quic-transport-30 == Outdated reference: A later version (-02) exists of draft-iyengar-quic-delayed-ack-00 Summary: 0 errors (**), 0 flaws (~~), 4 warnings (==), 1 comment (--). Run idnits with the --verbose option for more detailed information about the items above. -------------------------------------------------------------------------------- 2 Internet Engineering Task Force G. Fairhurst 3 Internet-Draft A. Custura 4 Intended status: Informational T. Jones 5 Expires: March 18, 2021 University of Aberdeen 6 September 14, 2020 8 Changing the Default QUIC ACK Policy 9 draft-fairhurst-quic-ack-scaling-03 11 Abstract 13 Acknowledgement packets (ACKs) are used by transport protocols to 14 confirm the delivery of packets, and their reception is used in a 15 variety of other ways (to measure path round trip time, to gauge path 16 congestion, etc). However, the transmission of ACKs also consumes 17 resources at the receiver, forwarding resource in the network and 18 processing resources at the sender. 20 On network paths with significant path asymmetry, transmission of 21 ACKs can limit the available throughput or can reduce the efficient 22 use of network capacity. This occurs when the return capacity is 23 significantly more constrained than the forward capacity, and/or the 24 cost of transmission per packet is a significant component of the 25 total transmission cost. In these cases, reducing the ratio of ACK 26 packets to data packets can improve link utilisation and reduce link 27 transmission costs. It can also reduce processing overhead at the 28 sender and receiver. 30 This document proposes a change to the default acknowledgement policy 31 of the QUIC transport protocol to improve performance over paths with 32 appreciable asymmetry. 34 Status of This Memo 36 This Internet-Draft is submitted in full conformance with the 37 provisions of BCP 78 and BCP 79. 39 Internet-Drafts are working documents of the Internet Engineering 40 Task Force (IETF). Note that other groups may also distribute 41 working documents as Internet-Drafts. The list of current Internet- 42 Drafts is at https://datatracker.ietf.org/drafts/current/. 44 Internet-Drafts are draft documents valid for a maximum of six months 45 and may be updated, replaced, or obsoleted by other documents at any 46 time. It is inappropriate to use Internet-Drafts as reference 47 material or to cite them other than as "work in progress." 48 This Internet-Draft will expire on March 18, 2021. 50 Copyright Notice 52 Copyright (c) 2020 IETF Trust and the persons identified as the 53 document authors. All rights reserved. 55 This document is subject to BCP 78 and the IETF Trust's Legal 56 Provisions Relating to IETF Documents 57 (https://trustee.ietf.org/license-info) in effect on the date of 58 publication of this document. Please review these documents 59 carefully, as they describe your rights and restrictions with respect 60 to this document. Code Components extracted from this document must 61 include Simplified BSD License text as described in Section 4.e of 62 the Trust Legal Provisions and are provided without warranty as 63 described in the Simplified BSD License. 65 Table of Contents 67 1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . 2 68 2. Motivation . . . . . . . . . . . . . . . . . . . . . . . . . 3 69 2.1. ACK Ratio Impact on Asymmetric Paths . . . . . . . . . . 3 70 2.2. Terminology . . . . . . . . . . . . . . . . . . . . . . . 4 71 2.3. Adapting the ACK Ratio in Current Transport 72 Specifications . . . . . . . . . . . . . . . . . . . . . 4 73 2.4. Adapting the TCP ACK Ratio in the Network . . . . . . . . 4 74 2.5. QUIC ACKs . . . . . . . . . . . . . . . . . . . . . . . . 5 75 3. Updating the Default ACK Ratio for QUIC . . . . . . . . . . . 6 76 3.1. Considerations Updating the Default QUIC ACK Policy . . . 6 77 3.2. Mitigating the Impact of ACKs during Slow Start . . . . . 7 78 3.3. ACKs after Slow Start . . . . . . . . . . . . . . . . . . 8 79 3.4. ACKs after re-ordering is Detected . . . . . . . . . . . 8 80 3.5. Further Tuning of the ACK Policy . . . . . . . . . . . . 8 81 4. IANA Considerations . . . . . . . . . . . . . . . . . . . . . 9 82 5. Security Considerations . . . . . . . . . . . . . . . . . . . 9 83 6. References . . . . . . . . . . . . . . . . . . . . . . . . . 9 84 6.1. Normative References . . . . . . . . . . . . . . . . . . 9 85 6.2. Informative References . . . . . . . . . . . . . . . . . 9 86 Appendix A. Summary of Recommended ACK Policy for TCP . . . . . 11 87 Appendix B. Understanding ACKs in Slow Start . . . . . . . . . . 12 88 Appendix C. Experiments Exploring an ACK Ratio of 1:10 . . . . . 16 89 Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . 17 91 1. Introduction 93 Acknowledgement packets (ACKs) are used by transport protocols to 94 confirm the delivery of packets, and their reception is used in a 95 variety of other ways (to measure path round trip time, to gauge path 96 congestion, etc) [Cust20]. However, the transmission of ACKs also 97 consumes resources at the receiver, forwarding resource in the 98 network and processing resources at the sender. 100 The current design of QUIC [I-D.ietf-quic-transport] currently 101 proposes a default acknowledgement (ACK) ratio of 1:2 (at least one 102 ACK for every 2 ack-eliciting packets) inspired by current 103 recommendations for TCP, see Appendix A. 105 This document motivates an increase in the ratio of ACK packets to 106 data packets from 1:2 to 1:10 for QUIC flows. 108 2. Motivation 110 When the characteristics of the forward and return paths are not 111 symmetric [RFC3449], the transmission of ACKs can adversely impact 112 either transport performance and/or the cost of sending data across a 113 link. 115 2.1. ACK Ratio Impact on Asymmetric Paths 117 TCP Performance Implications of Network Path Asymmetry [RFC3449] 118 describes a series of problems and mitigations when transports use an 119 asymmetric path. Performance problems arise in several access 120 networks, including bandwidth-asymmetric networks (such as broadband 121 satellite access, DOCSIS cable networks, cellular mobile, WiFi, etc) 122 [Cust20]. 124 Where the ACK rate is limited by the capacity of the return path, 125 this constrains the maximum throughput for the forward path. ACK 126 traffic also competes for capacity and/or transmission opportunities 127 with other traffic that shares a constrained return path. This 128 motivates a need to reduce the volume of ACK traffic (increase the 129 number of segments/packets that are acknowledged by each ACK). 131 Capacity is not the only asymmetric path constraint. Sending ACKs 132 can consume significant transmission resources and the cost of 133 transmitting ACKs can become a significant part of the cost of 134 transmission when using a network segment. In many wireless 135 technologies, there is appreciable overhead for the transmission of 136 each packet burst (data and ACK). There can also be associated costs 137 (e.g., in radio resource management and transmission scheduling) that 138 are often different for the forward and return paths because they use 139 different technologies or configurations. 141 These provides an incentive to reduce the rate of ACK traffic 142 generated by a transport protocol. 144 2.2. Terminology 146 The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", 147 "SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and 148 "OPTIONAL" in this document are to be interpreted as described in BCP 149 14 [RFC2119] [RFC8174] when, and only when, they appear in all 150 capitals, as shown here. 152 2.3. Adapting the ACK Ratio in Current Transport Specifications 154 We define the ACK Ratio as the ratio between the number of packets 155 that are received by a transport endpoint (on the forward path) and 156 the number of ACK packets that are returned (on the return path). A 157 simple protocol may send an ACK for each data packet, resulting in an 158 ACK Ratio of 1:1. One that acknowledges alternate data packets has a 159 ratio of 1:2. Most IETF-defined protocols specify a default ACK 160 Ratio. Note on the use of ACKs in TCP and the resulting ACK Ratio 161 are provided in Appendix A. 163 Various methods have been proposed to modify the ACK Ratio used by 164 transport protocols. Two examples follow, based on a receiver 165 understanding whether the return path has become congested: 167 Various proposals and analyses methods to allow TCP to dynamically 168 adjust the AR (e.g., [Lan08]). ACK-CC [RFC5690] proposed a method to 169 control the rate of ACKs to avoid the return path from becoming 170 congested, but this did not achieve wide-scale deployment. 172 The Datagram Congestion Control Protocol (DCCP) [RFC4340] has methods 173 to control the ACK ratio at the receiver. DCCP specifies a TCP- 174 friendly congestion control [RFC4341], which includes the ability to 175 use signalling that allows this sender to adjust the receiver ACK 176 ratio within certain parameters. This also was not widely used. 178 These methods are suggested to help when there is congestion on the 179 return path. However, end-to-end transport methods do not have a way 180 to detect and account for the cost of ACK transmission on a link 181 along the return path, and are difficult to tune to adapt to delays 182 introduced by links (e.g., the variations produced by radio resource 183 management). Transport adaptation therefore can only provide a 184 partial mitigation to the impacts when sending ACKs over an 185 asymmetric paths. 187 2.4. Adapting the TCP ACK Ratio in the Network 189 An alternative approach has been deployed for TCP that uses a 190 middlebox in the network to "thin" the rate of ACKs. This method is 191 used with paths that exhibit significant ACK asymmetry to improve 192 performance of TCP. They can modify the ACK Ratio experienced by a 193 network link from the default TCP ACK Ratio of 1:2 [RFC3449]. 194 Performance Enhancing Proxies (PEPs) implement these functions when 195 they detect/know an upstream link is (or is likely to be) filled with 196 TCP ACKs, or on cross-layer information provided by the physical 197 layer. 199 Removing redundant TCP ACKs (also known as "ACK Thinning") leads to 200 stretch ACKs (where a single ACK acknowledges more than two TCP 201 segments). Stretch ACKs have been observed on Internet paths [All98] 202 and are are now common [Din15] for various reasons. 204 The introduction of TCP Appropriate Byte Counting (ABC) [RFC3465] 205 mostly mitigates the impact of stretch ACKs, and also recommends 206 burst mitigation techniques at a TCP sender. 208 2.5. QUIC ACKs 210 QUIC, like other transports, generates ACK information and sends this 211 in QUIC packets on the return path. 213 ACK Thinning methods can only be used when ACKs can be observed by a 214 network device on the path. In contrast, QUIC uses an encrypted 215 feedback packet to communicate an ACK. This use of encryption 216 intentionally prevents such in-network optimisations. 218 A typical QUIC ACK is around 25% larger than a corresponding TCP ACK, 219 and can still be larger when there is loss/reordering. This means 220 additional processing overhead and link usage for all Internet paths, 221 with a significant impact on asymmetric links, where this can also 222 limit throughput: 224 o The smallest IPv4 TCP ACK is 20 bytes, resulting in a 40 byte IPv4 225 TCP ACK packet, although this could be up to 40 bytes larger when 226 TCP options are present. 228 o The TCP ACK size often increases following a reordering event due 229 to the presence of SACK option blocks. The maximum size for IPv4 230 is max 60 bytes. 232 o The smallest IPv4 QUIC ACK packet is 51 bytes. The ACK frame is 233 only 4 bytes, but is sent using a QUIC packet. For IPv4 this is: 234 20 (IP) + 8 (UDP) + 3 (1+1+1 QUIC header ... Assuming a 1 B CID) + 235 4 (minimum ACK frame size) + 16 (crypto overhead). Other types of 236 frames may also be sent in the same packet, increasing this size. 237 The QUIC ACK packet size also becomes larger for very long 238 connections due to the varint encoding. 240 o The packet size for an IPv4 QUIC ACK increases after a reordering 241 event due to the inclusion of ACK range information. The maximum 242 size of this information is limited to the size of a QUIC packet. 243 Implementation may further limit the size (often observed as 100's 244 of bytes depending on the pattern of loss/reordering). 246 Compared to TCP, the performance of QUIC is therefore disadvantaged 247 when QUIC uses an ACK Ratio of 1:2. 249 3. Updating the Default ACK Ratio for QUIC 251 Any ACK policy that changes the ACK ratio from 1:2 needs to consider 252 three issues: 254 o A reduced frequency of feedback can increase the time to detect 255 loss, impacting the loss recovery algorithm, and potentially 256 leading to cases of spurious retransmission. QUIC mitigates this 257 by using timer-based retransmission (using the PTO). 259 o A reduced frequency of feedback can increase the time to detect 260 and react to congestion, impacting the congestion control 261 algorithm. The speed of loss detection can be impacted by delayed 262 acknowledgements. Timer-based methods (e.g., using the QUIC Probe 263 Timeout (PTO), see section 5.1 of [I-D.ietf-quic-recovery]) can 264 reduce the reliance on ACKs to detect loss. 266 o A reduced ACK rate can lead to bursts of acknowledged packets, and 267 introduces a need for burst mitigation at the sender. 269 3.1. Considerations Updating the Default QUIC ACK Policy 271 A QUIC receiver can generate one ACK frame for every received ack- 272 eliciting packet, but normally aggregates the ACK information into a 273 cumulative ACK. The QUIC transport protocol currently recommends a 274 default ACK Ratio of 1:2 [I-D.ietf-quic-transport]. Additionally, 275 QUIC relies on pacing, rather than ACK-Clocking as a burst 276 mitigation. Hence reception of larger cumulative ACKs does not 277 normally have a significant impact on the sender's traffic. 279 Since the introduction of the specification to allow a larger TCP 280 Initial Window (IW) [RFC6928], there has been deployment experience 281 using TCP with an IW of 10 segments at startup. QUIC continues this 282 practice, which in part motivates the need for a QUIC transport 283 protocol to operate when a burst of acknowledgements for up to ten 284 packets is received. QUIC transport recommends the use of pacing to 285 mitigate packet bursts generated by a sender (see section 6.8 of 286 [I-D.ietf-quic-recovery]) 287 This document proposes changing the QUIC behaviour to send an ACK for 288 at least every 10 received packets. Further background to the 289 proposed method is detailed in the Annexe (Appendix C). 291 The QUIC transport protocol also currently specifies a maximum ACK 292 Delay, which is communicated by the sender at connection setup to 293 indicate the maximum time an endpoint will delay sending 294 acknowledgements. A default of 25 milliseconds is recommended 295 [I-D.ietf-quic-transport]. The ACK Delay timer ensures ACKs are not 296 unduly delayed (e.g., when data packets are spaced in time, or at the 297 end of a packet burst). The effect of a large delay could be 298 significant when a stretch ACK acknowledges more QUIC packets. 300 When the ACK Ratio is increased, an appropriate choice of the maximum 301 ACK Delay becomes more important - because it could otherwise 302 withhold ACK information for a time long enough to impact protocol 303 performance or operation. The proposed therefore method also 304 triggers ACK feedback at least four times per RTT (this additional 305 rule is less important when the RTT is greater than 100ms and the ACK 306 Delay forms a small part of the total RTT). 308 3.2. Mitigating the Impact of ACKs during Slow Start 310 A sender during slow start is often cwnd-limited, so any additional 311 delay in returning an ACK has the effect of increasing the duration 312 of the slow-start phase (see Appendix B). The requested change is: 314 The receiver can provide a higher rate of acknowledge for the 315 first 100 ACK-eliciting packets, where it acknowledges at least 316 every second received ACK-eliciting packet. A suitable method 317 might send an ACK frame for every two received ack-eliciting 318 packets for the first 100 received packets if max_ack_delay time 319 has passed since the oldest unacknowledged data was received. 321 NOTE: The original design of TCP used each ACK as a trigger to 322 increase the congestion window, motivating an initial ACK Ratio of 323 1:1 (as in DAASS Appendix A). 325 NOTE: A receiver typically has no understanding of the sender's 326 congestion control state, so the number 100 reflects a trade-off, 327 corresponding to an appreciable opening of the sender's congestion 328 window. 330 NOTE: A congestion controller typically re-enters slow start after 331 congestion is detected, if the congestion window is collapsed to a 332 small value, this could motivate again using this method. However, 333 the receiver is typically unaware of this event, and we do not 334 propose this is considered. 336 3.3. ACKs after Slow Start 338 We recommend an ACK frame should be generated for at least every 339 tenth ack-eliciting packet. The requested change is: 341 The receiver SHOULD send an ACK frame if a period more than 342 MIN(max_ack_delay,min_rtt/4) has passed since receiving the oldest 343 unacknowledged data or when the received has accumulated at most 344 10 unacknowledged ack-eliciting packets. 346 NOTE: The maximum of receiving not more than 10 ack-eliciting packets 347 is derived from the recommended TCP Initial Window [RFC6928]. 349 NOTE: This utilises the receiver's estimated min_rtt, when this is 350 known. 352 3.4. ACKs after re-ordering is Detected 354 Prompt and reliable communication of ACK ranges after loss is 355 important for efficient loss recovery. This suggests that additional 356 ACKs may be needed after a reordering event to protect the sender 357 from potential ACK loss in the return direction. However, such ACKs 358 also consume return path capacity and the number of additional 359 packets needs to be considered. 361 Following detected loss or congestion, a receiver sends ACKs 362 according to section 13.2.1 of QUIC transport 363 [I-D.ietf-quic-transport]. 365 NOTE: A future recommendation could recommend reducing the ACK 366 frequency after loss/re-ordering. 368 3.5. Further Tuning of the ACK Policy 370 In situations where the default method is not sufficient, the ACK 371 Ratio might be further tuned by server, as described in 372 [I-D.iyengar-quic-delayed-ack]. This could permit the ACK method to 373 be adapted to match the behaviour of a new congestion control 374 algorithm. Reducing the rate of ACKs can also lower the 375 computational effort required to process ACKs at the sender and 376 receiver, important for some high speed connections. For instance, 377 this could reduce the workload for high speed network interfaces by 378 reducing the rate of cache ejection for Generic Receiver Offload 379 (GRO). 381 The change to the default motivated in this document is not related 382 to such further tuning of the ACK policy. 384 4. IANA Considerations 386 This memo includes no request to IANA. 388 5. Security Considerations 390 The security considerations for the QUIC transport protocol are 391 described in [I-D.ietf-quic-transport]. 393 6. References 395 6.1. Normative References 397 [I-D.ietf-quic-recovery] 398 Iyengar, J. and I. Swett, "QUIC Loss Detection and 399 Congestion Control", draft-ietf-quic-recovery-30 (work in 400 progress), September 2020. 402 [I-D.ietf-quic-transport] 403 Iyengar, J. and M. Thomson, "QUIC: A UDP-Based Multiplexed 404 and Secure Transport", draft-ietf-quic-transport-30 (work 405 in progress), September 2020. 407 [RFC2119] Bradner, S., "Key words for use in RFCs to Indicate 408 Requirement Levels", BCP 14, RFC 2119, 409 DOI 10.17487/RFC2119, March 1997, 410 . 412 [RFC8174] Leiba, B., "Ambiguity of Uppercase vs Lowercase in RFC 413 2119 Key Words", BCP 14, RFC 8174, DOI 10.17487/RFC8174, 414 May 2017, . 416 6.2. Informative References 418 [All98] Allman, M., "On the Generation and use of TCP 419 Acknowledgments; ACM SIGCOMM CCR, 28 (5) p4-21.", October 420 1998. 422 [Cust20] Custura, A., Jones, T., and G. Fairhurst, "Rethinking ACKs 423 at the Transport Layer; FIT Workshop, IEEE IFIP 424 Networking", June 2020. 426 [Din15] Ding, H., "TCP Stretch Acknowledgements and Timestamps, 427 Findings and Implications for Passive RTT Measurement; ACM 428 SIGCOMM CCR, 45 (3) p20-27.", July 2015. 430 [I-D.iyengar-quic-delayed-ack] 431 Iyengar, J. and I. Swett, "Sender Control of 432 Acknowledgement Delays in QUIC", draft-iyengar-quic- 433 delayed-ack-00 (work in progress), January 2020. 435 [Lan08] Landstroem, S., "TCP/IP Technology for Modern Network 436 Environments; PhD Thesis", 2008. 438 [RFC1122] Braden, R., Ed., "Requirements for Internet Hosts - 439 Communication Layers", STD 3, RFC 1122, 440 DOI 10.17487/RFC1122, October 1989, 441 . 443 [RFC2525] Paxson, V., Allman, M., Dawson, S., Fenner, W., Griner, 444 J., Heavens, I., Lahey, K., Semke, J., and B. Volz, "Known 445 TCP Implementation Problems", RFC 2525, 446 DOI 10.17487/RFC2525, March 1999, 447 . 449 [RFC3449] Balakrishnan, H., Padmanabhan, V., Fairhurst, G., and M. 450 Sooriyabandara, "TCP Performance Implications of Network 451 Path Asymmetry", BCP 69, RFC 3449, DOI 10.17487/RFC3449, 452 December 2002, . 454 [RFC3465] Allman, M., "TCP Congestion Control with Appropriate Byte 455 Counting (ABC)", RFC 3465, DOI 10.17487/RFC3465, February 456 2003, . 458 [RFC4340] Kohler, E., Handley, M., and S. Floyd, "Datagram 459 Congestion Control Protocol (DCCP)", RFC 4340, 460 DOI 10.17487/RFC4340, March 2006, 461 . 463 [RFC4341] Floyd, S. and E. Kohler, "Profile for Datagram Congestion 464 Control Protocol (DCCP) Congestion Control ID 2: TCP-like 465 Congestion Control", RFC 4341, DOI 10.17487/RFC4341, March 466 2006, . 468 [RFC5681] Allman, M., Paxson, V., and E. Blanton, "TCP Congestion 469 Control", RFC 5681, DOI 10.17487/RFC5681, September 2009, 470 . 472 [RFC5690] Floyd, S., Arcia, A., Ros, D., and J. Iyengar, "Adding 473 Acknowledgement Congestion Control to TCP", RFC 5690, 474 DOI 10.17487/RFC5690, February 2010, 475 . 477 [RFC6928] Chu, J., Dukkipati, N., Cheng, Y., and M. Mathis, 478 "Increasing TCP's Initial Window", RFC 6928, 479 DOI 10.17487/RFC6928, April 2013, 480 . 482 Appendix A. Summary of Recommended ACK Policy for TCP 484 A TCP sender regards reception of a TCP ACK as a positive indication 485 that data has been received across the path. The congestion control 486 algorithm uses this to increase the size of the congestion window, 487 cwnd RFC 5681 [RFC5681]. 489 To reduce the ACK Rate, a receiver can delay sending an ACK for a 490 period of time called the ACK Delay. This can increase network 491 efficiency. 493 RFC 5681 [RFC5681] clarifies the TCP ACK frequency described in 494 RFC 1122 [RFC1122]. This recommends that a receiver generates an ACK 495 corresponding to every second maximum segment size, MSS, of received 496 data, Section 4.2 of RFC 5681 [RFC5681], specifies the ACK policy for 497 a TCP receiver: "an ACK SHOULD be generated for at least every second 498 full-sized segment, and MUST be generated within 500 ms of the 499 arrival of the first unacknowledged packet". However, RFC 3449 500 [RFC3449] also notes the need for, and deployment of, methods to 501 further reduce the number of TCP ACKs in networks with asymmetric 502 paths. 504 When a receiver decides to delay ACKs, this can reduce the rate of 505 growth of the cwnd. 507 Some congestion controllers can benefit from frequent feedback during 508 an initial slow start period, where the sender is probing for 509 available path capacity. When a TCP sender uses each ACK to increase 510 the cwnd, this directly impacts the cwnd growth. TCP implementations 511 often use heuristics such as DAAS (Delayed ACK after Slow Start) to 512 mitigate this. This allows the receiver to estimate when the sender 513 could be in the slow start phase of cwnd growth, and for a period of 514 time sends an ACK for each received segment/packet (i.e., an ACK 515 Ratio of 1:1). In contrast, a TCP congestion controller can increase 516 its congestion window based on the number of bytes acknowledged, not 517 the number of ACK packets. (QUIC chooses this approach). 519 A delayed ACK can also increase the time to complete slow start, by 520 introducing an additional delay when returning ACKs. This effect is 521 different to a direct change to the cwnd, but never-the-less can 522 impact the performance, especially over paths where the ACK Delay 523 period is comparable to the RTT. 525 RFC 3465 [RFC3465] further discusses the issue of bursts that may be 526 caused by the interaction between ACK processing and congestion 527 control. This motivates a need to deal with bursts within TCP. TCP 528 senders can mitigate bursts by using Appropriate Byte Counting (ABC), 529 which increases the congestion window in proportion to the amount of 530 data sent into the network, rather than upon the arrival of each ACK. 531 (This also defends against ACK Splitting, where multiple ACKs are 532 received for parts of the same segment/packet). 534 ACKs can be lost on the return path (either through packet loss, or 535 by intentional thinning of the ACK stream). If a sender does not 536 receive an ACK for every second segment, a stretch ACK has occurred, 537 similar to using a larger ACK Ratio. RFC2525 [RFC2525] describes the 538 significance of stretch ACK violations: 540 "This behavior will cause TCP senders to generate burstier 541 traffic, which can degrade performance in congested environments. 542 In addition, generating fewer ACKs increases the amount of time 543 needed by the slow start algorithm to open the congestion window 544 to an appropriate point, which diminishes performance in 545 environments with large bandwidth-delay products. Finally, 546 generating fewer ACKs may cause needless retransmission timeouts 547 in lossy environments, as it increases the possibility that an 548 entire window of ACKs is lost, forcing a retransmission timeout." 550 There are other scenarios where a change to the TCP ACK policy could 551 have improved performance. However, the design of TCP, and 552 ossification of the protocol has made it hard for new mechanisms to 553 be deployed. QUIC does not suffer from these design constraints. 555 Appendix B. Understanding ACKs in Slow Start 557 This section provides diagrams to help explain the way ACKs are using 558 during slow start. This assumes the sender uses volume-based methods 559 to increase cwnd, as in QUIC. 561 Some congestion controllers can benefit from frequent feedback during 562 an initial slow start period, where the sender is probing for 563 available path capacity (see Appendix A). Since a QUIC CC is 564 expected to work in terms of the volume of data acknowledged, rather 565 than the number of ACKs received, this is not expected to be an 566 issue. 568 Consider a burst of 10 packets sent over a forward path. If the path 569 RTT is shorter than the ACK Delay value, then all packets are 570 cumulatively acknowledged by a single ACK. This is therefore the ACK 571 pattern with an ACK Ratio of 1:10, where 10 packets are sent as a 572 burst: 1 2 ... 10. 574 1 2 ... 10 575 \ \ \ \ / 576 \ \ \ \ / 577 \ \ \ \ / 578 \ \ \ \ / 579 \ \ \ \ / 580 \ \ \ \ / 581 \ \ \ \ / 582 \ \ \ \ / 583 \ \ \ \ / 584 \ \ \ \ / 585 \ \ \ \ / 586 \ \ \ X 587 ACK 589 Figure B1: Burst; ACK Delay 25ms; RTT 10ms; ACK Ratio 1:10. 591 The single ACK both cumulatively acknowledges all the data, and 592 increases the cwnd corresponding to reception of the 10 packets. An 593 ACK policy that generates more frequent ACKs (e.g., a lower ACK Delay 594 or ACK Ratio 1:1 or 1:2, would send multiple ACKs when receiving the 595 10 packets, the duration of the ACK burst is the same as the duration 596 of the transmission burst: 1 2 ... 10. 598 1 2 ... 10 599 \ \ \ \ / ..... / 600 \ \ \ \ / / 601 \ \ \ \ / / 602 \ \ \ \ / / 603 \ \ \ \ / / 604 \ \ \ \ / / 605 \ \ \ \ / / 606 \ \ \ X / 607 \ \ \ / \ / 608 \ \ X \ / 609 \ \ / \ \ / 610 \ X \ X 611 ACK ACK 613 Figure : Burst; ACK Delay 25ms; RTT 10ms; ACK Ratio 1:2. 615 The result of using this ACK policy is that it reduces the time to 616 the first ACK by about the burst size. This is not usually a 617 significant benefit, because often the RTT is greater than the burst 618 size. A similar result occurs when ACKs are generated by either the 619 ACK Ratio or the ACK Delay timer. This also has an advantage that it 620 generates more than one ACK, preventing progress being a hostage to 621 fortune on a single ACK. QUIC recommends pacing. If the sender were 622 to pace the 10 packets over the RTT, and the receiver ACK Ratio was 623 1:1, then this is the ACK pattern for the 10 packets paced over the 624 RTT: 626 1 .............. 9 10 627 \ \ \ \ \ \ \ / \/ \/ \/ / / / / / 628 \ \ \ \ \ \ X /\ /\ /\ / / / / / 629 \ \ \ \ \ \ / \ / \/ \/ X / / / / 630 \ \ \ \ \ X X /\ /\ / \ / / / / 631 \ \ \ \ \/ \ / \ / \/ X X / / / 632 \ \ \ \ /\ X X /\ / \ / \ / / / 633 \ \ \ \/ \/ \ / \ / X X X / / 634 \ \ \ /\ /\ X X / \ / \ / \ / / 635 \ \ \/ \/ \/ \ / \/ X X \/ / 636 \ \ /\ /\ /\ X /\ / \ / \ /\ / 637 \ \/ \/ \/ \/ \/ \/ X \/ X / 638 \ /\ /\ /\ /\ /\ /\ / \ /\ / \/ 639 ACK ACK ACK ..................... ACK 641 Figure B2: Pacing; ACK Delay not relevant; RTT 10ms; ACK Ratio 1:1. 643 The time to receive the first ACK is similar to that without pacing, 644 and all later ACKs arrive paced. For the same path (RTT shorter than 645 the ACK delay), but with the ACK Ratio set to 1:10, then the pattern 646 for 10 packets paced over an RTT is: 648 1 2 3 4 5 6 7 8 9 10 11 12 13 649 \ \ \ \ \ \ \ \ \ \ X \ \ 650 \ \ \ \ \ \ \ \ \ \ / \ \ \ 651 \ \ \ \ \ \ \ \ \ \ / \ \ \ 652 \ \ \ \ \ \ \ \ \ \ / 653 \ \ \ \ \ \ \ \ \ \ / 654 \ \ \ \ \ \ \ \ \ \ / 655 \ \ \ \ \ \ \ \ \ \ / 656 \ \ \ \ \ \ \ \ \ \ / 657 \ \ \ \ \ \ \ \ \ \ / 658 \ \ \ \ \ \ \ \ \ \ / 659 \ \ \ \ \ \ \ \ \ X 660 ACK 662 Figure B3: Pacing; ACK Delay 25ms; RTT 25ms; ACK Ratio 1;10. 664 The effect is similar to a very large ACK Delay. The time to receive 665 the first ACK is increased by an RTT (because that was the pacing 666 target). This causes the sender to wait an additional RTT (during 667 which the sender becomes idle) before it receives an ACK for the 668 series of 10 packets, and resumes pacing new data (at the new higher 669 rate, following the ACK). Because this increases the time for 670 receiver to send the first ACK, it therefore slows the window growth. 671 If the path is much longer than the ACK delay, this pathology does 672 not occur. The patterns of ACKs is primarily the result of the ACK 673 Delay setting. The first ACK is returned after waiting for the ACK 674 Delay, and arrives shortly after one RTT: 676 1 .............. 9 10 677 \ \ \ \ \ \ \ / \/ \/ \/ / / / / / 678 \ \ \ \ \ \ X /\ /\ /\ / / / / / 679 \ \ \ \ \ \ / \ / \/ \/ X / / / / 680 \ \ \ \ \ X X /\ /\ / \ / / / / 681 \ \ \ \ \/ \ / \ / \/ X X / / / 682 \ \ \ \ /\ X X /\ / \ / \ / / / 683 \ \ \ \/ \/ \ / \ / X X X / / 684 \ \ \ /\ /\ X X / \ / \ / \ / / 685 \ \ \/ \/ \/ \ / \/ X X \/ / 686 \ \ /\ /\ /\ X /\ / \ / \ /\ / 687 \ \/ \/ \/ \/ \/ \/ X \/ X / 688 \ /\ /\ /\ /\ /\ /\ / \ /\ / \/ 689 >ACK >ACK >ACK ................. >ACK 691 Figure B4: Pacing; ACK Delay 25ms; RTT 650ms; ACK Ratio - not 692 relevant. 694 The window growth is slowed only by the additional time of the ACK 695 delay period. In this case the role of the ACK Ratio becomes 696 apparent only as the rate increases so that multiple ACKs are 697 received with the ACK Delay interval. 699 Although QUIC uses cumulative ACKs to inflate the cwnd, and does not 700 rely upon ACK-clocking to time the transmission of new packets, it is 701 still influenced by the rate of ACKs in the time it is building the 702 congestion window. This could be mitigated by choosing an 703 appropriate ACK Delay, relative to the RTT, but the RTT is often 704 unknown at the time when ACK Delay is negotiated. Another way to 705 mitigate this is to ensure sufficient ACKs are sent for the first "n" 706 packets. 708 For example, setting the ACK Ratio as 1:2 for the first 100 received 709 packets. A value of 1:2 is expected to be a reasonable compromise 710 between reducing delay and conserving return path capacity, since 711 QUIC uses volume-based accounting to increase cwnd, it does not need 712 to ACK each received packet as in DAAS. 714 Appendix C. Experiments Exploring an ACK Ratio of 1:10 716 We used an experimental approach to examine a change to QUIC's ACK 717 Policy . This 718 section provides a few of the many results. These experiments were 719 performed in January 2020 based on available implementations at that 720 time. 722 Our tests show that the proposed change to the ACK Ratio did not 723 negatively impact the protocol. It reduced the amount of IP, UDP and 724 ACK overhead by a factor of approximately 5. The implemented 725 congestion control, was also not negatively impacted. Unlike TCP, 726 QUIC sends other types of data frames in addition to ACK frames, 727 increasing the total overhead on the return path. On asymmetrical 728 paths an ACK Ratio of 1:10 may still reduce the ACK traffic, helping 729 to avoid return path capacity limits impacting the ability to use the 730 forward path capacity. 732 Figure 1 presents a table with a set of asymmetric scenarios. The 733 columns present the rate of ACK traffic required (in kbps) to fill 734 each of the forward paths. The table shows the results for TCP 735 (without a PEP), both for lossless communication. It considers the 736 period after loss, when ACKs communicate the loss information. It 737 also shows the impact of using an ACK Ratio of 1:10 with QUIC. 739 An ACK Ratio of 1:10 reduces the utilisation of the return path. 740 Scenarios where the ACK traffic exceeds the return link capacity 741 (i.e. where this limits the forward path capacity that can be used) 742 are marked with a star. Note that the QUIC figure does not include 743 the encryption overhead, which would be dependent on the ciphers 744 chosen. This would add several additional bytes for every QUIC 745 packet. 747 +-------------------+-------------+---------------+-----------------+ 748 | | 10/2 Mbps | 50/10 Mbps | 250/3 Mbps | 749 +-------------------+-------------+---------------+-----------------+ 750 | TCP no loss | 133 - 346 | 650 -1,730 | 3250 - 8,650* | 751 | TCP loss | 346 - 560 | 1,730 - 2,800 | 8,650 - 14,000* | 752 | QUIC 1:2 no loss | 144 - 438 | 720 - 2,190 | 3,600 - 10,950* | 753 | QUIC 1:2 loss | 290 - * | 1450 - * | 7,250 - * | 754 | QUIC 1:10 no loss | 28.8 - 87.6 | 144 - 438 | 720 - 2,190 | 755 +-------------------+-------------+---------------+-----------------+ 757 Figure 1: ACK traffic required to fill the forward path in different 758 loss and asymmetry scenarios. The QUIC figures do not include 759 encryption overhead. 761 Figure 2 presents a table with the numbers of packets sent by two 762 QUIC implementations using a 1:2 and a 1:10 ACK Ratio. This shows 763 between a four and five time reduction in the number of packets sent. 765 +-------------------------------+----------------+----------------+ 766 | | Chromium | Quicly | 767 | | Draft 23 | Draft 23 | 768 +-------------------------------+----------------+----------------+ 769 | Packets sent | 77,419 | 83,238 | 770 | Packets on return path (1:2) | 39,089 (50.4%) | 41,108 (49.3%) | 771 | Packets on return path (1:10) | 10,409 (13.4%) | 9650 (11.5%) | 772 +-------------------------------+----------------+----------------+ 774 Figure 2: Number of packets sent and received during a 100MB QUIC 775 transfer using different ACK ratios, for two implementations 777 Figure 3 presents a table with the number of bytes sent by one of the 778 QUIC implementations using a 1:2 and a 1:10 ACK Ratio. This shows a 779 reduction in the number of bytes on the return path from 2.7 to 0.7% 780 of the total bytes sent. In these scenarios, this is sufficient to 781 take full advantage of the forward path capacity. 783 +---------------------------------+----------------+ 784 | | Chromium | 785 | | Draft 23 | 786 +---------------------------------+----------------+ 787 | Bytes sent | 110M | 788 | Bytes on return path (1:2 AR) | 3,056KB (2.7%) | 789 | Bytes on return path (1:10 AR) | 810KB (0.7%) | 790 +---------------------------------+----------------+ 792 Figure 3: Number of bytes sent and received during a 100MB Chromium 793 QUIC transfer using different ACK ratios 795 The number of bytes and packets reduces, as expected, when using an 796 ACK Ratio of 1:10, without any increase in loss, on a high-latency 797 path with an asymmetry of 5:1. This offers a clear benefit for paths 798 that are capacity-constrained, as well as paths which would benefit 799 from a reduction in the ACK Rate. 801 Additional data and analysis is provided in [Cust20]. 803 Authors' Addresses 804 Godred Fairhurst 805 University of Aberdeen 806 School of Engineering 807 Fraser Noble Building 808 Aberdeen AB24 3UE 809 UK 811 Email: gorry@erg.abdn.ac.uk 813 Ana Custura 814 University of Aberdeen 815 School of Engineering 816 Fraser Noble Building 817 Aberdeen AB24 3UE 818 UK 820 Email: ana@erg.abdn.ac.uk 822 Tom Jones 823 University of Aberdeen 824 School of Engineering 825 Fraser Noble Building 826 Aberdeen AB24 3UE 827 UK 829 Email: tom@erg.abdn.ac.uk