idnits 2.17.00 (12 Aug 2021) /tmp/idnits36321/draft-ietf-tcpm-newcwv-13.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 : ---------------------------------------------------------------------------- -- The draft header indicates that this document obsoletes RFC2861, but the abstract doesn't seem to directly say this. It does mention RFC2861 though, so this could be OK. Miscellaneous warnings: ---------------------------------------------------------------------------- == The copyright year in the IETF Trust and authors Copyright Line does not match the current year -- The document date (June 25, 2015) is 2515 days in the past. Is this intentional? Checking references for intended status: Experimental ---------------------------------------------------------------------------- ** Obsolete normative reference: RFC 2861 (Obsoleted by RFC 7661) Summary: 1 error (**), 0 flaws (~~), 1 warning (==), 2 comments (--). Run idnits with the --verbose option for more detailed information about the items above. -------------------------------------------------------------------------------- 2 TCPM Working Group G. Fairhurst 3 Internet-Draft A. Sathiaseelan 4 Obsoletes: 2861 (if approved) R. Secchi 5 Intended status: Experimental University of Aberdeen 6 Expires: December 27, 2015 June 25, 2015 8 Updating TCP to support Rate-Limited Traffic 9 draft-ietf-tcpm-newcwv-13 11 Abstract 13 This document provides a mechanism to address issues that arise when 14 TCP is used for traffic that exhibits periods where the sending rate 15 is limited by the application rather than the congestion window. It 16 provides an experimental update to TCP that allows a TCP sender to 17 restart quickly following a rate-limited interval. This method is 18 expected to benefit applications that send rate-limited traffic using 19 TCP, while also providing an appropriate response if congestion is 20 experienced. 22 It also evaluates the Experimental specification of TCP Congestion 23 Window Validation, CWV, defined in RFC 2861, and concludes that RFC 24 2861 sought to address important issues, but failed to deliver a 25 widely used solution. This document therefore recommends that the 26 status of RFC 2861 is moved from Experimental to Historic, and that 27 it is replaced by the current specification. 29 Status of This Memo 31 This Internet-Draft is submitted in full conformance with the 32 provisions of BCP 78 and BCP 79. 34 Internet-Drafts are working documents of the Internet Engineering 35 Task Force (IETF). Note that other groups may also distribute 36 working documents as Internet-Drafts. The list of current Internet- 37 Drafts is at http://datatracker.ietf.org/drafts/current/. 39 Internet-Drafts are draft documents valid for a maximum of six months 40 and may be updated, replaced, or obsoleted by other documents at any 41 time. It is inappropriate to use Internet-Drafts as reference 42 material or to cite them other than as "work in progress." 44 This Internet-Draft will expire on December 27, 2015. 46 Copyright Notice 48 Copyright (c) 2015 IETF Trust and the persons identified as the 49 document authors. All rights reserved. 51 This document is subject to BCP 78 and the IETF Trust's Legal 52 Provisions Relating to IETF Documents 53 (http://trustee.ietf.org/license-info) in effect on the date of 54 publication of this document. Please review these documents 55 carefully, as they describe your rights and restrictions with respect 56 to this document. Code Components extracted from this document must 57 include Simplified BSD License text as described in Section 4.e of 58 the Trust Legal Provisions and are provided without warranty as 59 described in the Simplified BSD License. 61 Table of Contents 63 1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . 3 64 1.1. Implementation of new CWV . . . . . . . . . . . . . . . . 5 65 1.2. Standards Status of this Document . . . . . . . . . . . . 5 66 2. Reviewing experience with TCP-CWV . . . . . . . . . . . . . . 5 67 3. Terminology . . . . . . . . . . . . . . . . . . . . . . . . . 7 68 4.1. Initialisation . . . . . . . . . . . . . . . . . . . . . 8 69 4.2. Estimating the validated capacity supported by a path . . 8 70 4.3. Preserving cwnd during a rate-limited period. . . . . . . 10 71 4.4. TCP congestion control during the non-validated phase . . 11 72 4.4.1. Response to congestion in the non-validated phase . . 12 73 4.4.2. Sender burst control during the non-validated phase . 13 74 4.4.3. Adjustment at the end of the Non-Validated Period 75 (NVP) . . . . . . . . . . . . . . . . . . . . . . . . 14 76 4.5. Examples of Implementation . . . . . . . . . . . . . . . 15 77 4.5.1. Implementing the pipeACK measurement . . . . . . . . 15 78 4.5.2. Measurement of the NVP and pipeACK samples . . . . . 16 79 4.5.3. Implementing detection of the cwnd-limited condition 16 80 5. Determining a safe period to preserve cwnd . . . . . . . . . 17 81 6. Security Considerations . . . . . . . . . . . . . . . . . . . 18 82 7. IANA Considerations . . . . . . . . . . . . . . . . . . . . . 18 83 8. Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . 18 84 9. Author Notes . . . . . . . . . . . . . . . . . . . . . . . . 18 85 9.1. Other related work . . . . . . . . . . . . . . . . . . . 18 86 10. Revision notes . . . . . . . . . . . . . . . . . . . . . . . 20 87 11. References . . . . . . . . . . . . . . . . . . . . . . . . . 24 88 11.1. Normative References . . . . . . . . . . . . . . . . . . 24 89 11.2. Informative References . . . . . . . . . . . . . . . . . 25 90 Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . 26 92 1. Introduction 94 TCP is used for traffic with a range of application behaviours. The 95 TCP congestion window (cwnd) controls the maximum number of 96 unacknowledged packets/bytes that a TCP flow may have in the network 97 at any time, a value known as the FlightSize [RFC5681]. FlightSize 98 is a measure of the volume of data that is unacknowledged at a 99 specific time. A bulk application will always have data available to 100 transmit. The rate at which it sends is therefore limited by the 101 maximum permitted by the receiver advertised window and the sender 102 congestion window (cwnd). The FlightSize of a bulk flow increases 103 with the cwnd, and tracks the volume of data acknowledged in the last 104 Round Trip Time (RTT). 106 In contrast, a rate-limited application will experience periods when 107 the sender is either idle or is unable to send at the maximum rate 108 permitted by the cwnd. In this case, the volume of data sent 109 (FlightSize) can change significantly from one RTT to another, and 110 can be much less than the cwnd. Hence, it is possible that the 111 FlightSize could significantly exceed the recently used capacity. 112 The update in this document targets the operation of TCP in such 113 rate-limited cases. 115 Standard TCP [RFC5681] states that a TCP sender SHOULD set cwnd to no 116 more than the Restart Window (RW) before beginning transmission, if 117 the TCP sender has not sent data in an interval exceeding the 118 retransmission timeout, i.e., when an application becomes idle. 119 [RFC2861] noted that this TCP behaviour was not always observed in 120 current implementations. Experiments [Bis08] confirm this to still 121 be the case. 123 Congestion Window Validation, CWV, introduced the terminology of 124 "application limited periods". RFC2861 describes any time that an 125 application limits the sending rate, rather than being limited by the 126 transport, as "rate-limited". This update improves support for 127 applications that vary their transmission rate, either with (short) 128 idle periods between transmission or by changing the rate at which 129 the application sends. These applications are characterised by the 130 TCP FlightSize often being less than cwnd. Many Internet 131 applications exhibit this behaviour, including web browsing, http- 132 based adaptive streaming, applications that support query/response 133 type protocols, network file sharing, and live video transmission. 134 Many such applications currently avoid using long-lived (persistent) 135 TCP connections (e.g., [RFC7230] servers typically support persistent 136 HTTP connections, but do not enable this by default). Such 137 applications often instead either use a succession of short TCP 138 transfers or use UDP. 140 Standard TCP does not impose additional restrictions on the growth of 141 the congestion window when a TCP sender is unable to send at the 142 maximum rate allowed by the cwnd. In this case, the rate-limited 143 sender may grow a cwnd far beyond that corresponding to the current 144 transmit rate, resulting in a value that does not reflect current 145 information about the state of the network path the flow is using. 146 Use of such an invalid cwnd may result in reduced application 147 performance and/or could significantly contribute to network 148 congestion. 150 [RFC2861] proposed a solution to these issues in an experimental 151 method known as CWV. CWV was intended to help reduce cases where TCP 152 accumulated an invalid (inappropriately large) cwnd. The use and 153 drawbacks of using the CWV algorithm in RFC 2861 with an application 154 are discussed in Section 2. 156 Section 3 defines relevant terminology. 158 Section 4 specifies an alternative to CWV that seeks to address the 159 same issues, but does so in a way that is expected to mitigate the 160 impact on an application that varies its sending rate. The updated 161 method applies to the rate-limited conditions (including both 162 application-limited and idle senders). 164 The goals of this update are: 166 o To not change the behaviour of a TCP sender that performs bulk 167 transfers that fully use the cwnd. 169 o To provide a method that co-exists with Standard TCP and other 170 flows that use this updated method. 172 o To reduce transfer latency for applications that change their rate 173 over short intervals of time. 175 o To avoid a TCP sender growing a large "non-validated" cwnd, when 176 it has not recently sent using this cwnd. 178 o To remove the incentive for ad-hoc application or network stack 179 methods (such as "padding") solely to maintain a large cwnd for 180 future transmission. 182 o To provide an incentive for the use of long-lived connections, 183 rather than a succession of short-lived flows, benefiting both the 184 flows and other flows sharing the network path when actual 185 congestion is encountered. 187 Section 5 describes the rationale for selecting the safe period to 188 preserve the cwnd. 190 1.1. Implementation of new CWV 192 The method specified in Section 4 of this document is a sender-side 193 only change to the the TCP congestion control behaviour of TCP. 195 The method creates a new protocol state, and requires a sender to 196 determine when the cwnd is validated or non-validated to control the 197 entry and exit from this state Section 4.3. It defines how a TCP 198 sender manages the growth of the cwnd using the set of rules defined 199 in Section 4. 201 Implementation of this specification requires an implementor to 202 define a method to measure the available capacity using the pipeACK 203 samples. The details of this measurement are implementation- 204 specific. An example is provided in Section 4.5.1, but other methods 205 are permitted. A sender also needs to provide a method to determine 206 when it becomes cwnd-limited. Implementation of this may require 207 consideration of other TCP methods (see Section 4.5.3). 209 A sender is also recommended to provide a method that controls the 210 maximum burst size, Section 4.4.2. However, implementors are allowed 211 flexibility in how this method is implemented and the choice of an 212 appropriate method is expected to depend on the way in which the 213 sender stack implements other TCP methods (such as TCP Segment 214 Offload, TSO). 216 1.2. Standards Status of this Document 218 The document obsoletes the methods described in [RFC2861]. It 219 recommends a set of mechanisms, including the use of pacing during a 220 non-validated period. The updated mechanisms are intended to have a 221 less aggressive congestion impact than would be exhibited by a 222 standard TCP sender. 224 The specification in this draft is classified as "Experimental" 225 pending experience with deployed implementations of the methods. 227 2. Reviewing experience with TCP-CWV 229 [RFC2861] described a simple modification to the TCP congestion 230 control algorithm that decayed the cwnd after the transition to a 231 "sufficiently-long" idle period. This used the slow-start threshold 232 (ssthresh) to save information about the previous value of the 233 congestion window. The approach relaxed the standard TCP behaviour 234 [RFC5681] for an idle session, intended to improve application 235 performance. CWV also modified the behaviour when a sender 236 transmitted at a rate less than allowed by cwnd. 238 [RFC2861] proposed two set of responses, one after an "application- 239 limited" and one after an "idle period". Although this distinction 240 was argued, in practice differentiating the two conditions was found 241 problematic in actual networks (e.g., [Bis10]). While this offers 242 predictable performance for long on-off periods (>>1 RTT), or slowly 243 varying rate-based traffic, the performance could be unpredictable 244 for variable-rate traffic and depended both upon whether an accurate 245 RTT had been obtained and the pattern of application traffic relative 246 to the measured RTT. 248 Many applications can and often do vary their transmission over a 249 wide range of rates. Using [RFC2861] such applications often 250 experienced varying performance, which made it hard for application 251 developers to predict the TCP latency even when using a path with 252 stable network characteristics. We argue that an attempt to classify 253 application behaviour as application-limited or idle is problematic 254 and also inappropriate. This document therefore explicitly avoids 255 trying to differentiate these two cases, instead treating all rate- 256 limited traffic uniformly. 258 [RFC2861] has been implemented in some mainstream operating systems 259 as the default behaviour [Bis08]. Analysis (e.g., [Bis10] [Fai12]) 260 has shown that a TCP sender using CWV is able to use available 261 capacity on a shared path after an idle period. This can benefit 262 variable-rate applications, especially over long delay paths, when 263 compared to the slow-start restart specified by standard TCP. 264 However, CWV would only benefit an application if the idle period 265 were less than several Retransmission Time Out (RTO) intervals 266 [RFC6298], since the behaviour would otherwise be the same as for 267 standard TCP, which resets the cwnd to the TCP Restart Window after 268 this period. 270 To enable better performance for variable-rate applications with TCP, 271 some operating systems have chosen to support non-standard methods, 272 or applications have resorted to "padding" streams by sending dummy 273 data to maintain their sending rate when they have no data to 274 transmit. Although transmitting redundant data across a network path 275 provides good evidence that the path can sustain data at the offered 276 rate, padding also consumes network capacity and reduces the 277 opportunity for congestion-free statistical multiplexing. For 278 variable-rate flows, the benefits of statistical multiplexing can be 279 significant and it is therefore a goal to find a viable alternative 280 to padding streams. 282 Experience with [RFC2861] suggests that although the CWV method 283 benefited the network in a rate-limited scenario (reducing the 284 probability of network congestion), the behaviour was too 285 conservative for many common rate-limited applications. This 286 mechanism did not therefore offer the desirable increase in 287 application performance for rate-limited applications and it is 288 unclear whether applications actually use this mechanism in the 289 general Internet. 291 It is therefore concluded that CWV, as defined in [RFC2861], was 292 often a poor solution for many rate-limited applications. It had the 293 correct motivation, but had the wrong approach to solving this 294 problem. 296 3. Terminology 298 The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", 299 "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this 300 document are to be interpreted as described in [RFC2119]. 302 The document assumes familiarity with the terminology of TCP 303 congestion control [RFC5681]. 305 The following additional terminology is introduced in this document: 307 cwnd-limited: A TCP flow that has sent the maximum number of segments 308 permitted by the cwnd, where the application utilises the allowed 309 sending rate (see Section 4.5.3). 311 pipeACK sample: A measure of the volume of data acknowledged by the 312 network within an RTT. 314 pipeACK variable: A variable that measures the available capacity 315 using the set of pipeACK samples. 317 pipeACK Sampling Period: The maximum period that a measured pipeACK 318 sample may influence the pipeACK variable. 320 Non-validated phase: The phase where the cwnd reflects a previous 321 measurement of the available path capacity. 323 Non-validated period, NVP: The maximum period for which cwnd is 324 preserved in the non-validated phase. 326 Rate-limited: A TCP flow that does not consume more than one half of 327 cwnd, and hence operates in the non-validated phase. This includes 328 periods when an application is either idle or chooses to send at a 329 rate less than the maximum permitted by the cwnd. 331 Validated phase: The phase where the cwnd reflects a current estimate 332 of the available path capacity. 334 4. A New Congestion Window Validation method 336 This section proposes an update to the TCP congestion control 337 behaviour during a rate-limited interval. This new method 338 intentionally does not differentiate between times when the sender 339 has become idle or chooses to send at a rate less than the maximum 340 allowed by the cwnd. 342 The period where actual usage is less than allowed by cwnd, is named 343 the non-validated phase. The update allows an application in the 344 non-validated phase to resume transmission at a previous rate without 345 incurring the delay of slow-start. However, if the TCP sender 346 experiences congestion using the preserved cwnd, it is required to 347 immediately reset the cwnd to an appropriate value specified by the 348 method. If a sender does not take advantage of the preserved cwnd 349 within the Non-validated period, NVP, the value of cwnd is reduced, 350 ensuring the value better reflects the capacity that was recently 351 actually used. 353 It is expected that this update will satisfy the requirements of many 354 rate-limited applications and at the same time provide an appropriate 355 method for use in the Internet. New-CWV reduces this incentive for 356 an application to send "padding" data simply to keep transport 357 congestion state. 359 The method is specified in following subsections and is expected to 360 encourage applications and TCP stacks to use standards-based 361 congestion control methods. It may also encourage the use of long- 362 lived connections where this offers benefit (such as persistent 363 http). 365 4.1. Initialisation 367 A sender starts a TCP connection in the validated phase and 368 initialises the pipeACK variable to the "undefined" value. This 369 value inhibits use of the value in cwnd calculations. 371 4.2. Estimating the validated capacity supported by a path 373 [RFC6675] defines a variable, FlightSize, that indicates the 374 instantaneous amount of data that has been sent, but not cumulatively 375 acknowledged. In this method a new variable "pipeACK" is introduced 376 to measure the acknowledged size of the network pipe. This is used 377 to determine if the sender has validated the cwnd. pipeACK differs 378 from FlightSize in that it is evaluated over a window of acknowledged 379 data, rather than reflecting the amount of data outstanding. 381 A sender determines a pipeACK sample by measuring the volume of data 382 that was acknowledged by the network over the period of a measured 383 Round Trip Time (RTT). Using the variables defined in [RFC6675], a 384 value could be measured by caching the value of HighACK and after one 385 RTT measuring the difference between the cached HighACK value and the 386 current HighACK value. A sender MAY count TCP DupACKs that 387 acknowledge new data when collecting the pipeACK sample. Other 388 equivalent methods may be used. 390 A sender is not required to continuously update the pipeACK variable 391 after each received ACK, but SHOULD perform a pipeACK sample at least 392 once per RTT when it has sent unacknowledged segments. 394 The pipeACK variable MAY consider multiple pipeACK samples over the 395 pipeACK Sampling Period. The value of the pipeACK variable MUST NOT 396 exceed the maximum (highest value) within the sampling period. This 397 specification defines the pipeACK Sampling Period as Max(3*RTT, 1 398 second). This period enables a sender to compensate for large 399 fluctuations in the sending rate, where there may be pauses in 400 transmission, and allows the pipeACK variable to reflect the largest 401 recently measured pipeACK sample. 403 When no measurements are available (e.g., a sender that has just 404 started transmission or immediately after loss recovery), the pipeACK 405 variable is set to the "undefined value". This value is used to 406 inhibit entering the non-validated phase until the first new 407 measurement of a pipeACK sample. (Section 4.5 provides examples of 408 implementation.) 410 The pipeACK variable MUST NOT be updated during TCP Fast Recovery. 411 That is, the sender stops collecting pipeACK samples during loss 412 recovery. The method RECOMMENDS enabling the TCP SACK option 413 [RFC2018] and RECOMMENDS the method defined in [RFC6675] to recover 414 missing segments. This allows the sender to more accurately 415 determine the number of missing bytes during the loss recovery phase, 416 and using this method will result in a more appropriate cwnd 417 following loss. 419 NOTE: The use of pipeACK rather than FlightSize can change the 420 behaviour of a TCP when a sender does not always have data available 421 to send. One example arises when there is a pause in transmission 422 after sending a sequence of many packets, and the sender experiences 423 loss at or near the end of its transmission sequence. In this case, 424 the TCP flow may have used a significant amount of capacity just 425 prior to the loss (which would be reflected in the volume of data 426 acknowledged, recorded in the pipeACK variable), but at the actual 427 time of loss the number of unacknowledged packets in flight (at the 428 end of the sequence) may be small, i.e., there is a small FlightSize. 429 After loss recovery, the sender resets its congestion control state. 431 [Fai12] explored the benefits of different responses to congestion 432 for application-limited streams. If the response is based only on 433 the Loss FlightSize, the sender would assign a small cwnd and 434 ssthresh, based only on the volume of data sent after the loss. When 435 the sender next starts to transmit it can incur may RTTs of delay in 436 slow start before it reacquires its previous rate. When the pipeACK 437 value is also usedto calculate the cwnd and ssthresh (as specified in 438 this update in Section 4.4.1), the sender can use a value that also 439 reflects the recently used capacity before the loss. This prevents a 440 variable-rate application from being unduly penalised. When the 441 sender resumes, it starts at one half its previous rate, similar to 442 the behaviour of a bulk TCP flow [Hos15]. To ensure an appropriate 443 reaction to on-going congestion, this method requires that the 444 pipeACK variable is reset after it is used in this way. 446 4.3. Preserving cwnd during a rate-limited period. 448 The updated method creates a new TCP sender phase that captures 449 whether the cwnd reflects a validated or non-validated value. The 450 phases are defined as: 452 o Validated phase: pipeACK >=(1/2)*cwnd, or pipeACK is undefined 453 (i.e., at the start or directly after loss recovery). This is the 454 normal phase, where cwnd is expected to be an approximate 455 indication of the capacity currently available along the network 456 path, and the standard methods are used to increase cwnd 457 (currently [RFC5681]). 459 o Non-validated phase: pipeACK <(1/2)*cwnd. This is the phase where 460 the cwnd has a value based on a previous measurement of the 461 available capacity, and the usage of this capacity has not been 462 validated in the pipeACK Sampling Period. That is, when it is not 463 known whether the cwnd reflects the currently available capacity 464 along the network path. The mechanisms to be used in this phase 465 seek to determine a safe value for cwnd and an appropriate 466 reaction to congestion. 468 Note: A threshold is needed to determine whether a sender is in the 469 validated or non-validated phase. A standard TCP sender in slow- 470 start is permitted to double its FlightSize from one RTT to the next. 471 This motivated the choice of a threshold value of 1/2. This 472 threshold ensures a sender does not further increase the cwnd as long 473 as the FlightSize is less than (1/2*cwnd). Furthermore, a sender 474 with a FlightSize less than (1/2*cwnd) may in the next RTT be 475 permitted by the cwnd to send at a rate that more than doubles the 476 FlightSize, and hence this case needs to be regarded as non-validated 477 and a sender therefore needs to employ additional mechanisms while in 478 this phase. 480 4.4. TCP congestion control during the non-validated phase 482 A TCP sender implementing this specification MUST enter the non- 483 validated phase when the pipeACK is less than (1/2)*cwnd. (The note 484 at the end of section 4.4.1 describes why pipeACK<=(1/2)*cwnd is 485 expected to be a safe value.) 487 A TCP sender that enters the non-validated phase preserves the cwnd 488 (i.e., the cwnd only increases after a sender fully uses the cwnd in 489 this phase, otherwise the cwnd neither grows nor reduces). The phase 490 is concluded when the sender transmits sufficient data so that 491 pipeACK > (1/2)*cwnd (i.e., the sender is no longer rate-limited), or 492 when the sender receives an indication of congestion. 494 After a fixed period of time (the non-validated period, NVP), the 495 sender adjusts the cwnd Section 4.4.3). The NVP SHOULD NOT exceed 5 496 minutes.Section 5 discusses the rationale for choosing a safe value 497 for this period. 499 The behaviour in the non-validated phase is specified as: 501 o A sender determines whether to increase the cwnd based upon 502 whether it is cwnd-limited (see Section 4.5.3): 504 * A sender that is cwnd-limited MAY use the standard TCP method 505 to increase cwnd (i.e., a TCP sender that fully utilises the 506 cwnd is permitted to increase cwnd each received ACK using 507 standard methods). 509 * A sender that is not cwnd-limited MUST NOT increase the cwnd 510 when ACK packets are received in this phase (i.e., needs to 511 avoid growing the cwnd when it has not recently sent using the 512 current size of cwnd). 514 o If the sender receives an indication of congestion while in the 515 non-validated phase (i.e., detects loss), the sender MUST exit the 516 non-validated phase (reducing the cwnd as defined in 517 Section 4.4.1). 519 o If the Retransmission Time Out (RTO) expires while in the non- 520 validated phase, the sender MUST exit the non-validated phase. It 521 then resumes using the standard TCP RTO mechanism [RFC5681]. 523 o A sender with a pipeACK variable greater than (1/2)*cwnd SHOULD 524 enter the validated phase. (A rate-limited sender will not 525 normally be impacted by whether it is in a validated or non- 526 validated phase, since it will normally not increase FlightSize to 527 use the entire cwnd. However, a change to the validated phase 528 will release the sender from constraints on the growth of cwnd, 529 and result in using the standard congestion response.) 531 The cwnd-limited behaviour may be triggered during a transient 532 condition that occurs when a sender is in the non-validated phase and 533 receives an ACK that acknowledges received data, the cwnd was fully 534 utilised, and more data is awaiting transmission than may be sent 535 with the current cwnd. The sender MAY then use the standard method 536 to increase the cwnd. (Note, if the sender succeeds in sending these 537 new segments, the updated cwnd and pipeACK variables will eventually 538 result in a transition to the validated phase.) 540 4.4.1. Response to congestion in the non-validated phase 542 Reception of congestion feedback while in the non-validated phase is 543 interpreted as an indication that it was inappropriate for the sender 544 to use the preserved cwnd. The sender is therefore required to 545 quickly reduce the rate to avoid further congestion. Since the cwnd 546 does not have a validated value, a new cwnd value needs to be 547 selected based on the utilised rate. 549 A sender that detects a packet-drop MUST record the current 550 FlightSize in the variable LossFlightSize and MUST calculate a safe 551 cwnd for loss recovery using the method below: 553 cwnd = (Max(pipeACK,LossFlightSize))/2. 555 The pipeACK value is not updated during loss recovery (see 556 Section 4.2). If there is a valid pipeACK value, the new cwnd is 557 adjusted to reflect that a non-validated cwnd may be larger than the 558 actual FlightSize, or recently used FlightSize (recorded in pipeACK). 559 The updated cwnd therefore prevents overshoot by a sender 560 significantly increasing its transmission rate during the recovery 561 period. 563 At the end of the recovery phase, the TCP sender MUST reset the cwnd 564 using the method below: 566 cwnd = (Max(pipeACK,LossFlightSize) - R)/2. 568 Where R is the volume of data that was successfully retransmitted 569 during the recovery phase. This corresponds to segments 570 retransmitted and considered lost by the pipe estimation algorithm at 571 the end of recovery. It does not include the additional cost of 572 multiple retransmission of the same data. The loss of segments 573 indicates that the path capacity was exceeded by at least R, and 574 hence the calculated cwnd is reduced by at least R before the window 575 is halved. 577 The calculated cwnd value MUST NOT be reduced below 1 TCP Maximum 578 Segment Size (MSS). 580 After completing the loss recovery phase, the sender MUST re- 581 initialise the pipeACK variable to the "undefined" value. This 582 ensures that standard TCP methods are used immediately after 583 completing loss recovery until a new pipeACK value can be determined. 585 The ssthresh is adjusted using the standard TCP method (Step 6 in 586 Section 3.2 of RFC 5681 assigns the ssthresh a value equal to cwnd at 587 the end of the loss recovery). 589 Note: The adjustment by reducing cwnd by the volume of data not sent 590 (R) follows the method proposed for Jump Start [Liu07]. The 591 inclusion of the term R makes the adjustment more conservative than 592 standard TCP. This is required, since a sender in the non-validated 593 state may increase the rate more than a standard TCP would have done 594 relative to what was sent in the last RTT (i.e., more than doubled 595 the number of segments in flight relative to what it sent in the last 596 RTT). The additional reduction after congestion is beneficial when 597 the LossFlightSize has significantly overshot the available path 598 capacity incurring significant loss (e.g., following a change of path 599 characteristics or when additional traffic has taken a larger share 600 of the network bottleneck during a period when the sender transmits 601 less). 603 Note: The pipeACK value is only valid during a non-validated phase, 604 and therefore this does not exceed cwnd/2. If LossFlightSize and R 605 were small, then this can result in the final cwnd after loss 606 recovery being at most one quarter of the cwnd on detection of 607 congestion. This reduction is conservative, and pipeACK is then 608 reset to undefined, hence cwnd updates after a congestion event do 609 not depend upon the pipeACK history before congestion was detected. 611 4.4.2. Sender burst control during the non-validated phase 613 TCP congestion control allows a sender to accumulate a cwnd that 614 would allow it to send a burst of segments with a total size up to 615 the difference between the FlightsSize and cwnd. Such bursts can 616 impact other flows that share a network bottleneck and/or may induce 617 congestion when buffering is limited. 619 Various methods have been proposed to control the sender burstiness 620 [Hug01], [All05]. For example, TCP can limit the number of new 621 segments it sends per received ACK. This is effective when a flow of 622 ACKs is received, but can not be used to control a sender that has 623 not send appreciable data in the previous RTT [All05]. 625 This document recommends using a method to avoid line-rate bursts 626 after an idle or rate-limited interval when there is less reliable 627 information about the capacity of the network path: A TCP sender in 628 the non-validated phase SHOULD control the maximum burst size, e.g., 629 using a rate-based pacing algorithm in which a sender paces out the 630 cwnd over its estimate of the RTT, or some other method, to prevent 631 many segments being transmitted contiguously at line-rate. The most 632 appropriate method(s) to implement pacing depend on the design of the 633 TCP/IP stack, speed of interface and whether hardware support (such 634 as TCP Segment Offload, TSO) is used. The present document does not 635 recommend any specific method. 637 4.4.3. Adjustment at the end of the Non-Validated Period (NVP) 639 An application that remains in the non-validated phase for a period 640 greater than the NVP is required to adjust its congestion control 641 state. If the sender exits the non-validated phase after this 642 period, it MUST update the ssthresh: 644 ssthresh = max(ssthresh, 3*cwnd/4). 646 (This adjustment of ssthresh ensures that the sender records that it 647 has safely sustained the present rate. The change is beneficial to 648 rate-limited flows that encounter occasional congestion, and could 649 otherwise suffer an unwanted additional delay in recovering the 650 sending rate.) 652 The sender MUST then update cwnd to be not greater than: 654 cwnd = max((1/2)*cwnd, IW). 656 Where IW is the appropriate TCP initial window, used by the TCP 657 sender (e.g., [RFC5681]). 659 Note: These cwnd and ssthresh adjustments cause the sender to enter 660 slow-start (since ssthresh > cwnd). This adjustment ensures that the 661 sender responds conservatively after remaining in the non-validated 662 phase for more than the non-validated period. In this case, it 663 reduces the cwnd by a factor of two from the preserved value. This 664 adjustment is helpful when flows accumulate but do not use a large 665 cwnd, and seeks to mitigate the impact when these flows later resume 666 transmission. This could for instance mitigate the impact if 667 multiple high-rate application flows were to become idle over an 668 extended period of time and then were simultaneously awakened by an 669 external event. 671 4.5. Examples of Implementation 673 This section provides informative examples of implementation methods. 674 Implementations may choose to use other methods that comply with the 675 normative requirements. 677 4.5.1. Implementing the pipeACK measurement 679 A pipeACK sample may be measured once each RTT. This reduces the 680 sender processing burden for calculating after each acknowledgement 681 and also reduces storage requirements at the sender. 683 Since application behaviour can be bursty using CWV, it may be 684 desirable to implement a maximum filter to accumulate the measured 685 values so that the pipeACK variable records the largest pipeACK 686 sample within the pipeACK Sampling Period. One simple way to 687 implement this is to divide the pipeACK Sampling Period into several 688 (e.g., 5) equal length measurement periods. The sender then records 689 the start time for each measurement period and the highest measured 690 pipeACK sample. At the end of the measurement period, any 691 measurement(s) that are older than the pipeACK Sampling Period are 692 discarded. The pipeACK variable is then assigned the largest of the 693 set of the highest measured values. 695 pipeACK sample (Bytes) 696 ^ 697 | +----------+----------+ +----------+---...... 698 | | Sample A | Sample B | No | Sample C | Sample D 699 | | | | Sample | | 700 | | |\ 5 | | | | 701 | | | | | | | /\ 4 | 702 | | | | | |\ 3 | | | \ | 703 | | | \ | | \--- | | / \ | /| 2 704 | |/ \------| - | | / \------/ \... 705 +//-+----------+---------\+----/ /----+/---------+-------------> Time 707 <------------------------------------------------| 708 Sampling Period Current Time 710 Figure 1: Example of measuring pipeACK samples 711 Figure 1 shows an example of how measurement samples may be 712 collected. At the time represented by the figure new samples are 713 being accumulated into sample D. Three previous samples also fall 714 within the pipeACK Sampling Period: A, B, and C. There was also a 715 period of inactivity between samples B and C during which no 716 measurements were taken (because no new data segments were 717 acknowledged). The current value of the pipeACK variable will be 5, 718 the maximum across all samples. During this period, the pipeACK 719 samples may be regarded as zero, and hence do not contribute to the 720 calculated pipeACK value. 722 After one further measurement period, Sample A will be discarded, 723 since it then is older than the pipeACK Sampling Period and the 724 pipeACK variable will be recalculated, Its value will be the larger 725 of Sample C or the final value accumulated in Sample D. 727 4.5.2. Measurement of the NVP and pipeACK samples 729 The mechanism requires a number of measurements of time. These 730 measurements could be implemented using protocol timers, but do not 731 necessarily require a new timer to be implemented. Avoiding the use 732 of dedicated timers can save operating system resources, especially 733 when there may be large numbers of TCP flows. 735 The NVP could be measured by recording a timestamp when the sender 736 enters the non-validated phase. Each time a sender transmits a new 737 segment, this timestamp can be used to determine if the NVP has 738 expired. If the measured period exceeds the NVP, the sender can then 739 take into account how many units of the NVP have passed and make one 740 reduction (defined in Section 4.4.3) for each NVP. 742 Similarly, the time measurements for collecting pipeACK samples and 743 determining the Sampling Period could be derived by using a timestamp 744 to record when each sample was measured, and to use this to calculate 745 how much time has passed when each new ACK is received. 747 4.5.3. Implementing detection of the cwnd-limited condition 749 A sender needs to implement a method that detects the cwnd-limited 750 condition (see Section 4.4). This detects a condition where a sender 751 in the non-validated phase receives an ACK, but the size of cwnd 752 prevents sending more new data. 754 In simple terms, this condition is true only when the FlightSize of a 755 TCP sender is equal to or larger than the current cwnd. However, an 756 implementation also needs to consider constraints on the way in which 757 the cwnd variable can be used, for instance implementations need to 758 support other TCP methods such as the Nagle Algorithm and TCP Segment 759 Offload (TSO) that also use cwnd to control transmission. These 760 other methods can result in a sender becoming cwnd-limited when the 761 cwnd is nearly, rather than completely, equal to the FlightSize. 763 5. Determining a safe period to preserve cwnd 765 This section documents the rationale for selecting the maximum period 766 that cwnd may be preserved, known as the NVP. 768 Limiting the period that cwnd may be preserved avoids undesirable 769 side effects that would result if the cwnd were to be kept 770 unnecessarily high for an arbitrary long period, which was a part of 771 the problem that CWV originally attempted to address. The period a 772 sender may safely preserve the cwnd, is a function of the period that 773 a network path is expected to sustain the capacity reflected by cwnd. 774 There is no ideal choice for this time. 776 A period of five minutes was chosen for this NVP. This is a 777 compromise that was larger than the idle intervals of common 778 applications, but not sufficiently larger than the period for which 779 the capacity of an Internet path may commonly be regarded as stable. 780 The capacity of wired networks is usually relatively stable for 781 periods of several minutes and that load stability increases with the 782 capacity. This suggests that cwnd may be preserved for at least a 783 few minutes. 785 There are cases where the TCP throughput exhibits significant 786 variability over a time less than five minutes. Examples could 787 include wireless topologies, where TCP rate variations may fluctuate 788 on the order of a few seconds as a consequence of medium access 789 protocol instabilities. Mobility changes may also impact TCP 790 performance over short time scales. Senders that observe such rapid 791 changes in the path characteristic may also experience increased 792 congestion with the new method, however such variation would likely 793 also impact TCP's behaviour when supporting interactive and bulk 794 applications. 796 Routing algorithms may change the the network path that is used by a 797 transport. Although a change of path can in turn disrupt the RTT 798 measurement and may result in a change of the capacity available to a 799 TCP connection, we assume these path changes do not usually occur 800 frequently (compared to a time frame of a few minutes). 802 The value of five minutes is therefore expected to be sufficient for 803 most current applications. Simulation studies (e.g., [Bis11]) also 804 suggest that for many practical applications, the performance using 805 this value will not be significantly different to that observed using 806 a non-standard method that does not reset the cwnd after idle. 808 Finally, other TCP sender mechanisms have used a 5 minute timer, and 809 there could be simplifications in some implementations by reusing the 810 same interval. TCP defines a default user timeout of 5 minutes 811 [RFC0793] i.e., how long transmitted data may remain unacknowledged 812 before a connection is forcefully closed. 814 6. Security Considerations 816 General security considerations concerning TCP congestion control are 817 discussed in [RFC5681]. This document describes an algorithm that 818 updates one aspect of the congestion control procedures, and so the 819 considerations described in RFC 5681 also apply to this algorithm. 821 7. IANA Considerations 823 There are no IANA considerations. 825 8. Acknowledgments 827 This document was produced by the TCP Maintenance and Minor 828 Extensions (tcpm) working group. 830 The authors acknowledge the contributions of Dr I Biswas, Dr Ziaul 831 Hossain in supporting the evaluation of CWV and for their help in 832 developing the mechanisms proposed in this draft. We also 833 acknowledge comments received from the Internet Congestion Control 834 Research Group, in particular Yuchung Cheng, Mirja Kuehlewind, Joe 835 Touch, and Mark Allman. This work was part-funded by the European 836 Community under its Seventh Framework Programme through the Reducing 837 Internet Transport Latency (RITE) project (ICT-317700). 839 9. Author Notes 841 RFC-Editor note: please remove this section prior to publication. 843 9.1. Other related work 845 RFC-Editor note: please remove this section prior to publication. 847 There are several issues to be discussed more widely: 849 o There are potential interactions with the Experimental update in 850 RFC 6928 that raises the TCP initial Window to ten segments, do 851 these cases need to be elaborated? 852 This relates to the Experimental specification for increasing 853 the TCP IW defined in RFC 6928. 855 The two methods have different functions and different response 856 to loss/congestion. 858 RFC 6928 proposes an experimental update to TCP that would 859 increase the IW to ten segments. This would allow faster 860 opening of the cwnd, and also a large (same size) restart 861 window. This approach is based on the assumption that many 862 forward paths can sustain bursts of up to ten segments without 863 (appreciable) loss. Such a significant increase in cwnd must 864 be matched with an equally large reduction of cwnd if loss/ 865 congestion is detected, and such a congestion indication is 866 likely to require future use of IW=10 to be disabled for this 867 path for some time. This guards against the unwanted behaviour 868 of a series of short flows continuously flooding a network path 869 without network congestion feedback. 871 In contrast, this document proposes an update with a rationale 872 that relies on recent previous path history to select an 873 appropriate cwnd after restart. 875 The behaviour differs in three ways: 877 1) For applications that send little initially, new-cwv may 878 constrain more than RFC 6928, but would not require the 879 connection to reset any path information when a restart 880 incurred loss. In contrast, new-cwv would allow the TCP 881 connection to preserve the cached cwnd, any loss, would impact 882 cwnd, but not impact other flows. 884 2) For applications that utilise more capacity than provided by 885 a cwnd of 10 segments, this method would permit a larger 886 restart window compared to a restart using the method in RFC 887 6928. This is justified by the recent path history. 889 3) new-CWV is attended to also be used for rate-limited 890 applications, where the application sends, but does not seek to 891 fully utilise the cwnd. In this case, new-cwv constrains the 892 cwnd to that justified by the recent path history. The 893 performance trade-offs are hence different, and it would be 894 possible to enable new-cwv when also using the method in RFC 895 6928, and yield benefits. 897 o There is potential overlap with the Laminar proposal (draft- 898 mathis-tcpm-tcp-laminar) 899 The current draft was intended as a standards-track update to 900 TCP, rather than a new transport variant. At least, it would 901 be good to understand how the two interact and whether there is 902 a possibility of a single method. 904 o There is potential performance loss in loss of a short burst 905 (off list with M Allman) 907 A sender can transmit several segments then become idle. If 908 the first set of segments are all Acknowledged, the ssthresh 909 collapses to a small value (no new data is sent by the idle 910 sender). Loss of the later data results in congestion (e.g., 911 maybe a RED drop or some other cause, rather than the maximum 912 rate of this flow). When the sender performs loss recovery it 913 may have an appreciable pipeACK and cwnd, but a very low 914 FlightSize - the Standard algorithm therefore results in an 915 unusually low cwnd ((1/2)* FlightSize). 917 A constant rate flow would have maintained a FlightSize 918 appropriate to pipeACK (cwnd, if it is a bulk flow). 920 This could be fixed by adding a new state variable? It could 921 also be argued this is a corner case (e.g., loss of only the 922 last segments would have resulted in RTO), the impact could be 923 significant. 925 o There is potential interaction with TCP Control Block Sharing(M 926 Welzl) 928 An application that is non-validated can accumulate a cwnd that 929 is larger than the actual capacity. Is this a fair value to 930 use in TCB sharing? 932 We propose that TCB sharing should use the pipeACK in place of 933 cwnd when a TCP sender is in the Non-validated phase. This 934 value better reflects the capacity that the flow has utilised 935 in the network path. 937 10. Revision notes 939 RFC-Editor note: please remove this section prior to publication. 941 Draft 03 was submitted to ICCRG to receive comments and feedback. 943 Draft 04 contained the first set of clarifications after feedback: 945 o Changed name to application limited and used the term rate-limited 946 in all places. 948 o Added justification and many minor changes suggested on the list. 950 o Added text to tie-in with more accurate ECN marking. 952 o Added ref to Hug01 954 Draft 05 contained various updates: 956 o New text to redefine how to measure the acknowledged pipe, 957 differentiating this from the FlightSize, and hence avoiding 958 previous issues with infrequent large bursts of data not being 959 validated. A key point new feature is that pipeACK only triggers 960 leaving the NVP after the size of the pipe has been acknowledged. 961 This removed the need for hysteresis. 963 o Reduction values were changed to 1/2, following analysis of 964 suggestions from ICCRG. This also sets the "target" cwnd as twice 965 the used rate for non-validated case. 967 o Introduced a symbolic name (NVP) to denote the 5 minute period. 969 Draft 06 contained various updates: 971 o Required reset of pipeACK after congestion. 973 o Added comment on the effect of congestion after a short burst (M. 974 Allman). 976 o Correction of minor Typos. 978 WG draft 00 contained various updates: 980 o Updated initialisation of pipeACK to maximum value. 982 o Added note on intended status still to be determined. 984 WG draft 01 contained: 986 o Added corrections from Richard Scheffenegger. 988 o Raffaello Secchi added to the mechanism, based on implementation 989 experience. 991 o Removed that the requirement for the method to use TCP SACK option 993 o Although it may be desirable to use SACK, this is not essential to 994 the algorithm. 996 o Added the notion of the sampling period to accommodate large rate 997 variations and ensure that the method is stable. This algorithm 998 to be validated through implementation. 1000 WG draft 02 contained: 1002 o Clarified language around pipeACK variable and pipeACK sample - 1003 Feedback from Aris Angelogiannopoulos. 1005 WG draft 03 contained: 1007 o Editorial corrections - Feedback from Anna Brunstrom. 1009 o An adjustment to the procedure at the start and end of Reoloss 1010 recovery to align the two equations. 1012 o Further clarification of the "undefined" value of the pipeACK 1013 variable. 1015 WG draft 04 contained: 1017 o Editorial corrections. 1019 o Introduced the "cwnd-limited" term. 1021 o An adjustment to the procedure at the start of a cwnd-limited 1022 phase - the new text is intended to ensure that new-cwv is not 1023 unnecessarily more conservative than standard TCP when the flow is 1024 cwnd-limited. This resolves two issues: first it prevents 1025 pathologies in which pipeACK increases slowly and erratically. It 1026 also ensures that performance of bulk applications is not 1027 significantly impacted when using the method. 1029 o Clearly identifies that pacing (or equivalent) is requiring during 1030 the NVP to control burstiness. New section added. 1032 WG draft 05 contained: 1034 o Clarification to first two bullets in Section 4.4 describing cwnd- 1035 limited, to explain these are really alternates to the same case. 1037 o Section giving implementation examples was restructured to clarify 1038 there are two methods described. 1040 o Cross References to sections updated - thanks to comments from 1041 Martin Winbjoerk and Tim Wicinski. 1043 WG draft 06 contained: 1045 o The section giving implementation examples was restructured to 1046 clarify there are two methods described. 1048 o Justification of design decisions. 1050 o Re-organised text to improve clarity of argument. 1052 WG draft 07 contained: 1054 o Updated publication date. 1056 o Text on noting that cwnd shouldn't ever be made negative. 1058 o Updated text on ECN to clarify the process where R is a reduction 1059 based on ECN marks. 1061 WG draft 08 contained: 1063 o Removed description of how to use Accurate ECN feedback. It is 1064 not clear that this document should specify a usage of a mechanism 1065 that has not been fully defined. Accurate ECN may lead to 1066 different congestion responses and these will need to be defined 1067 in the CC specifications for using Accurate ECN. 1069 WG draft 09 contained: 1071 o Removed update to RFC 5681 - the status of the present document is 1072 Experimental, and hence this document does not update RFC 5681. 1074 WG draft 10 contained edits following WGLC: 1076 o Section 1.1 Implementation of new CWV: New section added to 1077 introduce the places where there are implementation flexibility. 1079 o Section 4.4: Clarified that the MUST is to satisfy the goal to 1080 avoid a TCP sender growing a large "non-validated" cwnd, when it 1081 has not recently sent using the current size of cwnd, and fixed 1082 format of bullet 2 in 4.4. 1084 o Section 4.5.2: rewritten section text. 1086 WG draft 11 contained edits following IETF LC: 1088 o Updated text in section 1.1. 1090 o Updated text in response to AD, Gen-ART, & Sec reviews. 1092 o LC call comments from Mirja Kuehlewind 1094 WG draft 12 contained edits following IETF LC (Mirja Kuehlewind): 1096 o Additional text (based on text in annexe notes) to clarify use of 1097 pipeACK rather than FlightSize. 1099 o Corrected text on undefined pipeACK - to be consistent. 1101 o Added text on standard TCP method (reference to RFC 5681). 1103 o Separated text on implementation experience of "timers" into a new 1104 implementation subsection (4.5.2), to avoid this common 1105 implementation method being overlooked. 1107 WG draft 13 contained edits following IESG Review: 1109 o Jari/Gen-ART (note: MSS was defined) 1111 o Kathleen Moriarty (SecDir) 1113 o Ben Campbell 1115 o Barry Leiba (note: reference added to section 4, rather than new 1116 wording to requirement). 1118 11. References 1120 11.1. Normative References 1122 [RFC0793] Postel, J., "Transmission Control Protocol", September 1123 1981. 1125 [RFC2018] Mathis, M., Mahdavi, J., Floyd, S., and A. Romanow, "TCP 1126 Selective Acknowledgment Options", RFC 2018, October 1996. 1128 [RFC2119] Bradner, S., "Key words for use in RFCs to Indicate 1129 Requirement Levels", BCP 14, RFC 2119, March 1997. 1131 [RFC2861] Handley, M., Padhye, J., and S. Floyd, "TCP Congestion 1132 Window Validation", RFC 2861, June 2000. 1134 [RFC5681] Allman, M., Paxson, V., and E. Blanton, "TCP Congestion 1135 Control", September 2009. 1137 [RFC6298] Paxson, V., Allman, M., Chu, J., and M. Sargent, 1138 "Computing TCP's Retransmission Timer", June 2011. 1140 [RFC6675] Blanton, E., Allman, M., Wang, L., Jarvinen, I., Kojo, M., 1141 and Y. Nishida, "A Conservative Loss Recovery Algorithm 1142 Based on Selective Acknowledgment (SACK) for TCP", RFC 1143 6675, August 2012. 1145 11.2. Informative References 1147 [All05] Allman, M. and E. Blanton, "Notes on burst mitigation for 1148 transport protocols", March 2005. 1150 [Bis08] Biswas, I. and G. Fairhurst, "A Practical Evaluation of 1151 Congestion Window Validation Behaviour, 9th Annual 1152 Postgraduate Symposium in the Convergence of 1153 Telecommunications, Networking and Broadcasting (PGNet), 1154 Liverpool, UK", June 2008. 1156 [Bis10] Biswas, I., Sathiaseelan, A., Secchi, R., and G. 1157 Fairhurst, "Analysing TCP for Bursty Traffic, Int'l J. of 1158 Communications, Network and System Sciences, 7(3)", June 1159 2010. 1161 [Bis11] Biswas, I., "PhD Thesis, Internet congestion control for 1162 variable rate TCP traffic, School of Engineering, 1163 University of Aberdeen", June 2011. 1165 [Fai12] Sathiaseelan, A., Secchi, R., Fairhurst, G., and I. 1166 Biswas, "Enhancing TCP Performance to support Variable- 1167 Rate Traffic, 2nd Capacity Sharing Workshop, ACM CoNEXT, 1168 Nice, France, 10th December 2012.", June 2008. 1170 [Hos15] Hossain, Z., "PhD Thesis, A Study of Mechanisms to Support 1171 Variable-rate Internet Applications over a Multi-service 1172 Satellite Platform, School of Engineering, University of 1173 Aberdeen", January 2015. 1175 [Hug01] Hughes, A., Touch, J., and J. Heidemann, "Issues in TCP 1176 Slow-Start Restart After Idle (Work-in-Progress)", 1177 December 2001. 1179 [Liu07] Liu, D., Allman, M., Jiny, S., and L. Wang, "Congestion 1180 Control without a Startup Phase, 5th International 1181 Workshop on Protocols for Fast Long-Distance Networks 1182 (PFLDnet), Los Angeles, California, USA", February 2007. 1184 [RFC7230] Fielding, R. and J. Reschke, "Hypertext Transfer Protocol 1185 (HTTP/1.1): Message Syntax and Routing", RFC 7230, June 1186 2014. 1188 Authors' Addresses 1190 Godred Fairhurst 1191 University of Aberdeen 1192 School of Engineering 1193 Fraser Noble Building 1194 Aberdeen, Scotland AB24 3UE 1195 UK 1197 Email: gorry@erg.abdn.ac.uk 1198 URI: http://www.erg.abdn.ac.uk 1200 Arjuna Sathiaseelan 1201 University of Aberdeen 1202 School of Engineering 1203 Fraser Noble Building 1204 Aberdeen, Scotland AB24 3UE 1205 UK 1207 Email: arjuna@erg.abdn.ac.uk 1208 URI: http://www.erg.abdn.ac.uk 1210 Raffaello Secchi 1211 University of Aberdeen 1212 School of Engineering 1213 Fraser Noble Building 1214 Aberdeen, Scotland AB24 3UE 1215 UK 1217 Email: raffaello@erg.abdn.ac.uk 1218 URI: http://www.erg.abdn.ac.uk