idnits 2.17.1 draft-ietf-aqm-eval-guidelines-08.txt: Checking boilerplate required by RFC 5378 and the IETF Trust (see https://trustee.ietf.org/license-info): ---------------------------------------------------------------------------- No issues found here. Checking nits according to https://www.ietf.org/id-info/1id-guidelines.txt: ---------------------------------------------------------------------------- No issues found here. Checking nits according to https://www.ietf.org/id-info/checklist : ---------------------------------------------------------------------------- No issues found here. Miscellaneous warnings: ---------------------------------------------------------------------------- == The copyright year in the IETF Trust and authors Copyright Line does not match the current year -- The document date (September 21, 2015) is 3133 days in the past. Is this intentional? Checking references for intended status: Informational ---------------------------------------------------------------------------- == Missing Reference: 'Byte' is mentioned on line 295, but not defined == Outdated reference: A later version (-07) exists of draft-ietf-tcpm-cubic-00 ** Obsolete normative reference: RFC 793 (Obsoleted by RFC 9293) ** Obsolete normative reference: RFC 2679 (Obsoleted by RFC 7679) ** Obsolete normative reference: RFC 2680 (Obsoleted by RFC 7680) -- Obsolete informational reference (is this intentional?): RFC 2309 (Obsoleted by RFC 7567) Summary: 3 errors (**), 0 flaws (~~), 3 warnings (==), 2 comments (--). Run idnits with the --verbose option for more detailed information about the items above. -------------------------------------------------------------------------------- 2 Internet Engineering Task Force N. Kuhn, Ed. 3 Internet-Draft Telecom Bretagne 4 Intended status: Informational P. Natarajan, Ed. 5 Expires: March 24, 2016 Cisco Systems 6 N. Khademi, Ed. 7 University of Oslo 8 D. Ros 9 Simula Research Laboratory AS 10 September 21, 2015 12 AQM Characterization Guidelines 13 draft-ietf-aqm-eval-guidelines-08 15 Abstract 17 Unmanaged large buffers in today's networks have given rise to a slew 18 of performance issues. These performance issues can be addressed by 19 some form of Active Queue Management (AQM) mechanism, optionally in 20 combination with a packet scheduling scheme such as fair queuing. 21 The IETF Active Queue Management and Packet Scheduling working group 22 was formed to standardize AQM schemes that are robust, easily 23 implementable, and successfully deployable in today's networks. This 24 document describes various criteria for performing precautionary 25 characterizations of AQM proposals. This document also helps in 26 ascertaining whether any given AQM proposal should be taken up for 27 standardization by the AQM WG. 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 March 24, 2016. 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 . . . . . . . . . . . . . . . . . . . . . . . . 4 64 1.1. Reducing the latency and maximizing the goodput . . . . . 5 65 1.2. Guidelines for AQM evaluation . . . . . . . . . . . . . . 5 66 1.3. Requirements Language . . . . . . . . . . . . . . . . . . 6 67 1.4. Glossary . . . . . . . . . . . . . . . . . . . . . . . . 6 68 2. End-to-end metrics . . . . . . . . . . . . . . . . . . . . . 6 69 2.1. Flow completion time . . . . . . . . . . . . . . . . . . 7 70 2.2. Flow start up time . . . . . . . . . . . . . . . . . . . 7 71 2.3. Packet loss . . . . . . . . . . . . . . . . . . . . . . . 7 72 2.4. Packet loss synchronization . . . . . . . . . . . . . . . 8 73 2.5. Goodput . . . . . . . . . . . . . . . . . . . . . . . . . 8 74 2.6. Latency and jitter . . . . . . . . . . . . . . . . . . . 9 75 2.7. Discussion on the trade-off between latency and goodput . 9 76 3. Generic setup for evaluations . . . . . . . . . . . . . . . . 10 77 3.1. Topology and notations . . . . . . . . . . . . . . . . . 10 78 3.2. Buffer size . . . . . . . . . . . . . . . . . . . . . . . 12 79 3.3. Congestion controls . . . . . . . . . . . . . . . . . . . 12 80 4. Methodology, Metrics, AQM Comparisons, Packet Sizes, 81 Scheduling and ECN . . . . . . . . . . . . . . . . . . . . . 13 82 4.1. Methodology . . . . . . . . . . . . . . . . . . . . . . . 13 83 4.2. Comments on metrics measurement . . . . . . . . . . . . . 13 84 4.3. Comparing AQM schemes . . . . . . . . . . . . . . . . . . 14 85 4.3.1. Performance comparison . . . . . . . . . . . . . . . 14 86 4.3.2. Deployment comparison . . . . . . . . . . . . . . . . 15 87 4.4. Packet sizes and congestion notification . . . . . . . . 15 88 4.5. Interaction with ECN . . . . . . . . . . . . . . . . . . 15 89 4.6. Interaction with Scheduling . . . . . . . . . . . . . . . 16 90 5. Transport Protocols . . . . . . . . . . . . . . . . . . . . . 16 91 5.1. TCP-friendly sender . . . . . . . . . . . . . . . . . . . 17 92 5.1.1. TCP-friendly sender with the same initial congestion 93 window . . . . . . . . . . . . . . . . . . . . . . . 17 95 5.1.2. TCP-friendly sender with different initial congestion 96 windows . . . . . . . . . . . . . . . . . . . . . . . 17 97 5.2. Aggressive transport sender . . . . . . . . . . . . . . . 18 98 5.3. Unresponsive transport sender . . . . . . . . . . . . . . 18 99 5.4. Less-than Best Effort transport sender . . . . . . . . . 19 100 6. Round Trip Time Fairness . . . . . . . . . . . . . . . . . . 19 101 6.1. Motivation . . . . . . . . . . . . . . . . . . . . . . . 19 102 6.2. Recommended tests . . . . . . . . . . . . . . . . . . . . 20 103 6.3. Metrics to evaluate the RTT fairness . . . . . . . . . . 20 104 7. Burst Absorption . . . . . . . . . . . . . . . . . . . . . . 20 105 7.1. Motivation . . . . . . . . . . . . . . . . . . . . . . . 20 106 7.2. Recommended tests . . . . . . . . . . . . . . . . . . . . 21 107 8. Stability . . . . . . . . . . . . . . . . . . . . . . . . . . 22 108 8.1. Motivation . . . . . . . . . . . . . . . . . . . . . . . 22 109 8.2. Recommended tests . . . . . . . . . . . . . . . . . . . . 23 110 8.2.1. Definition of the congestion Level . . . . . . . . . 23 111 8.2.2. Mild congestion . . . . . . . . . . . . . . . . . . . 24 112 8.2.3. Medium congestion . . . . . . . . . . . . . . . . . . 24 113 8.2.4. Heavy congestion . . . . . . . . . . . . . . . . . . 24 114 8.2.5. Varying the congestion level . . . . . . . . . . . . 24 115 8.2.6. Varying available capacity . . . . . . . . . . . . . 24 116 8.3. Parameter sensitivity and stability analysis . . . . . . 25 117 9. Various Traffic Profiles . . . . . . . . . . . . . . . . . . 26 118 9.1. Traffic mix . . . . . . . . . . . . . . . . . . . . . . . 26 119 9.2. Bi-directional traffic . . . . . . . . . . . . . . . . . 27 120 10. Multi-AQM Scenario . . . . . . . . . . . . . . . . . . . . . 27 121 10.1. Motivation . . . . . . . . . . . . . . . . . . . . . . . 27 122 10.2. Details on the evaluation scenario . . . . . . . . . . . 27 123 11. Implementation cost . . . . . . . . . . . . . . . . . . . . . 28 124 11.1. Motivation . . . . . . . . . . . . . . . . . . . . . . . 28 125 11.2. Recommended discussion . . . . . . . . . . . . . . . . . 28 126 12. Operator Control and Auto-tuning . . . . . . . . . . . . . . 28 127 12.1. Motivation . . . . . . . . . . . . . . . . . . . . . . . 29 128 12.2. Recommended discussion . . . . . . . . . . . . . . . . . 29 129 13. Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . 30 130 14. Acknowledgements . . . . . . . . . . . . . . . . . . . . . . 31 131 15. Contributors . . . . . . . . . . . . . . . . . . . . . . . . 31 132 16. IANA Considerations . . . . . . . . . . . . . . . . . . . . . 31 133 17. Security Considerations . . . . . . . . . . . . . . . . . . . 31 134 18. References . . . . . . . . . . . . . . . . . . . . . . . . . 31 135 18.1. Normative References . . . . . . . . . . . . . . . . . . 31 136 18.2. Informative References . . . . . . . . . . . . . . . . . 33 137 Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . 34 139 1. Introduction 141 Active Queue Management (AQM) [RFC7567] addresses the concerns 142 arising from using unnecessarily large and unmanaged buffers to 143 improve network and application performance. Several AQM algorithms 144 have been proposed in the past years, most notably Random Early 145 Detection (RED), BLUE, and Proportional Integral controller (PI), and 146 more recently CoDel [NICH2012] and PIE [PAN2013]. In general, these 147 algorithms actively interact with the Transmission Control Protocol 148 (TCP) and any other transport protocol that deploys a congestion 149 control scheme to manage the amount of data they keep in the network. 150 The available buffer space in the routers and switches should be 151 large enough to accommodate the short-term buffering requirements. 152 AQM schemes aim at reducing buffer occupancy, and therefore the end- 153 to-end delay. Some of these algorithms, notably RED, have also been 154 widely implemented in some network devices. However, the potential 155 benefits of the RED scheme have not been realized since RED is 156 reported to be usually turned off. The main reason of this 157 reluctance to use RED in today's deployments comes from its 158 sensitivity to the operating conditions in the network and the 159 difficulty of tuning its parameters. 161 A buffer is a physical volume of memory in which a queue or set of 162 queues are stored. When speaking of a specific queue in this 163 document, "buffer occupancy" refers to the amount of data (measured 164 in bytes or packets) that are in the queue, and the "maximum buffer 165 size" refers to the maximum buffer occupancy. In real 166 implementations of switches, a global memory is often shared between 167 the available devices, and thus, the maximum buffer size may vary 168 over the time. 170 Bufferbloat [BB2011] is the consequence of deploying large unmanaged 171 buffers on the Internet -- the buffering has often been measured to 172 be ten times or hundred times larger than needed. Large buffer sizes 173 in combination with TCP and/or unresponsive flows increases end-to- 174 end delay. This results in poor performance for latency-sensitive 175 applications such as real-time multimedia (e.g., voice, video, 176 gaming, etc). The degree to which this affects modern networking 177 equipment, especially consumer-grade equipment's, produces problems 178 even with commonly used web services. Active queue management is 179 thus essential to control queuing delay and decrease network latency. 181 The Active Queue Management and Packet Scheduling Working Group (AQM 182 WG) was chartered to address the problems with large unmanaged 183 buffers in the Internet. Specifically, the AQM WG is tasked with 184 standardizing AQM schemes that not only address concerns with such 185 buffers, but also are robust under a wide variety of operating 186 conditions. 188 In order to ascertain whether the WG should undertake standardizing 189 an AQM proposal, the WG requires guidelines for assessing AQM 190 proposals. This document provides the necessary characterization 191 guidelines. [RFC7567] separately describes the AQM algorithm 192 implemented in a router from the scheduling of packets sent by the 193 router. The rest of this memo refers to the AQM as a dropping/ 194 marking policy as a separate feature to any interface scheduling 195 scheme. This document may be complemented with another one on 196 guidelines for assessing combination of packet scheduling and AQM. 197 We note that such a document will inherit all the guidelines from 198 this document plus any additional scenarios relevant for packet 199 scheduling such as flow starvation evaluation or impact of the number 200 of hash buckets. 202 1.1. Reducing the latency and maximizing the goodput 204 The trade-off between reducing the latency and maximizing the goodput 205 is intrinsically linked to each AQM scheme and is key to evaluating 206 its performance. This trade-off MUST be considered in a variety of 207 scenarios to ensure the safety of an AQM deployment. Whenever 208 possible, solutions ought to aim at both maximizing goodput and 209 minimizing latency. 211 1.2. Guidelines for AQM evaluation 213 The guidelines help to quantify performance of AQM schemes in terms 214 of latency reduction, goodput maximization and the trade-off between 215 these two. The guidelines also discuss methods to understand the 216 various aspects associated with safely deploying and operating the 217 AQM scheme. These guidelines discuss methods to understand ease of 218 development, deployment and operational aspects of the AQM scheme 219 verses the potential gain in performance from the introduction of the 220 proposed scheme. 222 This memo details generic characterization scenarios against which 223 any AQM proposal should be evaluated, irrespective of whether or not 224 an AQM is standardized by the IETF. This documents recommends the 225 relevant scenarios and metrics to be considered. The document 226 presents central aspects of an AQM algorithm that must be considered 227 whatever the context, such as burst absorption capacity, RTT fairness 228 or resilience to fluctuating network conditions. 230 These guidelines do not cover every possible aspect of a particular 231 algorithm. In addition, it is worth noting that the proposed 232 criteria are not bound to a particular evaluation toolset. These 233 guidelines do not present context-dependent scenarios (such as 802.11 234 WLANs, data-centers or rural broadband networks). 236 1.3. Requirements Language 238 The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", 239 "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this 240 document are to be interpreted as described in RFC 2119 [RFC2119]. 242 1.4. Glossary 244 o AQM: [RFC7567] separately describes the Active Queue Managment 245 (AQM) algorithm implemented in a router from the scheduling of 246 packets sent by the router. The rest of this memo refers to the 247 AQM as a dropping/marking policy as a separate feature to any 248 interface scheduling scheme. 250 o buffer: a physical volume of memory in which a queue or set of 251 queues are stored. 253 o buffer occupancy: amount of data that are stored in a buffer, 254 measured in bytes or packets. 256 o buffer size: maximum buffer occupancy, that is the maximum amount 257 of data that may be stored in a buffer, measured in bytes or 258 packets. 260 o goodput: goodput is defined as the number of bits per unit of time 261 forwarded to the correct destination minus any bits lost or 262 retransmitted [RFC2647]. 264 2. End-to-end metrics 266 End-to-end delay is the result of propagation delay, serialization 267 delay, service delay in a switch, medium-access delay and queuing 268 delay, summed over the network elements along the path. AQM schemes 269 may reduce the queuing delay by providing signals to the sender on 270 the emergence of congestion, but any impact on the goodput must be 271 carefully considered. This section presents the metrics that could 272 be used to better quantify (1) the reduction of latency, (2) 273 maximization of goodput and (3) the trade-off between these two. 274 This section provides normative requirements for metrics that can be 275 used to assess the performance of an AQM scheme. 277 Some metrics listed in this section are not suited to every type of 278 traffic detailed in the rest of this document. It is therefore not 279 necessary to measure all of the following metrics: the chosen metric 280 may not be relevant to the context of the evaluation scenario (e.g., 281 latency vs. goodput trade-off in application-limited traffic 282 scenarios). Guidance is provided for each metric. 284 2.1. Flow completion time 286 The flow completion time is an important performance metric for the 287 end-user when the flow size is finite. Considering the fact that an 288 AQM scheme may drop/mark packets, the flow completion time is 289 directly linked to the dropping/marking policy of the AQM scheme. 290 This metric helps to better assess the performance of an AQM 291 depending on the flow size. The Flow Completion Time (FCT) is 292 related to the flow size (Fs) and the goodput for the flow (G) as 293 follows: 295 FCT [s] = Fs [Byte] / ( G [Bit/s] / 8 [Bit/Byte] ) 297 If this metric is used to evaluate the performance of web transfers, 298 it is suggested to rather consider the time needed to download all 299 the objects that compose the web page, as this makes more sense in 300 terms of user experience than assessing the time needed to download 301 each object. 303 2.2. Flow start up time 305 The flow start up time is the time between the request has been sent 306 from the client and the server starts to transmit data. The amount 307 of packets dropped by an AQM may seriously affect the waiting period 308 during which the data transfer has not started. This metric would 309 specifically focus on the operations such as DNS lookups, TCP opens 310 of SSL handshakes. 312 2.3. Packet loss 314 Packet loss can occur en-route, this can impact the end-to-end 315 performance measured at receiver. 317 The tester SHOULD evaluate loss experienced at the receiver using one 318 of the two metrics: 320 o the packet loss ratio: this metric is to be frequently measured 321 during the experiment. The long-term loss ratio is of interest 322 for steady-state scenarios only; 324 o the interval between consecutive losses: the time between two 325 losses is to be measured. 327 The packet loss ratio can be assessed by simply evaluating the loss 328 ratio as a function of the number of lost packets and the total 329 number of packets sent. This might not be easily done in laboratory 330 testing, for which these guidelines advice the tester: 332 o to check that for every packet, a corresponding packet was 333 received within a reasonable time, as explained in [RFC2680]. 335 o to keep a count of all packets sent, and a count of the non- 336 duplicate packets received, as explained in the section 10 of 337 [RFC2544]. 339 The interval between consecutive losses, which is also called a gap, 340 is a metric of interest for VoIP traffic and, as a result, has been 341 further specified in [RFC3611]. 343 2.4. Packet loss synchronization 345 One goal of an AQM algorithm is to help to avoid global 346 synchronization of flows sharing a bottleneck buffer on which the AQM 347 operates ([RFC2309],[RFC7567]). The "degree" of packet-loss 348 synchronization between flows SHOULD be assessed, with and without 349 the AQM under consideration. 351 As discussed e.g., in [HASS2008], loss synchronization among flows 352 may be quantified by several slightly different metrics that capture 353 different aspects of the same issue. However, in real-world 354 measurements the choice of metric could be imposed by practical 355 considerations -- e.g., whether fine-grained information on packet 356 losses in the bottleneck available or not. For the purpose of AQM 357 characterization, a good candidate metric is the global 358 synchronization ratio, measuring the proportion of flows losing 359 packets during a loss event. [JAY2006] used this metric in real- 360 world experiments to characterize synchronization along arbitrary 361 Internet paths; the full methodology is described in [JAY2006]. 363 If an AQM scheme is evaluated using real-life network environments, 364 it is worth pointing out that some network events, such as failed 365 link restoration may cause synchronized losses between active flows 366 and thus confuse the meaning of this metric. 368 2.5. Goodput 370 The goodput has been defined in section 3.17 of [RFC2647] as the 371 number of bits per unit of time forwarded to the correct destination 372 interface, minus any bits lost or retransmitted. This definition 373 induces that the test setup needs to be qualified to assure that it 374 is not generating losses on its own. 376 Measuring the end-to-end goodput provides an appreciation of how well 377 an AQM scheme improves transport and application performance. The 378 measured end-to-end goodput is linked to the dropping/marking policy 379 of the AQM scheme -- e.g., the fewer the number of packet drops, the 380 fewer packets need retransmission, minimizing the impact of AQM on 381 transport and application performance. Additionally, an AQM scheme 382 may resort to Explicit Congestion Notification (ECN) marking as an 383 initial means to control delay. Again, marking packets instead of 384 dropping them reduces the number of packet retransmissions and 385 increases goodput. End-to-end goodput values help to evaluate the 386 AQM scheme's effectiveness of an AQM scheme in minimizing packet 387 drops that impact application performance and to estimate how well 388 the AQM scheme works with ECN. 390 The measurement of the goodput allows the tester evaluate to which 391 extent an AQM is able to maintain a high bottleneck utilization. 392 This metric should be also obtained frequently during an experiment 393 as the long-term goodput is relevant for steady-state scenarios only 394 and may not necessarily reflect how the introduction of an AQM 395 actually impacts the link utilization during at a certain period of 396 time. Fluctuations in the values obtained from these measurements 397 may depend on other factors than the introduction of an AQM, such as 398 link layer losses due to external noise or corruption, fluctuating 399 bandwidths (802.11 WLANs), heavy congestion levels or transport 400 layer's rate reduction by congestion control mechanism. 402 2.6. Latency and jitter 404 The latency, or the one-way delay metric, is discussed in [RFC2679]. 405 There is a consensus on an adequate metric for the jitter, that 406 represents the one-way delay variations for packets from the same 407 flow: the Packet Delay Variation (PDV), detailed in [RFC5481], serves 408 well all use cases. 410 The end-to-end latency includes components other than just the 411 queuing delay, such as the signal processing delay, transmission 412 delay and the processing delay. Moreover, the jitter is caused by 413 variations in queuing and processing delay (e.g., scheduling 414 effects). The introduction of an AQM scheme would impact these 415 metrics (end-to-end latency and jitter) and therefore they should be 416 considered in the end-to-end evaluation of performance. 418 2.7. Discussion on the trade-off between latency and goodput 420 The metrics presented in this section may be considered as explained 421 in the rest of this document, in order to discuss and quantify the 422 trade-off between latency and goodput. 424 With regards to the goodput, and in addition to the long-term 425 stationary goodput value, it is RECOMMENDED to take measurements 426 every multiple of the minimum RTT (minRTT) between A and B. It is 427 suggested to take measurements at least every K x minRTT (to smooth 428 out the fluctuations), with K=10. Higher values for K are encouraged 429 whenever it is more appropriate for the presentation of the results. 430 The value for K may depend on the network's path characteristics. 431 The measurement period MUST be disclosed for each experiment and when 432 results/values are compared across different AQM schemes, the 433 comparisons SHOULD use exactly the same measurement periods. With 434 regards to latency, it is RECOMMENDED to take the samples on per- 435 packet basis whenever possible depending on the features provided by 436 hardware/software and the impact of sampling itself on the hardware 437 performance. It is generally RECOMMENDED to provide at least 10 438 samples per RTT. 440 From each of these sets of measurements, the cumulative density 441 function (CDF) of the considered metrics SHOULD be computed. If the 442 considered scenario introduces dynamically varying parameters, 443 temporal evolution of the metrics could also be generated. For each 444 scenario, the following graph may be generated: the x-axis shows 445 queuing delay (that is the average per-packet delay in excess of 446 minimum RTT), the y-axis the goodput. Ellipses are computed such as 447 detailed in [WINS2014]: "We take each individual [...] run [...] as 448 one point, and then compute the 1-epsilon elliptic contour of the 449 maximum-likelihood 2D Gaussian distribution that explains the points. 450 [...] we plot the median per-sender throughput and queueing delay as 451 a circle. [...] The orientation of an ellipse represents the 452 covariance between the throughput and delay measured for the 453 protocol." This graph provides part of a better understanding of (1) 454 the delay/goodput trade-off for a given congestion control mechanism 455 Section 5, and (2) how the goodput and average queue delay vary as a 456 function of the traffic load Section 8.2. 458 3. Generic setup for evaluations 460 This section presents the topology that can be used for each of the 461 following scenarios, the corresponding notations and discusses 462 various assumptions that have been made in the document. 464 3.1. Topology and notations 465 +---------+ +-----------+ 466 |senders A| |receivers B| 467 +---------+ +-----------+ 469 +--------------+ +--------------+ 470 |traffic class1| |traffic class1| 471 |--------------| |--------------| 472 | SEN.Flow1.1 +---------+ +-----------+ REC.Flow1.1 | 473 | + | | | | + | 474 | | | | | | | | 475 | + | | | | + | 476 | SEN.Flow1.X +-----+ | | +--------+ REC.Flow1.X | 477 +--------------+ | | | | +--------------+ 478 + +-+---+---+ +--+--+---+ + 479 | |Router L | |Router R | | 480 | |---------| |---------| | 481 | | AQM | | | | 482 | | BuffSize| | BuffSize| | 483 | | (Bsize) +-----+ (Bsize) | | 484 | +-----+--++ ++-+------+ | 485 + | | | | + 486 +--------------+ | | | | +--------------+ 487 |traffic classN| | | | | |traffic classN| 488 |--------------| | | | | |--------------| 489 | SEN.FlowN.1 +---------+ | | +-----------+ REC.FlowN.1 | 490 | + | | | | + | 491 | | | | | | | | 492 | + | | | | + | 493 | SEN.FlowN.Y +------------+ +-------------+ REC.FlowN.Y | 494 +--------------+ +--------------+ 496 Figure 1: Topology and notations 498 Figure 1 is a generic topology where: 500 o sender with different traffic characteristics (i.e., traffic 501 profiles) can be introduced; 503 o the timing of each flow could be different (i.e., when does each 504 flow start and stop); 506 o each traffic profile can comprise various number of flows; 508 o each link is characterized by a couple (one-way delay, capacity); 510 o flows are generated at A and sent to B, sharing a bottleneck (the 511 link between routers L and R); 513 o the tester SHOULD consider both scenarios of asymmetric and 514 symmetric bottleneck links in terms of bandwidth. In case of 515 asymmetric link, the capacity from senders to receivers is higher 516 than the one from receivers to senders; the symmetric link 517 scenario provides a basic understanding of the operation of the 518 AQM mechanism whereas the asymmetric link scenario evaluates an 519 AQM mechanism in a more realistic setup; 521 o in asymmetric link scenarios, the tester SHOULD study the bi- 522 directional traffic between A and B (downlink and uplink) with the 523 AQM mechanism deployed on one direction only. The tester MAY 524 additionally consider a scenario with AQM mechanism being deployed 525 on both directions. In each scenario, the tester SHOULD 526 investigate the impact of drop policy of the AQM on TCP ACK 527 packets and its impact on the performance. 529 Although this topology may not perfectly reflect actual topologies, 530 the simple topology is commonly used in the world of simulations and 531 small testbeds. It can be considered as adequate to evaluate AQM 532 proposals, similarly to the topology proposed in 533 [I-D.irtf-iccrg-tcpeval]. Testers ought to pay attention to the 534 topology that has been used to evaluate an AQM scheme when comparing 535 this scheme with a newly proposed AQM scheme. 537 3.2. Buffer size 539 The size of the buffers should be carefully chosen, and is to be set 540 to the bandwidth-delay product; the bandwidth being the bottleneck 541 capacity and the delay the largest RTT in the considered network. 542 The size of the buffer can impact the AQM performance and is a 543 dimensioning parameter that will be considered when comparing AQM 544 proposals. 546 If a specific buffer size is required, the tester MUST justify and 547 detail the way the maximum queue size is set. Indeed, the maximum 548 size of the buffer may affect the AQM's performance and its choice 549 SHOULD be elaborated for a fair comparison between AQM proposals. 550 While comparing AQM schemes the buffer size SHOULD remain the same 551 across the tests. 553 3.3. Congestion controls 555 This document considers running three different congestion control 556 algorithms between A and B 558 o Standard TCP congestion control: the base-line congestion control 559 is TCP NewReno with SACK, as explained in [RFC5681]. 561 o Aggressive congestion controls: a base-line congestion control for 562 this category is TCP Cubic [I-D.ietf-tcpm-cubic]. 564 o Less-than Best Effort (LBE) congestion controls: an LBE congestion 565 control 'results in smaller bandwidth and/or delay impact on 566 standard TCP than standard TCP itself, when sharing a bottleneck 567 with it.' [RFC6297] 569 Other transport congestion controls can OPTIONALLY be evaluated in 570 addition. Recent transport layer protocols are not mentioned in the 571 following sections, for the sake of simplicity. 573 4. Methodology, Metrics, AQM Comparisons, Packet Sizes, Scheduling and 574 ECN 576 4.1. Methodology 578 One key objective behind formulating the guidelines is to help 579 ascertain whether a specific AQM is not only better than drop-tail 580 (with BDP-sized buffer) but also safe to deploy. Testers therefore 581 need to provide a reference document for their proposal discussing 582 performance and deployment compared to those of drop-tail. 584 A description of each test setup SHOULD be detailed to allow this 585 test to be compared with other tests. This also allows others to 586 replicate the tests if needed. This test setup SHOULD detail 587 software and hardware versions. The tester could make its data 588 available. 590 The proposals SHOULD be evaluated on real-life systems, or they MAY 591 be evaluated with event-driven simulations (such as ns-2, ns-3, 592 OMNET, etc). The proposed scenarios are not bound to a particular 593 evaluation toolset. 595 The tester is encouraged to make the detailed test setup and the 596 results publicly available. 598 4.2. Comments on metrics measurement 600 The document presents the end-to-end metrics that ought to be used to 601 evaluate the trade-off between latency and goodput in Section 2. In 602 addition to the end-to-end metrics, the queue-level metrics (normally 603 collected at the device operating the AQM) provide a better 604 understanding of the AQM behavior under study and the impact of its 605 internal parameters. Whenever it is possible (e.g., depending on the 606 features provided by the hardware/software), these guidelines advice 607 to consider queue-level metrics, such as link utilization, queuing 608 delay, queue size or packet drop/mark statistics in addition to the 609 AQM-specific parameters. However, the evaluation MUST be primarily 610 based on externally observed end-to-end metrics. 612 These guidelines do not aim to detail on the way these metrics can be 613 measured, since the way these metrics are measured is expected to 614 depend on the evaluation toolset. 616 4.3. Comparing AQM schemes 618 This document recognizes that these guidelines may be used for 619 comparing AQM schemes. 621 AQM schemes need to be compared against both performance and 622 deployment categories. In addition, this section details how best to 623 achieve a fair comparison of AQM schemes by avoiding certain 624 pitfalls. 626 4.3.1. Performance comparison 628 AQM schemes MUST be compared against all the generic scenarios 629 presented in this memo. AQM schemes MAY be compared for specific 630 network environments such as data centers, home networks, etc. If an 631 AQM scheme has parameter(s) that were externally tuned for 632 optimization or other purposes, these values MUST be disclosed. 634 AQM schemes belong to different varieties such as queue-length based 635 schemes (ex. RED) or queueing-delay based scheme (ex. CoDel, PIE). 636 AQM schemes expose different control knobs associated with different 637 semantics. For example, while both PIE and CoDel are queueing-delay 638 based schemes and each expose a knob to control the queueing delay -- 639 PIE's "queueing delay reference" vs. CoDel's "queueing delay target", 640 the two tuning parameters of the two schemes have different 641 semantics, resulting in different control points. Such differences 642 in AQM schemes can be easily overlooked while making comparisons. 644 This document RECOMMENDS the following procedures for a fair 645 performance comparison between the AQM schemes: 647 1. comparable control parameters and comparable input values: 648 carefully identify the set of parameters that control similar 649 behavior between the two AQM schemes and ensure these parameters 650 have comparable input values. For example, to compare how well a 651 queue-length based AQM scheme controls queueing delay vs. a 652 queueing-delay based AQM scheme, a tester can identify the 653 parameters of the schemes that control queue delay and ensure 654 that their input values are comparable. Similarly, to compare 655 how well two AQM schemes accommodate packet bursts, the tester 656 can identify burst-related control parameters and ensure they are 657 configured with similar values. Additionally, it would be 658 preferable if an AQM proposal listed such parameters and 659 discussed how each relates to network characteristics such as 660 capacity, average RTT etc. 662 2. compare over a range of input configurations: there could be 663 situations when the set of control parameters that affect a 664 specific behavior have different semantics between the two AQM 665 schemes. As mentioned above, PIE has tuning parameters to 666 control queue delay that has a different semantics from those 667 used in CoDel. In such situations, these schemes need to be 668 compared over a range of input configurations. For example, 669 compare PIE vs. CoDel over the range of target delay input 670 configurations. 672 4.3.2. Deployment comparison 674 AQM schemes MUST be compared against deployment criteria such as the 675 parameter sensitivity (Section 8.3), auto-tuning (Section 12) or 676 implementation cost (Section 11). 678 4.4. Packet sizes and congestion notification 680 An AQM scheme may be considering packet sizes while generating 681 congestion signals. [RFC7141] discusses the motivations behind this. 682 For example, control packets such as DNS requests/responses, TCP 683 SYNs/ACKs are small, but their loss can severely impact the 684 application performance. An AQM scheme may therefore be biased 685 towards small packets by dropping them with smaller probability 686 compared to larger packets. However, such an AQM scheme is unfair to 687 data senders generating larger packets. Data senders, malicious or 688 otherwise, are motivated to take advantage of such AQM scheme by 689 transmitting smaller packets, and could result in unsafe deployments 690 and unhealthy transport and/or application designs. 692 An AQM scheme SHOULD adhere to the recommendations outlined in 693 [RFC7141], and SHOULD NOT provide undue advantage to flows with 694 smaller packets [RFC7567]. 696 4.5. Interaction with ECN 698 Deployed AQM algorithms SHOULD implement Explicit Congestion 699 Notification (ECN) as well as loss to signal congestion to endpoints 700 [RFC7567]. ECN [RFC3168] is an alternative that allows AQM schemes 701 to signal receivers about network congestion that does not use packet 702 drop. The benefits of providing ECN support for an AQM scheme are 703 described in [WELZ2015]. Section 3 of [WELZ2015] describes expected 704 operation of routers enabling ECN. AQM schemes SHOULD NOT drop or 705 remark packets solely because the ECT(0) or ECT(1) codepoints are 706 used, and when ECN-capable SHOULD set a CE-mark on ECN-capable 707 packets in the presence of incipient congestion. 709 If the tested AQM scheme can support ECN [RFC7567], the testers MUST 710 discuss and describe the support of ECN. Since these guidelines can 711 be used to evaluate the performance of the tested AQM with and 712 without ECN markings, they could also be used to quantify the 713 interest of enabling ECN. 715 4.6. Interaction with Scheduling 717 A network device may use per-flow or per-class queuing with a 718 scheduling algorithm to either prioritize certain applications or 719 classes of traffic, limit the rate of transmission, or to provide 720 isolation between different traffic flows within a common class 721 [RFC7567]. 723 The scheduling and the AQM conjointly impact on the end-to-end 724 performance. Therefore, the AQM proposal MUST discuss the 725 feasibility to add scheduling combined with the AQM algorithm. This 726 discussion as an instance, MAY explain whether the dropping policy is 727 applied when packets are being enqueued or dequeued. 729 These guidelines do not propose guidelines to assess the performance 730 of scheduling algorithms. Indeed, as opposed to characterizing AQM 731 schemes that is related to their capacity to control the queuing 732 delay in a queue, characterizing scheduling schemes is related to the 733 scheduling itself and its interaction with the AQM scheme. As one 734 example, the scheduler may create sub-queues and the AQM scheme may 735 be applied on each of the sub-queues, and/or the AQM could be applied 736 on the whole queue. Also, schedulers might, such as FQ-CoDel 737 [HOEI2015] or FavorQueue [ANEL2014], introduce flow prioritization. 738 In these cases, specific scenarios should be proposed to ascertain 739 that these scheduler schemes not only helps in tackling the 740 bufferbloat, but also are robust under a wide variety of operating 741 conditions. This is out of the scope of this document that focus on 742 dropping and/or marking AQM schemes. 744 5. Transport Protocols 746 Network and end-devices need to be configured with a reasonable 747 amount of buffer space to absorb transient bursts. In some 748 situations, network providers tend to configure devices with large 749 buffers to avoid packet drops triggered by a full buffer and to 750 maximize the link utilization for standard loss-based TCP traffic. 752 AQM algorithms are often evaluated by considering Transmission 753 Control Protocol (TCP) [RFC0793] with a limited number of 754 applications. TCP is a widely deployed transport. It fills up 755 available buffers until a sender transfering a bulk flow with TCP 756 receives a signal (packet drop) that reduces the sending rate. The 757 larger the buffer, the higher the buffer occupancy, and therefore the 758 queuing delay. An efficient AQM scheme sends out early congestion 759 signals to TCP to bring the queuing delay under control. 761 Not all endpoints (or applications) using TCP use the same flavor of 762 TCP. Variety of senders generate different classes of traffic which 763 may not react to congestion signals (aka non-responsive flows 764 [RFC7567]) or may not reduce their sending rate as expected (aka 765 Transport Flows that are less responsive than TCP[RFC7567], also 766 called "aggressive flows"). In these cases, AQM schemes seek to 767 control the queuing delay. 769 This section provides guidelines to assess the performance of an AQM 770 proposal for various traffic profiles -- different types of senders 771 (with different TCP congestion control variants, unresponsive, 772 aggressive). 774 5.1. TCP-friendly sender 776 5.1.1. TCP-friendly sender with the same initial congestion window 778 This scenario helps to evaluate how an AQM scheme reacts to a TCP- 779 friendly transport sender. A single long-lived, non application- 780 limited, TCP NewReno flow, with an Initial congestion Window (IW) set 781 to 3 packets, transfers data between sender A and receiver B. Other 782 TCP friendly congestion control schemes such as TCP-friendly rate 783 control [RFC5348] etc MAY also be considered. 785 For each TCP-friendly transport considered, the graph described in 786 Section 2.7 could be generated. 788 5.1.2. TCP-friendly sender with different initial congestion windows 790 This scenario can be used to evaluate how an AQM scheme adapts to a 791 traffic mix consisting of TCP flows with different values of the IW. 793 For this scenario, two types of flows MUST be generated between 794 sender A and receiver B: 796 o A single long-lived non application-limited TCP NewReno flow; 797 o A single application-limited TCP NewReno flow, with an IW set to 3 798 or 10 packets. The size of the data transferred must be strictly 799 higher than 10 packets and should be lower than 100 packets. 801 The transmission of the non application-limited flow must start 802 before the transmission of the application-limited flow and only 803 after the steady state has been reached by non application-limited 804 flow. 806 For each of these scenarios, the graph described in Section 2.7 could 807 be generated for each class of traffic (application-limited and non 808 application-limited). The completion time of the application-limited 809 TCP flow could be measured. 811 5.2. Aggressive transport sender 813 This scenario helps testers to evaluate how an AQM scheme reacts to a 814 transport sender that is more aggressive than a single TCP-friendly 815 sender. We define 'aggressiveness' as a higher increase factor than 816 standard upon a successful transmission and/or a lower than standard 817 decrease factor upon a unsuccessful transmission (e.g., in case of 818 congestion controls with Additive-Increase Multiplicative-Decrease 819 (AIMD) principle, a larger AI and/or MD factors). A single long- 820 lived, non application-limited, TCP Cubic flow transfers data between 821 sender A and receiver B. Other aggressive congestion control schemes 822 MAY also be considered. 824 For each flavor of aggressive transports, the graph described in 825 Section 2.7 could be generated. 827 5.3. Unresponsive transport sender 829 This scenario helps testers to evaluate how an AQM scheme reacts to a 830 transport sender that is less responsive than TCP. Note that faulty 831 transport implementations on an end host and/or faulty network 832 elements en-route that "hide" congestion signals in packet headers 833 [RFC7567] may also lead to a similar situation, such that the AQM 834 scheme needs to adapt to unresponsive traffic. To this end, these 835 guidelines propose the two following scenarios. 837 The first scenario can be used to evaluate queue build up. It 838 considers unresponsive flow(s) whose sending rate is greater than the 839 bottleneck link capacity between routers L and R. This scenario 840 consists of a long-lived non application limited UDP flow transmits 841 data between sender A and receiver B. Graphs described in 842 Section 2.7 could be generated. 844 The second scenario can be used to evaluate if the AQM scheme is able 845 to keep the responsive fraction under control. This scenario 846 considers a mixture of TCP-friendly and unresponsive traffics. It 847 consists of a long-lived UDP flow from unresponsive application and a 848 single long-lived, non application-limited (unlimited data available 849 to the transport sender from application layer), TCP New Reno flow 850 that transmit data between sender A and receiver B. As opposed to 851 the first scenario, the rate of the UDP traffic should not be greater 852 than the bottleneck capacity, and should be higher than half of the 853 bottleneck capacity. For each type of traffic, the graph described 854 in Section 2.7 could be generated. 856 5.4. Less-than Best Effort transport sender 858 This scenario helps to evaluate how an AQM scheme reacts to LBE 859 congestion controls that 'results in smaller bandwidth and/or delay 860 impact on standard TCP than standard TCP itself, when sharing a 861 bottleneck with it.' [RFC6297]. The potential fateful interaction 862 when AQM and LBE techniques are combined has been shown in 863 [GONG2014]; this scenario helps to evaluate whether the coexistence 864 of the proposed AQM and LBE techniques may be possible. 866 A single long-lived non application-limited TCP NewReno flow 867 transfers data between sender A and receiver B. Other TCP-friendly 868 congestion control schemes MAY also be considered. Single long-lived 869 non application-limited LEDBAT [RFC6817] flows transfer data between 870 sender A and receiver B. We recommend to set the target delay and 871 gain values of LEDBAT respectively to 5 ms and 10 [TRAN2014]. Other 872 LBE congestion control schemes, any of those listed in [RFC6297], MAY 873 also be considered. 875 For each of the TCP-friendly and LBE transports, the graph described 876 in Section 2.7 could be generated. 878 6. Round Trip Time Fairness 880 6.1. Motivation 882 An AQM scheme's congestion signals (via drops or ECN marks) must 883 reach the transport sender so that a responsive sender can initiate 884 its congestion control mechanism and adjust the sending rate. This 885 procedure is thus dependent on the end-to-end path RTT. When the RTT 886 varies, the onset of congestion control is impacted, and in turn 887 impacts the ability of an AQM scheme to control the queue. It is 888 therefore important to assess the AQM schemes for a set of RTTs 889 between A and B (e.g., from 5 ms to 200 ms). 891 The asymmetry in terms of difference in intrinsic RTT between various 892 paths sharing the same bottleneck SHOULD be considered so that the 893 fairness between the flows can be discussed since in this scenario, a 894 flow traversing on shorter RTT path may react faster to congestion 895 and recover faster from it compared to another flow on a longer RTT 896 path. The introduction of AQM schemes may potentially improve this 897 type of fairness. 899 Introducing an AQM scheme may cause the unfairness between the flows, 900 even if the RTTs are identical. This potential unfairness SHOULD be 901 investigated as well. 903 6.2. Recommended tests 905 The RECOMMENDED topology is detailed in Figure 1. 907 To evaluate the RTT fairness, for each run, two flows divided into 908 two categories. Category I which RTT between sender A and receiver B 909 SHOULD be 100ms. Category II which RTT between sender A and receiver 910 B should be in [5ms;560ms]. The maximum value for the RTT represents 911 the RTT of a satellite link that, according to section 2 of [RFC2488] 912 should be at least 558ms. 914 A set of evaluated flows MUST use the same congestion control 915 algorithm: all the generated flows could be single long-lived non 916 application-limited TCP NewReno flows. 918 6.3. Metrics to evaluate the RTT fairness 920 The outputs that MUST be measured are: (1) the cumulative average 921 goodput of the flow from Category I, goodput_Cat_I (Section 2.5); (2) 922 the cumulative average goodput of the flow from Category II, 923 goodput_Cat_II (Section 2.5); (3) the ratio goodput_Cat_II/ 924 goodput_Cat_I; (4) the average packet drop rate for each category 925 (Section 2.3). 927 7. Burst Absorption 929 "AQM mechanisms need to control the overall queue sizes, to ensure 930 that arriving bursts can be accommodated without dropping packets" 931 [RFC7567]. 933 7.1. Motivation 935 An AQM scheme can face bursts of packet arrivals due to various 936 reasons. Dropping one or more packets from a burst can result in 937 performance penalties for the corresponding flows, since dropped 938 packets have to be retransmitted. Performance penalties can result 939 in failing to meet SLAs and be a disincentive to AQM adoption. 941 The ability to accommodate bursts translates to larger queue length 942 and hence more queuing delay. On the one hand, it is important that 943 an AQM scheme quickly brings bursty traffic under control. On the 944 other hand, a peak in the packet drop rates to bring a packet burst 945 quickly under control could result in multiple drops per flow and 946 severely impact transport and application performance. Therefore, an 947 AQM scheme ought to bring bursts under control by balancing both 948 aspects -- (1) queuing delay spikes are minimized and (2) performance 949 penalties for ongoing flows in terms of packet drops are minimized. 951 An AQM scheme that maintains short queues allows some remaining space 952 in the buffer for bursts of arriving packets. The tolerance to 953 bursts of packets depends upon the number of packets in the queue, 954 which is directly linked to the AQM algorithm. Moreover, an AQM 955 scheme may implement a feature controlling the maximum size of 956 accepted bursts, that can depend on the buffer occupancy or the 957 currently estimated queuing delay. The impact of the buffer size on 958 the burst allowance may be evaluated. 960 7.2. Recommended tests 962 For this scenario, tester MUST evaluate how the AQM performs with the 963 following traffic generated from sender A to receiver B: 965 o Web traffic with IW10; 967 o Bursty video frames; 969 o Constant Bit Rate (CBR) UDP traffic. 971 o A single non application-limited bulk TCP flow as background 972 traffic. 974 Figure 2 presents the various cases for the traffic that MUST be 975 generated between sender A and receiver B. 977 +-------------------------------------------------+ 978 |Case| Traffic Type | 979 | +-----+------------+----+--------------------+ 980 | |Video|Web (IW 10)| CBR| Bulk TCP Traffic | 981 +----|-----|------------|----|--------------------| 982 |I | 0 | 1 | 1 | 0 | 983 +----|-----|------------|----|--------------------| 984 |II | 0 | 1 | 1 | 1 | 985 |----|-----|------------|----|--------------------| 986 |III | 1 | 1 | 1 | 0 | 987 +----|-----|------------|----|--------------------| 988 |IV | 1 | 1 | 1 | 1 | 989 +----+-----+------------+----+--------------------+ 991 Figure 2: Bursty traffic scenarios 993 A new web page download could start after the previous web page 994 download is finished. Each web page could be composed by at least 50 995 objects and the size of each object should be at least 1kB. 6 TCP 996 parallel connections SHOULD be generated to download the objects, 997 each parallel connections having an initial congestion window set to 998 10 packets. 1000 For each of these scenarios, the graph described in Section 2.7 could 1001 be generated for each application. Metrics such as end-to-end 1002 latency, jitter, flow completion time MAY be generated. For the 1003 cases of frame generation of bursty video traffic as well as the 1004 choice of web traffic pattern, these details and their presentation 1005 are left to the testers. 1007 8. Stability 1009 8.1. Motivation 1011 The safety of an AQM scheme is directly related to its stability 1012 under varying operating conditions such as varying traffic profiles 1013 and fluctuating network conditions. Since operating conditions can 1014 vary often the AQM needs to remain stable under these conditions 1015 without the need for additional external tuning. 1017 Network devices can experience varying operating conditions depending 1018 on factors such as time of the day, deployment scenario, etc. For 1019 example: 1021 o Traffic and congestion levels are higher during peak hours than 1022 off-peak hours. 1024 o In the presence of a scheduler, the draining rate of a queue can 1025 vary depending on the occupancy of other queues: a low load on a 1026 high priority queue implies a higher draining rate for the lower 1027 priority queues. 1029 o The capacity available can vary over time (e.g., a lossy channel, 1030 a link supporting traffic in a higher diffserv class). 1032 Whether the target context is a not stable environment, the ability 1033 of an AQM scheme to maintain its control over the queuing delay and 1034 buffer occupancy can be challenged. This document proposes 1035 guidelines to assess the behavior of AQM schemes under varying 1036 congestion levels and varying draining rates. 1038 8.2. Recommended tests 1040 Note that the traffic profiles explained below comprises non 1041 application-limited TCP flows. For each of the below scenarios, the 1042 graphs described in Section 2.7 SHOULD be generated, and the goodput 1043 of the various flows should be cumulated. For Section 8.2.5 and 1044 Section 8.2.6 they SHOULD incorporate the results in per-phase basis 1045 as well. 1047 Wherever the notion of time has explicitly mentioned in this 1048 subsection, time 0 starts from the moment all TCP flows have already 1049 reached their congestion avoidance phase. 1051 8.2.1. Definition of the congestion Level 1053 In these guidelines, the congestion levels are represented by the 1054 projected packet drop rate, had a drop-tail queue was chosen instead 1055 of an AQM scheme. When the bottleneck is shared among non 1056 application-limited TCP flows. l_r, the loss rate projection can be 1057 expressed as a function of N, the number of bulk TCP flows, and S, 1058 the sum of the bandwidth-delay product and the maximum buffer size, 1059 both expressed in packets, based on Eq. 3 of [MORR2000]: 1061 l_r = 0.76 * N^2 / S^2 1063 N = S * sqrt(1/0.76) * sqrt (l_r) 1065 These guidelines use the loss rate to define the different congestion 1066 levels, but they do not stipulate that in other circumstances, 1067 measuring the congestion level gives you an accurate estimation of 1068 the loss rate or vice-versa. 1070 8.2.2. Mild congestion 1072 This scenario can be used to evaluate how an AQM scheme reacts to a 1073 light load of incoming traffic resulting in mild congestion -- packet 1074 drop rates around 0.1%. The number of bulk flows required to achieve 1075 this congestion level, N_mild, is then: 1077 N_mild = round(0.036*S) 1079 8.2.3. Medium congestion 1081 This scenario can be used to evaluate how an AQM scheme reacts to 1082 incoming traffic resulting in medium congestion -- packet drop rates 1083 around 0.5%. The number of bulk flows required to achieve this 1084 congestion level, N_med, is then: 1086 N_med = round (0.081*S) 1088 8.2.4. Heavy congestion 1090 This scenario can be used to evaluate how an AQM scheme reacts to 1091 incoming traffic resulting in heavy congestion -- packet drop rates 1092 around 1%. The number of bulk flows required to achieve this 1093 congestion level, N_heavy, is then: 1095 N_heavy = round (0.114*S) 1097 8.2.5. Varying the congestion level 1099 This scenario can be used to evaluate how an AQM scheme reacts to 1100 incoming traffic resulting in various levels of congestion during the 1101 experiment. In this scenario, the congestion level varies within a 1102 large time-scale. The following phases may be considered: phase I - 1103 mild congestion during 0-20s; phase II - medium congestion during 1104 20-40s; phase III - heavy congestion during 40-60s; phase I again, 1105 and so on. 1107 8.2.6. Varying available capacity 1109 This scenario can be used to help characterize how the AQM behaves 1110 and adapts to bandwidth changes. The experiments are not meant to 1111 reflect the exact conditions of Wi-Fi environments since it is hard 1112 to design repetitive experiments or accurate simulations for such 1113 scenarios. 1115 To emulate varying draining rates, the bottleneck capacity between 1116 nodes 'Router L' and 'Router R' varies over the course of the 1117 experiment as follows: 1119 o Experiment 1: the capacity varies between two values within a 1120 large time-scale. As an example, the following phases may be 1121 considered: phase I - 100Mbps during 0-20s; phase II - 10Mbps 1122 during 20-40s; phase I again, and so on. 1124 o Experiment 2: the capacity varies between two values within a 1125 short time-scale. As an example, the following phases may be 1126 considered: phase I - 100Mbps during 0-100ms; phase II - 10Mbps 1127 during 100-200ms; phase I again, and so on. 1129 The tester MAY choose a phase time-interval value different than what 1130 is stated above, if the network's path conditions (such as bandwidth- 1131 delay product) necessitate. In this case the choice of such time- 1132 interval value SHOULD be stated and elaborated. 1134 The tester MAY additionally evaluate the two mentioned scenarios 1135 (short-term and long-term capacity variations), during and/or 1136 including TCP slow-start phase. 1138 More realistic fluctuating capacity patterns MAY be considered. The 1139 tester MAY choose to incorporate realistic scenarios with regards to 1140 common fluctuation of bandwidth in state-of-the-art technologies. 1142 The scenario consists of TCP NewReno flows between sender A and 1143 receiver B. To better assess the impact of draining rates on the AQM 1144 behavior, the tester MUST compare its performance with those of drop- 1145 tail and SHOULD provide a reference document for their proposal 1146 discussing performance and deployment compared to those of drop-tail. 1147 Burst traffic, such as presented in Section 7.2, could also be 1148 considered to assess the impact of varying available capacity on the 1149 burst absorption of the AQM. 1151 8.3. Parameter sensitivity and stability analysis 1153 The control law used by an AQM is the primary means by which the 1154 queuing delay is controlled. Hence understanding the control law is 1155 critical to understanding the behavior of the AQM scheme. The 1156 control law could include several input parameters whose values 1157 affect the AQM scheme's output behavior and its stability. 1158 Additionally, AQM schemes may auto-tune parameter values in order to 1159 maintain stability under different network conditions (such as 1160 different congestion levels, draining rates or network environments). 1161 The stability of these auto-tuning techniques is also important to 1162 understand. 1164 Transports operating under the control of AQM experience the effect 1165 of multiple control loops that react over different timescales. It 1166 is therefore important that proposed AQM schemes are seen to be 1167 stable when they are deployed at multiple points of potential 1168 congestion along an Internet path. The pattern of congestion signals 1169 (loss or ECN-marking) arising from AQM methods also need to not 1170 adversely interact with the dynamics of the transport protocols that 1171 they control. 1173 AQM proposals SHOULD provide background material showing control 1174 theoretic analysis of the AQM control law and the input parameter 1175 space within which the control law operates as expected; or could use 1176 another way to discuss the stability of the control law. For 1177 parameters that are auto-tuned, the material SHOULD include stability 1178 analysis of the auto-tuning mechanism(s) as well. Such analysis 1179 helps to understand an AQM control law better and the network 1180 conditions/deployments under which the AQM is stable. 1182 9. Various Traffic Profiles 1184 This section provides guidelines to assess the performance of an AQM 1185 proposal for various traffic profiles such as traffic with different 1186 applications or bi-directional traffic. 1188 9.1. Traffic mix 1190 This scenario can be used to evaluate how an AQM scheme reacts to a 1191 traffic mix consisting of different applications such as: 1193 o Bulk TCP transfer 1195 o Web traffic 1197 o VoIP 1199 o Constant Bit Rate (CBR) UDP traffic 1201 o Adaptive video streaming 1203 Various traffic mixes can be considered. These guidelines RECOMMEND 1204 to examine at least the following example: 1 bi-directional VoIP; 6 1205 Web pages download (such as detailed in Section 7.2); 1 CBR; 1 1206 Adaptive Video; 5 bulk TCP. Any other combinations could be 1207 considered and should be carefully documented. 1209 For each scenario, the graph described in Section 2.7 could be 1210 generated for each class of traffic. Metrics such as end-to-end 1211 latency, jitter and flow completion time MAY be reported. 1213 9.2. Bi-directional traffic 1215 Control packets such as DNS requests/responses, TCP SYNs/ACKs are 1216 small, but their loss can severely impact the application 1217 performance. The scenario proposed in this section will help in 1218 assessing whether the introduction of an AQM scheme increases the 1219 loss probability of these important packets. 1221 For this scenario, traffic MUST be generated in both downlink and 1222 uplink, such as defined in Section 3.1. These guidelines RECOMMEND 1223 to consider a mild congestion level and the traffic presented in 1224 Section 8.2.2 in both directions. In this case, the metrics reported 1225 MUST be the same as in Section 8.2 for each direction. 1227 The traffic mix presented in Section 9.1 MAY also be generated in 1228 both directions. 1230 10. Multi-AQM Scenario 1232 10.1. Motivation 1234 Transports operating under the control of AQM experience the effect 1235 of multiple control loops that react over different timescales. It 1236 is therefore important that proposed AQM schemes are seen to be 1237 stable when they are deployed at multiple points of potential 1238 congestion along an Internet path. The pattern of congestion signals 1239 (loss or ECN-marking) arising from AQM methods also need to not 1240 adversely interact with the dynamics of the transport protocols that 1241 they control. 1243 10.2. Details on the evaluation scenario 1245 +---------+ +-----------+ 1246 |senders A|---+ +---|receivers A| 1247 +---------+ | | +-----------+ 1248 +-----+---+ +---------+ +--+-----+ 1249 |Router L |--|Router M |--|Router R| 1250 |AQM | |AQM | |No AQM | 1251 +---------+ +--+------+ +--+-----+ 1252 +---------+ | | +-----------+ 1253 |senders B|-------------+ +---|receivers B| 1254 +---------+ +-----------+ 1256 Figure 3: Topology for the Multi-AQM scenario 1258 This scenario can be used to evaluate how having AQM schemes in 1259 sequence impact the induced latency reduction, the induced goodput 1260 maximization and the trade-off between these two. The topology 1261 presented in Figure 3 could be used. AQM schemes introduced in 1262 Router L and Router M should be the same; any other configurations 1263 could be considered. For this scenario, it is recommended to 1264 consider a mild congestion level, the number of flows specified in 1265 Section 8.2.2 being equally shared among senders A and B. Any other 1266 relevant combination of congestion levels could be considered. We 1267 recommend to measure the metrics presented in Section 8.2. 1269 11. Implementation cost 1271 11.1. Motivation 1273 Successful deployment of AQM is directly related to its cost of 1274 implementation. Network devices can need hardware or software 1275 implementations of the AQM mechanism. Depending on a device's 1276 capabilities and limitations, the device may or may not be able to 1277 implement some or all parts of their AQM logic. 1279 AQM proposals SHOULD provide pseudo-code for the complete AQM scheme, 1280 highlighting generic implementation-specific aspects of the scheme 1281 such as "drop-tail" vs. "drop-head", inputs (e.g., current queuing 1282 delay, queue length), computations involved, need for timers, etc. 1283 This helps to identify costs associated with implementing the AQM 1284 scheme on a particular hardware or software device. This also 1285 facilitates discsusions around which kind of devices can easily 1286 support the AQM and which cannot. 1288 11.2. Recommended discussion 1290 AQM proposals SHOULD highlight parts of their AQM logic that are 1291 device dependent and discuss if and how AQM behavior could be 1292 impacted by the device. For example, a queueing-delay based AQM 1293 scheme requires current queuing delay as input from the device. If 1294 the device already maintains this value, then it can be trivial to 1295 implement the their AQM logic on the device. If the device provides 1296 indirect means to estimate the queuing delay (for example: 1297 timestamps, dequeuing rate), then the AQM behavior is sensitive to 1298 the precision of the queuing delay estimations are for that device. 1299 Highlighting the sensitivity of an AQM scheme to queuing delay 1300 estimations helps implementers to identify appropriate means of 1301 implementing the mechanism on a device. 1303 12. Operator Control and Auto-tuning 1304 12.1. Motivation 1306 One of the biggest hurdles of RED deployment was/is its parameter 1307 sensitivity to operating conditions -- how difficult it is to tune 1308 RED parameters for a deployment to achieve acceptable benefit from 1309 using RED. Fluctuating congestion levels and network conditions add 1310 to the complexity. Incorrect parameter values lead to poor 1311 performance. 1313 Any AQM scheme is likely to have parameters whose values affect the 1314 control law and behaviour of an AQM. Exposing all these parameters 1315 as control parameters to a network operator (or user) can easily 1316 result in a unsafe AQM deployment. Unexpected AQM behavior ensues 1317 when parameter values are set improperly. A minimal number of 1318 control parameters minimizes the number of ways a user can break a 1319 system where an AQM scheme is deployed at. Fewer control parameters 1320 make the AQM scheme more user-friendly and easier to deploy and 1321 debug. 1323 [RFC7567] states "AQM algorithms SHOULD NOT require tuning of initial 1324 or configuration parameters in common use cases." A scheme ought to 1325 expose only those parameters that control the macroscopic AQM 1326 behavior such as queue delay threshold, queue length threshold, etc. 1328 Additionally, the safety of an AQM scheme is directly related to its 1329 stability under varying operating conditions such as varying traffic 1330 profiles and fluctuating network conditions, as described in 1331 Section 8. Operating conditions vary often and hence the AQM needs 1332 to remain stable under these conditions without the need for 1333 additional external tuning. If AQM parameters require tuning under 1334 these conditions, then the AQM must self-adapt necessary parameter 1335 values by employing auto-tuning techniques. 1337 12.2. Recommended discussion 1339 In order to understand an AQM's deployment considerations and 1340 performance under a specific environment, AQM proposals SHOULD 1341 describe the parameters that control the macroscopic AQM behavior, 1342 and identify any parameters that require tuning to operational 1343 conditions. It could be interesting to also discuss that even if an 1344 AQM scheme may not adequately auto-tune its parameters, the resulting 1345 performance may not be optimal, but close to something reasonable. 1347 If there are any fixed parameters within the AQM, their setting 1348 SHOULD be discussed and justified, to help understand whether a fixed 1349 parameter value is applicable for a particular environment. 1351 If an AQM scheme is evaluated with parameter(s) that were externally 1352 tuned for optimization or other purposes, these values MUST be 1353 disclosed. 1355 13. Conclusion 1357 Figure 4 lists the scenarios and their requirements. 1359 +------------------------------------------------------------------+ 1360 |Scenario |Sec. |Requirement | 1361 +------------------------------------------------------------------+ 1362 +------------------------------------------------------------------+ 1363 |Interaction with ECN | 4.5 |MUST be discussed if supported | 1364 +------------------------------------------------------------------+ 1365 |Interaction with Scheduling| 4.6 |Feasibility MUST be discussed | 1366 +------------------------------------------------------------------+ 1367 |Transport Protocols |5. | | 1368 | TCP-friendly sender | 5.1 |Scenario MUST be considered | 1369 | Aggressive sender | 5.2 |Scenario MUST be considered | 1370 | Unresponsive sender | 5.3 |Scenario MUST be considered | 1371 | LBE sender | 5.4 |Scenario MAY be considered | 1372 +------------------------------------------------------------------+ 1373 |Round Trip Time Fairness | 6.2 |Scenario MUST be considered | 1374 +------------------------------------------------------------------+ 1375 |Burst Absorption | 7.2 |Scenario MUST be considered | 1376 +------------------------------------------------------------------+ 1377 |Stability |8. | | 1378 | Varying congestion levels | 8.2.5|Scenario MUST be considered | 1379 | Varying available capacity| 8.2.6|Scenario MUST be considered | 1380 | Parameters and stability | 8.3 |This SHOULD be discussed | 1381 +------------------------------------------------------------------+ 1382 |Various Traffic Profiles |9. | | 1383 | Traffic mix | 9.1 |Scenario is RECOMMENDED | 1384 | Bi-directional traffic | 9.2 |Scenario MAY be considered | 1385 +------------------------------------------------------------------+ 1386 |Multi-AQM | 10.2 |Scenario MAY be considered | 1387 +------------------------------------------------------------------+ 1388 |Implementation Cost | 11.2 |Pseudo-code SHOULD be provided | 1389 +------------------------------------------------------------------+ 1390 |Operator Control | 12.2 |Tuning SHOULD NOT be required | 1391 +------------------------------------------------------------------+ 1393 Figure 4: Summary of the scenarios and their requirements 1395 14. Acknowledgements 1397 This work has been partially supported by the European Community 1398 under its Seventh Framework Programme through the Reducing Internet 1399 Transport Latency (RITE) project (ICT-317700). 1401 15. Contributors 1403 Many thanks to S. Akhtar, A.B. Bagayoko, F. Baker, R. Bless, D. 1404 Collier-Brown, G. Fairhurst, J. Gettys, T. Hoiland-Jorgensen, K. 1405 Kilkki, C. Kulatunga, W. Lautenschlager, A.C. Morton, R. Pan, D. 1406 Taht and M. Welzl for detailed and wise feedback on this document. 1408 16. IANA Considerations 1410 This memo includes no request to IANA. 1412 17. Security Considerations 1414 Some security considerations for AQM are identified in [RFC7567].This 1415 document, by itself, presents no new privacy nor security issues. 1417 18. References 1419 18.1. Normative References 1421 [I-D.ietf-tcpm-cubic] 1422 Rhee, I., Xu, L., Ha, S., Zimmermann, A., Eggert, L., and 1423 R. Scheffenegger, "CUBIC for Fast Long-Distance Networks", 1424 draft-ietf-tcpm-cubic-00 (work in progress), June 2015. 1426 [I-D.irtf-iccrg-tcpeval] 1427 Hayes, D., Ros, D., Andrew, L., and S. Floyd, "Common TCP 1428 Evaluation Suite", draft-irtf-iccrg-tcpeval-01 (work in 1429 progress), July 2014. 1431 [RFC0793] Postel, J., "Transmission Control Protocol", STD 7, RFC 1432 793, DOI 10.17487/RFC0793, September 1981, 1433 . 1435 [RFC2119] Bradner, S., "Key words for use in RFCs to Indicate 1436 Requirement Levels", RFC 2119, 1997. 1438 [RFC2488] Allman, M., Glover, D., and L. Sanchez, "Enhancing TCP 1439 Over Satellite Channels using Standard Mechanisms", BCP 1440 28, RFC 2488, DOI 10.17487/RFC2488, January 1999, 1441 . 1443 [RFC2544] Bradner, S. and J. McQuaid, "Benchmarking Methodology for 1444 Network Interconnect Devices", RFC 2544, DOI 10.17487/ 1445 RFC2544, March 1999, 1446 . 1448 [RFC2647] Newman, D., "Benchmarking Terminology for Firewall 1449 Performance", RFC 2647, DOI 10.17487/RFC2647, August 1999, 1450 . 1452 [RFC2679] Almes, G., Kalidindi, S., and M. Zekauskas, "A One-way 1453 Delay Metric for IPPM", RFC 2679, DOI 10.17487/RFC2679, 1454 September 1999, . 1456 [RFC2680] Almes, G., Kalidindi, S., and M. Zekauskas, "A One-way 1457 Packet Loss Metric for IPPM", RFC 2680, DOI 10.17487/ 1458 RFC2680, September 1999, 1459 . 1461 [RFC3168] Ramakrishnan, K., Floyd, S., and D. Black, "The Addition 1462 of Explicit Congestion Notification (ECN) to IP", RFC 1463 3168, DOI 10.17487/RFC3168, September 2001, 1464 . 1466 [RFC3611] Friedman, T., Ed., Caceres, R., Ed., and A. Clark, Ed., 1467 "RTP Control Protocol Extended Reports (RTCP XR)", RFC 1468 3611, DOI 10.17487/RFC3611, November 2003, 1469 . 1471 [RFC5348] Floyd, S., Handley, M., Padhye, J., and J. Widmer, "TCP 1472 Friendly Rate Control (TFRC): Protocol Specification", RFC 1473 5348, DOI 10.17487/RFC5348, September 2008, 1474 . 1476 [RFC5481] Morton, A. and B. Claise, "Packet Delay Variation 1477 Applicability Statement", RFC 5481, DOI 10.17487/RFC5481, 1478 March 2009, . 1480 [RFC5681] Allman, M., Paxson, V., and E. Blanton, "TCP Congestion 1481 Control", RFC 5681, DOI 10.17487/RFC5681, September 2009, 1482 . 1484 [RFC6297] Welzl, M. and D. Ros, "A Survey of Lower-than-Best-Effort 1485 Transport Protocols", RFC 6297, DOI 10.17487/RFC6297, June 1486 2011, . 1488 [RFC6817] Shalunov, S., Hazel, G., Iyengar, J., and M. Kuehlewind, 1489 "Low Extra Delay Background Transport (LEDBAT)", RFC 6817, 1490 DOI 10.17487/RFC6817, December 2012, 1491 . 1493 [RFC7141] Briscoe, B. and J. Manner, "Byte and Packet Congestion 1494 Notification", RFC 7141, 2014. 1496 [RFC7567] Baker, F., Ed. and G. Fairhurst, Ed., "IETF 1497 Recommendations Regarding Active Queue Management", BCP 1498 197, RFC 7567, DOI 10.17487/RFC7567, July 2015, 1499 . 1501 18.2. Informative References 1503 [ANEL2014] 1504 Anelli, P., Diana, R., and E. Lochin, "FavorQueue: a 1505 Parameterless Active Queue Management to Improve TCP 1506 Traffic Performance", Computer Networks vol. 60, 2014. 1508 [BB2011] "BufferBloat: what's wrong with the internet?", ACM Queue 1509 vol. 9, 2011. 1511 [GONG2014] 1512 Gong, Y., Rossi, D., Testa, C., Valenti, S., and D. Taht, 1513 "Fighting the bufferbloat: on the coexistence of AQM and 1514 low priority congestion control", Computer Networks, 1515 Elsevier, 2014, 60, pp.115 - 128 , 2014. 1517 [HASS2008] 1518 Hassayoun, S. and D. Ros, "Loss Synchronization and Router 1519 Buffer Sizing with High-Speed Versions of TCP", IEEE 1520 INFOCOM Workshops , 2008. 1522 [HOEI2015] 1523 Hoeiland-Joergensen, T., McKenney, P., Taht, D., Gettys, 1524 J., and E. Dumazet, "FlowQueue-Codel", IETF (Work-in- 1525 Progress) , January 2015. 1527 [JAY2006] Jay, P., Fu, Q., and G. Armitage, "A preliminary analysis 1528 of loss synchronisation between concurrent TCP flows", 1529 Australian Telecommunication Networks and Application 1530 Conference (ATNAC) , 2006. 1532 [MORR2000] 1533 Morris, R., "Scalable TCP congestion control", IEEE 1534 INFOCOM , 2000. 1536 [NICH2012] 1537 Nichols, K. and V. Jacobson, "Controlling Queue Delay", 1538 ACM Queue , 2012. 1540 [PAN2013] Pan, R., Natarajan, P., Piglione, C., Prabhu, MS., 1541 Subramanian, V., Baker, F., and B. VerSteeg, "PIE: A 1542 lightweight control scheme to address the bufferbloat 1543 problem", IEEE HPSR , 2013. 1545 [RFC2309] Braden, B., Clark, D., Crowcroft, J., Davie, B., Deering, 1546 S., Estrin, D., Floyd, S., Jacobson, V., Minshall, G., 1547 Partridge, C., Peterson, L., Ramakrishnan, K., Shenker, 1548 S., Wroclawski, J., and L. Zhang, "Recommendations on 1549 Queue Management and Congestion Avoidance in the 1550 Internet", RFC 2309, April 1998. 1552 [TRAN2014] 1553 Trang, S., Kuhn, N., Lochin, E., Baudoin, C., Dubois, E., 1554 and P. Gelard, "On The Existence Of Optimal LEDBAT 1555 Parameters", IEEE ICC 2014 - Communication QoS, 1556 Reliability and Modeling Symposium , 2014. 1558 [WELZ2015] 1559 Welzl, M. and G. Fairhurst, "The Benefits to Applications 1560 of using Explicit Congestion Notification (ECN)", IETF 1561 (Work-in-Progress) , June 2015. 1563 [WINS2014] 1564 Winstein, K., "Transport Architectures for an Evolving 1565 Internet", PhD thesis, Massachusetts Institute of 1566 Technology , 2014. 1568 Authors' Addresses 1570 Nicolas Kuhn (editor) 1571 Telecom Bretagne 1572 2 rue de la Chataigneraie 1573 Cesson-Sevigne 35510 1574 France 1576 Phone: +33 2 99 12 70 46 1577 Email: nicolas.kuhn@telecom-bretagne.eu 1578 Preethi Natarajan (editor) 1579 Cisco Systems 1580 510 McCarthy Blvd 1581 Milpitas, California 1582 United States 1584 Email: prenatar@cisco.com 1586 Naeem Khademi (editor) 1587 University of Oslo 1588 Department of Informatics, PO Box 1080 Blindern 1589 N-0316 Oslo 1590 Norway 1592 Phone: +47 2285 24 93 1593 Email: naeemk@ifi.uio.no 1595 David Ros 1596 Simula Research Laboratory AS 1597 P.O. Box 134 1598 Lysaker, 1325 1599 Norway 1601 Phone: +33 299 25 21 21 1602 Email: dros@simula.no