idnits 2.17.1 draft-ietf-aqm-eval-guidelines-12.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 (June 10, 2016) is 2876 days in the past. Is this intentional? Checking references for intended status: Informational ---------------------------------------------------------------------------- == Missing Reference: 'Byte' is mentioned on line 336, but not defined ** Obsolete normative reference: RFC 2679 (Obsoleted by RFC 7679) ** Obsolete normative reference: RFC 2680 (Obsoleted by RFC 7680) == Outdated reference: A later version (-10) exists of draft-ietf-aqm-codel-04 == Outdated reference: A later version (-10) exists of draft-ietf-aqm-pie-08 == Outdated reference: A later version (-07) exists of draft-ietf-tcpm-cubic-01 -- Obsolete informational reference (is this intentional?): RFC 793 (Obsoleted by RFC 9293) -- Obsolete informational reference (is this intentional?): RFC 2309 (Obsoleted by RFC 7567) Summary: 2 errors (**), 0 flaws (~~), 5 warnings (==), 3 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 CNES, Telecom Bretagne 4 Intended status: Informational P. Natarajan, Ed. 5 Expires: December 12, 2016 Cisco Systems 6 N. Khademi, Ed. 7 University of Oslo 8 D. Ros 9 Simula Research Laboratory AS 10 June 10, 2016 12 AQM Characterization Guidelines 13 draft-ietf-aqm-eval-guidelines-12 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 This document describes various criteria for performing 22 characterizations of AQM schemes, that can be used in lab testing 23 during development, prior to deployment. 25 Status of This Memo 27 This Internet-Draft is submitted in full conformance with the 28 provisions of BCP 78 and BCP 79. 30 Internet-Drafts are working documents of the Internet Engineering 31 Task Force (IETF). Note that other groups may also distribute 32 working documents as Internet-Drafts. The list of current Internet- 33 Drafts is at http://datatracker.ietf.org/drafts/current/. 35 Internet-Drafts are draft documents valid for a maximum of six months 36 and may be updated, replaced, or obsoleted by other documents at any 37 time. It is inappropriate to use Internet-Drafts as reference 38 material or to cite them other than as "work in progress." 40 This Internet-Draft will expire on December 12, 2016. 42 Copyright Notice 44 Copyright (c) 2016 IETF Trust and the persons identified as the 45 document authors. All rights reserved. 47 This document is subject to BCP 78 and the IETF Trust's Legal 48 Provisions Relating to IETF Documents 49 (http://trustee.ietf.org/license-info) in effect on the date of 50 publication of this document. Please review these documents 51 carefully, as they describe your rights and restrictions with respect 52 to this document. Code Components extracted from this document must 53 include Simplified BSD License text as described in Section 4.e of 54 the Trust Legal Provisions and are provided without warranty as 55 described in the Simplified BSD License. 57 Table of Contents 59 1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . 3 60 1.1. Reducing the latency and maximizing the goodput . . . . . 5 61 1.2. Goals of this document . . . . . . . . . . . . . . . . . 5 62 1.3. Requirements Language . . . . . . . . . . . . . . . . . . 6 63 1.4. Glossary . . . . . . . . . . . . . . . . . . . . . . . . 6 64 2. End-to-end metrics . . . . . . . . . . . . . . . . . . . . . 7 65 2.1. Flow completion time . . . . . . . . . . . . . . . . . . 7 66 2.2. Flow start up time . . . . . . . . . . . . . . . . . . . 8 67 2.3. Packet loss . . . . . . . . . . . . . . . . . . . . . . . 8 68 2.4. Packet loss synchronization . . . . . . . . . . . . . . . 9 69 2.5. Goodput . . . . . . . . . . . . . . . . . . . . . . . . . 9 70 2.6. Latency and jitter . . . . . . . . . . . . . . . . . . . 10 71 2.7. Discussion on the trade-off between latency and goodput . 10 72 3. Generic setup for evaluations . . . . . . . . . . . . . . . . 11 73 3.1. Topology and notations . . . . . . . . . . . . . . . . . 11 74 3.2. Buffer size . . . . . . . . . . . . . . . . . . . . . . . 13 75 3.3. Congestion controls . . . . . . . . . . . . . . . . . . . 13 76 4. Methodology, Metrics, AQM Comparisons, Packet Sizes, 77 Scheduling and ECN . . . . . . . . . . . . . . . . . . . . . 14 78 4.1. Methodology . . . . . . . . . . . . . . . . . . . . . . . 14 79 4.2. Comments on metrics measurement . . . . . . . . . . . . . 14 80 4.3. Comparing AQM schemes . . . . . . . . . . . . . . . . . . 15 81 4.3.1. Performance comparison . . . . . . . . . . . . . . . 15 82 4.3.2. Deployment comparison . . . . . . . . . . . . . . . . 16 83 4.4. Packet sizes and congestion notification . . . . . . . . 16 84 4.5. Interaction with ECN . . . . . . . . . . . . . . . . . . 17 85 4.6. Interaction with Scheduling . . . . . . . . . . . . . . . 17 86 5. Transport Protocols . . . . . . . . . . . . . . . . . . . . . 18 87 5.1. TCP-friendly sender . . . . . . . . . . . . . . . . . . . 18 88 5.1.1. TCP-friendly sender with the same initial congestion 89 window . . . . . . . . . . . . . . . . . . . . . . . 18 90 5.1.2. TCP-friendly sender with different initial congestion 91 windows . . . . . . . . . . . . . . . . . . . . . . . 19 92 5.2. Aggressive transport sender . . . . . . . . . . . . . . . 19 93 5.3. Unresponsive transport sender . . . . . . . . . . . . . . 19 94 5.4. Less-than Best Effort transport sender . . . . . . . . . 20 95 6. Round Trip Time Fairness . . . . . . . . . . . . . . . . . . 21 96 6.1. Motivation . . . . . . . . . . . . . . . . . . . . . . . 21 97 6.2. Recommended tests . . . . . . . . . . . . . . . . . . . . 21 98 6.3. Metrics to evaluate the RTT fairness . . . . . . . . . . 21 99 7. Burst Absorption . . . . . . . . . . . . . . . . . . . . . . 22 100 7.1. Motivation . . . . . . . . . . . . . . . . . . . . . . . 22 101 7.2. Recommended tests . . . . . . . . . . . . . . . . . . . . 22 102 8. Stability . . . . . . . . . . . . . . . . . . . . . . . . . . 23 103 8.1. Motivation . . . . . . . . . . . . . . . . . . . . . . . 23 104 8.2. Recommended tests . . . . . . . . . . . . . . . . . . . . 24 105 8.2.1. Definition of the congestion Level . . . . . . . . . 24 106 8.2.2. Mild congestion . . . . . . . . . . . . . . . . . . . 25 107 8.2.3. Medium congestion . . . . . . . . . . . . . . . . . . 25 108 8.2.4. Heavy congestion . . . . . . . . . . . . . . . . . . 25 109 8.2.5. Varying the congestion level . . . . . . . . . . . . 25 110 8.2.6. Varying available capacity . . . . . . . . . . . . . 25 111 8.3. Parameter sensitivity and stability analysis . . . . . . 26 112 9. Various Traffic Profiles . . . . . . . . . . . . . . . . . . 27 113 9.1. Traffic mix . . . . . . . . . . . . . . . . . . . . . . . 27 114 9.2. Bi-directional traffic . . . . . . . . . . . . . . . . . 28 115 10. Example of multi-AQM scenario . . . . . . . . . . . . . . . . 28 116 10.1. Motivation . . . . . . . . . . . . . . . . . . . . . . . 28 117 10.2. Details on the evaluation scenario . . . . . . . . . . . 28 118 11. Implementation cost . . . . . . . . . . . . . . . . . . . . . 29 119 11.1. Motivation . . . . . . . . . . . . . . . . . . . . . . . 29 120 11.2. Recommended discussion . . . . . . . . . . . . . . . . . 29 121 12. Operator Control and Auto-tuning . . . . . . . . . . . . . . 30 122 12.1. Motivation . . . . . . . . . . . . . . . . . . . . . . . 30 123 12.2. Recommended discussion . . . . . . . . . . . . . . . . . 30 124 13. Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . 31 125 14. Acknowledgements . . . . . . . . . . . . . . . . . . . . . . 32 126 15. IANA Considerations . . . . . . . . . . . . . . . . . . . . . 32 127 16. Security Considerations . . . . . . . . . . . . . . . . . . . 32 128 17. References . . . . . . . . . . . . . . . . . . . . . . . . . 32 129 17.1. Normative References . . . . . . . . . . . . . . . . . . 32 130 17.2. Informative References . . . . . . . . . . . . . . . . . 33 131 Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . 35 133 1. Introduction 135 Active Queue Management (AQM) addresses the concerns arising from 136 using unnecessarily large and unmanaged buffers to improve network 137 and application performance, such as presented in the section 1.2 of 138 the AQM recommendations document [RFC7567]. Several AQM algorithms 139 have been proposed in the past years, most notably Random Early 140 Detection (RED), BLUE, and Proportional Integral controller (PI), and 141 more recently CoDel [I-D.ietf-aqm-codel] and PIE [I-D.ietf-aqm-pie]. 142 In general, these algorithms actively interact with the Transmission 143 Control Protocol (TCP) and any other transport protocol that deploys 144 a congestion control scheme to manage the amount of data they keep in 145 the network. The available buffer space in the routers and switches 146 should be large enough to accommodate the short-term buffering 147 requirements. AQM schemes aim at reducing buffer occupancy, and 148 therefore the end-to-end delay. Some of these algorithms, notably 149 RED, have also been widely implemented in some network devices. 150 However, the potential benefits of the RED scheme have not been 151 realized since RED is reported to be usually turned off. 153 A buffer is a physical volume of memory in which a queue or set of 154 queues are stored. When speaking of a specific queue in this 155 document, "buffer occupancy" refers to the amount of data (measured 156 in bytes or packets) that are in the queue, and the "maximum buffer 157 size" refers to the maximum buffer occupancy. In switches and 158 routers, a global memory space is often shared between the available 159 interfaces, and thus, the maximum buffer size for any given interface 160 may vary over the time. 162 Bufferbloat [BB2011] is the consequence of deploying large unmanaged 163 buffers on the Internet -- the buffering has often been measured to 164 be ten times or hundred times larger than needed. Large buffer sizes 165 in combination with TCP and/or unresponsive flows increases end-to- 166 end delay. This results in poor performance for latency-sensitive 167 applications such as real-time multimedia (e.g., voice, video, 168 gaming, etc). The degree to which this affects modern networking 169 equipment, especially consumer-grade equipment's, produces problems 170 even with commonly used web services. Active queue management is 171 thus essential to control queuing delay and decrease network latency. 173 The Active Queue Management and Packet Scheduling Working Group (AQM 174 WG) was chartered to address the problems with large unmanaged 175 buffers in the Internet. Specifically, the AQM WG is tasked with 176 standardizing AQM schemes that not only address concerns with such 177 buffers, but also are robust under a wide variety of operating 178 conditions. This document provides characterization guidelines that 179 can be used to assess the applicability, performance and 180 deployability of an AQM, whether it is candidate for standardization 181 at IETF or not. 183 AQM algorithm implemented in a router can be separated from the 184 scheduling of packets sent out by the router as discussed in the AQM 185 recommendations document [RFC7567]. The rest of this memo refers to 186 the AQM as a dropping/marking policy as a separate feature to any 187 interface scheduling scheme. This document may be complemented with 188 another one on guidelines for assessing combination of packet 189 scheduling and AQM. We note that such a document will inherit all 190 the guidelines from this document plus any additional scenarios 191 relevant for packet scheduling such as flow starvation evaluation or 192 impact of the number of hash buckets. 194 1.1. Reducing the latency and maximizing the goodput 196 The trade-off between reducing the latency and maximizing the goodput 197 is intrinsically linked to each AQM scheme and is key to evaluating 198 its performance. To ensure the safety deployment of an AQM, its 199 behaviour should be assessed in a variety of scenarios. Whenever 200 possible, solutions ought to aim at both maximizing goodput and 201 minimizing latency. 203 1.2. Goals of this document 205 This document recommends a generic list of scenarios against which an 206 AQM proposal should be evaluated, considering both potential 207 performance gain and safety of deployment. The guidelines help to 208 quantify performance of AQM schemes in terms of latency reduction, 209 goodput maximization and the trade-off between these two. The 210 document presents central aspects of an AQM algorithm that should be 211 considered whatever the context, such as burst absorption capacity, 212 RTT fairness or resilience to fluctuating network conditions. The 213 guidelines also discuss methods to understand the various aspects 214 associated with safely deploying and operating the AQM scheme. Thus, 215 one of the key objectives behind formulating the guidelines is to 216 help ascertain whether a specific AQM is not only better than drop- 217 tail (i.e. without AQM and with a BDP-sized buffer) but also safe to 218 deploy: the guidelines can be used to compare several AQM proposals 219 with each other, but should be used to compare a proposal with drop- 220 tail. 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 should be 227 considered whatever the context, such as burst absorption capacity, 228 RTT fairness or resilience to fluctuating network conditions. 230 These guidelines do not define and are not bound to a particular 231 deployment scenario or evaluation toolset. Instead the guidelines 232 can be used to assert the potential gain of introducing an AQM for 233 the particular environment, which is of interest to the testers. 234 These guidelines do not cover every possible aspect of a particular 235 algorithm. These guidelines do not present context-dependent 236 scenarios (such as 802.11 WLANs, data-centers or rural broadband 237 networks). To keep the guidelines generic, a number of potential 238 router components and algorithms (such as DiffServ) are omitted. 240 The goals of this document can thus be summarized as follows: 242 o The present characterization guidelines provide a non-exhaustive 243 list of scenarios to help ascertain whether an AQM is not only 244 better than drop-tail (with a BDP-sized buffer), but also safe to 245 deploy; the guidelines can also be used to compare several AQM 246 proposals with each other. 248 o The present characterization guidelines (1) are not bound to a 249 particular evaluation toolset and (2) can be used for various 250 deployment contexts; testers are free to select a toolset that is 251 best suited for the environment in which their proposal will be 252 deployed. 254 o The present characterization guidelines are intended to provide 255 guidance for better selecting an AQM for a specific environment; 256 it is not required that an AQM proposal is evaluated following 257 these guidelines for its standardization. 259 1.3. Requirements Language 261 The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", 262 "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this 263 document are to be interpreted as described in RFC 2119 [RFC2119]. 265 1.4. Glossary 267 o application-limited traffic: a type of traffic that does not have 268 an unlimited amount of data to transmit. 270 o AQM: the Active Queue Managment (AQM) algorithm implemented in a 271 router can be separated from the scheduling of packets sent by the 272 router. The rest of this memo refers to the AQM as a dropping/ 273 marking policy as a separate feature to any interface scheduling 274 scheme [RFC7567]. 276 o BDP: Bandwidth Delay Product. 278 o buffer: a physical volume of memory in which a queue or set of 279 queues are stored. 281 o buffer occupancy: amount of data that are stored in a buffer, 282 measured in bytes or packets. 284 o buffer size: maximum buffer occupancy, that is the maximum amount 285 of data that may be stored in a buffer, measured in bytes or 286 packets. 288 o IW10: TCP initial congestion window set to 10 packets. 290 o latency: one-way delay of packets across Internet paths. This 291 definition suits transport layer definition of the latency, that 292 shall not be confused with an application layer view of the 293 latency. 295 o goodput: goodput is defined as the number of bits per unit of time 296 forwarded to the correct destination minus any bits lost or 297 retransmitted [RFC2647]. The goodput should be determined for 298 each flow and not for aggregates of flows. 300 o SQRT: the square root function. 302 o ROUND: the round function. 304 2. End-to-end metrics 306 End-to-end delay is the result of propagation delay, serialization 307 delay, service delay in a switch, medium-access delay and queuing 308 delay, summed over the network elements along the path. AQM schemes 309 may reduce the queuing delay by providing signals to the sender on 310 the emergence of congestion, but any impact on the goodput must be 311 carefully considered. This section presents the metrics that could 312 be used to better quantify (1) the reduction of latency, (2) 313 maximization of goodput and (3) the trade-off between these two. 314 This section provides normative requirements for metrics that can be 315 used to assess the performance of an AQM scheme. 317 Some metrics listed in this section are not suited to every type of 318 traffic detailed in the rest of this document. It is therefore not 319 necessary to measure all of the following metrics: the chosen metric 320 may not be relevant to the context of the evaluation scenario (e.g., 321 latency vs. goodput trade-off in application-limited traffic 322 scenarios). Guidance is provided for each metric. 324 2.1. Flow completion time 326 The flow completion time is an important performance metric for the 327 end-user when the flow size is finite. The definition of the flow 328 size may be source of contradictions, thus, this metric can consider 329 a flow as a single file. Considering the fact that an AQM scheme may 330 drop/mark packets, the flow completion time is directly linked to the 331 dropping/marking policy of the AQM scheme. This metric helps to 332 better assess the performance of an AQM depending on the flow size. 333 The Flow Completion Time (FCT) is related to the flow size (Fs) and 334 the goodput for the flow (G) as follows: 336 FCT [s] = Fs [Byte] / ( G [Bit/s] / 8 [Bit/Byte] ) 337 Where flow size is the size of the application-level flow in bits and 338 goodput is the application-level transfer time (described in 339 Section 2.5). 341 If this metric is used to evaluate the performance of web transfers, 342 it is suggested to rather consider the time needed to download all 343 the objects that compose the web page, as this makes more sense in 344 terms of user experience than assessing the time needed to download 345 each object. 347 2.2. Flow start up time 349 The flow start up time is the time between the request has been sent 350 from the client and the server starts to transmit data. The amount 351 of packets dropped by an AQM may seriously affect the waiting period 352 during which the data transfer has not started. This metric would 353 specifically focus on the operations such as DNS lookups, TCP opens 354 and SSL handshakes. 356 2.3. Packet loss 358 Packet loss can occur en-route, this can impact the end-to-end 359 performance measured at receiver. 361 The tester should evaluate loss experienced at the receiver using one 362 of the two metrics: 364 o the packet loss ratio: this metric is to be frequently measured 365 during the experiment. The long-term loss ratio is of interest 366 for steady-state scenarios only; 368 o the interval between consecutive losses: the time between two 369 losses is to be measured. 371 The packet loss ratio can be assessed by simply evaluating the loss 372 ratio as a function of the number of lost packets and the total 373 number of packets sent. This might not be easily done in laboratory 374 testing, for which these guidelines advice the tester: 376 o to check that for every packet, a corresponding packet was 377 received within a reasonable time, as presented in the document 378 that proposes a metric for one-way packet loss across Internet 379 paths [RFC2680]. 381 o to keep a count of all packets sent, and a count of the non- 382 duplicate packets received, as discussed in RFC that presents a 383 benchmarking methodology [RFC2544]. 385 The interval between consecutive losses, which is also called a gap, 386 is a metric of interest for VoIP traffic [RFC3611]. 388 2.4. Packet loss synchronization 390 One goal of an AQM algorithm is to help to avoid global 391 synchronization of flows sharing a bottleneck buffer on which the AQM 392 operates ([RFC2309],[RFC7567]). The "degree" of packet-loss 393 synchronization between flows should be assessed, with and without 394 the AQM under consideration. 396 Loss synchronization among flows may be quantified by several 397 slightly different metrics that capture different aspects of the same 398 issue [HASS2008]. However, in real-world measurements the choice of 399 metric could be imposed by practical considerations -- e.g., whether 400 fine-grained information on packet losses at the bottleneck is 401 available or not. For the purpose of AQM characterization, a good 402 candidate metric is the global synchronization ratio, measuring the 403 proportion of flows losing packets during a loss event. This metric 404 can be used in real-world experiments to characterize synchronization 405 along arbitrary Internet paths [JAY2006]. 407 If an AQM scheme is evaluated using real-life network environments, 408 it is worth pointing out that some network events, such as failed 409 link restoration may cause synchronized losses between active flows 410 and thus confuse the meaning of this metric. 412 2.5. Goodput 414 The goodput has been defined as the number of bits per unit of time 415 forwarded to the correct destination interface, minus any bits lost 416 or retransmitted, such as proposed in the secton 3.17 of the RFC 417 describing the benchmarking terminology for firewall performances 418 [RFC2647]. This definition requires that the test setup needs to be 419 qualified to assure that it is not generating losses on its own. 421 Measuring the end-to-end goodput provides an appreciation of how well 422 an AQM scheme improves transport and application performance. The 423 measured end-to-end goodput is linked to the dropping/marking policy 424 of the AQM scheme -- e.g., the fewer the number of packet drops, the 425 fewer packets need retransmission, minimizing the impact of AQM on 426 transport and application performance. Additionally, an AQM scheme 427 may resort to Explicit Congestion Notification (ECN) marking as an 428 initial means to control delay. Again, marking packets instead of 429 dropping them reduces the number of packet retransmissions and 430 increases goodput. End-to-end goodput values help to evaluate the 431 AQM scheme's effectiveness of an AQM scheme in minimizing packet 432 drops that impact application performance and to estimate how well 433 the AQM scheme works with ECN. 435 The measurement of the goodput allows the tester to evaluate to which 436 extent an AQM is able to maintain a high bottleneck utilization. 437 This metric should also be obtained frequently during an experiment 438 as the long-term goodput is relevant for steady-state scenarios only 439 and may not necessarily reflect how the introduction of an AQM 440 actually impacts the link utilization during at a certain period of 441 time. Fluctuations in the values obtained from these measurements 442 may depend on other factors than the introduction of an AQM, such as 443 link layer losses due to external noise or corruption, fluctuating 444 bandwidths (802.11 WLANs), heavy congestion levels or transport 445 layer's rate reduction by congestion control mechanism. 447 2.6. Latency and jitter 449 The latency, or the one-way delay metric, is discussed in [RFC2679]. 450 There is a consensus on an adequate metric for the jitter, that 451 represents the one-way delay variations for packets from the same 452 flow: the Packet Delay Variation (PDV) serves well all use cases 453 [RFC5481]. 455 The end-to-end latency includes components other than just the 456 queuing delay, such as the signal processing delay, transmission 457 delay and the processing delay. Moreover, the jitter is caused by 458 variations in queuing and processing delay (e.g., scheduling 459 effects). The introduction of an AQM scheme would impact end-to-end 460 latency and jitter, and therefore these metrics should be considered 461 in the end-to-end evaluation of performance. 463 2.7. Discussion on the trade-off between latency and goodput 465 The metrics presented in this section may be considered in order to 466 discuss and quantify the trade-off between latency and goodput. 468 With regards to the goodput, and in addition to the long-term 469 stationary goodput value, it is recommended to take measurements 470 every multiple of the minimum RTT (minRTT) between A and B. It is 471 suggested to take measurements at least every K x minRTT (to smooth 472 out the fluctuations), with K=10. Higher values for K can be 473 considered whenever it is more appropriate for the presentation of 474 the results, since the value for K may depend on the network's path 475 characteristics. The measurement period must be disclosed for each 476 experiment and when results/values are compared across different AQM 477 schemes, the comparisons should use exactly the same measurement 478 periods. With regards to latency, it is recommended to take the 479 samples on per-packet basis whenever possible depending on the 480 features provided by hardware/software and the impact of sampling 481 itself on the hardware performance. 483 From each of these sets of measurements, the cumulative density 484 function (CDF) of the considered metrics should be computed. If the 485 considered scenario introduces dynamically varying parameters, 486 temporal evolution of the metrics could also be generated. For each 487 scenario, the following graph may be generated: the x-axis shows 488 queuing delay (that is the average per-packet delay in excess of 489 minimum RTT), the y-axis the goodput. Ellipses are computed such as 490 detailed in [WINS2014]: "We take each individual [...] run [...] as 491 one point, and then compute the 1-epsilon elliptic contour of the 492 maximum-likelihood 2D Gaussian distribution that explains the points. 493 [...] we plot the median per-sender throughput and queueing delay as 494 a circle. [...] The orientation of an ellipse represents the 495 covariance between the throughput and delay measured for the 496 protocol." This graph provides part of a better understanding of (1) 497 the delay/goodput trade-off for a given congestion control mechanism 498 (Section 5), and (2) how the goodput and average queue delay vary as 499 a function of the traffic load (Section 8.2). 501 3. Generic setup for evaluations 503 This section presents the topology that can be used for each of the 504 following scenarios, the corresponding notations and discusses 505 various assumptions that have been made in the document. 507 3.1. Topology and notations 508 +--------------+ +--------------+ 509 |sender A_i | |receive B_i | 510 |--------------| |--------------| 511 | SEN.Flow1.1 +---------+ +-----------+ REC.Flow1.1 | 512 | + | | | | + | 513 | | | | | | | | 514 | + | | | | + | 515 | SEN.Flow1.X +-----+ | | +--------+ REC.Flow1.X | 516 +--------------+ | | | | +--------------+ 517 + +-+---+---+ +--+--+---+ + 518 | |Router L | |Router R | | 519 | |---------| |---------| | 520 | | AQM | | | | 521 | | BuffSize| | BuffSize| | 522 | | (Bsize) +-----+ (Bsize) | | 523 | +-----+--++ ++-+------+ | 524 + | | | | + 525 +--------------+ | | | | +--------------+ 526 |sender A_n | | | | | |receive B_n | 527 |--------------| | | | | |--------------| 528 | SEN.FlowN.1 +---------+ | | +-----------+ REC.FlowN.1 | 529 | + | | | | + | 530 | | | | | | | | 531 | + | | | | + | 532 | SEN.FlowN.Y +------------+ +-------------+ REC.FlowN.Y | 533 +--------------+ +--------------+ 535 Figure 1: Topology and notations 537 Figure 1 is a generic topology where: 539 o traffic profile is a set of flows with similar characteristics - 540 RTT, congestion control scheme, transport protocol, etc.; 542 o senders with different traffic characteristics (i.e., traffic 543 profiles) can be introduced; 545 o the timing of each flow could be different (i.e., when does each 546 flow start and stop); 548 o each traffic profile can comprise various number of flows; 550 o each link is characterized by a couple (one-way delay, capacity); 552 o sender A_i is instantiated for each traffic profile. A 553 corresponding receiver B_i is instantiated for receiving the flows 554 in the profile; 556 o flows sharing a bottleneck (the link between routers L and R); 558 o the tester should consider both scenarios of asymmetric and 559 symmetric bottleneck links in terms of bandwidth. In case of 560 asymmetric link, the capacity from senders to receivers is higher 561 than the one from receivers to senders; the symmetric link 562 scenario provides a basic understanding of the operation of the 563 AQM mechanism whereas the asymmetric link scenario evaluates an 564 AQM mechanism in a more realistic setup; 566 o in asymmetric link scenarios, the tester should study the bi- 567 directional traffic between A and B (downlink and uplink) with the 568 AQM mechanism deployed on one direction only. The tester may 569 additionally consider a scenario with AQM mechanism being deployed 570 on both directions. In each scenario, the tester should 571 investigate the impact of drop policy of the AQM on TCP ACK 572 packets and its impact on the performance (Section 9.2). 574 Although this topology may not perfectly reflect actual topologies, 575 the simple topology is commonly used in the world of simulations and 576 small testbeds. It can be considered as adequate to evaluate AQM 577 proposals [I-D.irtf-iccrg-tcpeval]. Testers ought to pay attention 578 to the topology that has been used to evaluate an AQM scheme when 579 comparing this scheme with a newly proposed AQM scheme. 581 3.2. Buffer size 583 The size of the buffers should be carefully chosen, and may be set to 584 the bandwidth-delay product; the bandwidth being the bottleneck 585 capacity and the delay the largest RTT in the considered network. 586 The size of the buffer can impact the AQM performance and is a 587 dimensioning parameter that will be considered when comparing AQM 588 proposals. 590 If a specific buffer size is required, the tester must justify and 591 detail the way the maximum queue size is set. Indeed, the maximum 592 size of the buffer may affect the AQM's performance and its choice 593 should be elaborated for a fair comparison between AQM proposals. 594 While comparing AQM schemes the buffer size should remain the same 595 across the tests. 597 3.3. Congestion controls 599 This document considers running three different congestion control 600 algorithms between A and B 602 o Standard TCP congestion control: the base-line congestion control 603 is TCP NewReno with SACK [RFC5681]. 605 o Aggressive congestion controls: a base-line congestion control for 606 this category is TCP Cubic [I-D.ietf-tcpm-cubic]. 608 o Less-than Best Effort (LBE) congestion controls: an LBE congestion 609 control 'results in smaller bandwidth and/or delay impact on 610 standard TCP than standard TCP itself, when sharing a bottleneck 611 with it.': a base-line congestion control for this category is 612 LEDBAT [RFC6817]. 614 Other transport congestion controls can OPTIONALLY be evaluated in 615 addition. Recent transport layer protocols are not mentioned in the 616 following sections, for the sake of simplicity. 618 4. Methodology, Metrics, AQM Comparisons, Packet Sizes, Scheduling and 619 ECN 621 4.1. Methodology 623 A description of each test setup should be detailed to allow this 624 test to be compared with other tests. This also allows others to 625 replicate the tests if needed. This test setup should detail 626 software and hardware versions. The tester could make its data 627 available. 629 The proposals should be evaluated on real-life systems, or they may 630 be evaluated with event-driven simulations (such as ns-2, ns-3, 631 OMNET, etc). The proposed scenarios are not bound to a particular 632 evaluation toolset. 634 The tester is encouraged to make the detailed test setup and the 635 results publicly available. 637 4.2. Comments on metrics measurement 639 The document presents the end-to-end metrics that ought to be used to 640 evaluate the trade-off between latency and goodput in Section 2. In 641 addition to the end-to-end metrics, the queue-level metrics (normally 642 collected at the device operating the AQM) provide a better 643 understanding of the AQM behavior under study and the impact of its 644 internal parameters. Whenever it is possible (e.g., depending on the 645 features provided by the hardware/software), these guidelines advise 646 to consider queue-level metrics, such as link utilization, queuing 647 delay, queue size or packet drop/mark statistics in addition to the 648 AQM-specific parameters. However, the evaluation must be primarily 649 based on externally observed end-to-end metrics. 651 These guidelines do not aim to detail on the way these metrics can be 652 measured, since the way these metrics are measured is expected to 653 depend on the evaluation toolset. 655 4.3. Comparing AQM schemes 657 This document recognizes that these guidelines may be used for 658 comparing AQM schemes. 660 AQM schemes need to be compared against both performance and 661 deployment categories. In addition, this section details how best to 662 achieve a fair comparison of AQM schemes by avoiding certain 663 pitfalls. 665 4.3.1. Performance comparison 667 AQM schemes should be compared against the generic scenarios that are 668 summarized in Section 13. AQM schemes may be compared for specific 669 network environments such as data centers, home networks, etc. If an 670 AQM scheme has parameter(s) that were externally tuned for 671 optimization or other purposes, these values must be disclosed. 673 AQM schemes belong to different varieties such as queue-length based 674 schemes (ex. RED) or queueing-delay based scheme (ex. CoDel, PIE). 675 AQM schemes expose different control knobs associated with different 676 semantics. For example, while both PIE and CoDel are queueing-delay 677 based schemes and each expose a knob to control the queueing delay -- 678 PIE's "queueing delay reference" vs. CoDel's "queueing delay target", 679 the two tuning parameters of the two schemes have different 680 semantics, resulting in different control points. Such differences 681 in AQM schemes can be easily overlooked while making comparisons. 683 This document recommends the following procedures for a fair 684 performance comparison between the AQM schemes: 686 1. similar control parameters and implications: Testers should be 687 aware of the control parameters of the different schemes that 688 control similar behavior. Testers should also be aware of the 689 input value ranges and corresponding implications. For example, 690 consider two different schemes - (A) queue-length based AQM 691 scheme, and (B) queueing-delay based scheme. A and B are likely 692 to have different kinds of control inputs to control the target 693 delay - target queue length in A vs. target queuing delay in B, 694 for example. Setting parameter values such as 100MB for A vs. 695 10ms for B will have different implications depending on 696 evaluation context. Such context-dependent implications must be 697 considered before drawing conclusions on performance comparisons. 698 Also, it would be preferable if an AQM proposal listed such 699 parameters and discussed how each relates to network 700 characteristics such as capacity, average RTT etc. 702 2. compare over a range of input configurations: there could be 703 situations when the set of control parameters that affect a 704 specific behavior have different semantics between the two AQM 705 schemes. As mentioned above, PIE has tuning parameters to 706 control queue delay that has a different semantics from those 707 used in CoDel. In such situations, these schemes need to be 708 compared over a range of input configurations. For example, 709 compare PIE vs. CoDel over the range of target delay input 710 configurations. 712 4.3.2. Deployment comparison 714 AQM schemes must be compared against deployment criteria such as the 715 parameter sensitivity (Section 8.3), auto-tuning (Section 12) or 716 implementation cost (Section 11). 718 4.4. Packet sizes and congestion notification 720 An AQM scheme may be considering packet sizes while generating 721 congestion signals [RFC7141]. For example, control packets such as 722 DNS requests/responses, TCP SYNs/ACKs are small, but their loss can 723 severely impact application performance. An AQM scheme may therefore 724 be biased towards small packets by dropping them with lower 725 probability compared to larger packets. However, such an AQM scheme 726 is unfair to data senders generating larger packets. Data senders, 727 malicious or otherwise, are motivated to take advantage of such AQM 728 scheme by transmitting smaller packets, and could result in unsafe 729 deployments and unhealthy transport and/or application designs. 731 An AQM scheme should adhere to the recommendations outlined in the 732 best current practive for dropping and marking packets document 733 [RFC7141], and should not provide undue advantage to flows with 734 smaller packets, such as discussed in the section 4.4 of the AQM 735 recommendation document [RFC7567]. In order to evaluate if an AQM 736 scheme is biased towards flows with smaller size packets, traffic can 737 be generated, such as defined in Section 8.2.2, where half of the 738 flows have smaller packets (e.g. 500 bytes packets) than the other 739 half of the flow (e.g. 1500 bytes packets). In this case, the 740 metrics reported could be the same as in Section 6.3, where Category 741 I is the set of flows with smaller packets and Category II the one 742 with larger packets. The bidirectional scenario could also be 743 considered (Section 9.2). 745 4.5. Interaction with ECN 747 ECN [RFC3168] is an alternative that allows AQM schemes to signal 748 receivers about network congestion that does not use packet drop. 749 There are benefits of providing ECN support for an AQM scheme 750 [WELZ2015]. 752 If the tested AQM scheme can support ECN, the testers must discuss 753 and describe the support of ECN, such as discussed in the AQM 754 recommendation [RFC7567]. Also, the AQM's ECN support can be studied 755 and verified by replicating tests in Section 8.1 with ECN turned ON 756 at the TCP senders. The results can be used to not only evaluate the 757 performance of the tested AQM with and without ECN markings, but also 758 quantify the interest of enabling ECN. 760 4.6. Interaction with Scheduling 762 A network device may use per-flow or per-class queuing with a 763 scheduling algorithm to either prioritize certain applications or 764 classes of traffic, limit the rate of transmission, or to provide 765 isolation between different traffic flows within a common class, such 766 as discussed in the section 2.1 of the AQM recommendation document 767 [RFC7567]. 769 The scheduling and the AQM conjointly impact on the end-to-end 770 performance. Therefore, the AQM proposal must discuss the 771 feasibility to add scheduling combined with the AQM algorithm. It 772 can be explained whether the dropping policy is applied when packets 773 are being enqueued or dequeued. 775 These guidelines do not propose guidelines to assess the performance 776 of scheduling algorithms. Indeed, as opposed to characterizing AQM 777 schemes that is related to their capacity to control the queuing 778 delay in a queue, characterizing scheduling schemes is related to the 779 scheduling itself and its interaction with the AQM scheme. As one 780 example, the scheduler may create sub-queues and the AQM scheme may 781 be applied on each of the sub-queues, and/or the AQM could be applied 782 on the whole queue. Also, schedulers might, such as FQ-CoDel 783 [HOEI2015] or FavorQueue [ANEL2014], introduce flow prioritization. 784 In these cases, specific scenarios should be proposed to ascertain 785 that these scheduler schemes not only helps in tackling the 786 bufferbloat, but also are robust under a wide variety of operating 787 conditions. This is out of the scope of this document that focus on 788 dropping and/or marking AQM schemes. 790 5. Transport Protocols 792 Network and end-devices need to be configured with a reasonable 793 amount of buffer space to absorb transient bursts. In some 794 situations, network providers tend to configure devices with large 795 buffers to avoid packet drops triggered by a full buffer and to 796 maximize the link utilization for standard loss-based TCP traffic. 798 AQM algorithms are often evaluated by considering Transmission 799 Control Protocol (TCP) [RFC0793] with a limited number of 800 applications. TCP is a widely deployed transport. It fills up 801 available buffers until a sender transfering a bulk flow with TCP 802 receives a signal (packet drop) that reduces the sending rate. The 803 larger the buffer, the higher the buffer occupancy, and therefore the 804 queuing delay. An efficient AQM scheme sends out early congestion 805 signals to TCP to bring the queuing delay under control. 807 Not all endpoints (or applications) using TCP use the same flavor of 808 TCP. Variety of senders generate different classes of traffic which 809 may not react to congestion signals (aka non-responsive flows in the 810 section 3 of the AQM recommendation document [RFC7567]) or may not 811 reduce their sending rate as expected (aka Transport Flows that are 812 less responsive than TCP, such as proposed in the section 3 of the 813 AQM recommendation document [RFC7567], also called "aggressive 814 flows"). In these cases, AQM schemes seek to control the queuing 815 delay. 817 This section provides guidelines to assess the performance of an AQM 818 proposal for various traffic profiles -- different types of senders 819 (with different TCP congestion control variants, unresponsive, 820 aggressive). 822 5.1. TCP-friendly sender 824 5.1.1. TCP-friendly sender with the same initial congestion window 826 This scenario helps to evaluate how an AQM scheme reacts to a TCP- 827 friendly transport sender. A single long-lived, non application- 828 limited, TCP NewReno flow, with an Initial congestion Window (IW) set 829 to 3 packets, transfers data between sender A and receiver B. Other 830 TCP friendly congestion control schemes such as TCP-friendly rate 831 control [RFC5348] etc may also be considered. 833 For each TCP-friendly transport considered, the graph described in 834 Section 2.7 could be generated. 836 5.1.2. TCP-friendly sender with different initial congestion windows 838 This scenario can be used to evaluate how an AQM scheme adapts to a 839 traffic mix consisting of TCP flows with different values of the IW. 841 For this scenario, two types of flows must be generated between 842 sender A and receiver B: 844 o A single long-lived non application-limited TCP NewReno flow; 846 o A single application-limited TCP NewReno flow, with an IW set to 3 847 or 10 packets. The size of the data transferred must be strictly 848 higher than 10 packets and should be lower than 100 packets. 850 The transmission of the non application-limited flow must start first 851 and the transmission of the application-limited flow starts after the 852 non application-limited flow has reached steady state. The steady 853 state can be assumed when the goodput is stable. 855 For each of these scenarios, the graph described in Section 2.7 could 856 be generated for each class of traffic (application-limited and non 857 application-limited). The completion time of the application-limited 858 TCP flow could be measured. 860 5.2. Aggressive transport sender 862 This scenario helps testers to evaluate how an AQM scheme reacts to a 863 transport sender that is more aggressive than a single TCP-friendly 864 sender. We define 'aggressiveness' as a higher increase factor than 865 standard upon a successful transmission and/or a lower than standard 866 decrease factor upon a unsuccessful transmission (e.g., in case of 867 congestion controls with Additive-Increase Multiplicative-Decrease 868 (AIMD) principle, a larger AI and/or MD factors). A single long- 869 lived, non application-limited, TCP Cubic flow transfers data between 870 sender A and receiver B. Other aggressive congestion control schemes 871 may also be considered. 873 For each flavor of aggressive transports, the graph described in 874 Section 2.7 could be generated. 876 5.3. Unresponsive transport sender 878 This scenario helps testers to evaluate how an AQM scheme reacts to a 879 transport sender that is less responsive than TCP. Note that faulty 880 transport implementations on an end host and/or faulty network 881 elements en-route that "hide" congestion signals in packet headers 882 may also lead to a similar situation, such that the AQM scheme needs 883 to adapt to unresponsive traffic (see the section 3 of the AQM 884 recommendation document [RFC7567]). To this end, these guidelines 885 propose the two following scenarios. 887 The first scenario can be used to evaluate queue build up. It 888 considers unresponsive flow(s) whose sending rate is greater than the 889 bottleneck link capacity between routers L and R. This scenario 890 consists of a long-lived non application limited UDP flow transmits 891 data between sender A and receiver B. Graphs described in 892 Section 2.7 could be generated. 894 The second scenario can be used to evaluate if the AQM scheme is able 895 to keep the responsive fraction under control. This scenario 896 considers a mixture of TCP-friendly and unresponsive traffics. It 897 consists of a long-lived UDP flow from unresponsive application and a 898 single long-lived, non application-limited (unlimited data available 899 to the transport sender from application layer), TCP New Reno flow 900 that transmit data between sender A and receiver B. As opposed to 901 the first scenario, the rate of the UDP traffic should not be greater 902 than the bottleneck capacity, and should be higher than half of the 903 bottleneck capacity. For each type of traffic, the graph described 904 in Section 2.7 could be generated. 906 5.4. Less-than Best Effort transport sender 908 This scenario helps to evaluate how an AQM scheme reacts to LBE 909 congestion controls that 'results in smaller bandwidth and/or delay 910 impact on standard TCP than standard TCP itself, when sharing a 911 bottleneck with it.' [RFC6297]. There are potential fateful 912 interactions when AQM and LBE techniques are combined [GONG2014]; 913 this scenario helps to evaluate whether the coexistence of the 914 proposed AQM and LBE techniques may be possible. 916 A single long-lived non application-limited TCP NewReno flow 917 transfers data between sender A and receiver B. Other TCP-friendly 918 congestion control schemes may also be considered. Single long-lived 919 non application-limited LEDBAT [RFC6817] flows transfer data between 920 sender A and receiver B. We recommend to set the target delay and 921 gain values of LEDBAT respectively to 5 ms and 10 [TRAN2014]. Other 922 LBE congestion control schemes may also be considered and are listed 923 in the IETF survey of LBE protocols [RFC6297]. 925 For each of the TCP-friendly and LBE transports, the graph described 926 in Section 2.7 could be generated. 928 6. Round Trip Time Fairness 930 6.1. Motivation 932 An AQM scheme's congestion signals (via drops or ECN marks) must 933 reach the transport sender so that a responsive sender can initiate 934 its congestion control mechanism and adjust the sending rate. This 935 procedure is thus dependent on the end-to-end path RTT. When the RTT 936 varies, the onset of congestion control is impacted, and in turn 937 impacts the ability of an AQM scheme to control the queue. It is 938 therefore important to assess the AQM schemes for a set of RTTs 939 between A and B (e.g., from 5 ms to 200 ms). 941 The asymmetry in terms of difference in intrinsic RTT between various 942 paths sharing the same bottleneck should be considered, so that the 943 fairness between the flows can be discussed. In this scenario, a 944 flow traversing on shorter RTT path may react faster to congestion 945 and recover faster from it compared to another flow on a longer RTT 946 path. The introduction of AQM schemes may potentially improve the 947 RTT fairness. 949 Introducing an AQM scheme may cause the unfairness between the flows, 950 even if the RTTs are identical. This potential unfairness should be 951 investigated as well. 953 6.2. Recommended tests 955 The recommended topology is detailed in Figure 1. 957 To evaluate the RTT fairness, for each run, two flows are divided 958 into two categories. Category I whose RTT between sender A and 959 receiver B should be 100ms. Category II which RTT between sender A 960 and receiver B should be in the range [5ms;560ms] inclusive. The 961 maximum value for the RTT represents the RTT of a satellite link 962 [RFC2488]. 964 A set of evaluated flows must use the same congestion control 965 algorithm: all the generated flows could be single long-lived non 966 application-limited TCP NewReno flows. 968 6.3. Metrics to evaluate the RTT fairness 970 The outputs that must be measured are: (1) the cumulative average 971 goodput of the flow from Category I, goodput_Cat_I (Section 2.5); (2) 972 the cumulative average goodput of the flow from Category II, 973 goodput_Cat_II (Section 2.5); (3) the ratio goodput_Cat_II/ 974 goodput_Cat_I; (4) the average packet drop rate for each category 975 (Section 2.3). 977 7. Burst Absorption 979 "AQM mechanisms need to control the overall queue sizes, to ensure 980 that arriving bursts can be accommodated without dropping packets" 981 [RFC7567]. 983 7.1. Motivation 985 An AQM scheme can face bursts of packet arrivals due to various 986 reasons. Dropping one or more packets from a burst can result in 987 performance penalties for the corresponding flows, since dropped 988 packets have to be retransmitted. Performance penalties can result 989 in failing to meet SLAs and be a disincentive to AQM adoption. 991 The ability to accommodate bursts translates to larger queue length 992 and hence more queuing delay. On the one hand, it is important that 993 an AQM scheme quickly brings bursty traffic under control. On the 994 other hand, a peak in the packet drop rates to bring a packet burst 995 quickly under control could result in multiple drops per flow and 996 severely impact transport and application performance. Therefore, an 997 AQM scheme ought to bring bursts under control by balancing both 998 aspects -- (1) queuing delay spikes are minimized and (2) performance 999 penalties for ongoing flows in terms of packet drops are minimized. 1001 An AQM scheme that maintains short queues allows some remaining space 1002 in the buffer for bursts of arriving packets. The tolerance to 1003 bursts of packets depends upon the number of packets in the queue, 1004 which is directly linked to the AQM algorithm. Moreover, an AQM 1005 scheme may implement a feature controlling the maximum size of 1006 accepted bursts, that can depend on the buffer occupancy or the 1007 currently estimated queuing delay. The impact of the buffer size on 1008 the burst allowance may be evaluated. 1010 7.2. Recommended tests 1012 For this scenario, tester must evaluate how the AQM performs with a 1013 traffic mixed that could be composed of (from sender A to receiver 1014 B): 1016 o Burst of packets at the beginning of a transmission, such as web 1017 traffic with IW10; 1019 o Applications that send large bursts of data, such as bursty video 1020 frames; 1022 o Background traffic, such as Constant Bit Rate (CBR) UDP traffic 1023 and/or A single non application-limited bulk TCP flow as 1024 background traffic. 1026 Figure 2 presents the various cases for the traffic that must be 1027 generated between sender A and receiver B. 1029 +-------------------------------------------------+ 1030 |Case| Traffic Type | 1031 | +-----+------------+----+--------------------+ 1032 | |Video|Web (IW 10)| CBR| Bulk TCP Traffic | 1033 +----|-----|------------|----|--------------------| 1034 |I | 0 | 1 | 1 | 0 | 1035 +----|-----|------------|----|--------------------| 1036 |II | 0 | 1 | 1 | 1 | 1037 |----|-----|------------|----|--------------------| 1038 |III | 1 | 1 | 1 | 0 | 1039 +----|-----|------------|----|--------------------| 1040 |IV | 1 | 1 | 1 | 1 | 1041 +----+-----+------------+----+--------------------+ 1043 Figure 2: Bursty traffic scenarios 1045 A new web page download could start after the previous web page 1046 download is finished. Each web page could be composed by at least 50 1047 objects and the size of each object should be at least 1kB. 6 TCP 1048 parallel connections should be generated to download the objects, 1049 each parallel connections having an initial congestion window set to 1050 10 packets. 1052 For each of these scenarios, the graph described in Section 2.7 could 1053 be generated for each application. Metrics such as end-to-end 1054 latency, jitter, flow completion time may be generated. For the 1055 cases of frame generation of bursty video traffic as well as the 1056 choice of web traffic pattern, these details and their presentation 1057 are left to the testers. 1059 8. Stability 1061 8.1. Motivation 1063 The safety of an AQM scheme is directly related to its stability 1064 under varying operating conditions such as varying traffic profiles 1065 and fluctuating network conditions. Since operating conditions can 1066 vary often the AQM needs to remain stable under these conditions 1067 without the need for additional external tuning. 1069 Network devices can experience varying operating conditions depending 1070 on factors such as time of the day, deployment scenario, etc. For 1071 example: 1073 o Traffic and congestion levels are higher during peak hours than 1074 off-peak hours. 1076 o In the presence of a scheduler, the draining rate of a queue can 1077 vary depending on the occupancy of other queues: a low load on a 1078 high priority queue implies a higher draining rate for the lower 1079 priority queues. 1081 o The capacity available can vary over time (e.g., a lossy channel, 1082 a link supporting traffic in a higher diffserv class). 1084 Whether the target context is a not stable environment, the ability 1085 of an AQM scheme to maintain its control over the queuing delay and 1086 buffer occupancy can be challenged. This document proposes 1087 guidelines to assess the behavior of AQM schemes under varying 1088 congestion levels and varying draining rates. 1090 8.2. Recommended tests 1092 Note that the traffic profiles explained below comprises non 1093 application-limited TCP flows. For each of the below scenarios, the 1094 graphs described in Section 2.7 should be generated, and the goodput 1095 of the various flows should be cumulated. For Section 8.2.5 and 1096 Section 8.2.6 they should incorporate the results in per-phase basis 1097 as well. 1099 Wherever the notion of time has explicitly mentioned in this 1100 subsection, time 0 starts from the moment all TCP flows have already 1101 reached their congestion avoidance phase. 1103 8.2.1. Definition of the congestion Level 1105 In these guidelines, the congestion levels are represented by the 1106 projected packet drop rate, had a drop-tail queue was chosen instead 1107 of an AQM scheme. When the bottleneck is shared among non 1108 application-limited TCP flows. l_r, the loss rate projection can be 1109 expressed as a function of N, the number of bulk TCP flows, and S, 1110 the sum of the bandwidth-delay product and the maximum buffer size, 1111 both expressed in packets, based on Eq. 3 of [MORR2000]: 1113 l_r = 0.76 * N^2 / S^2 1115 N = S * SQRT(1/0.76) * SQRT (l_r) 1117 These guidelines use the loss rate to define the different congestion 1118 levels, but they do not stipulate that in other circumstances, 1119 measuring the congestion level gives you an accurate estimation of 1120 the loss rate or vice-versa. 1122 8.2.2. Mild congestion 1124 This scenario can be used to evaluate how an AQM scheme reacts to a 1125 light load of incoming traffic resulting in mild congestion -- packet 1126 drop rates around 0.1%. The number of bulk flows required to achieve 1127 this congestion level, N_mild, is then: 1129 N_mild = ROUND (0.036*S) 1131 8.2.3. Medium congestion 1133 This scenario can be used to evaluate how an AQM scheme reacts to 1134 incoming traffic resulting in medium congestion -- packet drop rates 1135 around 0.5%. The number of bulk flows required to achieve this 1136 congestion level, N_med, is then: 1138 N_med = ROUND (0.081*S) 1140 8.2.4. Heavy congestion 1142 This scenario can be used to evaluate how an AQM scheme reacts to 1143 incoming traffic resulting in heavy congestion -- packet drop rates 1144 around 1%. The number of bulk flows required to achieve this 1145 congestion level, N_heavy, is then: 1147 N_heavy = ROUND (0.114*S) 1149 8.2.5. Varying the congestion level 1151 This scenario can be used to evaluate how an AQM scheme reacts to 1152 incoming traffic resulting in various levels of congestion during the 1153 experiment. In this scenario, the congestion level varies within a 1154 large time-scale. The following phases may be considered: phase I - 1155 mild congestion during 0-20s; phase II - medium congestion during 1156 20-40s; phase III - heavy congestion during 40-60s; phase I again, 1157 and so on. 1159 8.2.6. Varying available capacity 1161 This scenario can be used to help characterize how the AQM behaves 1162 and adapts to bandwidth changes. The experiments are not meant to 1163 reflect the exact conditions of Wi-Fi environments since it is hard 1164 to design repetitive experiments or accurate simulations for such 1165 scenarios. 1167 To emulate varying draining rates, the bottleneck capacity between 1168 nodes 'Router L' and 'Router R' varies over the course of the 1169 experiment as follows: 1171 o Experiment 1: the capacity varies between two values within a 1172 large time-scale. As an example, the following phases may be 1173 considered: phase I - 100Mbps during 0-20s; phase II - 10Mbps 1174 during 20-40s; phase I again, and so on. 1176 o Experiment 2: the capacity varies between two values within a 1177 short time-scale. As an example, the following phases may be 1178 considered: phase I - 100Mbps during 0-100ms; phase II - 10Mbps 1179 during 100-200ms; phase I again, and so on. 1181 The tester may choose a phase time-interval value different than what 1182 is stated above, if the network's path conditions (such as bandwidth- 1183 delay product) necessitate. In this case the choice of such time- 1184 interval value should be stated and elaborated. 1186 The tester may additionally evaluate the two mentioned scenarios 1187 (short-term and long-term capacity variations), during and/or 1188 including TCP slow-start phase. 1190 More realistic fluctuating capacity patterns may be considered. The 1191 tester may choose to incorporate realistic scenarios with regards to 1192 common fluctuation of bandwidth in state-of-the-art technologies. 1194 The scenario consists of TCP NewReno flows between sender A and 1195 receiver B. To better assess the impact of draining rates on the AQM 1196 behavior, the tester must compare its performance with those of drop- 1197 tail and should provide a reference document for their proposal 1198 discussing performance and deployment compared to those of drop-tail. 1199 Burst traffic, such as presented in Section 7.2, could also be 1200 considered to assess the impact of varying available capacity on the 1201 burst absorption of the AQM. 1203 8.3. Parameter sensitivity and stability analysis 1205 The control law used by an AQM is the primary means by which the 1206 queuing delay is controlled. Hence understanding the control law is 1207 critical to understanding the behavior of the AQM scheme. The 1208 control law could include several input parameters whose values 1209 affect the AQM scheme's output behavior and its stability. 1210 Additionally, AQM schemes may auto-tune parameter values in order to 1211 maintain stability under different network conditions (such as 1212 different congestion levels, draining rates or network environments). 1213 The stability of these auto-tuning techniques is also important to 1214 understand. 1216 Transports operating under the control of AQM experience the effect 1217 of multiple control loops that react over different timescales. It 1218 is therefore important that proposed AQM schemes are seen to be 1219 stable when they are deployed at multiple points of potential 1220 congestion along an Internet path. The pattern of congestion signals 1221 (loss or ECN-marking) arising from AQM methods also need to not 1222 adversely interact with the dynamics of the transport protocols that 1223 they control. 1225 AQM proposals should provide background material showing control 1226 theoretic analysis of the AQM control law and the input parameter 1227 space within which the control law operates as expected; or could use 1228 another way to discuss the stability of the control law. For 1229 parameters that are auto-tuned, the material should include stability 1230 analysis of the auto-tuning mechanism(s) as well. Such analysis 1231 helps to understand an AQM control law better and the network 1232 conditions/deployments under which the AQM is stable. 1234 9. Various Traffic Profiles 1236 This section provides guidelines to assess the performance of an AQM 1237 proposal for various traffic profiles such as traffic with different 1238 applications or bi-directional traffic. 1240 9.1. Traffic mix 1242 This scenario can be used to evaluate how an AQM scheme reacts to a 1243 traffic mix consisting of different applications such as: 1245 o Bulk TCP transfer 1247 o Web traffic 1249 o VoIP 1251 o Constant Bit Rate (CBR) UDP traffic 1253 o Adaptive video streaming (either unidirectional or bidirectional) 1255 Various traffic mixes can be considered. These guidelines recommend 1256 to examine at least the following example: 1 bi-directional VoIP; 6 1257 Web pages download (such as detailed in Section 7.2); 1 CBR; 1 1258 Adaptive Video; 5 bulk TCP. Any other combinations could be 1259 considered and should be carefully documented. 1261 For each scenario, the graph described in Section 2.7 could be 1262 generated for each class of traffic. Metrics such as end-to-end 1263 latency, jitter and flow completion time may be reported. 1265 9.2. Bi-directional traffic 1267 Control packets such as DNS requests/responses, TCP SYNs/ACKs are 1268 small, but their loss can severely impact the application 1269 performance. The scenario proposed in this section will help in 1270 assessing whether the introduction of an AQM scheme increases the 1271 loss probability of these important packets. 1273 For this scenario, traffic must be generated in both downlink and 1274 uplink, such as defined in Section 3.1. The amount of asymmetry 1275 between the uplink and the downlink depends on the context. These 1276 guidelines recommend to consider a mild congestion level and the 1277 traffic presented in Section 8.2.2 in both directions. In this case, 1278 the metrics reported must be the same as in Section 8.2 for each 1279 direction. 1281 The traffic mix presented in Section 9.1 may also be generated in 1282 both directions. 1284 10. Example of multi-AQM scenario 1286 10.1. Motivation 1288 Transports operating under the control of AQM experience the effect 1289 of multiple control loops that react over different timescales. It 1290 is therefore important that proposed AQM schemes are seen to be 1291 stable when they are deployed at multiple points of potential 1292 congestion along an Internet path. The pattern of congestion signals 1293 (loss or ECN-marking) arising from AQM methods also need to not 1294 adversely interact with the dynamics of the transport protocols that 1295 they control. 1297 10.2. Details on the evaluation scenario 1299 +---------+ +-----------+ 1300 |senders A|---+ +---|receivers A| 1301 +---------+ | | +-----------+ 1302 +-----+---+ +---------+ +--+-----+ 1303 |Router L |--|Router M |--|Router R| 1304 |AQM A | |AQM M | |No AQM | 1305 +---------+ +--+------+ +--+-----+ 1306 +---------+ | | +-----------+ 1307 |senders B|-------------+ +---|receivers B| 1308 +---------+ +-----------+ 1310 Figure 3: Topology for the Multi-AQM scenario 1312 Figure Figure 3 describes topology options for evaluating multi-AQM 1313 scenarios. The AQM schemes are applied in sequence and impact the 1314 induced latency reduction, the induced goodput maximization and the 1315 trade-off between these two. Note that AQM schemes A and B 1316 introduced in Routers L and M could be (I) same scheme with identical 1317 parameter values, (ii) same scheme with different parameter values, 1318 or (iii) two different schemed. To best understand the interactions 1319 and implications, the mild congestion scenario as described in 1320 Section 8.2.2 is recommended such that the number of flows is equally 1321 shared among senders A and B. Other relevant combination of 1322 congestion levels could also be considered. We recommend to measure 1323 the metrics presented in Section 8.2. 1325 11. Implementation cost 1327 11.1. Motivation 1329 Successful deployment of AQM is directly related to its cost of 1330 implementation. Network devices can need hardware or software 1331 implementations of the AQM mechanism. Depending on a device's 1332 capabilities and limitations, the device may or may not be able to 1333 implement some or all parts of their AQM logic. 1335 AQM proposals should provide pseudo-code for the complete AQM scheme, 1336 highlighting generic implementation-specific aspects of the scheme 1337 such as "drop-tail" vs. "drop-head", inputs (e.g., current queuing 1338 delay, queue length), computations involved, need for timers, etc. 1339 This helps to identify costs associated with implementing the AQM 1340 scheme on a particular hardware or software device. This also 1341 facilitates discsusions around which kind of devices can easily 1342 support the AQM and which cannot. 1344 11.2. Recommended discussion 1346 AQM proposals should highlight parts of their AQM logic that are 1347 device dependent and discuss if and how AQM behavior could be 1348 impacted by the device. For example, a queueing-delay based AQM 1349 scheme requires current queuing delay as input from the device. If 1350 the device already maintains this value, then it can be trivial to 1351 implement the their AQM logic on the device. If the device provides 1352 indirect means to estimate the queuing delay (for example: 1353 timestamps, dequeuing rate), then the AQM behavior is sensitive to 1354 the precision of the queuing delay estimations are for that device. 1355 Highlighting the sensitivity of an AQM scheme to queuing delay 1356 estimations helps implementers to identify appropriate means of 1357 implementing the mechanism on a device. 1359 12. Operator Control and Auto-tuning 1361 12.1. Motivation 1363 One of the biggest hurdles of RED deployment was/is its parameter 1364 sensitivity to operating conditions -- how difficult it is to tune 1365 RED parameters for a deployment to achieve acceptable benefit from 1366 using RED. Fluctuating congestion levels and network conditions add 1367 to the complexity. Incorrect parameter values lead to poor 1368 performance. 1370 Any AQM scheme is likely to have parameters whose values affect the 1371 control law and behaviour of an AQM. Exposing all these parameters 1372 as control parameters to a network operator (or user) can easily 1373 result in a unsafe AQM deployment. Unexpected AQM behavior ensues 1374 when parameter values are set improperly. A minimal number of 1375 control parameters minimizes the number of ways a user can break a 1376 system where an AQM scheme is deployed at. Fewer control parameters 1377 make the AQM scheme more user-friendly and easier to deploy and 1378 debug. 1380 "AQM algorithms should not require tuning of initial or configuration 1381 parameters in common use cases." such as stated in the section 4.3 of 1382 the AQM recommendation document [RFC7567]. A scheme ought to expose 1383 only those parameters that control the macroscopic AQM behavior such 1384 as queue delay threshold, queue length threshold, etc. 1386 Additionally, the safety of an AQM scheme is directly related to its 1387 stability under varying operating conditions such as varying traffic 1388 profiles and fluctuating network conditions, as described in 1389 Section 8. Operating conditions vary often and hence the AQM needs 1390 to remain stable under these conditions without the need for 1391 additional external tuning. If AQM parameters require tuning under 1392 these conditions, then the AQM must self-adapt necessary parameter 1393 values by employing auto-tuning techniques. 1395 12.2. Recommended discussion 1397 In order to understand an AQM's deployment considerations and 1398 performance under a specific environment, AQM proposals should 1399 describe the parameters that control the macroscopic AQM behavior, 1400 and identify any parameters that require tuning to operational 1401 conditions. It could be interesting to also discuss that even if an 1402 AQM scheme may not adequately auto-tune its parameters, the resulting 1403 performance may not be optimal, but close to something reasonable. 1405 If there are any fixed parameters within the AQM, their setting 1406 should be discussed and justified, to help understand whether a fixed 1407 parameter value is applicable for a particular environment. 1409 If an AQM scheme is evaluated with parameter(s) that were externally 1410 tuned for optimization or other purposes, these values must be 1411 disclosed. 1413 13. Summary 1415 Figure 4 lists the scenarios and their requirements for an extended 1416 characterization of an AQM scheme. 1418 +------------------------------------------------------------------+ 1419 |Scenario |Sec. |Requirement | 1420 +------------------------------------------------------------------+ 1421 +------------------------------------------------------------------+ 1422 |Interaction with ECN | 4.5 |MUST be discussed if supported | 1423 +------------------------------------------------------------------+ 1424 |Interaction with Scheduling| 4.6 |Feasibility MUST be discussed | 1425 +------------------------------------------------------------------+ 1426 |Transport Protocols |5. | | 1427 | TCP-friendly sender | 5.1 |Scenario MUST be considered | 1428 | Aggressive sender | 5.2 |Scenario MUST be considered | 1429 | Unresponsive sender | 5.3 |Scenario MUST be considered | 1430 | LBE sender | 5.4 |Scenario MAY be considered | 1431 +------------------------------------------------------------------+ 1432 |Round Trip Time Fairness | 6.2 |Scenario MUST be considered | 1433 +------------------------------------------------------------------+ 1434 |Burst Absorption | 7.2 |Scenario MUST be considered | 1435 +------------------------------------------------------------------+ 1436 |Stability |8. | | 1437 | Varying congestion levels | 8.2.5|Scenario MUST be considered | 1438 | Varying available capacity| 8.2.6|Scenario MUST be considered | 1439 | Parameters and stability | 8.3 |This SHOULD be discussed | 1440 +------------------------------------------------------------------+ 1441 |Various Traffic Profiles |9. | | 1442 | Traffic mix | 9.1 |Scenario is RECOMMENDED | 1443 | Bi-directional traffic | 9.2 |Scenario MAY be considered | 1444 +------------------------------------------------------------------+ 1445 |Multi-AQM | 10.2 |Scenario MAY be considered | 1446 +------------------------------------------------------------------+ 1448 Figure 4: Summary of the scenarios and their requirements 1450 14. Acknowledgements 1452 This work has been partially supported by the European Community 1453 under its Seventh Framework Programme through the Reducing Internet 1454 Transport Latency (RITE) project (ICT-317700). 1456 Many thanks to S. Akhtar, A.B. Bagayoko, F. Baker, R. Bless, D. 1457 Collier-Brown, G. Fairhurst, J. Gettys, P. Goltsman, T. Hoiland- 1458 Jorgensen, K. Kilkki, C. Kulatunga, W. Lautenschlager, A.C. 1459 Morton, R. Pan, G. Skinner, D. Taht and M. Welzl for detailed and 1460 wise feedback on this document. 1462 15. IANA Considerations 1464 This memo includes no request to IANA. 1466 16. Security Considerations 1468 Some security considerations for AQM are identified in [RFC7567].This 1469 document, by itself, presents no new privacy nor security issues. 1471 17. References 1473 17.1. Normative References 1475 [RFC2119] Bradner, S., "Key words for use in RFCs to Indicate 1476 Requirement Levels", RFC 2119, 1997. 1478 [RFC2544] Bradner, S. and J. McQuaid, "Benchmarking Methodology for 1479 Network Interconnect Devices", RFC 2544, 1480 DOI 10.17487/RFC2544, March 1999, 1481 . 1483 [RFC2647] Newman, D., "Benchmarking Terminology for Firewall 1484 Performance", RFC 2647, DOI 10.17487/RFC2647, August 1999, 1485 . 1487 [RFC2679] Almes, G., Kalidindi, S., and M. Zekauskas, "A One-way 1488 Delay Metric for IPPM", RFC 2679, DOI 10.17487/RFC2679, 1489 September 1999, . 1491 [RFC2680] Almes, G., Kalidindi, S., and M. Zekauskas, "A One-way 1492 Packet Loss Metric for IPPM", RFC 2680, 1493 DOI 10.17487/RFC2680, September 1999, 1494 . 1496 [RFC5481] Morton, A. and B. Claise, "Packet Delay Variation 1497 Applicability Statement", RFC 5481, DOI 10.17487/RFC5481, 1498 March 2009, . 1500 [RFC7567] Baker, F., Ed. and G. Fairhurst, Ed., "IETF 1501 Recommendations Regarding Active Queue Management", 1502 BCP 197, RFC 7567, DOI 10.17487/RFC7567, July 2015, 1503 . 1505 17.2. Informative References 1507 [ANEL2014] 1508 Anelli, P., Diana, R., and E. Lochin, "FavorQueue: a 1509 Parameterless Active Queue Management to Improve TCP 1510 Traffic Performance", Computer Networks vol. 60, 2014. 1512 [BB2011] "BufferBloat: what's wrong with the internet?", ACM 1513 Queue vol. 9, 2011. 1515 [GONG2014] 1516 Gong, Y., Rossi, D., Testa, C., Valenti, S., and D. Taht, 1517 "Fighting the bufferbloat: on the coexistence of AQM and 1518 low priority congestion control", Computer Networks, 1519 Elsevier, 2014, 60, pp.115 - 128 , 2014. 1521 [HASS2008] 1522 Hassayoun, S. and D. Ros, "Loss Synchronization and Router 1523 Buffer Sizing with High-Speed Versions of TCP", IEEE 1524 INFOCOM Workshops , 2008. 1526 [HOEI2015] 1527 Hoeiland-Joergensen, T., McKenney, P., Taht, D., Gettys, 1528 J., and E. Dumazet, "FlowQueue-Codel", IETF (Work-in- 1529 Progress) , January 2015. 1531 [I-D.ietf-aqm-codel] 1532 Nichols, K., Jacobson, V., McGregor, A., and J. Iyengar, 1533 "Controlled Delay Active Queue Management", draft-ietf- 1534 aqm-codel-04 (work in progress), June 2016. 1536 [I-D.ietf-aqm-pie] 1537 Pan, R., Natarajan, P., Baker, F., and G. White, "PIE: A 1538 Lightweight Control Scheme To Address the Bufferbloat 1539 Problem", draft-ietf-aqm-pie-08 (work in progress), June 1540 2016. 1542 [I-D.ietf-tcpm-cubic] 1543 Rhee, I., Xu, L., Ha, S., Zimmermann, A., Eggert, L., and 1544 R. Scheffenegger, "CUBIC for Fast Long-Distance Networks", 1545 draft-ietf-tcpm-cubic-01 (work in progress), January 2016. 1547 [I-D.irtf-iccrg-tcpeval] 1548 Hayes, D., Ros, D., Andrew, L., and S. Floyd, "Common TCP 1549 Evaluation Suite", draft-irtf-iccrg-tcpeval-01 (work in 1550 progress), July 2014. 1552 [JAY2006] Jay, P., Fu, Q., and G. Armitage, "A preliminary analysis 1553 of loss synchronisation between concurrent TCP flows", 1554 Australian Telecommunication Networks and Application 1555 Conference (ATNAC) , 2006. 1557 [MORR2000] 1558 Morris, R., "Scalable TCP congestion control", IEEE 1559 INFOCOM , 2000. 1561 [RFC0793] Postel, J., "Transmission Control Protocol", STD 7, 1562 RFC 793, DOI 10.17487/RFC0793, September 1981, 1563 . 1565 [RFC2309] Braden, B., Clark, D., Crowcroft, J., Davie, B., Deering, 1566 S., Estrin, D., Floyd, S., Jacobson, V., Minshall, G., 1567 Partridge, C., Peterson, L., Ramakrishnan, K., Shenker, 1568 S., Wroclawski, J., and L. Zhang, "Recommendations on 1569 Queue Management and Congestion Avoidance in the 1570 Internet", RFC 2309, April 1998. 1572 [RFC2488] Allman, M., Glover, D., and L. Sanchez, "Enhancing TCP 1573 Over Satellite Channels using Standard Mechanisms", 1574 BCP 28, RFC 2488, DOI 10.17487/RFC2488, January 1999, 1575 . 1577 [RFC3168] Ramakrishnan, K., Floyd, S., and D. Black, "The Addition 1578 of Explicit Congestion Notification (ECN) to IP", 1579 RFC 3168, DOI 10.17487/RFC3168, September 2001, 1580 . 1582 [RFC3611] Friedman, T., Ed., Caceres, R., Ed., and A. Clark, Ed., 1583 "RTP Control Protocol Extended Reports (RTCP XR)", 1584 RFC 3611, DOI 10.17487/RFC3611, November 2003, 1585 . 1587 [RFC5348] Floyd, S., Handley, M., Padhye, J., and J. Widmer, "TCP 1588 Friendly Rate Control (TFRC): Protocol Specification", 1589 RFC 5348, DOI 10.17487/RFC5348, September 2008, 1590 . 1592 [RFC5681] Allman, M., Paxson, V., and E. Blanton, "TCP Congestion 1593 Control", RFC 5681, DOI 10.17487/RFC5681, September 2009, 1594 . 1596 [RFC6297] Welzl, M. and D. Ros, "A Survey of Lower-than-Best-Effort 1597 Transport Protocols", RFC 6297, DOI 10.17487/RFC6297, June 1598 2011, . 1600 [RFC6817] Shalunov, S., Hazel, G., Iyengar, J., and M. Kuehlewind, 1601 "Low Extra Delay Background Transport (LEDBAT)", RFC 6817, 1602 DOI 10.17487/RFC6817, December 2012, 1603 . 1605 [RFC7141] Briscoe, B. and J. Manner, "Byte and Packet Congestion 1606 Notification", RFC 7141, 2014. 1608 [TRAN2014] 1609 Trang, S., Kuhn, N., Lochin, E., Baudoin, C., Dubois, E., 1610 and P. Gelard, "On The Existence Of Optimal LEDBAT 1611 Parameters", IEEE ICC 2014 - Communication QoS, 1612 Reliability and Modeling Symposium , 2014. 1614 [WELZ2015] 1615 Welzl, M. and G. Fairhurst, "The Benefits to Applications 1616 of using Explicit Congestion Notification (ECN)", IETF 1617 (Work-in-Progress) , June 2015. 1619 [WINS2014] 1620 Winstein, K., "Transport Architectures for an Evolving 1621 Internet", PhD thesis, Massachusetts Institute of 1622 Technology , 2014. 1624 Authors' Addresses 1626 Nicolas Kuhn (editor) 1627 CNES, Telecom Bretagne 1628 18 avenue Edouard Belin 1629 Toulouse 31400 1630 France 1632 Phone: +33 5 61 27 32 13 1633 Email: nicolas.kuhn@cnes.fr 1634 Preethi Natarajan (editor) 1635 Cisco Systems 1636 510 McCarthy Blvd 1637 Milpitas, California 1638 United States 1640 Email: prenatar@cisco.com 1642 Naeem Khademi (editor) 1643 University of Oslo 1644 Department of Informatics, PO Box 1080 Blindern 1645 N-0316 Oslo 1646 Norway 1648 Phone: +47 2285 24 93 1649 Email: naeemk@ifi.uio.no 1651 David Ros 1652 Simula Research Laboratory AS 1653 P.O. Box 134 1654 Lysaker, 1325 1655 Norway 1657 Phone: +33 299 25 21 21 1658 Email: dros@simula.no