idnits 2.17.1 draft-ietf-tsvwg-nqb-07.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 (28 July 2021) is 1003 days in the past. Is this intentional? Checking references for intended status: Proposed Standard ---------------------------------------------------------------------------- (See RFCs 3967 and 4897 for information about using normative references to lower-maturity documents in RFCs) ** Downref: Normative reference to an Informational RFC: RFC 2983 == Outdated reference: A later version (-07) exists of draft-briscoe-docsis-q-protection-00 == Outdated reference: A later version (-25) exists of draft-ietf-tsvwg-aqm-dualq-coupled-16 == Outdated reference: A later version (-13) exists of draft-ietf-tsvwg-dscp-considerations-00 == Outdated reference: A later version (-29) exists of draft-ietf-tsvwg-ecn-l4s-id-19 == Outdated reference: A later version (-20) exists of draft-ietf-tsvwg-l4s-arch-10 Summary: 1 error (**), 0 flaws (~~), 6 warnings (==), 1 comment (--). Run idnits with the --verbose option for more detailed information about the items above. -------------------------------------------------------------------------------- 2 Transport Area Working Group G. White 3 Internet-Draft CableLabs 4 Intended status: Standards Track T. Fossati 5 Expires: 29 January 2022 ARM 6 28 July 2021 8 A Non-Queue-Building Per-Hop Behavior (NQB PHB) for Differentiated 9 Services 10 draft-ietf-tsvwg-nqb-07 12 Abstract 14 This document specifies properties and characteristics of a Non- 15 Queue-Building Per-Hop Behavior (NQB PHB). The purpose of this NQB 16 PHB is to provide a separate queue that enables smooth, low-data- 17 rate, application-limited traffic flows, which would ordinarily share 18 a queue with bursty and capacity-seeking traffic, to avoid the 19 latency, latency variation and loss caused by such traffic. This PHB 20 is implemented without prioritization and without rate policing, 21 making it suitable for environments where the use of either these 22 features may be restricted. The NQB PHB has been developed primarily 23 for use by access network segments, where queuing delays and queuing 24 loss caused by Queue-Building protocols are manifested, but its use 25 is not limited to such segments. In particular, applications to 26 cable broadband links, Wi-Fi links, and mobile network radio and core 27 segments are discussed. This document recommends a specific 28 Differentiated Services Code Point (DSCP) to identify Non-Queue- 29 Building flows. 31 Status of This Memo 33 This Internet-Draft is submitted in full conformance with the 34 provisions of BCP 78 and BCP 79. 36 Internet-Drafts are working documents of the Internet Engineering 37 Task Force (IETF). Note that other groups may also distribute 38 working documents as Internet-Drafts. The list of current Internet- 39 Drafts is at https://datatracker.ietf.org/drafts/current/. 41 Internet-Drafts are draft documents valid for a maximum of six months 42 and may be updated, replaced, or obsoleted by other documents at any 43 time. It is inappropriate to use Internet-Drafts as reference 44 material or to cite them other than as "work in progress." 46 This Internet-Draft will expire on 29 January 2022. 48 Copyright Notice 50 Copyright (c) 2021 IETF Trust and the persons identified as the 51 document authors. All rights reserved. 53 This document is subject to BCP 78 and the IETF Trust's Legal 54 Provisions Relating to IETF Documents (https://trustee.ietf.org/ 55 license-info) in effect on the date of publication of this document. 56 Please review these documents carefully, as they describe your rights 57 and restrictions with respect to this document. Code Components 58 extracted from this document must include Simplified BSD License text 59 as described in Section 4.e of the Trust Legal Provisions and are 60 provided without warranty as described in the Simplified BSD License. 62 Table of Contents 64 1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . 3 65 2. Requirements Language . . . . . . . . . . . . . . . . . . . . 4 66 3. Context . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 67 3.1. Non-Queue-Building Behavior . . . . . . . . . . . . . . . 4 68 3.2. Relationship to the Diffserv Architecture . . . . . . . . 4 69 3.3. Relationship to L4S . . . . . . . . . . . . . . . . . . . 6 70 4. DSCP Marking of NQB Traffic . . . . . . . . . . . . . . . . . 6 71 4.1. Non-Queue-Building Sender Requirements . . . . . . . . . 6 72 4.2. Aggregation of the NQB DSCP with other Diffserv PHBs . . 7 73 4.3. End-to-end usage and DSCP Re-marking . . . . . . . . . . 8 74 4.4. The NQB DSCP and Tunnels . . . . . . . . . . . . . . . . 9 75 5. Non-Queue-Building PHB Requirements . . . . . . . . . . . . . 9 76 5.1. Primary Requirements . . . . . . . . . . . . . . . . . . 10 77 5.2. Traffic Protection . . . . . . . . . . . . . . . . . . . 11 78 6. Impact on Higher Layer Protocols . . . . . . . . . . . . . . 12 79 7. Configuration and Management . . . . . . . . . . . . . . . . 12 80 8. Example Use Cases . . . . . . . . . . . . . . . . . . . . . . 12 81 8.1. DOCSIS Access Networks . . . . . . . . . . . . . . . . . 12 82 8.2. Mobile Networks . . . . . . . . . . . . . . . . . . . . . 13 83 8.3. WiFi Networks . . . . . . . . . . . . . . . . . . . . . . 13 84 8.3.1. Interoperability with Existing WiFi Networks . . . . 14 85 9. Acknowledgements . . . . . . . . . . . . . . . . . . . . . . 15 86 10. IANA Considerations . . . . . . . . . . . . . . . . . . . . . 15 87 11. Security Considerations . . . . . . . . . . . . . . . . . . . 15 88 12. References . . . . . . . . . . . . . . . . . . . . . . . . . 16 89 12.1. Normative References . . . . . . . . . . . . . . . . . . 16 90 12.2. Informative References . . . . . . . . . . . . . . . . . 17 91 Appendix A. DSCP Remarking Pathologies . . . . . . . . . . . . . 19 92 Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . 20 94 1. Introduction 96 This document defines a Differentiated Services per-hop behavior 97 (PHB) called "Non-Queue-Building Per-Hop Behavior" (NQB PHB), which 98 isolates traffic flows that are relatively low data rate and that do 99 not themselves materially contribute to queueing delay and loss, 100 allowing them to avoid the queuing delays and losses caused by other 101 traffic. Such Non-Queue-Building flows (for example: interactive 102 voice, gaming, machine-to-machine applications) are application 103 limited flows that are distinguished from traffic flows managed by an 104 end-to-end congestion control algorithm. 106 The vast majority of packets that are carried by broadband access 107 networks are managed by an end-to-end congestion control algorithm, 108 such as Reno, Cubic or BBR. These congestion control algorithms 109 attempt to seek the available capacity of the end-to-end path (which 110 can frequently be the access network link capacity), and in doing so 111 generally overshoot the available capacity, causing a queue to build- 112 up at the bottleneck link. This queue build up results in queuing 113 delay (variable latency) and possibly packet loss that can affect all 114 of the applications that are sharing the bottleneck link. 116 In contrast to traditional congestion-controlled applications, there 117 are a variety of relatively low data rate applications that do not 118 materially contribute to queueing delay and loss, but are nonetheless 119 subjected to it by sharing the same bottleneck link in the access 120 network. Many of these applications may be sensitive to latency or 121 latency variation, as well as packet loss, and thus produce a poor 122 quality of experience in such conditions. 124 Active Queue Management (AQM) mechanisms (such as PIE [RFC8033], 125 DOCSIS-PIE [RFC8034], or CoDel [RFC8289]) can improve the quality of 126 experience for latency sensitive applications, but there are 127 practical limits to the amount of improvement that can be achieved 128 without impacting the throughput of capacity-seeking applications. 129 For example, AQMs generally allow a significant amount of queue depth 130 variation in order to accommodate the behaviors of congestion control 131 algorithms such as Reno and Cubic. If the AQM attempted to control 132 the queue much more tightly, applications using those algorithms 133 would not perform well. Alternatively, flow queueuing systems, such 134 as fq_codel [RFC8290] can be employed to isolate flows from one 135 another, but these are not appropriate for all bottleneck links, due 136 to complexity or other reasons. 138 The NQB PHB supports differentiating between these two classes of 139 traffic in bottleneck links and queuing them separately in order that 140 both classes can deliver satisfactory quality of experience for their 141 applications. 143 To be clear, a network implementing the NQB PHB solely provides 144 isolation for traffic classified as behaving in conformance with the 145 NQB DSCP (and optionally enforces that behavior). It is the NQB 146 senders' behavior itself which results in low latency and low loss. 148 2. Requirements Language 150 The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", 151 "SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and 152 "OPTIONAL" in this document are to be interpreted as described in BCP 153 14 [RFC2119] [RFC8174] when, and only when, they appear in all 154 capitals, as shown here. 156 3. Context 158 3.1. Non-Queue-Building Behavior 160 There are many applications that send traffic at relatively low data 161 rates and/or in a fairly smooth and consistent manner such that they 162 are highly unlikely to exceed the available capacity of the network 163 path between source and sink. These applications may themselves only 164 cause very small, transient queues to form in network buffers, but 165 nonetheless they can be subjected to packet delay and delay variation 166 as a result of sharing a network buffer with applications that tend 167 to cause large and/or standing queues to form. Many of these 168 applications are negatively affected by excessive packet delay and 169 delay variation. Such applications are ideal candidates to be queued 170 separately from the applications that are the cause of queue buildup, 171 latency and loss. 173 In contrast, Queue-building (QB) flows include those that use TCP or 174 QUIC, with Cubic, Reno or other TCP congestion control algorithms 175 that probe for the link capacity and induce latency and loss as a 176 result. Other types of QB flows include those that frequently send 177 at a high burst rate (e.g. several consecutive packets sent well in 178 excess of 1 Mbps) even if the long-term average data rate is much 179 lower. 181 3.2. Relationship to the Diffserv Architecture 183 The IETF has defined the Differentiated Services architecture 184 [RFC2475] with the intention that it allows traffic to be marked in a 185 manner that conveys the performance requirements of that traffic 186 either quantitatively or in a relative sense (i.e. priority). The 187 architecture defines the use of the Diffserv field [RFC2474] for this 188 purpose, and numerous RFCs have been written that describe 189 recommended interpretations of the values (Diffserv Code Points) of 190 the field, and standardized treatments (traffic conditioning and per- 191 hop-behaviors) that can be implemented to satisfy the performance 192 requirements of traffic so marked. 194 While this architecture is powerful, and can be configured to meet 195 the performance requirements of a variety of applications and traffic 196 categories, or to achieve differentiated service offerings, it has 197 proven problematic to enable its use for these purposes end-to-end 198 across the Internet. 200 This difficulty is in part due to the fact that meeting (in an end- 201 to-end context) the performance requirements of an application 202 involves all of the networks in the path agreeing on what those 203 requirements are, and sharing an interest in meeting them. In many 204 cases this is made more difficult due to the fact that the 205 performance "requirements" are not strict ones (e.g. applications 206 will degrade in some manner as loss/latency/jitter increase), so the 207 importance of meeting them for any particular application in some 208 cases involves a judgment as to the value of avoiding some amount of 209 degradation in quality for that application in exchange for an 210 increase in the degradation of another application. 212 Further, in many cases the implementation of Diffserv PHBs has 213 historically involved prioritization of service classes with respect 214 to one another, which sets up the zero-sum game alluded to in the 215 previous paragraph, and results in the need to limit access to higher 216 priority classes via mechanisms such as access control, admission 217 control, traffic conditioning and rate policing, and/or to meter and 218 bill for carriage of such traffic. These mechanisms can be difficult 219 or impossible to implement in an end-to-end context. 221 Finally, some jurisdictions impose regulations that limit the ability 222 of networks to provide differentiation of services, in large part 223 based on the belief that doing so necessarily involves prioritization 224 or privileged access to bandwidth, and thus a benefit to one class of 225 traffic always comes at the expense of another. 227 In contrast, the NQB PHB has been designed with the goal that it 228 avoids many of these issues, and thus could conceivably be deployed 229 end-to-end across the Internet. The intent of the NQB DSCP is that 230 it signals verifiable behavior rather than simply a desire for 231 differentiated treatment. Also, the NQB traffic is to be given a 232 separate queue with priority equal to default traffic, and given no 233 reserved bandwidth other than the bandwidth that it shares with 234 default traffic. As a result, the NQB PHB does not aim to meet 235 specific application performance requirements. Instead the goal of 236 the NQB PHB is to provide statistically better loss, latency, and 237 jitter performance for traffic that is itself only an insignificant 238 contributor to those degradations. The PHB is also designed to 239 minimize any incentives for a sender to mismark its traffic, since 240 neither higher priority nor reserved bandwith are being offered. 241 These attributes eliminate many of the tradeoffs that underlie the 242 handling of differentiated service classes in the Diffserv 243 architecture as it has traditionally been defined. They also 244 significantly simplify access control and admission control 245 functions, reducing them to simple verification of behavior. 247 3.3. Relationship to L4S 249 The NQB DSCP and PHB described in this draft have been defined to 250 operate independently of the experimental L4S Architecture 251 [I-D.ietf-tsvwg-l4s-arch]. Nonetheless, the NQB traffic flows are 252 intended to be compatible with [I-D.ietf-tsvwg-l4s-arch], with the 253 result being that NQB traffic and L4S traffic can share the low- 254 latency queue in an L4S DualQ node 255 [I-D.ietf-tsvwg-aqm-dualq-coupled]. Compliance with the DualQ 256 Coupled AQM requirements (Section 2.5 of 257 [I-D.ietf-tsvwg-aqm-dualq-coupled]) is considered sufficient to 258 support the NQB PHB requirement of fair allocation of bandwidth 259 between the QB and NQB queues (Section 5). 261 4. DSCP Marking of NQB Traffic 263 4.1. Non-Queue-Building Sender Requirements 265 Non-queue-building (NQB) flows are typically UDP flows that don't 266 seek the maximum capacity of the link (examples: online games, voice 267 chat, DNS lookups, real-time IoT analytics data). Here the data rate 268 is limited by the application itself rather than by network capacity 269 - these applications send, at most, the equivalent of a few well- 270 spaced packets per RTT, even if the packets are not actually RTT- 271 clocked. In today's network this corresponds to an instantaneous 272 data rate (packet size divided by packet inter-arrival time) of no 273 more than about 1 Mbps (e.g. no more than one 1250 B packet every 10 274 ms), but there is no precise bound since it depends on the conditions 275 in which the application is operating. 277 Note that, while such flows ordinarily don't implement a traditional 278 congestion control mechanism, they nonetheless are expected to comply 279 with existing guidance for safe deployment on the Internet, for 280 example the requirements in [RFC8085] and Section 2 of [RFC3551] 281 (also see the circuit breaker limits in Section 4.3 of [RFC8083] and 282 the description of inelastic pseudowires in Section 4 of [RFC7893]). 283 To be clear, the description of NQB flows in this document should not 284 be interpreted as suggesting that such flows are in any way exempt 285 from this responsibility. 287 Applications that align with the description of NQB behavior in the 288 preceding paragraphs SHOULD identify themselves to the network using 289 a Diffserv Code Point (DSCP) of 45 (decimal) so that their packets 290 can be queued separately from QB flows. The choice of the value 45 291 is motivated in part by the desire to achieve separate queuing in 292 existing WiFi networks (see Section 8.3). In networks where another 293 (e.g. a local-use) codepoint is designated for NQB traffic, or where 294 specialized PHBs are available that can meet specific application 295 requirements (e.g. a guaranteed-latency path for voice traffic), it 296 may be preferred to use another DSCP. 298 If the application's traffic exceeds more than a few packets per RTT, 299 or exceeds approximately 1 Mbps on an instantaneous (inter-packet) 300 basis, the application SHOULD NOT mark its traffic with the NQB DSCP. 301 In such a case, the application has to instead implement a relevant 302 congestion control mechanism, for example as described in Section 3.1 303 of [RFC8085] or [I-D.ietf-tsvwg-ecn-l4s-id]. 305 4.2. Aggregation of the NQB DSCP with other Diffserv PHBs 307 It is RECOMMENDED that networks and nodes that do not support the NQB 308 PHB be configured to treat NQB marked traffic the same as traffic 309 marked "Default". It is additionally RECOMMENDED that such networks 310 and nodes simply classify the NQB DSCP into the same treatment 311 aggregate as Default traffic, or encapsulate the NQB marked packet, 312 rather than re-marking NQB traffic as Default. This preservation of 313 the NQB marking enables hops further along the path to provide the 314 NQB PHB successfully. 316 In backbone and core network switches (particularly if shallow- 317 buffered), and nodes that do not typically experience congestion, 318 treating NQB marked traffic the same as Default may be sufficient to 319 preserve loss/latency/jitter performance for NQB traffic. In other 320 nodes, treating NQB marked traffic as Default could result in 321 degradation of loss/latency/jitter performance but is recommended 322 nonetheless in order to preserve the incentives described in 323 Section 5. An alternative, in controlled environments where there is 324 no risk of mismarking of traffic, would be to aggregate NQB marked 325 traffic with real-time, latency sensitive traffic. Similarly, 326 networks and nodes that aggregate service classes as discussed in 327 [RFC5127] and [RFC8100] may not be able to provide a PDB/PHB that 328 meets the requirements of this document. In these cases it is 329 RECOMMENDED that NQB-marked traffic be aggregated into the Elastic 330 Treatment Aggregate (for [RFC5127] networks) or the Default / Elastic 331 Treatment Aggregate (for [RFC8100] networks), although in some cases 332 a network operator may instead choose to aggregate NQB traffic into 333 the (Bulk) Real-Time Treatment Aggregate. Either approach comes with 334 trade-offs: when the aggregated traffic encounters a bottleneck, 335 aggregating with Default/Elastic traffic could result in a 336 degradation of loss/latency/jitter performance for NQB traffic, while 337 aggregating with Real-Time (assuming such traffic is provided a 338 prioritized PHB) risks creating an incentive for mismarking of non- 339 compliant traffic as NQB (except in controlled environments). In 340 either case, the NQB DSCP SHOULD be preserved (possibly via 341 encapsulation) in order to limit the negative impact that such 342 networks would have on end-to-end performance for NQB traffic. This 343 aligns with recommendations in [RFC5127]. 345 Nodes that support the NQB PHB may choose to aggregate other service 346 classes into the NQB queue. Candidate service classes for this 347 aggregation would include those that carry inelastic traffic that has 348 low to very-low tolerance for loss, latency and/or jitter as 349 discussed in [RFC4594]. These could include Telephony (EF/VA), 350 Signaling (CS5), Real-Time Interactive (CS4) and Broadcast Video 351 (CS3). 353 4.3. End-to-end usage and DSCP Re-marking 355 In contrast to some existing standard PHBs, many of which are 356 typically only meaningful within a Diffserv Domain (e.g. an AS or an 357 enterprise network), this PHB is expected to be used end-to-end 358 across the Internet, wherever suitable operator agreements apply. 359 Under the [RFC2474] model, this requires that the corresponding DSCP 360 is recognized by all operators and mapped across their boundaries 361 accordingly. 363 To support NQB, networks MUST preserve a DSCP marking distinction 364 between NQB traffic and Default traffic when forwarding via an 365 interconnect from or to another network. To facilitate the default 366 treatment of NQB traffic in backbones and core networks discussed in 367 the previous section (where IP Precedence may be deployed), networks 368 that support NQB SHOULD remap NQB traffic (DSCP 45) to DSCP 5 prior 369 to interconnection, unless agreed otherwise between the 370 interconnecting partners. The fact that this PHB is intended for 371 end-to-end usage does not preclude networks from mapping the NQB DSCP 372 to a value other than 45 or 5 for internal usage, as long as the 373 appropriate NQB DSCP is restored when forwarding to another network. 374 Additionally, interconnecting networks are not precluded from 375 negotiating (via an SLA or some other agreement) a different DSCP to 376 use to signal NQB across the interconnect. 378 Furthermore, in other network environments where IP Precedence is 379 deployed, it is RECOMMENDED that the network operator re-mark NQB 380 traffic to DSCP 5 in order to ensure that it is aggregated with 381 Default traffic. 383 In order to enable interoperability with WiFi equipment as described 384 in Section 8.3.1, networks SHOULD re-mark NQB traffic (e.g. DSCP 5) 385 to DSCP 45 prior to a customer access link, subject to the safeguards 386 described in that section. 388 Thus, this document recommends two DSCPs to designate NQB, the value 389 45 for use by hosts and in WiFi networks, and the value 5 for use 390 across network interconnections. 392 4.4. The NQB DSCP and Tunnels 394 [RFC2983] discusses tunnel models that support Diffserv. It 395 describes a "uniform model" in which the inner DSCP is copied to the 396 outer header at encapsulation, and the outer DSCP is copied to the 397 inner header at decapsulation. It also describes a "pipe model" in 398 which the outer DSCP is not copied to the inner header at 399 decapsulation. Both models can be used in conjunction with the NQB 400 PHB. In the case of the pipe model, any DSCP manipulation (re- 401 marking) of the outer header by intermediate nodes would be discarded 402 at tunnel egress, potentially improving the possibility of achieving 403 NQB treatment in subsequent nodes. 405 As is discussed in [RFC2983], tunnel protocols that are sensitive to 406 reordering can result in undesirable interactions if multiple DSCP 407 PHBs are signaled for traffic within a tunnel instance. This is true 408 for NQB marked traffic as well. If a tunnel contains a mix of QB and 409 NQB traffic, and this is reflected in the outer DSCP in a network 410 that supports the NQB PHB, it would be necessary to avoid a 411 reordering-sensitive tunnel protocol. 413 5. Non-Queue-Building PHB Requirements 415 It is worthwhile to note again that the NQB designation and marking 416 is intended to convey verifiable traffic behavior, as opposed to 417 simply a desire for differentiated treatment. Also, it is important 418 that incentives are aligned correctly, i.e. that there is a benefit 419 to the application in marking its packets correctly, and a 420 disadvantage (or at least no benefit) to an application in 421 intentionally mismarking its traffic. Thus, a useful property of 422 nodes (i.e. network switches and routers) that support separate 423 queues for NQB and QB flows is that for NQB flows, the NQB queue 424 provides better performance than the QB queue; and for QB flows, the 425 QB queue provides better performance than the NQB queue (this is 426 discussed further in this section and Section 11). By adhering to 427 these principles, there is no incentive for senders to mismark their 428 traffic as NQB, and further, any mismarking can be identified by the 429 network. 431 5.1. Primary Requirements 433 A node supporting the NQB PHB makes no guarantees on latency or data 434 rate for NQB marked flows, but instead aims to provide a bound on 435 queuing delay for as many such marked flows as it can, and shed load 436 when needed. 438 A node supporting the NQB PHB MUST provide a queue for non-queue- 439 building traffic separate from any queue used for queue-building 440 traffic. 442 NQB traffic, in aggregate, SHOULD NOT be rate limited or rate policed 443 separately from queue-building traffic of equivalent importance. 445 The NQB queue SHOULD be given equivalent forwarding preference 446 compared to queue-building traffic of equivalent importance. The 447 node SHOULD provide a scheduler that allows QB and NQB traffic of 448 equivalent importance to share the link in a fair manner, e.g. a 449 deficit round-robin scheduler with equal weights. Compliance with 450 these recommendations helps to ensure that there are no incentives 451 for QB traffic to be mismarked as NQB. In environments where 452 mismarking is not a potential issue (e.g. a network where a marking 453 policy is enforced by other means), these requirements may not be 454 necessary. 456 A node supporting the NQB PHB SHOULD treat traffic marked as Default 457 (DSCP=0) as QB traffic having equivalent importance to the NQB marked 458 traffic. A node supporting the NQB DSCP MUST support the ability to 459 configure the classification criteria that are used to identify QB 460 and NQB traffic of equivalent importance. 462 The NQB queue SHOULD have a buffer size that is significantly smaller 463 than the buffer provided for QB traffic (e.g. single-digit 464 milliseconds). It is expected that most QB traffic is engineered to 465 work well when the network provides a relatively deep buffer (e.g. on 466 the order of tens or hundreds of ms) in nodes where support for the 467 NQB PHB is advantageous (i.e. bottleneck nodes). Providing a 468 similarly deep buffer for the NQB queue would be at cross purposes to 469 providing very low queueing delay, and would erode the incentives for 470 QB traffic to be marked correctly. 472 5.2. Traffic Protection 474 It is possible that due to an implementation error or 475 misconfiguration, a QB flow would end up getting mismarked as NQB, or 476 vice versa. In the case of an NQB flow that isn't marked as NQB and 477 ends up in the QB queue, it would only impact its own quality of 478 service, and so it seems to be of lesser concern. However, a QB flow 479 that is mismarked as NQB would cause queuing delays and/or loss for 480 all of the other flows that are sharing the NQB queue. 482 To prevent this situation from harming the performance of the real 483 NQB flows, network elements that support differentiating NQB traffic 484 SHOULD support a "traffic protection" function that can identify QB 485 flows that are mismarked as NQB, and either reclassify those flows/ 486 packets to the QB queue or discard the offending traffic. Such a 487 function SHOULD be implemented in an objective and verifiable manner, 488 basing its decisions upon the behavior of the flow rather than on 489 application-layer constructs. It may be advantageous for a traffic 490 protection function to employ hysteresis to prevent borderline flows 491 from being reclassified capriciously. 493 One example traffic protection algorithm can be found in 494 [I-D.briscoe-docsis-q-protection]. 496 There are some situations where such function may not be necessary. 497 For example, a network element designed for use in controlled 498 environments (e.g. enterprise LAN) may not require a traffic 499 protection function. Additionally, some networks may prefer to 500 police the application of the NQB DSCP at the ingress edge, so that 501 in-network traffic protection is not needed. 503 6. Impact on Higher Layer Protocols 505 Network elements that support the NQB PHB and that support traffic 506 protection as discussed in the previous section introduce the 507 possibility that flows classified into the NQB queue could experience 508 out of order delivery or packet loss if their behavior is not 509 consistent with NQB. This is particularly true if the traffic 510 protection algorithm makes decisions on a packet-by-packet basis. In 511 this scenario, a flow that is (mis)marked as NQB and that causes a 512 queue to form in this bottleneck link could see some of its packets 513 forwarded by the NQB queue, and some of them either discarded or 514 redirected to the QB queue. In the case of redirection, depending on 515 the queueing latency and scheduling within the network element, this 516 could result in packets being delivered out of order. As a result, 517 the use of the NQB DSCP by a higher layer protocol carries some risk 518 that an increased amount of out of order delivery or packet loss will 519 be experienced. This characteristic provides one disincentive for 520 mis-marking of traffic. 522 7. Configuration and Management 524 As required above, nodes supporting the NQB PHB provide for the 525 configuration of classifiers that can be used to differentiate 526 between QB and NQB traffic of equivalent importance. The default for 527 such classifiers is recommended to be the assigned NQB DSCP (to 528 identify NQB traffic) and the Default (0) DSCP (to identify QB 529 traffic). 531 8. Example Use Cases 533 8.1. DOCSIS Access Networks 535 Residential cable broadband Internet services are commonly configured 536 with a single bottleneck link (the access network link) upon which 537 the service definition is applied. The service definition, typically 538 an upstream/downstream data rate tuple, is implemented as a 539 configured pair of rate shapers that are applied to the user's 540 traffic. In such networks, the quality of service that each 541 application receives, and as a result, the quality of experience that 542 it generates for the user is influenced by the characteristics of the 543 access network link. 545 To support the NQB PHB, cable broadband services MUST be configured 546 to provide a separate queue for NQB marked traffic. The NQB queue 547 MUST be configured to share the service's rate shaped bandwidth with 548 the queue for QB traffic. 550 8.2. Mobile Networks 552 Historically, 3GPP mobile networks have utilised "bearers" to 553 encapsulate each user's user plane traffic through the radio and core 554 networks. A "dedicated bearer" may be allocated a Quality of Service 555 (QoS) to apply any prioritisation to its flows at queues and radio 556 schedulers. Typically an LTE operator provides a dedicated bearer 557 for IMS VoLTE (Voice over LTE) traffic, which is prioritised in order 558 to meet regulatory obligations for call completion rates; and a "best 559 effort" default bearer, for Internet traffic. The "best effort" 560 bearer provides no guarantees, and hence its buffering 561 characteristics are not compatible with low-latency traffic. The 5G 562 radio and core systems offer more flexibility over bearer allocation, 563 meaning bearers can be allocated per traffic type (e.g. loss- 564 tolerant, low-latency etc.) and hence support more suitable treatment 565 of Internet real-time flows. 567 To support the NQB PHB, the mobile network SHOULD be configured to 568 give UEs a dedicated, low-latency, non-GBR, EPS bearer, e.g. one with 569 QCI 7, in addition to the default EPS bearer; or a Data Radio Bearer 570 with 5QI 7 in a 5G system (see Table 5.7.4-1: Standardized 5QI to QoS 571 characteristics mapping in [SA-5G]). 573 A packet carrying the NQB DSCP SHOULD be routed through the dedicated 574 low-latency EPS bearer. A packet that has no associated NQB marking 575 SHOULD NOT be routed through the dedicated low-latency EPS bearer. 577 8.3. WiFi Networks 579 WiFi networking equipment compliant with 802.11e/n/ac/ax [IEEE802-11] 580 generally supports either four or eight transmit queues and four sets 581 of associated Enhanced Multimedia Distributed Control Access (EDCA) 582 parameters (corresponding to the four WiFi Multimedia (WMM) Access 583 Categories) that are used to enable differentiated media access 584 characteristics. As discussed in [RFC8325], most existing WiFi 585 implementations use a default DSCP to User Priority mapping that 586 utilizes the most significant three bits of the Diffserv Field to 587 select "User Priority" which is then mapped to the four WMM Access 588 Categories. [RFC8325] also provides an alternative mapping that more 589 closely aligns with the DSCP recommendations provided by the IETF. 591 In addition to the requirements provided in other sections of this 592 document, to support the NQB PHB, WiFi equipment SHOULD map the NQB 593 codepoint 45 into a separate queue in the same Access Category as the 594 queue that carries default traffic (i.e. the Best Effort Access 595 Category). 597 8.3.1. Interoperability with Existing WiFi Networks 599 While some existing WiFi equipment may be capable (in some cases via 600 firmware update) of supporting the NQB PHB requirements, many 601 currently deployed devices cannot be configured in this way. As a 602 result the remainder of this section discusses interoperability with 603 these existing WiFi networks, as opposed to PHB compliance. 605 In order to increase the likelihood that NQB traffic is provided a 606 separate queue from QB traffic in existing WiFi equipment that uses 607 the default mapping, the 45 code point is recommended for NQB. This 608 maps NQB to UP_5 which is in the "Video" Access Category. While this 609 DSCP to User Priority mapping enables these WiFi systems to support 610 the NQB PHB requirement for segregated queuing, it does not support 611 the remaining NQB PHB requirements in Section 5. The ramifications 612 of, and remedies for this are discussed further below. 614 Existing WiFi devices are unlikely to support a traffic protection 615 algorithm, so traffic mismarked as NQB is not likely to be detected 616 and remedied by such devices. 618 Furthermore, in their default configuration, existing WiFi devices 619 utilize EDCA parameters that result in statistical prioritization of 620 the "Video" Access Category above the "Best Effort" Access Category. 621 If left unchanged, this would violate the NQB PHB requirement for 622 equal prioritization, and could erode the principle of alignment of 623 incentives. In order to preserve the incentives principle for NQB, 624 WiFi systems SHOULD configure the EDCA parameters for the Video 625 Access Category to match those of the Best Effort Access Category. 627 In cases where a network operator is delivering traffic into an 628 unmanaged WiFi network outside of their control (e.g. a residential 629 ISP delivering traffic to a customer's home network), the network 630 operator should presume that the existing WiFi equipment does not 631 support the safeguards that are provided by the NQB PHB requirements, 632 and thus should take precautions to prevent issues. When the data 633 rate of the access network segment is less than the expected data 634 rate of the WiFi network, this is unlikely to be an issue. However, 635 if the access network rate exceeds the expected rate of the WiFi 636 network, the operator SHOULD deploy a policing function on NQB marked 637 traffic that minimizes the potential for negative impacts on traffic 638 marked Default, for example by limiting the rate of such traffic to a 639 set fraction of the customer's service rate, with excess traffic 640 either dropped or re-marked as Default. 642 As an additional safeguard, and to prevent the inadvertent 643 introduction of problematic traffic into unmanaged WiFi networks, 644 network equipment that is intended to deliver traffic into unmanaged 645 WiFi networks (e.g. an access network gateway for a residential ISP) 646 MUST by default ensure that NQB traffic is marked with a DSCP that 647 selects the "Best Effort" Access Category. Such equipment MUST 648 support the ability to configure the remapping, so that (when 649 appropriate safeguards are in place) traffic can be delivered as NQB- 650 marked. 652 Similarly, systems that utilize [RFC8325] but that are unable to 653 fully support the PHB requirements, SHOULD map the recommended NQB 654 code point 45 (or the locally determined alternative) to UP_5 in the 655 "Video" Access Category. 657 9. Acknowledgements 659 Thanks to Diego Lopez, Stuart Cheshire, Brian Carpenter, Bob Briscoe, 660 Greg Skinner, Toke Hoeiland-Joergensen, Luca Muscariello, David 661 Black, Sebastian Moeller, Ruediger Geib, Jerome Henry, Steven Blake, 662 Jonathan Morton, Roland Bless, Kevin Smith, Martin Dolly, and Kyle 663 Rose for their review comments. Thanks also to Gorry Fairhurst, Ana 664 Custura, and Ruediger Geib for their input on selection of 665 appropriate DSCPs. 667 10. IANA Considerations 669 This document requests that IANA assign the Differentiated Services 670 Field Codepoints (DSCP) 5 ('0b000101', 0x05) and 45 ('0b101101', 671 0x2D) from the "Differentiated Services Field Codepoints (DSCP)" 672 registry (https://www.iana.org/assignments/dscp-registry/) ("DSCP 673 Pool 3 Codepoints", Codepoint Space xxxx01, Standards Action) as the 674 RECOMMENDED codepoints for Non-Queue-Building behavior. 676 11. Security Considerations 678 When the NQB PHB is fully supported in bottleneck links, there is no 679 incentive for an application to mismark its packets as NQB (or vice 680 versa). If a queue-building flow were to mark its packets as NQB, it 681 would be unlikely to receive a benefit by doing so, and it could 682 experience excessive packet loss, excessive latency variation and/or 683 excessive out-of-order delivery (depending on the nature of the 684 traffic protection function). If a non-queue-building flow were to 685 fail to mark its packets as NQB, it could suffer the latency and loss 686 typical of sharing a queue with capacity seeking traffic. 688 In order to preserve low latency performance for NQB traffic, 689 networks that support the NQB PHB will need to ensure that mechanisms 690 are in place to prevent malicious NQB-marked traffic from causing 691 excessive queue delays. This document recommends the implementation 692 of a traffic protection mechanism to achieve this goal, but 693 recognizes that other options may be more desirable in certain 694 situations. 696 Notwithstanding the above, the choice of DSCP for NQB does allow 697 existing WiFi networks to readily (and by default) support some of 698 the PHB requirements, but without a traffic protection function, and 699 (when left in the default state) by giving NQB traffic higher 700 priority than QB traffic. This does open up the NQB marking to 701 potential abuse on these WiFi links, but since these existing WiFi 702 networks already give one quarter of the DSCP space this same 703 treatment, and further they give another quarter of the DSCP space 704 even higher priority, the NQB DSCP does not seem to be of any greater 705 risk for abuse than these others. 707 The NQB signal is not integrity protected and could be flipped by an 708 on-path attacker. This might negatively affect the QoS of the 709 tampered flow. 711 12. References 713 12.1. Normative References 715 [RFC2119] Bradner, S., "Key words for use in RFCs to Indicate 716 Requirement Levels", BCP 14, RFC 2119, 717 DOI 10.17487/RFC2119, March 1997, 718 . 720 [RFC2474] Nichols, K., Blake, S., Baker, F., and D. Black, 721 "Definition of the Differentiated Services Field (DS 722 Field) in the IPv4 and IPv6 Headers", RFC 2474, 723 DOI 10.17487/RFC2474, December 1998, 724 . 726 [RFC2983] Black, D., "Differentiated Services and Tunnels", 727 RFC 2983, DOI 10.17487/RFC2983, October 2000, 728 . 730 [RFC8085] Eggert, L., Fairhurst, G., and G. Shepherd, "UDP Usage 731 Guidelines", BCP 145, RFC 8085, DOI 10.17487/RFC8085, 732 March 2017, . 734 [RFC8174] Leiba, B., "Ambiguity of Uppercase vs Lowercase in RFC 735 2119 Key Words", BCP 14, RFC 8174, DOI 10.17487/RFC8174, 736 May 2017, . 738 [RFC8325] Szigeti, T., Henry, J., and F. Baker, "Mapping Diffserv to 739 IEEE 802.11", RFC 8325, DOI 10.17487/RFC8325, February 740 2018, . 742 12.2. Informative References 744 [Barik] Barik, R., Welzl, M., Elmokashfi, A., Dreibholz, T., and 745 S. Gjessing, "Can WebRTC QoS Work? A DSCP Measurement 746 Study", ITC 30, September 2018. 748 [Custura] Custura, A., Venne, A., and G. Fairhurst, "Exploring DSCP 749 modification pathologies in mobile edge networks", TMA , 750 2017. 752 [I-D.briscoe-docsis-q-protection] 753 Briscoe, B. and G. White, "Queue Protection to Preserve 754 Low Latency", Work in Progress, Internet-Draft, draft- 755 briscoe-docsis-q-protection-00, 8 July 2019, 756 . 759 [I-D.ietf-tsvwg-aqm-dualq-coupled] 760 Schepper, K. D., Briscoe, B., and G. White, "DualQ Coupled 761 AQMs for Low Latency, Low Loss and Scalable Throughput 762 (L4S)", Work in Progress, Internet-Draft, draft-ietf- 763 tsvwg-aqm-dualq-coupled-16, 7 July 2021, 764 . 767 [I-D.ietf-tsvwg-dscp-considerations] 768 Custura, A., Fairhurst, G., and R. Secchi, "Considerations 769 for Assigning a new Recommended DiffServ Codepoint 770 (DSCP)", Work in Progress, Internet-Draft, draft-ietf- 771 tsvwg-dscp-considerations-00, 26 July 2021, 772 . 775 [I-D.ietf-tsvwg-ecn-l4s-id] 776 Schepper, K. D. and B. Briscoe, "Explicit Congestion 777 Notification (ECN) Protocol for Very Low Queuing Delay 778 (L4S)", Work in Progress, Internet-Draft, draft-ietf- 779 tsvwg-ecn-l4s-id-19, 26 July 2021, 780 . 783 [I-D.ietf-tsvwg-l4s-arch] 784 Briscoe, B., Schepper, K. D., Bagnulo, M., and G. White, 785 "Low Latency, Low Loss, Scalable Throughput (L4S) Internet 786 Service: Architecture", Work in Progress, Internet-Draft, 787 draft-ietf-tsvwg-l4s-arch-10, 1 July 2021, 788 . 791 [IEEE802-11] 792 IEEE-SA, "IEEE 802.11-2020", IEEE 802, December 2020, 793 . 795 [RFC2475] Blake, S., Black, D., Carlson, M., Davies, E., Wang, Z., 796 and W. Weiss, "An Architecture for Differentiated 797 Services", RFC 2475, DOI 10.17487/RFC2475, December 1998, 798 . 800 [RFC3551] Schulzrinne, H. and S. Casner, "RTP Profile for Audio and 801 Video Conferences with Minimal Control", STD 65, RFC 3551, 802 DOI 10.17487/RFC3551, July 2003, 803 . 805 [RFC4594] Babiarz, J., Chan, K., and F. Baker, "Configuration 806 Guidelines for DiffServ Service Classes", RFC 4594, 807 DOI 10.17487/RFC4594, August 2006, 808 . 810 [RFC5127] Chan, K., Babiarz, J., and F. Baker, "Aggregation of 811 Diffserv Service Classes", RFC 5127, DOI 10.17487/RFC5127, 812 February 2008, . 814 [RFC7893] Stein, Y(J)., Black, D., and B. Briscoe, "Pseudowire 815 Congestion Considerations", RFC 7893, 816 DOI 10.17487/RFC7893, June 2016, 817 . 819 [RFC8033] Pan, R., Natarajan, P., Baker, F., and G. White, 820 "Proportional Integral Controller Enhanced (PIE): A 821 Lightweight Control Scheme to Address the Bufferbloat 822 Problem", RFC 8033, DOI 10.17487/RFC8033, February 2017, 823 . 825 [RFC8034] White, G. and R. Pan, "Active Queue Management (AQM) Based 826 on Proportional Integral Controller Enhanced PIE) for 827 Data-Over-Cable Service Interface Specifications (DOCSIS) 828 Cable Modems", RFC 8034, DOI 10.17487/RFC8034, February 829 2017, . 831 [RFC8083] Perkins, C. and V. Singh, "Multimedia Congestion Control: 832 Circuit Breakers for Unicast RTP Sessions", RFC 8083, 833 DOI 10.17487/RFC8083, March 2017, 834 . 836 [RFC8100] Geib, R., Ed. and D. Black, "Diffserv-Interconnection 837 Classes and Practice", RFC 8100, DOI 10.17487/RFC8100, 838 March 2017, . 840 [RFC8289] Nichols, K., Jacobson, V., McGregor, A., Ed., and J. 841 Iyengar, Ed., "Controlled Delay Active Queue Management", 842 RFC 8289, DOI 10.17487/RFC8289, January 2018, 843 . 845 [RFC8290] Hoeiland-Joergensen, T., McKenney, P., Taht, D., Gettys, 846 J., and E. Dumazet, "The Flow Queue CoDel Packet Scheduler 847 and Active Queue Management Algorithm", RFC 8290, 848 DOI 10.17487/RFC8290, January 2018, 849 . 851 [SA-5G] 3GPP, "System Architecture for 5G", TS 23.501, 2019. 853 Appendix A. DSCP Remarking Pathologies 855 Some network operators typically bleach (zero out) the Diffserv field 856 on ingress into their network 857 [I-D.ietf-tsvwg-dscp-considerations][Custura][Barik], and in some 858 cases apply their own DSCP for internal usage. Bleaching the NQB 859 DSCP is not expected to cause harm to default traffic, but it will 860 severely limit the ability to provide NQB treatment end-to-end. 861 Reports on existing deployments of DSCP manipulation [Custura][Barik] 862 categorize the re-marking behaviors into the following six policies: 863 bleach all traffic (set DSCP to zero), set the top three bits (the 864 former Precedence bits) on all traffic to 0b000, 0b001, or 0b010, set 865 the low three bits on all traffic to 0b000, or remark all traffic to 866 a particular (non-zero) DSCP value. 868 Regarding the DSCP values of 5 & 45, there were no observations of 869 DSCP manipulation reported in which traffic was marked 5 or 45 by any 870 of these policies. Thus it appears that these re-marking policies 871 would be unlikely to result in QB traffic being marked as NQB (45). 872 In terms of the fate of NQB-marked traffic that is subjected to one 873 of these policies, the result would be that NQB marked traffic would 874 be indistinguishable from some subset (possibly all) of other 875 traffic. In the policies where all traffic is remarked using the 876 same (zero or non-zero) DSCP, the ability for a subsequent network 877 hop to differentiate NQB traffic via DSCP would clearly be lost 878 entirely. 880 In the policies where the top three bits are overwritten, both NQB 881 values (5 & 45) would receive the same marking as would the currently 882 unassigned Pool 3 DSCPs 13,21,29,37,53,61, with all of these code 883 points getting mapped to DSCP=5, 13 or 21 (depending on the overwrite 884 value used). Since none of the DSCPs in the preceding lists are 885 currently assigned by IANA, and they all are set aside for Standards 886 Action, it is believed that they are not widely used currently, but 887 this may vary based on local-usage. 889 For the policy in which the low three bits are set to 0b000, the NQB 890 (45) value would be mapped to CS5 and would be indistinguishable from 891 CS5, VA, EF (and the unassigned DSCPs 41, 42, 43). Traffic marked 892 using the existing standardized DSCPs in this list are likely to 893 share the same general properties as NQB traffic (non capacity- 894 seeking, very low data rate or relatively low and consistent data 895 rate). Similarly, any future recommended usage for DSCPs 41, 42, 43 896 would likely be somewhat compatible with NQB treatment, assuming that 897 IP Precedence compatibility (see Section 1.5.4 of [RFC4594]) is 898 maintained in the future. Here there may be an opportunity for a 899 node to provide the NQB PHB or the CS5 PHB to CS5-marked traffic and 900 retain some of the benefits of NQB marking. This could be another 901 motivation to (as discussed in Section 4.2) classify CS5-marked 902 traffic into NQB queue. For this same re-marking policy, the NQB (5) 903 value would be mapped to CS0/default and would be indistinguishable 904 from CS0, LE (and the unassigned DSCPs 2,3,4,6,7). In this case, NQB 905 traffic is likely to be given default treatment in all subsequent 906 nodes, which would eliminate the ability to provide NQB treatment in 907 those nodes, but would be relatively harmless otherwise. 909 Authors' Addresses 911 Greg White 912 CableLabs 914 Email: g.white@cablelabs.com 916 Thomas Fossati 917 ARM 919 Email: Thomas.Fossati@arm.com