idnits 2.17.1 draft-ietf-tsvwg-aqm-dualq-coupled-05.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 (July 2, 2018) is 2124 days in the past. Is this intentional? Checking references for intended status: Experimental ---------------------------------------------------------------------------- -- Looks like a reference, but probably isn't: '0' on line 1528 -- Looks like a reference, but probably isn't: '1' on line 1528 == Outdated reference: A later version (-02) exists of draft-briscoe-tsvwg-l4s-diffserv-00 == Outdated reference: A later version (-29) exists of draft-ietf-tsvwg-ecn-l4s-id-02 == Outdated reference: A later version (-20) exists of draft-ietf-tsvwg-l4s-arch-02 -- Obsolete informational reference (is this intentional?): RFC 2309 (Obsoleted by RFC 7567) -- Obsolete informational reference (is this intentional?): RFC 8312 (Obsoleted by RFC 9438) Summary: 0 errors (**), 0 flaws (~~), 4 warnings (==), 5 comments (--). Run idnits with the --verbose option for more detailed information about the items above. -------------------------------------------------------------------------------- 2 Transport Area working group (tsvwg) K. De Schepper 3 Internet-Draft Nokia Bell Labs 4 Intended status: Experimental B. Briscoe, Ed. 5 Expires: January 3, 2019 CableLabs 6 O. Bondarenko 7 Simula Research Lab 8 I. Tsang 9 Nokia 10 July 2, 2018 12 DualQ Coupled AQMs for Low Latency, Low Loss and Scalable Throughput 13 (L4S) 14 draft-ietf-tsvwg-aqm-dualq-coupled-05 16 Abstract 18 Data Centre TCP (DCTCP) was designed to provide predictably low 19 queuing latency, near-zero loss, and throughput scalability using 20 explicit congestion notification (ECN) and an extremely simple 21 marking behaviour on switches. However, DCTCP does not co-exist with 22 existing TCP traffic---DCTCP is so aggressive that existing TCP 23 algorithms approach starvation. So, until now, DCTCP could only be 24 deployed where a clean-slate environment could be arranged, such as 25 in private data centres. This specification defines `DualQ Coupled 26 Active Queue Management (AQM)' to allow scalable congestion controls 27 like DCTCP to safely co-exist with classic Internet traffic. The 28 Coupled AQM ensures that a flow runs at about the same rate whether 29 it uses DCTCP or TCP Reno/Cubic, but without inspecting transport 30 layer flow identifiers. When tested in a residential broadband 31 setting, DCTCP achieved sub-millisecond average queuing delay and 32 zero congestion loss under a wide range of mixes of DCTCP and 33 `Classic' broadband Internet traffic, without compromising the 34 performance of the Classic traffic. The solution also reduces 35 network complexity and eliminates network configuration. 37 Status of This Memo 39 This Internet-Draft is submitted in full conformance with the 40 provisions of BCP 78 and BCP 79. 42 Internet-Drafts are working documents of the Internet Engineering 43 Task Force (IETF). Note that other groups may also distribute 44 working documents as Internet-Drafts. The list of current Internet- 45 Drafts is at https://datatracker.ietf.org/drafts/current/. 47 Internet-Drafts are draft documents valid for a maximum of six months 48 and may be updated, replaced, or obsoleted by other documents at any 49 time. It is inappropriate to use Internet-Drafts as reference 50 material or to cite them other than as "work in progress." 52 This Internet-Draft will expire on January 3, 2019. 54 Copyright Notice 56 Copyright (c) 2018 IETF Trust and the persons identified as the 57 document authors. All rights reserved. 59 This document is subject to BCP 78 and the IETF Trust's Legal 60 Provisions Relating to IETF Documents 61 (https://trustee.ietf.org/license-info) in effect on the date of 62 publication of this document. Please review these documents 63 carefully, as they describe your rights and restrictions with respect 64 to this document. Code Components extracted from this document must 65 include Simplified BSD License text as described in Section 4.e of 66 the Trust Legal Provisions and are provided without warranty as 67 described in the Simplified BSD License. 69 Table of Contents 71 1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . 3 72 1.1. Problem and Scope . . . . . . . . . . . . . . . . . . . . 3 73 1.2. Terminology . . . . . . . . . . . . . . . . . . . . . . . 5 74 1.3. Features . . . . . . . . . . . . . . . . . . . . . . . . 6 75 2. DualQ Coupled AQM . . . . . . . . . . . . . . . . . . . . . . 7 76 2.1. Coupled AQM . . . . . . . . . . . . . . . . . . . . . . . 7 77 2.2. Dual Queue . . . . . . . . . . . . . . . . . . . . . . . 8 78 2.3. Traffic Classification . . . . . . . . . . . . . . . . . 8 79 2.4. Overall DualQ Coupled AQM Structure . . . . . . . . . . . 9 80 2.5. Normative Requirements for a DualQ Coupled AQM . . . . . 11 81 2.5.1. Functional Requirements . . . . . . . . . . . . . . . 11 82 2.5.1.1. Requirements in Unexpected Cases . . . . . . . . 12 83 2.5.2. Management Requirements . . . . . . . . . . . . . . . 13 84 3. IANA Considerations . . . . . . . . . . . . . . . . . . . . . 14 85 4. Security Considerations . . . . . . . . . . . . . . . . . . . 14 86 4.1. Overload Handling . . . . . . . . . . . . . . . . . . . . 14 87 4.1.1. Avoiding Classic Starvation: Sacrifice L4S Throughput 88 or Delay? . . . . . . . . . . . . . . . . . . . . . . 15 89 4.1.2. Congestion Signal Saturation: Introduce L4S Drop or 90 Delay? . . . . . . . . . . . . . . . . . . . . . . . 16 91 4.1.3. Protecting against Unresponsive ECN-Capable Traffic . 17 92 5. Acknowledgements . . . . . . . . . . . . . . . . . . . . . . 17 93 6. References . . . . . . . . . . . . . . . . . . . . . . . . . 17 94 6.1. Normative References . . . . . . . . . . . . . . . . . . 18 95 6.2. Informative References . . . . . . . . . . . . . . . . . 18 96 Appendix A. Example DualQ Coupled PI2 Algorithm . . . . . . . . 21 97 A.1. Pass #1: Core Concepts . . . . . . . . . . . . . . . . . 21 98 A.2. Pass #2: Overload Details . . . . . . . . . . . . . . . . 27 99 Appendix B. Example DualQ Coupled Curvy RED Algorithm . . . . . 29 100 Appendix C. Guidance on Controlling Throughput Equivalence . . . 35 101 Appendix D. Open Issues . . . . . . . . . . . . . . . . . . . . 36 102 Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . 37 104 1. Introduction 106 1.1. Problem and Scope 108 Latency is becoming the critical performance factor for many (most?) 109 applications on the public Internet, e.g. interactive Web, Web 110 services, voice, conversational video, interactive video, interactive 111 remote presence, instant messaging, online gaming, remote desktop, 112 cloud-based applications, and video-assisted remote control of 113 machinery and industrial processes. In the developed world, further 114 increases in access network bit-rate offer diminishing returns, 115 whereas latency is still a multi-faceted problem. In the last decade 116 or so, much has been done to reduce propagation time by placing 117 caches or servers closer to users. However, queuing remains a major 118 component of latency. 120 The Diffserv architecture provides Expedited Forwarding [RFC3246], so 121 that low latency traffic can jump the queue of other traffic. 122 However, on access links dedicated to individual sites (homes, small 123 enterprises or mobile devices), often all traffic at any one time 124 will be latency-sensitive and, if all the traffic on a link is marked 125 as EF, Diffserv cannot reduce the delay of any of it. In contrast, 126 the Low Latency Low Loss Scalable throughput (L4S) approach removes 127 the causes of any unnecessary queuing delay. 129 The bufferbloat project has shown that excessively-large buffering 130 (`bufferbloat') has been introducing significantly more delay than 131 the underlying propagation time. These delays appear only 132 intermittently--only when a capacity-seeking (e.g. TCP) flow is long 133 enough for the queue to fill the buffer, making every packet in other 134 flows sharing the buffer sit through the queue. 136 Active queue management (AQM) was originally developed to solve this 137 problem (and others). Unlike Diffserv, which gives low latency to 138 some traffic at the expense of others, AQM controls latency for _all_ 139 traffic in a class. In general, AQMs introduce an increasing level 140 of discard from the buffer the longer the queue persists above a 141 shallow threshold. This gives sufficient signals to capacity-seeking 142 (aka. greedy) flows to keep the buffer empty for its intended 143 purpose: absorbing bursts. However, RED [RFC2309] and other 144 algorithms from the 1990s were sensitive to their configuration and 145 hard to set correctly. So, AQM was not widely deployed. 147 More recent state-of-the-art AQMs, e.g. fq_CoDel [RFC8290], 148 PIE [RFC8033], Adaptive RED [ARED01], are easier to configure, 149 because they define the queuing threshold in time not bytes, so it is 150 invariant for different link rates. However, no matter how good the 151 AQM, the sawtoothing rate of TCP will either cause queuing delay to 152 vary or cause the link to be under-utilized. Even with a perfectly 153 tuned AQM, the additional queuing delay will be of the same order as 154 the underlying speed-of-light delay across the network. Flow-queuing 155 can isolate one flow from another, but it cannot isolate a TCP flow 156 from the delay variations it inflicts on itself, and it has other 157 problems - it overrides the flow rate decisions of variable rate 158 video applications, it does not recognise the flows within IPSec VPN 159 tunnels and it is relatively expensive to implement. 161 It seems that further changes to the network alone will now yield 162 diminishing returns. Data Centre TCP (DCTCP [RFC8257]) teaches us 163 that a small but radical change to TCP is needed to cut two major 164 outstanding causes of queuing delay variability: 166 1. the `sawtooth' varying rate of TCP itself; 168 2. the smoothing delay deliberately introduced into AQMs to permit 169 bursts without triggering losses. 171 The former causes a flow's round trip time (RTT) to vary from about 1 172 to 2 times the base RTT between the machines in question. The latter 173 delays the system's response to change by a worst-case 174 (transcontinental) RTT, which could be hundreds of times the actual 175 RTT of typical traffic from localized CDNs. 177 Latency is not our only concern: 179 3. It was known when TCP was first developed that it would not scale 180 to high bandwidth-delay products. 182 Given regular broadband bit-rates over WAN distances are 183 already [RFC3649] beyond the scaling range of `classic' TCP Reno, 184 `less unscalable' Cubic [RFC8312] and 185 Compound [I-D.sridharan-tcpm-ctcp] variants of TCP have been 186 successfully deployed. However, these are now approaching their 187 scaling limits. Unfortunately, fully scalable TCPs such as DCTCP 188 cause `classic' TCP to starve itself, which is why they have been 189 confined to private data centres or research testbeds (until now). 191 This document specifies a `DualQ Coupled AQM' extension that solves 192 the problem of coexistence between scalable and classic flows, 193 without having to inspect flow identifiers. The AQM is not like 194 flow-queuing approaches [RFC8290] that classify packets by flow 195 identifier into numerous separate queues in order to isolate sparse 196 flows from the higher latency in the queues assigned to heavier flow. 197 In contrast, the AQM exploits the behaviour of scalable congestion 198 controls like DCTCP so that every packet in every flow sharing the 199 queue for DCTCP-like traffic can be served with very low latency. 201 This AQM extension can be combined with any single queue AQM that 202 generates a statistical or deterministic mark/drop probability driven 203 by the queue dynamics. In many cases it simplifies the basic control 204 algorithm, and requires little extra processing. Therefore it is 205 believed the Coupled AQM would be applicable and easy to deploy in 206 all types of buffers; buffers in cost-reduced mass-market residential 207 equipment; buffers in end-system stacks; buffers in carrier-scale 208 equipment including remote access servers, routers, firewalls and 209 Ethernet switches; buffers in network interface cards, buffers in 210 virtualized network appliances, hypervisors, and so on. 212 The overall L4S architecture is described in 213 [I-D.ietf-tsvwg-l4s-arch]. The supporting papers [PI2] and [DCttH15] 214 give the full rationale for the AQM's design, both discursively and 215 in more precise mathematical form. 217 1.2. Terminology 219 The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", 220 "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this 221 document are to be interpreted as described in [RFC2119]. In this 222 document, these words will appear with that interpretation only when 223 in ALL CAPS. Lower case uses of these words are not to be 224 interpreted as carrying RFC-2119 significance. 226 The DualQ Coupled AQM uses two queues for two services. Each of the 227 following terms identifies both the service and the queue that 228 provides the service: 230 Classic (denoted by subscript C): The `Classic' service is intended 231 for all the behaviours that currently co-exist with TCP Reno (TCP 232 Cubic, Compound, SCTP, etc). 234 Low-Latency, Low-Loss and Scalable (L4S, denoted by subscript L): 235 The `L4S' service is intended for a set of congestion controls 236 with scalable properties such as DCTCP (e.g. 237 Relentless [Mathis09]). 239 Either service can cope with a proportion of unresponsive or less- 240 responsive traffic as well (e.g. DNS, VoIP, etc), just as a single 241 queue AQM can. The DualQ Coupled AQM behaviour is similar to a 242 single FIFO queue with respect to unresponsive and overload traffic. 244 1.3. Features 246 The AQM couples marking and/or dropping across the two queues such 247 that a flow will get roughly the same throughput whichever it uses. 248 Therefore both queues can feed into the full capacity of a link and 249 no rates need to be configured for the queues. The L4S queue enables 250 scalable congestion controls like DCTCP to give stunningly low and 251 predictably low latency, without compromising the performance of 252 competing 'Classic' Internet traffic. Thousands of tests have been 253 conducted in a typical fixed residential broadband setting. Typical 254 experiments used base round trip delays up to 100ms between the data 255 centre and home network, and large amounts of background traffic in 256 both queues. For every L4S packet, the AQM kept the average queuing 257 delay below 1ms (or 2 packets if serialization delay is bigger for 258 slow links), and no losses at all were introduced by the AQM. 259 Details of the extensive experiments will be made available [PI2] 260 [DCttH15]. 262 Subjective testing was also conducted using a demanding panoramic 263 interactive video application run over a stack with DCTCP enabled and 264 deployed on the testbed. Each user could pan or zoom their own high 265 definition (HD) sub-window of a larger video scene from a football 266 match. Even though the user was also downloading large amounts of 267 L4S and Classic data, latency was so low that the picture appeared to 268 stick to their finger on the touchpad (all the L4S data achieved the 269 same ultra-low latency). With an alternative AQM, the video 270 noticeably lagged behind the finger gestures. 272 Unlike Diffserv Expedited Forwarding, the L4S queue does not have to 273 be limited to a small proportion of the link capacity in order to 274 achieve low delay. The L4S queue can be filled with a heavy load of 275 capacity-seeking flows like DCTCP and still achieve low delay. The 276 L4S queue does not rely on the presence of other traffic in the 277 Classic queue that can be 'overtaken'. It gives low latency to L4S 278 traffic whether or not there is Classic traffic, and the latency of 279 Classic traffic does not suffer when a proportion of the traffic is 280 L4S. The two queues are only necessary because DCTCP-like flows 281 cannot keep latency predictably low and keep utilization high if they 282 are mixed with legacy TCP flows, 284 The experiments used the Linux implementation of DCTCP that is 285 deployed in private data centres, without any modification despite 286 its known deficiencies. Nonetheless, certain modifications will be 287 necessary before DCTCP is safe to use on the Internet, which are 288 recorded in Appendix A of [I-D.ietf-tsvwg-ecn-l4s-id]. However, the 289 focus of this specification is to get the network service in place. 290 Then, without any management intervention, applications can exploit 291 it by migrating to scalable controls like DCTCP, which can then 292 evolve _while_ their benefits are being enjoyed by everyone on the 293 Internet. 295 2. DualQ Coupled AQM 297 There are two main aspects to the approach: 299 o the Coupled AQM that addresses throughput equivalence between 300 Classic (e.g. Reno, Cubic) flows and L4S (e.g. DCTCP) flows 302 o the Dual Queue structure that provides latency separation for L4S 303 flows to isolate them from the typically large Classic queue. 305 2.1. Coupled AQM 307 In the 1990s, the `TCP formula' was derived for the relationship 308 between TCP's congestion window, cwnd, and its drop probability, p. 309 To a first order approximation, cwnd of TCP Reno is inversely 310 proportional to the square root of p. 312 TCP Cubic implements a Reno-compatibility mode, which is the only 313 relevant mode for typical RTTs under 20ms as long as the throughput 314 of a single flow is less than about 500Mb/s. Therefore it can be 315 assumed that Cubic traffic behaves similarly to Reno (but with a 316 slightly different constant of proportionality), and the term 317 'Classic' will be used for the collection of Reno-friendly traffic 318 including Cubic in Reno mode. 320 The supporting paper [PI2] includes the derivation of the equivalent 321 rate equation for DCTCP, for which cwnd is inversely proportional to 322 p (not the square root), where in this case p is the ECN marking 323 probability. DCTCP is not the only congestion control that behaves 324 like this, so the term 'L4S' traffic will be used for all similar 325 behaviour. 327 In order to make a DCTCP flow run at roughly the same rate as a Reno 328 TCP flow (all other factors being equal), the drop or marking 329 probability for Classic traffic, p_C has to be distinct from the 330 marking probability for L4S traffic, p_L (in contrast to RFC3168 331 which requires them to be the same). It is necessary to make the 332 Classic drop probability p_C proportional to the square of the L4S 333 marking probability p_L. This makes the Reno flow rate roughly equal 334 the DCTCP flow rate, because it squares the square root of p_C in the 335 Reno rate equation to make it proportional to the straight p_L in the 336 DCTCP rate equation. 338 Stating this as a formula, the relation between Classic drop 339 probability, p_C, and L4S marking probability, p_L needs to take the 340 form: 342 p_C = ( p_L / k )^2 (1) 344 where k is the constant of proportionality. 346 2.2. Dual Queue 348 Classic traffic typically builds a large queue to prevent under- 349 utilization. Therefore a separate queue is provided for L4S traffic, 350 and it is scheduled with priority over Classic. Priority is 351 conditional to prevent starvation of Classic traffic. 353 Nonetheless, coupled marking ensures that giving priority to L4S 354 traffic still leaves the right amount of spare scheduling time for 355 Classic flows to each get equivalent throughput to DCTCP flows (all 356 other factors such as RTT being equal). The algorithm achieves this 357 without having to inspect flow identifiers. 359 2.3. Traffic Classification 361 Both the Coupled AQM and DualQ mechanisms need an identifier to 362 distinguish L and C packets. A separate draft 363 [I-D.ietf-tsvwg-ecn-l4s-id] recommends using the ECT(1) codepoint of 364 the ECN field as this identifier, having assessed various 365 alternatives. An additional process document has proved necessary to 366 make the ECT(1) codepoint available for experimentation [RFC8311]. 368 For policy reasons, an operator might choose to steer certain packets 369 (e.g. from certain flows or with certain addresses) out of the L 370 queue, even though they identify themselves as L4S by their ECN 371 codepoints. In such cases, the classifier MUST NOT alter the ECN 372 field, so that it is preserved end-to-end. The aim is that each 373 operator can choose how it treats L4S traffic locally, but an 374 individual operator does not alter the identification of L4S packets, 375 which would prevent other operators downstream from making their own 376 choices on how to treat L4S traffic. 378 In addition, other identifiers could be used to classify certain 379 additional packet types into the L queue, that are deemed not to risk 380 harming the L4S service. For instance addresses of specific 381 applications or hosts (see [I-D.ietf-tsvwg-ecn-l4s-id]), specific 382 Diffserv codepoints such as EF (Expedited Forwarding) and Voice-Admit 383 service classes (see [I-D.briscoe-tsvwg-l4s-diffserv]) or certain 384 protocols (e.g. ARP, DNS). 386 Note that the DualQ Coupled AQM only reads these classifiers, it MUST 387 NOT re-mark or alter these identifiers (except for marking the ECN 388 field with the CE codepoint - with increasing frequency to indicate 389 increasing congestion). 391 2.4. Overall DualQ Coupled AQM Structure 393 Figure 1 shows the overall structure that any DualQ Coupled AQM is 394 likely to have. This schematic is intended to aid understanding of 395 the current designs of DualQ Coupled AQMs. However, it is not 396 intended to preclude other innovative ways of satisfying the 397 normative requirements in Section 2.5 that minimally define a DualQ 398 Coupled AQM. 400 The classifier on the left separates incoming traffic between the two 401 queues (L and C). Each queue has its own AQM that determines the 402 likelihood of dropping or marking (p_L and p_C). Nonetheless, the 403 AQM for Classic traffic is implemented in two stages: i) a base stage 404 that outputs an internal probability p' (pronounced p-prime); and ii) 405 a squaring stage that outputs p_C, where 407 p_C = (p')^2. (2) 409 This allows p_L to be coupled to p_C by marking L4S traffic 410 proportionately to the intermediate output from the first stage. 411 Specifically, the output of the base AQM is coupled across to the L 412 queue in proportion to the output of the base AQM: 414 p_CL = k*p', (3) 416 where k is the constant coupling factor (see Appendix C) and p_CL is 417 the output from the coupling between the C queue and the L queue. 419 It can be seen in the following that these two transformations of p' 420 implement the required coupling given in equation (1) earlier. 421 Substituting for p' from equation (3) into (2): 423 p_C = ( p_CL / k )^2. 425 The actual L4S marking probability p_L is the maximum of the coupled 426 output (p_CL) and the output of a native L4S AQM (p'L), shown as 427 '(MAX)' in the schematic. While the output of the Native L4S AQM is 428 high (p'L > p_CL) it will dominate the way L traffic is marked. When 429 the native L4S AQM output is lower, the way L traffic is marked will 430 be driven by the coupling, that is p_L = p_CL. So, whenever the 431 coupling is needed, as required from equation (1): 433 p_C = ( p_L / k )^2. 435 _________ 436 | | ,------. 437 L4S queue | |===>| ECN | 438 ,'| _______|_| |marker|\ 439 <' | | `------'\\ 440 //`' v ^ p_L \\ 441 // ,-------. | \\ 442 // |Native |p'L | \\,. 443 // | L4S |-->(MAX) < | ___ 444 ,----------.// | AQM | ^ p_CL `\|.'Cond-`. 445 | IP-ECN |/ `-------' | / itional \ 446 ==>|Classifier| ,-------. (k*p') [ priority]==> 447 | |\ | Base | | \scheduler/ 448 `----------'\\ | AQM |--->: ,'|`-.___.-' 449 \\ | |p' | <' | 450 \\ `-------' (p'^2) //`' 451 \\ ^ | // 452 \\,. | v p_C // 453 < | _________ .------.// 454 `\| | | | Drop |/ 455 Classic |queue |===>|/mark | 456 __|______| `------' 458 Legend: ===> traffic flow; ---> control dependency. 460 Figure 1: DualQ Coupled AQM Schematic 462 After the AQMs have applied their dropping or marking, the scheduler 463 forwards their packets to the link, giving priority to L4S traffic. 464 Priority has to be conditional in some way (see Section 4.1). Simple 465 strict priority is inappropriate otherwise it could lead the L4S 466 queue to starve the Classic queue. For example, consider the case 467 where a continually busy L4S queue blocks a DNS request in the 468 Classic queue, arbitrarily delaying the start of a new Classic flow. 470 Example DualQ Coupled AQM algorithms called DualPI2 and Curvy RED are 471 given in Appendix A and Appendix B. Either example AQM can be used 472 to couple packet marking and dropping across a dual Q. 474 DualPI2 uses a Proportional-Integral (PI) controller as the Base AQM. 475 Indeed, this Base AQM with just the squared output and no L4S queue 476 can be used as a drop-in replacement for PIE [RFC8033], in which case 477 we call it just PI2 [PI2]. PI2 is a principled simplification of PIE 478 that is both more responsive and more stable in the face of 479 dynamically varying load. 481 Curvy RED is derived from RED [RFC2309], but its configuration 482 parameters are insensitive to link rate and it requires less 483 operations per packet. However, DualPI2 is more responsive and 484 stable over a wider range of RTTs than Curvy RED. As a consequence, 485 DualPI2 has attracted more development attention than Curvy RED, 486 leaving the Curvy RED design incomplete and not so fully evaluated. 488 Both AQMs regulate their queue in units of time not bytes. As 489 already explained, this ensures configuration can be invariant for 490 different drain rates. With AQMs in a dualQ structure this is 491 particularly important because the drain rate of each queue can vary 492 rapidly as flows for the two queues arrive and depart, even if the 493 combined link rate is constant. 495 It would be possible to control the queues with other alternative 496 AQMs, as long as the normative requirements (those expressed in 497 capitals) in Section 2.5 are observed. 499 2.5. Normative Requirements for a DualQ Coupled AQM 501 The following requirements are intended to capture only the essential 502 aspects of a DualQ Coupled AQM. They are intended to be independent 503 of the particular AQMs used for each queue. 505 2.5.1. Functional Requirements 507 In the Dual Queue, L4S packets MUST be given priority over Classic, 508 although priority MUST be bounded in order not to starve Classic 509 traffic. 511 Whatever identifier is used for L4S experiments, 512 [I-D.ietf-tsvwg-ecn-l4s-id] defines the meaning of an ECN marking on 513 L4S traffic, relative to drop of Classic traffic. In order to 514 prevent starvation of Classic traffic by scalable L4S traffic, it 515 says, "The likelihood that an AQM drops a Not-ECT Classic packet 516 (p_C) MUST be roughly proportional to the square of the likelihood 517 that it would have marked it if it had been an L4S packet (p_L)." In 518 other words, in any DualQ Coupled AQM, the power to which p_L is 519 raised in Eqn. (1) MUST be 2. The term 'likelihood' is used to allow 520 for marking and dropping to be either probabilistic or deterministic. 522 The constant of proportionality, k, in Eqn (1) determines the 523 relative flow rates of Classic and L4S flows when the AQM concerned 524 is the bottleneck (all other factors being equal). 526 [I-D.ietf-tsvwg-ecn-l4s-id] says, "The constant of proportionality 527 (k) does not have to be standardised for interoperability, but a 528 value of 2 is RECOMMENDED." 530 Assuming scalable congestion controls for the Internet will be as 531 aggressive as DCTCP, this will ensure their congestion window will be 532 roughly the same as that of a standards track TCP congestion control 533 (Reno) [RFC5681] and other so-called TCP-friendly controls, such as 534 TCP Cubic in its TCP-friendly mode. 536 {ToDo: The requirements for scalable congestion controls on the 537 Internet (termed the TCP Prague requirements) 538 [I-D.ietf-tsvwg-ecn-l4s-id] are not necessarily final. If the 539 aggressiveness of DCTCP is not defined as the benchmark for scalable 540 controls on the Internet, the recommended value of k will also be 541 subject to change.} 543 The choice of k is a matter of operator policy, and operators MAY 544 choose a different value using Table 1 and the guidelines in 545 Appendix C. 547 If multiple users share capacity at a bottleneck (e.g. in the 548 Internet access link of a campus network), the operator's choice of k 549 will determine capacity sharing between the flows of different users. 550 However, on the public Internet, access network operators typically 551 isolate customers from each other with some form of layer-2 552 multiplexing (TDM in DOCSIS, CDMA in 3G) or L3 scheduling (WRR in 553 DSL), rather than relying on TCP to share capacity between customers 554 [RFC0970]. In such cases, the choice of k will solely affect 555 relative flow rates within each customer's access capacity, not 556 between customers. Also, k will not affect relative flow rates at 557 any times when all flows are Classic or all L4S, and it will not 558 affect small flows. 560 2.5.1.1. Requirements in Unexpected Cases 562 The flexibility to allow operator-specific classifiers (Section 2.3) 563 leads to the need to specify what the AQM in each queue ought to do 564 with packets that do not carry the ECN field expected for that queue. 565 It is recommended that the AQM in each queue inspects the ECN field 566 to determine what sort of congestion notification to signal, then 567 decides whether to apply congestion notification to this particular 568 packet, as follows: 570 o If a packet that does not carry an ECT(1) or CE codepoint is 571 classified into the L queue: 573 * if the packet is ECT(0), the L AQM SHOULD apply CE-marking as 574 if the packet were ECT(1) 576 * if the packet is Not-ECT, the appropriate action depends on 577 whether some other function is protecting the L queue from 578 misbehaving flows (e.g. per-flow queue protection or policing): 580 + If separate queue protection is provided, the L AQM SHOULD 581 ignore the packet and forward it unchanged, meaning it 582 should not calculate whether to apply congestion 583 notification and it should neither drop nor CE-mark the 584 packet (for instance, the operator might classify EF traffic 585 that is unresponsive to drop into the L queue, alongside 586 responsive L4S-ECN traffic) 588 + if separate queue protection is not provided, the L AQM MUST 589 apply drop using the drop probability appropriate to the C 590 queue 592 o If a packet that carries an ECT(1) or CE codepoint is classified 593 into the C queue: 595 * the C AQM SHOULD apply CE-marking as if the packet were ECT(0). 597 If the DualQ Coupled AQM has detected overload, it will signal 598 congestion solely using drop, irrespective of the ECN field. 600 Most of the above requirements are worded as "SHOULDs", because 601 operator-specific classifiers are for flexibility, by definition. 602 Therefore, alternative actions might be appropriate in the operator's 603 specific circumstances. 605 2.5.2. Management Requirements 607 By default, a DualQ Coupled AQM SHOULD NOT need any configuration for 608 use at a bottleneck on the public Internet [RFC7567]. The following 609 parameters MAY be operator-configurable, e.g. to tune for non- 610 Internet settings: 612 o Optional packet classifier(s) to use in addition to the ECN field 613 (see Section 2.3); 615 o Expected typical RTT (a parameter for typical or target queuing 616 delay in each queue might be configurable instead); 618 o Expected maximum RTT (a stability parameter that depends on 619 maximum RTT might be configurable instead); 621 o Coupling factor, k; 623 o The limit to the conditional priority of L4S (scheduler-dependent, 624 e.g. the scheduler weight for WRR, or the time-shift for time- 625 shifted FIFO); 627 o The maximum Classic ECN marking probability, p_Cmax, before 628 switching over to drop. 630 An experimental DualQ Coupled AQM SHOULD allow the operator to 631 monitor the following operational statistics: 633 o Bits forwarded (total and per queue per sample interval), from 634 which utilization can be calculated 636 o Q delay (per queue over sample interval) 638 o Total packets arriving, enqueued and dequeued (per queue per 639 sample interval) 641 o ECN packets marked, non-ECN packets dropped, ECN packets dropped 642 (per queue per sample interval), from which marking and dropping 643 probabilities can be calculated 645 o Time and duration of each overload event. 647 The type of statistics produced for variables like Q delay (mean, 648 percentiles, etc.) will depend on implementation constraints. 650 3. IANA Considerations 652 This specification contains no IANA considerations. 654 4. Security Considerations 656 4.1. Overload Handling 658 Where the interests of users or flows might conflict, it could be 659 necessary to police traffic to isolate any harm to the performance of 660 individual flows. However it is hard to avoid unintended side- 661 effects with policing, and in a trusted environment policing is not 662 necessary. Therefore per-flow policing needs to be separable from a 663 basic AQM, as an option under policy control. 665 However, a basic DualQ AQM does at least need to handle overload. A 666 useful objective would be for the overload behaviour of the DualQ AQM 667 to be at least no worse than a single queue AQM. However, a trade- 668 off needs to be made between complexity and the risk of either 669 traffic class harming the other. In each of the following three 670 subsections, an overload issue specific to the DualQ is described, 671 followed by proposed solution(s). 673 Under overload the higher priority L4S service will have to sacrifice 674 some aspect of its performance. Alternative solutions are provided 675 below that each relax a different factor: e.g. throughput, delay, 676 drop. Some of these choices might need to be determined by operator 677 policy or by the developer, rather than by the IETF. {ToDo: Reach 678 consensus on which it is to be in each case.} 680 4.1.1. Avoiding Classic Starvation: Sacrifice L4S Throughput or Delay? 682 Priority of L4S is required to be conditional to avoid total 683 throughput starvation of Classic by heavy L4S traffic. This raises 684 the question of whether to sacrifice L4S throughput or L4S delay (or 685 some other policy) to mitigate starvation of Classic: 687 Sacrifice L4S throughput: By using weighted round robin as the 688 conditional priority scheduler, the L4S service can sacrifice some 689 throughput during overload to guarantee a minimum throughput 690 service for Classic traffic. The scheduling weight of the Classic 691 queue should be small (e.g. 1/16). Then, in most traffic 692 scenarios the scheduler will not interfere and it will not need to 693 - the coupling mechanism and the end-systems will share out the 694 capacity across both queues as if it were a single pool. However, 695 because the congestion coupling only applies in one direction 696 (from C to L), if L4S traffic is over-aggressive or unresponsive, 697 the scheduler weight for Classic traffic will at least be large 698 enough to ensure it does not starve. 700 In cases where the ratio of L4S to Classic flows (e.g. 19:1) is 701 greater than the ratio of their scheduler weights (e.g. 15:1), the 702 L4S flows will get less than an equal share of the capacity, but 703 only slightly. For instance, with the example numbers given, each 704 L4S flow will get (15/16)/19 = 4.9% when ideally each would get 705 1/20=5%. In the rather specific case of an unresponsive flow 706 taking up a large part of the capacity set aside for L4S, using 707 WRR could significantly reduce the capacity left for any 708 responsive L4S flows. 710 Sacrifice L4S Delay: To control milder overload of responsive 711 traffic, particularly when close to the maximum congestion signal, 712 the operator could choose to control overload of the Classic queue 713 by allowing some delay to 'leak' across to the L4S queue. The 714 scheduler can be made to behave like a single First-In First-Out 715 (FIFO) queue with different service times by implementing a very 716 simple conditional priority scheduler that could be called a 717 "time-shifted FIFO" (see the Modifier Earliest Deadline First 718 (MEDF) scheduler of [MEDF]). This scheduler adds tshift to the 719 queue delay of the next L4S packet, before comparing it with the 720 queue delay of the next Classic packet, then it selects the packet 721 with the greater adjusted queue delay. Under regular conditions, 722 this time-shifted FIFO scheduler behaves just like a strict 723 priority scheduler. But under moderate or high overload it 724 prevents starvation of the Classic queue, because the time-shift 725 (tshift) defines the maximum extra queuing delay of Classic 726 packets relative to L4S. 728 The example implementation in Appendix A can implement either policy. 730 4.1.2. Congestion Signal Saturation: Introduce L4S Drop or Delay? 732 To keep the throughput of both L4S and Classic flows roughly equal 733 over the full load range, a different control strategy needs to be 734 defined above the point where one AQM first saturates to a 735 probability of 100% leaving no room to push back the load any harder. 736 If k>1, L4S will saturate first, but saturation can be caused by 737 unresponsive traffic in either queue. 739 The term 'unresponsive' includes cases where a flow becomes 740 temporarily unresponsive, for instance, a real-time flow that takes a 741 while to adapt its rate in response to congestion, or a TCP-like flow 742 that is normally responsive, but above a certain congestion level it 743 will not be able to reduce its congestion window below the minimum of 744 2 segments, effectively becoming unresponsive. (Note that L4S 745 traffic ought to remain responsive below a window of 2 segments (see 746 [I-D.ietf-tsvwg-ecn-l4s-id]). 748 Saturation raises the question of whether to relieve congestion by 749 introducing some drop into the L4S queue or by allowing delay to grow 750 in both queues (which could eventually lead to tail drop too): 752 Drop on Saturation: Saturation can be avoided by setting a maximum 753 threshold for L4S ECN marking (assuming k>1) before saturation 754 starts to make the flow rates of the different traffic types 755 diverge. Above that the drop probability of Classic traffic is 756 applied to all packets of all traffic types. Then experiments 757 have shown that queueing delay can be kept at the target in any 758 overload situation, including with unresponsive traffic, and no 759 further measures are required. 761 Delay on Saturation: When L4S marking saturates, instead of 762 switching to drop, the drop and marking probabilities could be 763 capped. Beyond that, delay will grow either solely in the queue 764 with unresponsive traffic (if WRR is used), or in both queues (if 765 time-shifted FIFO is used). In either case, the higher delay 766 ought to control temporary high congestion. If the overload is 767 more persistent, eventually the combined DualQ will overflow and 768 tail drop will control congestion. 770 The example implementation in Appendix A applies only the "drop on 771 saturation" policy. 773 4.1.3. Protecting against Unresponsive ECN-Capable Traffic 775 Unresponsive traffic has a greater advantage if it is also ECN- 776 capable. The advantage is undetectable at normal low levels of drop/ 777 marking, but it becomes significant with the higher levels of drop/ 778 marking typical during overload. This is an issue whether the ECN- 779 capable traffic is L4S or Classic. 781 This raises the question of whether and when to switch off ECN 782 marking and use solely drop instead, as required by both Section 7 of 783 [RFC3168] and Section 4.2.1 of [RFC7567]. 785 Experiments with the DualPI2 AQM (Appendix A) have shown that 786 introducing 'drop on saturation' at 100% L4S marking addresses this 787 problem with unresponsive ECN as well as addressing the saturation 788 problem. It leaves only a small range of congestion levels where 789 unresponsive traffic gains any advantage from using the ECN 790 capability, and the advantage is hardly detectable [DualQ-Test]. 792 5. Acknowledgements 794 Thanks to Anil Agarwal, Sowmini Varadhan's and Gabi Bracha for 795 detailed review comments particularly of the appendices and 796 suggestions on how to make our explanation clearer. Thanks also to 797 Greg White and Tom Henderson for insights on the choice of schedulers 798 and queue delay measurement techniques. 800 The authors' contributions were originally part-funded by the 801 European Community under its Seventh Framework Programme through the 802 Reducing Internet Transport Latency (RITE) project (ICT-317700). Bob 803 Briscoe's contribution was also part-funded by the Research Council 804 of Norway through the TimeIn project. The views expressed here are 805 solely those of the authors. 807 6. References 808 6.1. Normative References 810 [RFC2119] Bradner, S., "Key words for use in RFCs to Indicate 811 Requirement Levels", BCP 14, RFC 2119, 812 DOI 10.17487/RFC2119, March 1997, 813 . 815 6.2. Informative References 817 [ARED01] Floyd, S., Gummadi, R., and S. Shenker, "Adaptive RED: An 818 Algorithm for Increasing the Robustness of RED's Active 819 Queue Management", ACIRI Technical Report , August 2001, 820 . 822 [CoDel] Nichols, K. and V. Jacobson, "Controlling Queue Delay", 823 ACM Queue 10(5), May 2012, 824 . 826 [CRED_Insights] 827 Briscoe, B., "Insights from Curvy RED (Random Early 828 Detection)", BT Technical Report TR-TUB8-2015-003, July 829 2015, 830 . 832 [DCttH15] De Schepper, K., Bondarenko, O., Briscoe, B., and I. 833 Tsang, "`Data Centre to the Home': Ultra-Low Latency for 834 All", 2015, . 837 (Under submission) 839 [DualQ-Test] 840 Steen, H., "Destruction Testing: Ultra-Low Delay using 841 Dual Queue Coupled Active Queue Management", Masters 842 Thesis, Dept of Informatics, Uni Oslo , May 2017. 844 [I-D.briscoe-tsvwg-l4s-diffserv] 845 Briscoe, B., "Interactions between Low Latency, Low Loss, 846 Scalable Throughput (L4S) and Differentiated Services", 847 draft-briscoe-tsvwg-l4s-diffserv-00 (work in progress), 848 March 2018. 850 [I-D.ietf-tsvwg-ecn-l4s-id] 851 Schepper, K., Briscoe, B., and I. Tsang, "Identifying 852 Modified Explicit Congestion Notification (ECN) Semantics 853 for Ultra-Low Queuing Delay", draft-ietf-tsvwg-ecn-l4s- 854 id-02 (work in progress), March 2018. 856 [I-D.ietf-tsvwg-l4s-arch] 857 Briscoe, B., Schepper, K., and M. Bagnulo, "Low Latency, 858 Low Loss, Scalable Throughput (L4S) Internet Service: 859 Architecture", draft-ietf-tsvwg-l4s-arch-02 (work in 860 progress), March 2018. 862 [I-D.sridharan-tcpm-ctcp] 863 Sridharan, M., Tan, K., Bansal, D., and D. Thaler, 864 "Compound TCP: A New TCP Congestion Control for High-Speed 865 and Long Distance Networks", draft-sridharan-tcpm-ctcp-02 866 (work in progress), November 2008. 868 [Mathis09] 869 Mathis, M., "Relentless Congestion Control", PFLDNeT'09 , 870 May 2009, . 873 [MEDF] Menth, M., Schmid, M., Heiss, H., and T. Reim, "MEDF - a 874 simple scheduling algorithm for two real-time transport 875 service classes with application in the UTRAN", Proc. IEEE 876 Conference on Computer Communications (INFOCOM'03) Vol.2 877 pp.1116-1122, March 2003. 879 [PI2] De Schepper, K., Bondarenko, O., Briscoe, B., and I. 880 Tsang, "PI2: A Linearized AQM for both Classic and 881 Scalable TCP", ACM CoNEXT'16 , December 2016, 882 . 885 (To appear) 887 [RFC0970] Nagle, J., "On Packet Switches With Infinite Storage", 888 RFC 970, DOI 10.17487/RFC0970, December 1985, 889 . 891 [RFC2309] Braden, B., Clark, D., Crowcroft, J., Davie, B., Deering, 892 S., Estrin, D., Floyd, S., Jacobson, V., Minshall, G., 893 Partridge, C., Peterson, L., Ramakrishnan, K., Shenker, 894 S., Wroclawski, J., and L. Zhang, "Recommendations on 895 Queue Management and Congestion Avoidance in the 896 Internet", RFC 2309, DOI 10.17487/RFC2309, April 1998, 897 . 899 [RFC3168] Ramakrishnan, K., Floyd, S., and D. Black, "The Addition 900 of Explicit Congestion Notification (ECN) to IP", 901 RFC 3168, DOI 10.17487/RFC3168, September 2001, 902 . 904 [RFC3246] Davie, B., Charny, A., Bennet, J., Benson, K., Le Boudec, 905 J., Courtney, W., Davari, S., Firoiu, V., and D. 906 Stiliadis, "An Expedited Forwarding PHB (Per-Hop 907 Behavior)", RFC 3246, DOI 10.17487/RFC3246, March 2002, 908 . 910 [RFC3649] Floyd, S., "HighSpeed TCP for Large Congestion Windows", 911 RFC 3649, DOI 10.17487/RFC3649, December 2003, 912 . 914 [RFC5681] Allman, M., Paxson, V., and E. Blanton, "TCP Congestion 915 Control", RFC 5681, DOI 10.17487/RFC5681, September 2009, 916 . 918 [RFC7567] Baker, F., Ed. and G. Fairhurst, Ed., "IETF 919 Recommendations Regarding Active Queue Management", 920 BCP 197, RFC 7567, DOI 10.17487/RFC7567, July 2015, 921 . 923 [RFC8033] Pan, R., Natarajan, P., Baker, F., and G. White, 924 "Proportional Integral Controller Enhanced (PIE): A 925 Lightweight Control Scheme to Address the Bufferbloat 926 Problem", RFC 8033, DOI 10.17487/RFC8033, February 2017, 927 . 929 [RFC8034] White, G. and R. Pan, "Active Queue Management (AQM) Based 930 on Proportional Integral Controller Enhanced PIE) for 931 Data-Over-Cable Service Interface Specifications (DOCSIS) 932 Cable Modems", RFC 8034, DOI 10.17487/RFC8034, February 933 2017, . 935 [RFC8257] Bensley, S., Thaler, D., Balasubramanian, P., Eggert, L., 936 and G. Judd, "Data Center TCP (DCTCP): TCP Congestion 937 Control for Data Centers", RFC 8257, DOI 10.17487/RFC8257, 938 October 2017, . 940 [RFC8290] Hoeiland-Joergensen, T., McKenney, P., Taht, D., Gettys, 941 J., and E. Dumazet, "The Flow Queue CoDel Packet Scheduler 942 and Active Queue Management Algorithm", RFC 8290, 943 DOI 10.17487/RFC8290, January 2018, 944 . 946 [RFC8311] Black, D., "Relaxing Restrictions on Explicit Congestion 947 Notification (ECN) Experimentation", RFC 8311, 948 DOI 10.17487/RFC8311, January 2018, 949 . 951 [RFC8312] Rhee, I., Xu, L., Ha, S., Zimmermann, A., Eggert, L., and 952 R. Scheffenegger, "CUBIC for Fast Long-Distance Networks", 953 RFC 8312, DOI 10.17487/RFC8312, February 2018, 954 . 956 Appendix A. Example DualQ Coupled PI2 Algorithm 958 As a first concrete example, the pseudocode below gives the DualPI2 959 algorithm. DualPI2 follows the structure of the DualQ Coupled AQM 960 framework in Figure 1. A simple step threshold (in units of queuing 961 time) is used for the Native L4S AQM, but a ramp is also described as 962 an alternative. And the PI2 algorithm [PI2] is used for the Classic 963 AQM. PI2 is an improved variant of the PIE AQM [RFC8033]. 965 We will introduce the pseudocode in two passes. The first pass 966 explains the core concepts, deferring handling of overload to the 967 second pass. To aid comparison, line numbers are kept in step 968 between the two passes by using letter suffixes where the longer code 969 needs extra lines. 971 A full open source implementation for Linux is available at: 972 https://github.com/olgabo/dualpi2. 974 A.1. Pass #1: Core Concepts 976 The pseudocode manipulates three main structures of variables: the 977 packet (pkt), the L4S queue (lq) and the Classic queue (cq). The 978 pseudocode consists of the following four functions: 980 o initialization code (Figure 2) that sets parameter defaults (the 981 API for setting non-default values is omitted for brevity) 983 o enqueue code (Figure 3) 985 o dequeue code (Figure 4) 987 o code to regularly update the base probability (p) used in the 988 dequeue code (Figure 5). 990 It also uses the following functions that are not shown in full here: 992 o scheduler(), which selects between the head packets of the two 993 queues; the choice of scheduler technology is discussed later; 995 o cq.len() or lq.len() returns the current length (aka. backlog) of 996 the relevant queue in bytes; 998 o cq.time() or lq.time() returns the current queuing delay (aka. 999 sojourn time or service time) of the relevant queue in units of 1000 time; 1002 Queuing delay could be measured directly by storing a per-packet 1003 time-stamp as each packet is enqueued, and subtracting this from the 1004 system time when the packet is dequeued. If time-stamping is not 1005 easy to introduce with certain hardware, queuing delay could be 1006 predicted indirectly by dividing the size of the queue by the 1007 predicted departure rate, which might be known precisely for some 1008 link technologies (see for example [RFC8034]). 1010 In our experiments so far (building on experiments with PIE) on 1011 broadband access links ranging from 4 Mb/s to 200 Mb/s with base RTTs 1012 from 5 ms to 100 ms, DualPI2 achieves good results with the default 1013 parameters in Figure 2. The parameters are categorised by whether 1014 they relate to the Base PI2 AQM, the L4S AQM or the framework 1015 coupling them together. Variables derived from these parameters are 1016 also included at the end of each category. Each parameter is 1017 explained as it is encountered in the walk-through of the pseudocode 1018 below. 1020 1: dualpi2_params_init(...) { % Set input parameter defaults 1021 2: % PI2 AQM parameters 1022 3: target = 15 ms % PI AQM Classic queue delay target 1023 4: Tupdate = 16 ms % PI Classic queue sampling interval 1024 5: alpha = 10 Hz^2 % PI integral gain 1025 6: beta = 100 Hz^2 % PI proportional gain 1026 7: p_Cmax = 1/4 % Max Classic drop/mark prob 1027 8: % Derived PI2 AQM variables 1028 9: alpha_U = alpha *Tupdate % PI integral gain per update interval 1029 10: beta_U = beta * Tupdate % PI prop'nal gain per update interval 1030 11: 1031 12: % DualQ Coupled framework parameters 1032 13: k = 2 % Coupling factor 1033 14: % scheduler weight or equival't parameter (scheduler-dependent) 1034 15: limit = MAX_LINK_RATE * 250 ms % Dual buffer size 1035 16: 1036 17: % L4S AQM parameters 1037 18: T_time = 1 ms % L4S marking threshold in time 1038 19: T_len = 2 * MTU % Min L4S marking threshold in bytes 1039 20: % Derived L4S AQM variables 1040 21: p_Lmax = min(k*sqrt(p_Cmax), 1) % Max L4S marking prob 1041 22: } 1043 Figure 2: Example Header Pseudocode for DualQ Coupled PI2 AQM 1045 The overall goal of the code is to maintain the base probability (p), 1046 which is an internal variable from which the marking and dropping 1047 probabilities for L4S and Classic traffic (p_L and p_C) are derived. 1048 The variable named p in the pseudocode and in this walk-through is 1049 the same as p' (p-prime) in Section 2.4. The probabilities p_L and 1050 p_C are derived in lines 3, 4 and 5 of the dualpi2_update() function 1051 (Figure 5) then used in the dualpi2_dequeue() function (Figure 4). 1052 The code walk-through below builds up to explaining that part of the 1053 code eventually, but it starts from packet arrival. 1055 1: dualpi2_enqueue(lq, cq, pkt) { % Test limit and classify lq or cq 1056 2: if ( lq.len() + cq.len() > limit ) 1057 3: drop(pkt) % drop packet if buffer is full 1058 4: else { % Packet classifier 1059 5: if ( ecn(pkt) modulo 2 == 1 ) % ECN bits = ECT(1) or CE 1060 6: lq.enqueue(pkt) 1061 7: else % ECN bits = not-ECT or ECT(0) 1062 8: cq.enqueue(pkt) 1063 9: } 1064 10: } 1066 Figure 3: Example Enqueue Pseudocode for DualQ Coupled PI2 AQM 1068 1: dualpi2_dequeue(lq, cq, pkt) { % Couples L4S & Classic queues 1069 2: while ( lq.len() + cq.len() > 0 ) 1070 3: if ( scheduler() == lq ) { 1071 4: lq.dequeue(pkt) % Scheduler chooses lq 1072 5: if ( ((lq.time() > T_time) % step marking ... 1073 6: AND (lq.len() > T_len)) 1074 7: OR (p_CL > rand()) ) % ...or linear marking 1075 8: mark(pkt) 1076 9: } else { 1077 10: cq.dequeue(pkt) % Scheduler chooses cq 1078 11: if ( p_C > rand() ) { % probability p_C = p^2 1079 12: if ( ecn(pkt) == 0 ) { % if ECN field = not-ECT 1080 13: drop(pkt) % squared drop 1081 14: continue % continue to the top of the while loop 1082 15: } 1083 16: mark(pkt) % squared mark 1084 17: } 1085 18: } 1086 19: return(pkt) % return the packet and stop 1087 20: } 1088 21: return(NULL) % no packet to dequeue 1089 22: } 1091 Figure 4: Example Dequeue Pseudocode for DualQ Coupled PI2 AQM 1093 When packets arrive, first a common queue limit is checked as shown 1094 in line 2 of the enqueuing pseudocode in Figure 3. Note that the 1095 limit is deliberately tested before enqueue to avoid any bias against 1096 larger packets (so the actual buffer has to be one MTU larger than 1097 limit). If limit is not exceeded, the packet will be classified and 1098 enqueued to the Classic or L4S queue dependent on the least 1099 significant bit of the ECN field in the IP header (line 5). Packets 1100 with a codepoint having an LSB of 0 (Not-ECT and ECT(0)) will be 1101 enqueued in the Classic queue. Otherwise, ECT(1) and CE packets will 1102 be enqueued in the L4S queue. Optional additional packet 1103 classification flexibility is omitted for brevity (see 1104 [I-D.ietf-tsvwg-ecn-l4s-id]). 1106 The dequeue pseudocode (Figure 4) is repeatedly called whenever the 1107 lower layer is ready to forward a packet. It schedules one packet 1108 for dequeuing (or zero if the queue is empty) then returns control to 1109 the caller, so that it does not block while that packet is being 1110 forwarded. While making this dequeue decision, it also makes the 1111 necessary AQM decisions on dropping or marking. The alternative of 1112 applying the AQMs at enqueue would shift some processing from the 1113 critical time when each packet is dequeued. However, it would also 1114 add a whole queue of delay to the control signals, making the control 1115 loop very sloppy. 1117 All the dequeue code is contained within a large while loop so that 1118 if it decides to drop a packet, it will continue until it selects a 1119 packet to schedule. Line 3 of the dequeue pseudocode is where the 1120 scheduler chooses between the L4S queue (lq) and the Classic queue 1121 (cq). Detailed implementation of the scheduler is not shown (see 1122 discussion later). 1124 o If an L4S packet is scheduled, lines 5 to 8 mark the packet if 1125 either the L4S threshold (T_time) is exceeded, or if a random 1126 marking decision is drawn according to p_CL (maintained by the 1127 dualpi2_update() function discussed below). This logical 'OR' on 1128 a per-packet basis implements the max() function shown in Figure 1 1129 to couple the outputs of the two AQMs together. The L4S threshold 1130 is usually in units of time (default T_time = 1 ms). However, on 1131 slow links the packet serialization time can approach the 1132 threshold T_time, so line 6 sets a floor of T_len (=2 MTU) to the 1133 threshold, otherwise marking is always too frequent on slow links. 1135 o If a Classic packet is scheduled, lines 10 to 17 drop or mark the 1136 packet based on the squared probability p_C. 1138 There is some concern that using a step function for the Native L4S 1139 AQM requires end-systems to smooth the signal for a lot longer - 1140 until its fidelity is sufficient. The latency benefits of a ramp are 1141 being investigated as a simple alternative to the step. This ramp 1142 would be similar to the RED algorithm, with the following 1143 differences: 1145 o The min and max of the ramp are defined in units of queuing delay, 1146 not bytes, so that configuration remains invariant as the queue 1147 departure rate varies. 1149 o It uses instantaneous queueing delay without smoothing (smoothing 1150 is done in the end-systems). 1152 o Determinism is being experimented with instead of randomness; to 1153 reduce the delay necessary to smooth out the noise of randomness 1154 from the signal. For each packet, the algorithm would accumulate 1155 p'_L in a counter and mark the packet that took the counter over 1156 1, then subtract 1 from the counter and continue. 1158 o The ramp rises linearly directly from 0 to 1, not to a an 1159 intermediate value of p'_L as RED would, because there is no need 1160 to keep ECN marking probability low. 1162 This ramp algorithm would require two configuration parameters (min 1163 and max threshold in units of queuing time), in contrast to the 1164 single parameter of a step. 1166 1: dualpi2_update(lq, cq, target) { % Update p every Tupdate 1167 2: curq = cq.time() % use queuing time of first-in Classic packet 1168 3: p = p + alpha_U * (curq - target) + beta_U * (curq - prevq) 1169 4: p_CL = p * k % Coupled L4S prob = base prob * coupling factor 1170 5: p_C = p^2 % Classic prob = (base prob)^2 1171 6: prevq = curq 1172 7: } 1174 Figure 5: Example PI-Update Pseudocode for DualQ Coupled PI2 AQM 1176 The base probability (p) is kept up to date by the core PI algorithm 1177 in Figure 5, which is executed every Tupdate. 1179 Note that p solely depends on the queuing time in the Classic queue. 1180 In line 2, the current queuing delay (curq) is evaluated from how 1181 long the head packet was in the Classic queue (cq). The function 1182 cq.time() (not shown) subtracts the time stamped at enqueue from the 1183 current time and implicitly takes the current queuing delay as 0 if 1184 the queue is empty. 1186 The algorithm centres on line 3, which is a classical Proportional- 1187 Integral (PI) controller that alters p dependent on: a) the error 1188 between the current queuing delay (curq) and the target queuing delay 1189 ('target' - see [RFC8033]); and b) the change in queuing delay since 1190 the last sample. The name 'PI' represents the fact that the second 1191 factor (how fast the queue is growing) is _P_roportional to load 1192 while the first is the _I_ntegral of the load (so it removes any 1193 standing queue in excess of the target). 1195 The two 'gain factors' in line 3, alpha_U and beta_U, respectively 1196 weight how strongly each of these elements ((a) and (b)) alters p. 1197 They are in units of 'per second of delay' or Hz, because they 1198 transform differences in queueing delay into changes in probability. 1200 alpha_U and beta_U are derived from the input parameters alpha and 1201 beta (see lines 5 and 6 of Figure 2). These recommended values of 1202 alpha and beta come from the stability analysis in [PI2] so that the 1203 AQM can change p as fast as possible in response to changes in load 1204 without over-compensating and therefore causing oscillations in the 1205 queue. 1207 alpha and beta determine how much p ought to change if it was updated 1208 every second. It is best to update p as frequently as possible, but 1209 the update interval (Tupdate) will probably be constrained by 1210 hardware performance. For link rates from 4 - 200 Mb/s, we found 1211 Tupdate=16ms (as recommended in [RFC8033]) is sufficient. However 1212 small the chosen value of Tupdate, p should change by the same amount 1213 per second, but in finer more frequent steps. So the gain factors 1214 used for updating p in Figure 5 need to be scaled by (Tupdate/1s), 1215 which is done in lines 9 and 10 of Figure 2). The suffix '_U' 1216 represents 'per update time' (Tupdate). 1218 In corner cases, p can overflow the range [0,1] so the resulting 1219 value of p has to be bounded (omitted from the pseudocode). Then, as 1220 already explained, the coupled and Classic probabilities are derived 1221 from the new p in lines 4 and 5 as p_CL = k*p and p_C = p^2. 1223 Because the coupled L4S marking probability (p_CL) is factored up by 1224 k, the dynamic gain parameters alpha and beta are also inherently 1225 factored up by k for the L4S queue, which is necessary to ensure that 1226 Classic TCP and DCTCP controls have the same stability. So, if alpha 1227 is 10 Hz^2, the effective gain factor for the L4S queue is k*alpha, 1228 which is 20 Hz^2 with the default coupling factor of k=2. 1230 Unlike in PIE [RFC8033], alpha_U and beta_U do not need to be tuned 1231 every Tupdate dependent on p. Instead, in PI2, alpha_U and beta_U 1232 are independent of p because the squaring applied to Classic traffic 1233 tunes them inherently. This is explained in [PI2], which also 1234 explains why this more principled approach removes the need for most 1235 of the heuristics that had to be added to PIE. 1237 {ToDo: Scaling beta with Tupdate and scaling both alpha & beta with 1238 RTT} 1240 A.2. Pass #2: Overload Details 1242 Figure 6 repeats the dequeue function of Figure 4, but with overload 1243 details added. Similarly Figure 7 repeats the core PI algorithm of 1244 Figure 5 with overload details added. The initialization and enqueue 1245 functions are unchanged. 1247 In line 7 of the initialization function (Figure 2), the default 1248 maximum Classic drop probability p_Cmax = 1/4 or 25%. This is the 1249 point at which it is deemed that the Classic queue has become 1250 persistently overloaded, so it switches to using solely drop, even 1251 for ECN-capable packets. This protects the queue against any 1252 unresponsive traffic that falsely claims that it is responsive to ECN 1253 marking, as required by [RFC3168] and [RFC7567]. 1255 Line 21 of the initialization function translates this into a maximum 1256 L4S marking probability (p_Lmax) by rearranging Equation (1). With a 1257 coupling factor of k=2 (the default) or greater, this translates to a 1258 maximum L4S marking probability of 1 (or 100%). This is intended to 1259 ensure that the L4S queue starts to introduce dropping once marking 1260 saturates and can rise no further. The 'TCP Prague' requirements 1261 [I-D.ietf-tsvwg-ecn-l4s-id] state that, when an L4S congestion 1262 control detects a drop, it falls back to a response that coexists 1263 with 'Classic' TCP. So it is correct that the L4S queue drops 1264 packets proportional to p^2, as if they are Classic packets. 1266 Both these switch-overs are triggered by the tests for overload 1267 introduced in lines 4b and 12b of the dequeue function (Figure 6). 1268 Lines 8c to 8g drop L4S packets with probability p^2. Lines 8h to 8i 1269 mark the remaining packets with probability p_CL. 1271 Lines 2c to 2d in the core PI algorithm (Figure 7) deal with overload 1272 of the L4S queue when there is no Classic traffic. This is 1273 necessary, because the core PI algorithm maintains the appropriate 1274 drop probability to regulate overload, but it depends on the length 1275 of the Classic queue. If there is no Classic queue the naive 1276 algorithm in Figure 5 drops nothing, even if the L4S queue is 1277 overloaded - so tail drop would have to take over (lines 3 and 4 of 1278 Figure 3). 1280 If the test at line 2a finds that the Classic queue is empty, line 2d 1281 measures the current queue delay using the L4S queue instead. While 1282 the L4S queue is not overloaded, its delay will always be tiny 1283 compared to the target Classic queue delay. So p_L will be driven to 1284 zero, and the L4S queue will naturally be governed solely by 1285 threshold marking (lines 5 and 6 of the dequeue algorithm in 1286 Figure 6). But, if unresponsive L4S source(s) cause overload, the 1287 DualQ transitions smoothly to L4S marking based on the PI algorithm. 1288 And as overload increases, it naturally transitions from marking to 1289 dropping by the switch-over mechanism already described. 1291 1: dualpi2_dequeue(lq, cq) { % Couples L4S & Classic queues, lq & cq 1292 2: while ( lq.len() + cq.len() > 0 ) 1293 3: if ( scheduler() == lq ) { 1294 4a: lq.dequeue(pkt) 1295 4b: if ( p_CL < p_Lmax ) { % Check for overload saturation 1296 5: if ( ((lq.time() > T_time) % step marking ... 1297 6: AND (lq.len > T_len)) 1298 7: OR (p_CL > rand()) ) % ...or linear marking 1299 8a: mark(pkt) 1300 8b: } else { % overload saturation 1301 8c: if ( p_C > rand() ) { % probability p_C = p^2 1302 8e: drop(pkt) % revert to Classic drop due to overload 1303 8f: continue % continue to the top of the while loop 1304 8g: } 1305 8h: if ( p_CL > rand() ) % probability p_CL = k * p 1306 8i: mark(pkt) % linear marking of remaining packets 1307 8j: } 1308 9: } else { 1309 10: cq.dequeue(pkt) 1310 11: if ( p_C > rand() ) { % probability p_C = p^2 1311 12a: if ( (ecn(pkt) == 0) % ECN field = not-ECT 1312 12b: OR (p_C >= p_Cmax) ) { % Overload disables ECN 1313 13: drop(pkt) % squared drop, redo loop 1314 14: continue % continue to the top of the while loop 1315 15: } 1316 16: mark(pkt) % squared mark 1317 17: } 1318 18: } 1319 19: return(pkt) % return the packet and stop 1320 20: } 1321 21: return(NULL) % no packet to dequeue 1322 22: } 1324 Figure 6: Example Dequeue Pseudocode for DualQ Coupled PI2 AQM 1325 (Including Integer Arithmetic and Overload Code) 1327 1: dualpi2_update(lq, cq, target) { % Update p every Tupdate 1328 2a: if ( cq.len() > 0 ) 1329 2b: curq = cq.time() %use queuing time of first-in Classic packet 1330 2c: else % Classic queue empty 1331 2d: curq = lq.time() % use queuing time of first-in L4S packet 1332 3: p = p + alpha_U * (curq - target) + beta_U * (curq - prevq) 1333 4: p_CL = p * k % Coupled L4S prob = base prob * coupling factor 1334 5: p_C = p^2 % Classic prob = (base prob)^2 1335 6: prevq = curq 1336 7: } 1338 Figure 7: Example PI-Update Pseudocode for DualQ Coupled PI2 AQM 1339 (Including Overload Code) 1341 The choice of scheduler technology is critical to overload protection 1342 (see Section 4.1). 1344 o A well-understood weighted scheduler such as weighted round robin 1345 (WRR) is recommended. The scheduler weight for Classic should be 1346 low, e.g. 1/16. 1348 o Alternatively, a time-shifted FIFO could be used. This is a very 1349 simple scheduler, but it does not fully isolate latency in the L4S 1350 queue from uncontrolled bursts in the Classic queue. It works by 1351 selecting the head packet that has waited the longest, biased 1352 against the Classic traffic by a time-shift of tshift. To 1353 implement time-shifted FIFO, the "if (scheduler() == lq )" test in 1354 line 3 of the dequeue code would simply be replaced by "if ( 1355 lq.time() + tshift >= cq.time() )". For the public Internet a 1356 good value for tshift is 50ms. For private networks with smaller 1357 diameter, about 4*target would be reasonable. 1359 o A strict priority scheduler would be inappropriate, because it 1360 would starve Classic if L4S was overloaded. 1362 Appendix B. Example DualQ Coupled Curvy RED Algorithm 1364 As another example of a DualQ Coupled AQM algorithm, the pseudocode 1365 below gives the Curvy RED based algorithm we used and tested. 1366 Although we designed the AQM to be efficient in integer arithmetic, 1367 to aid understanding it is first given using real-number arithmetic. 1368 Then, one possible optimization for integer arithmetic is given, also 1369 in pseudocode. To aid comparison, the line numbers are kept in step 1370 between the two by using letter suffixes where the longer code needs 1371 extra lines. 1373 1: dualq_dequeue(lq, cq) { % Couples L4S & Classic queues, lq & cq 1374 2: if ( lq.dequeue(pkt) ) { 1375 3a: p_L = cq.sec() / 2^S_L 1376 3b: if ( lq.byt() > T ) 1377 3c: mark(pkt) 1378 3d: elif ( p_L > maxrand(U) ) 1379 4: mark(pkt) 1380 5: return(pkt) % return the packet and stop here 1381 6: } 1382 7: while ( cq.dequeue(pkt) ) { 1383 8a: alpha = 2^(-f_C) 1384 8b: Q_C = alpha * pkt.sec() + (1-alpha)* Q_C % Classic Q EWMA 1385 9a: sqrt_p_C = Q_C / 2^S_C 1386 9b: if ( sqrt_p_C > maxrand(2*U) ) 1387 10: drop(pkt) % Squared drop, redo loop 1388 11: else 1389 12: return(pkt) % return the packet and stop here 1390 13: } 1391 14: return(NULL) % no packet to dequeue 1392 15: } 1394 16: maxrand(u) { % return the max of u random numbers 1395 17: maxr=0 1396 18: while (u-- > 0) 1397 19: maxr = max(maxr, rand()) % 0 <= rand() < 1 1398 20: return(maxr) 1399 21: } 1401 Figure 8: Example Dequeue Pseudocode for DualQ Coupled Curvy RED AQM 1403 Packet classification code is not shown, as it is no different from 1404 Figure 3. Potential classification schemes are discussed in 1405 Section 2.3. The Curvy RED algorithm has not been maintained to the 1406 same degree as the DualPI2 algorithm. Some ideas used in DualPI2 1407 would need to be translated into Curvy RED, such as i) the 1408 conditional priority scheduler instead of strict priority ii) the 1409 time-based L4S threshold; iii) turning off ECN as overload 1410 protection; iv) Classic ECN support. These are not shown in the 1411 Curvy RED pseudocode, but would need to be implemented for 1412 production. {ToDo} 1414 At the outer level, the structure of dualq_dequeue() implements 1415 strict priority scheduling. The code is written assuming the AQM is 1416 applied on dequeue (Note 1) . Every time dualq_dequeue() is called, 1417 the if-block in lines 2-6 determines whether there is an L4S packet 1418 to dequeue by calling lq.dequeue(pkt), and otherwise the while-block 1419 in lines 7-13 determines whether there is a Classic packet to 1420 dequeue, by calling cq.dequeue(pkt). (Note 2) 1421 In the lower priority Classic queue, a while loop is used so that, if 1422 the AQM determines that a classic packet should be dropped, it 1423 continues to test for classic packets deciding whether to drop each 1424 until it actually forwards one. Thus, every call to dualq_dequeue() 1425 returns one packet if at least one is present in either queue, 1426 otherwise it returns NULL at line 14. (Note 3) 1428 Within each queue, the decision whether to drop or mark is taken as 1429 follows (to simplify the explanation, it is assumed that U=1): 1431 L4S: If the test at line 2 determines there is an L4S packet to 1432 dequeue, the tests at lines 3a and 3c determine whether to mark 1433 it. The first is a simple test of whether the L4S queue (lq.byt() 1434 in bytes) is greater than a step threshold T in bytes (Note 4). 1435 The second test is similar to the random ECN marking in RED, but 1436 with the following differences: i) the marking function does not 1437 start with a plateau of zero marking until a minimum threshold, 1438 rather the marking probability starts to increase as soon as the 1439 queue is positive; ii) marking depends on queuing time, not bytes, 1440 in order to scale for any link rate without being reconfigured; 1441 iii) marking of the L4S queue does not depend on itself, it 1442 depends on the queuing time of the _other_ (Classic) queue, where 1443 cq.sec() is the queuing time of the packet at the head of the 1444 Classic queue (zero if empty); iv) marking depends on the 1445 instantaneous queuing time (of the other Classic queue), not a 1446 smoothed average; v) the queue is compared with the maximum of U 1447 random numbers (but if U=1, this is the same as the single random 1448 number used in RED). 1450 Specifically, in line 3a the marking probability p_L is set to the 1451 Classic queueing time qc.sec() in seconds divided by the L4S 1452 scaling parameter 2^S_L, which represents the queuing time (in 1453 seconds) at which marking probability would hit 100%. Then in line 1454 3d (if U=1) the result is compared with a uniformly distributed 1455 random number between 0 and 1, which ensures that marking 1456 probability will linearly increase with queueing time. The 1457 scaling parameter is expressed as a power of 2 so that division 1458 can be implemented as a right bit-shift (>>) in line 3 of the 1459 integer variant of the pseudocode (Figure 9). 1461 Classic: If the test at line 7 determines that there is at least one 1462 Classic packet to dequeue, the test at line 9b determines whether 1463 to drop it. But before that, line 8b updates Q_C, which is an 1464 exponentially weighted moving average (Note 5) of the queuing time 1465 in the Classic queue, where pkt.sec() is the instantaneous 1466 queueing time of the current Classic packet and alpha is the EWMA 1467 constant for the classic queue. In line 8a, alpha is represented 1468 as an integer power of 2, so that in line 8 of the integer code 1469 the division needed to weight the moving average can be 1470 implemented by a right bit-shift (>> f_C). 1472 Lines 9a and 9b implement the drop function. In line 9a the 1473 averaged queuing time Q_C is divided by the Classic scaling 1474 parameter 2^S_C, in the same way that queuing time was scaled for 1475 L4S marking. This scaled queuing time is given the variable name 1476 sqrt_p_C because it will be squared to compute Classic drop 1477 probability, so before it is squared it is effectively the square 1478 root of the drop probability. The squaring is done by comparing 1479 it with the maximum out of two random numbers (assuming U=1). 1480 Comparing it with the maximum out of two is the same as the 1481 logical `AND' of two tests, which ensures drop probability rises 1482 with the square of queuing time (Note 6). Again, the scaling 1483 parameter is expressed as a power of 2 so that division can be 1484 implemented as a right bit-shift in line 9 of the integer 1485 pseudocode. 1487 The marking/dropping functions in each queue (lines 3 & 9) are two 1488 cases of a new generalization of RED called Curvy RED, motivated as 1489 follows. When we compared the performance of our AQM with fq_CoDel 1490 and PIE, we came to the conclusion that their goal of holding queuing 1491 delay to a fixed target is misguided [CRED_Insights]. As the number 1492 of flows increases, if the AQM does not allow TCP to increase queuing 1493 delay, it has to introduce abnormally high levels of loss. Then loss 1494 rather than queuing becomes the dominant cause of delay for short 1495 flows, due to timeouts and tail losses. 1497 Curvy RED constrains delay with a softened target that allows some 1498 increase in delay as load increases. This is achieved by increasing 1499 drop probability on a convex curve relative to queue growth (the 1500 square curve in the Classic queue, if U=1). Like RED, the curve hugs 1501 the zero axis while the queue is shallow. Then, as load increases, 1502 it introduces a growing barrier to higher delay. But, unlike RED, it 1503 requires only one parameter, the scaling, not three. The diadvantage 1504 of Curvy RED is that it is not adapted to a wide range of RTTs. 1505 Curvy RED can be used as is when the RTT range to support is limited 1506 otherwise an adaptation mechanism is required. 1508 There follows a summary listing of the two parameters used for each 1509 of the two queues: 1511 Classic: 1513 S_C : The scaling factor of the dropping function scales Classic 1514 queuing times in the range [0, 2^(S_C)] seconds into a dropping 1515 probability in the range [0,1]. To make division efficient, it 1516 is constrained to be an integer power of two; 1518 f_C : To smooth the queuing time of the Classic queue and make 1519 multiplication efficient, we use a negative integer power of 1520 two for the dimensionless EWMA constant, which we define as 1521 alpha = 2^(-f_C). 1523 L4S : 1525 S_L (and k'): As for the Classic queue, the scaling factor of 1526 the L4S marking function scales Classic queueing times in the 1527 range [0, 2^(S_L)] seconds into a probability in the range 1528 [0,1]. Note that S_L = S_C + k', where k' is the coupling 1529 between the queues. So S_L and k' count as only one parameter; 1530 k' is related to k in Equation (1) (Section 2.1) by k=2^k', 1531 where both k and k' are constants. Then implementations can 1532 avoid costly division by shifting p_L by k' bits to the right. 1534 T : The queue size in bytes at which step threshold marking 1535 starts in the L4S queue. 1537 {ToDo: These are the raw parameters used within the algorithm. A 1538 configuration front-end could accept more meaningful parameters and 1539 convert them into these raw parameters.} 1541 From our experiments so far, recommended values for these parameters 1542 are: S_C = -1; f_C = 5; T = 5 * MTU for the range of base RTTs 1543 typical on the public Internet. [CRED_Insights] explains why these 1544 parameters are applicable whatever rate link this AQM implementation 1545 is deployed on and how the parameters would need to be adjusted for a 1546 scenario with a different range of RTTs (e.g. a data centre) {ToDo 1547 incorporate a summary of that report into this draft}. The setting of 1548 k depends on policy (see Section 2.5 and Appendix C respectively for 1549 its recommended setting and guidance on alternatives). 1551 There is also a cUrviness parameter, U, which is a small positive 1552 integer. It is likely to take the same hard-coded value for all 1553 implementations, once experiments have determined a good value. We 1554 have solely used U=1 in our experiments so far, but results might be 1555 even better with U=2 or higher. 1557 Note that the dropping function at line 9 calls maxrand(2*U), which 1558 gives twice as much curviness as the call to maxrand(U) in the 1559 marking function at line 3. This is the trick that implements the 1560 square rule in equation (1) (Section 2.1). This is based on the fact 1561 that, given a number X from 1 to 6, the probability that two dice 1562 throws will both be less than X is the square of the probability that 1563 one throw will be less than X. So, when U=1, the L4S marking 1564 function is linear and the Classic dropping function is squared. If 1565 U=2, L4S would be a square function and Classic would be quartic. 1566 And so on. 1568 The maxrand(u) function in lines 16-21 simply generates u random 1569 numbers and returns the maximum (Note 7). Typically, maxrand(u) 1570 could be run in parallel out of band. For instance, if U=1, the 1571 Classic queue would require the maximum of two random numbers. So, 1572 instead of calling maxrand(2*U) in-band, the maximum of every pair of 1573 values from a pseudorandom number generator could be generated out- 1574 of-band, and held in a buffer ready for the Classic queue to consume. 1576 1: dualq_dequeue(lq, cq) { % Couples L4S & Classic queues, lq & cq 1577 2: if ( lq.dequeue(pkt) ) { 1578 3: if ((lq.byt() > T) || ((cq.ns() >> (S_L-2)) > maxrand(U))) 1579 4: mark(pkt) 1580 5: return(pkt) % return the packet and stop here 1581 6: } 1582 7: while ( cq.dequeue(pkt) ) { 1583 8: Q_C += (pkt.ns() - Q_C) >> f_C % Classic Q EWMA 1584 9: if ( (Q_C >> (S_C-2) ) > maxrand(2*U) ) 1585 10: drop(pkt) % Squared drop, redo loop 1586 11: else 1587 12: return(pkt) % return the packet and stop here 1588 13: } 1589 14: return(NULL) % no packet to dequeue 1590 15: } 1592 Figure 9: Optimised Example Dequeue Pseudocode for Coupled DualQ AQM 1593 using Integer Arithmetic 1595 Notes: 1597 1. The drain rate of the queue can vary if it is scheduled relative 1598 to other queues, or to cater for fluctuations in a wireless 1599 medium. To auto-adjust to changes in drain rate, the queue must 1600 be measured in time, not bytes or packets [CoDel]. In our Linux 1601 implementation, it was easiest to measure queuing time at 1602 dequeue. Queuing time can be estimated when a packet is enqueued 1603 by measuring the queue length in bytes and dividing by the recent 1604 drain rate. 1606 2. An implementation has to use priority queueing, but it need not 1607 implement strict priority. 1609 3. If packets can be enqueued while processing dequeue code, an 1610 implementer might prefer to place the while loop around both 1611 queues so that it goes back to test again whether any L4S packets 1612 arrived while it was dropping a Classic packet. 1614 4. In order not to change too many factors at once, for now, we keep 1615 the marking function for DCTCP-only traffic as similar as 1616 possible to DCTCP. However, unlike DCTCP, all processing is at 1617 dequeue, so we determine whether to mark a packet at the head of 1618 the queue by the byte-length of the queue _behind_ it. We plan 1619 to test whether using queuing time will work in all 1620 circumstances, and if we find that the step can cause 1621 oscillations, we will investigate replacing it with a steep 1622 random marking curve. 1624 5. An EWMA is only one possible way to filter bursts; other more 1625 adaptive smoothing methods could be valid and it might be 1626 appropriate to decrease the EWMA faster than it increases. 1628 6. In practice at line 10 the Classic queue would probably test for 1629 ECN capability on the packet to determine whether to drop or mark 1630 the packet. However, for brevity such detail is omitted. All 1631 packets classified into the L4S queue have to be ECN-capable, so 1632 no dropping logic is necessary at line 3. Nonetheless, L4S 1633 packets could be dropped by overload code (see Section 4.1). 1635 7. In the integer variant of the pseudocode (Figure 9) real numbers 1636 are all represented as integers scaled up by 2^32. In lines 3 & 1637 9 the function maxrand() is arranged to return an integer in the 1638 range 0 <= maxrand() < 2^32. Queuing times are also scaled up by 1639 2^32, but in two stages: i) In lines 3 and 8 queuing times 1640 cq.ns() and pkt.ns() are returned in integer nanoseconds, making 1641 the values about 2^30 times larger than when the units were 1642 seconds, ii) then in lines 3 and 9 an adjustment of -2 to the 1643 right bit-shift multiplies the result by 2^2, to complete the 1644 scaling by 2^32. 1646 Appendix C. Guidance on Controlling Throughput Equivalence 1648 +---------------+------+-------+ 1649 | RTT_C / RTT_L | Reno | Cubic | 1650 +---------------+------+-------+ 1651 | 1 | k'=1 | k'=0 | 1652 | 2 | k'=2 | k'=1 | 1653 | 3 | k'=2 | k'=2 | 1654 | 4 | k'=3 | k'=2 | 1655 | 5 | k'=3 | k'=3 | 1656 +---------------+------+-------+ 1658 Table 1: Value of k' for which DCTCP throughput is roughly the same 1659 as Reno or Cubic, for some example RTT ratios 1661 k' is related to k in Equation (1) (Section 2.1) by k=2^k'. 1663 To determine the appropriate policy, the operator first has to judge 1664 whether it wants DCTCP flows to have roughly equal throughput with 1665 Reno or with Cubic (because, even in its Reno-compatibility mode, 1666 Cubic is about 1.4 times more aggressive than Reno). Then the 1667 operator needs to decide at what ratio of RTTs it wants DCTCP and 1668 Classic flows to have roughly equal throughput. For example choosing 1669 k'=0 (equivalent to k=1) will make DCTCP throughput roughly the same 1670 as Cubic, _if their RTTs are the same_. 1672 However, even if the base RTTs are the same, the actual RTTs are 1673 unlikely to be the same, because Classic (Cubic or Reno) traffic 1674 needs a large queue to avoid under-utilization and excess drop, 1675 whereas L4S (DCTCP) does not. The operator might still choose this 1676 policy if it judges that DCTCP throughput should be rewarded for 1677 keeping its own queue short. 1679 On the other hand, the operator will choose one of the higher values 1680 for k', if it wants to slow DCTCP down to roughly the same throughput 1681 as Classic flows, to compensate for Classic flows slowing themselves 1682 down by causing themselves extra queuing delay. 1684 The values for k' in the table are derived from the formulae, which 1685 was developed in [DCttH15]: 1687 2^k' = 1.64 (RTT_reno / RTT_dc) (2) 1688 2^k' = 1.19 (RTT_cubic / RTT_dc ) (3) 1690 For localized traffic from a particular ISP's data centre, we used 1691 the measured RTTs to calculate that a value of k'=3 (equivalant to 1692 k=8) would achieve throughput equivalence, and our experiments 1693 verified the formula very closely. 1695 For a typical mix of RTTs from local data centres and across the 1696 general Internet, a value of k'=1 (equivalent to k=2) is recommended 1697 as a good workable compromise. 1699 Appendix D. Open Issues 1701 Most of the following open issues are also tagged '{ToDo}' at the 1702 appropriate point in the document: 1704 Operational guidance to monitor L4S experiment 1706 PI2 appendix: scaling of alpha & beta, esp. dependence of beta_U 1707 on Tupdate 1709 Curvy RED appendix: complete the unfinished parts 1711 Authors' Addresses 1713 Koen De Schepper 1714 Nokia Bell Labs 1715 Antwerp 1716 Belgium 1718 Email: koen.de_schepper@nokia.com 1719 URI: https://www.bell-labs.com/usr/koen.de_schepper 1721 Bob Briscoe (editor) 1722 CableLabs 1723 UK 1725 Email: ietf@bobbriscoe.net 1726 URI: http://bobbriscoe.net/ 1728 Olga Bondarenko 1729 Simula Research Lab 1730 Lysaker 1731 Norway 1733 Email: olgabnd@gmail.com 1734 URI: https://www.simula.no/people/olgabo 1736 Ing-jyh Tsang 1737 Nokia 1738 Antwerp 1739 Belgium 1741 Email: ing-jyh.tsang@nokia.com