idnits 2.17.1 draft-ietf-tsvwg-ecn-encap-guidelines-08.txt: Checking boilerplate required by RFC 5378 and the IETF Trust (see https://trustee.ietf.org/license-info): ---------------------------------------------------------------------------- No issues found here. Checking nits according to https://www.ietf.org/id-info/1id-guidelines.txt: ---------------------------------------------------------------------------- No issues found here. Checking nits according to https://www.ietf.org/id-info/checklist : ---------------------------------------------------------------------------- -- The draft header indicates that this document updates RFC3819, but the abstract doesn't seem to mention this, which it should. Miscellaneous warnings: ---------------------------------------------------------------------------- == The copyright year in the IETF Trust and authors Copyright Line does not match the current year (Using the creation date from RFC3819, updated by this document, for RFC5378 checks: 1999-10-14) -- The document seems to lack a disclaimer for pre-RFC5378 work, but may have content which was first submitted before 10 November 2008. If you have contacted all the original authors and they are all willing to grant the BCP78 rights to the IETF Trust, then this is fine, and you can ignore this comment. If not, you may need to add the pre-RFC5378 disclaimer. (See the Legal Provisions document at https://trustee.ietf.org/license-info for more information.) -- The document date (March 27, 2017) is 2580 days in the past. Is this intentional? Checking references for intended status: Best Current Practice ---------------------------------------------------------------------------- (See RFCs 3967 and 4897 for information about using normative references to lower-maturity documents in RFCs) == Missing Reference: 'GF' is mentioned on line 1319, but not defined == Outdated reference: A later version (-16) exists of draft-ietf-nvo3-geneve-04 == Outdated reference: A later version (-07) exists of draft-ietf-trill-ecn-support-02 == Outdated reference: A later version (-23) exists of draft-ietf-tsvwg-rfc6040update-shim-00 -- Obsolete informational reference (is this intentional?): RFC 1323 (Obsoleted by RFC 7323) Summary: 0 errors (**), 0 flaws (~~), 5 warnings (==), 4 comments (--). Run idnits with the --verbose option for more detailed information about the items above. -------------------------------------------------------------------------------- 2 Transport Area Working Group B. Briscoe 3 Internet-Draft Simula Research Laboratory 4 Updates: 3819 (if approved) J. Kaippallimalil 5 Intended status: Best Current Practice Huawei 6 Expires: September 28, 2017 P. Thaler 7 Broadcom Corporation 8 March 27, 2017 10 Guidelines for Adding Congestion Notification to Protocols that 11 Encapsulate IP 12 draft-ietf-tsvwg-ecn-encap-guidelines-08 14 Abstract 16 The purpose of this document is to guide the design of congestion 17 notification in any lower layer or tunnelling protocol that 18 encapsulates IP. The aim is for explicit congestion signals to 19 propagate consistently from lower layer protocols into IP. Then the 20 IP internetwork layer can act as a portability layer to carry 21 congestion notification from non-IP-aware congested nodes up to the 22 transport layer (L4). Following these guidelines should assure 23 interworking between new lower layer congestion notification 24 mechanisms, whether specified by the IETF or other standards bodies. 26 Status of This Memo 28 This Internet-Draft is submitted in full conformance with the 29 provisions of BCP 78 and BCP 79. 31 Internet-Drafts are working documents of the Internet Engineering 32 Task Force (IETF). Note that other groups may also distribute 33 working documents as Internet-Drafts. The list of current Internet- 34 Drafts is at http://datatracker.ietf.org/drafts/current/. 36 Internet-Drafts are draft documents valid for a maximum of six months 37 and may be updated, replaced, or obsoleted by other documents at any 38 time. It is inappropriate to use Internet-Drafts as reference 39 material or to cite them other than as "work in progress." 41 This Internet-Draft will expire on September 28, 2017. 43 Copyright Notice 45 Copyright (c) 2017 IETF Trust and the persons identified as the 46 document authors. All rights reserved. 48 This document is subject to BCP 78 and the IETF Trust's Legal 49 Provisions Relating to IETF Documents 50 (http://trustee.ietf.org/license-info) in effect on the date of 51 publication of this document. Please review these documents 52 carefully, as they describe your rights and restrictions with respect 53 to this document. Code Components extracted from this document must 54 include Simplified BSD License text as described in Section 4.e of 55 the Trust Legal Provisions and are provided without warranty as 56 described in the Simplified BSD License. 58 Table of Contents 60 1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . 3 61 1.1. Scope . . . . . . . . . . . . . . . . . . . . . . . . . . 5 62 2. Terminology . . . . . . . . . . . . . . . . . . . . . . . . . 6 63 3. Guidelines in All Cases . . . . . . . . . . . . . . . . . . . 7 64 4. Modes of Operation . . . . . . . . . . . . . . . . . . . . . 8 65 4.1. Feed-Forward-and-Up Mode . . . . . . . . . . . . . . . . 8 66 4.2. Feed-Up-and-Forward Mode . . . . . . . . . . . . . . . . 10 67 4.3. Feed-Backward Mode . . . . . . . . . . . . . . . . . . . 11 68 4.4. Null Mode . . . . . . . . . . . . . . . . . . . . . . . . 13 69 5. Feed-Forward-and-Up Mode: Guidelines for Adding Congestion 70 Notification . . . . . . . . . . . . . . . . . . . . . . . . 13 71 5.1. IP-in-IP Tunnels with Tightly Coupled Shim Headers . . . 14 72 5.2. Wire Protocol Design: Indication of ECN Support . . . . . 14 73 5.3. Encapsulation Guidelines . . . . . . . . . . . . . . . . 16 74 5.4. Decapsulation Guidelines . . . . . . . . . . . . . . . . 18 75 5.5. Sequences of Similar Tunnels or Subnets . . . . . . . . . 19 76 5.6. Reframing and Congestion Markings . . . . . . . . . . . . 20 77 6. Feed-Up-and-Forward Mode: Guidelines for Adding Congestion 78 Notification . . . . . . . . . . . . . . . . . . . . . . . . 20 79 7. Feed-Backward Mode: Guidelines for Adding Congestion 80 Notification . . . . . . . . . . . . . . . . . . . . . . . . 22 81 8. IANA Considerations (to be removed by RFC Editor) . . . . . . 23 82 9. Security Considerations . . . . . . . . . . . . . . . . . . . 23 83 10. Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . 24 84 11. Acknowledgements . . . . . . . . . . . . . . . . . . . . . . 24 85 12. Comments Solicited . . . . . . . . . . . . . . . . . . . . . 24 86 13. References . . . . . . . . . . . . . . . . . . . . . . . . . 25 87 13.1. Normative References . . . . . . . . . . . . . . . . . . 25 88 13.2. Informative References . . . . . . . . . . . . . . . . . 25 89 Appendix A. Outstanding Document Issues . . . . . . . . . . . . 30 90 Appendix B. Changes in This Version (to be removed by RFC 91 Editor) . . . . . . . . . . . . . . . . . . . . . . 30 92 Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . 33 94 1. Introduction 96 The benefits of Explicit Congestion Notification (ECN) described 97 below can only be fully realised if support for ECN is added to the 98 relevant subnetwork technology, as well as to IP. When a lower layer 99 buffer drops a packet obviously it does not just drop at that layer; 100 the packet disappears from all layers. In contrast, when a lower 101 layer marks a packet with ECN, the marking needs to be explicitly 102 propagated up the layers. The same is true if a buffer marks the 103 outer header of a packet that encapsulates inner tunnelled headers. 104 Forwarding ECN is not as straightforward as other headers because it 105 has to be assumed ECN may be only partially deployed. If an egress 106 at any layer is not ECN-aware, or if the ultimate receiver or sender 107 is not ECN-aware, congestion needs to be indicated by dropping a 108 packet, not marking it. 110 The purpose of this document is to guide the addition of congestion 111 notification to any subnet technology or tunnelling protocol, so that 112 lower layer equipment can signal congestion explicitly and it will 113 propagate consistently into encapsulated (higher layer) headers, 114 otherwise the signals will not reach their ultimate destination. 116 ECN is defined in the IP header (v4 and v6) [RFC3168] to allow a 117 resource to notify the onset of queue build-up without having to drop 118 packets, by explicitly marking a proportion of packets with the 119 congestion experienced (CE) codepoint. 121 Given a suitable marking scheme, ECN removes nearly all congestion 122 loss and it cuts delays for two main reasons: 124 o It avoids the delay when recovering from congestion losses, which 125 particularly benefits small flows or real-time flows, making their 126 delivery time predictably short [RFC2884]; 128 o As ECN is used more widely by end-systems, it will gradually 129 remove the need to configure a degree of delay into buffers before 130 they start to notify congestion (the cause of bufferbloat). This 131 is because drop involves a trade-off between sending a timely 132 signal and trying to avoid impairment, whereas ECN is solely a 133 signal not an impairment, so there is no harm triggering it 134 earlier. 136 Some lower layer technologies (e.g. MPLS, Ethernet) are used to form 137 subnetworks with IP-aware nodes only at the edges. These networks 138 are often sized so that it is rare for interior queues to overflow. 139 However, until recently this was more due to the inability of TCP to 140 saturate the links. For many years, fixes such as window scaling 141 [RFC1323] proved hard to deploy. And the New Reno variant of TCP has 142 remained in widespread use despite its inability to scale to high 143 flow rates. However, now that modern operating systems are finally 144 capable of saturating interior links, even the buffers of well- 145 provisioned interior switches will need to signal episodes of 146 queuing. 148 Propagation of ECN is defined for MPLS [RFC5129], and is being 149 defined for TRILL [RFC7780], [I-D.ietf-trill-ecn-support], but it 150 remains to be defined for a number of other subnetwork technologies. 152 Similarly, ECN propagation is yet to be defined for many tunnelling 153 protocols. [RFC6040] defines how ECN should be propagated for IP-in- 154 IP [RFC2003] and IPsec [RFC4301] tunnels, and it is cited by more 155 recent tunnelling protocols, e.g. Generic UDP Encapsulation (GUE) 156 [I-D.ietf-nvo3-gue] and Geneve [I-D.ietf-nvo3-geneve]. However, as 157 Section 9.3 of RFC3168 pointed out, ECN support will need to be 158 defined for other tunnelling protocols, e.g. L2TP [RFC2661], GRE 159 [RFC1701], [RFC2784], PPTP [RFC2637] and GTP [GTPv1], [GTPv1-U], 160 [GTPv2-C], VXLAN [RFC7348]. 162 Incremental deployment is the most delicate aspect when adding 163 support for ECN. The original ECN protocol in IP [RFC3168] was 164 carefully designed so that a congested buffer would not mark a packet 165 (rather than drop it) unless both source and destination hosts were 166 ECN-capable. Otherwise its congestion markings would never be 167 detected and congestion would just build up further. However, to 168 support congestion marking below the IP layer, it is not sufficient 169 to only check that the two end-points support ECN; correct operation 170 also depends on the decapsulator at each subnet egress faithfully 171 propagating congestion notifications to the higher layer. Otherwise, 172 a legacy decapsulator might silently fail to propagate any ECN 173 signals from the outer to the forwarded header. Then the lost 174 signals would never be detected and again congestion would build up 175 further. The guidelines given later require protocol designers to 176 carefully consider incremental deployment, and suggest various safe 177 approaches for different circumstances. 179 Of course, the IETF does not have standards authority over every link 180 layer protocol. So this document gives guidelines for designing 181 propagation of congestion notification across the interface between 182 IP and protocols that may encapsulate IP (i.e. that can be layered 183 beneath IP). Each lower layer technology will exhibit different 184 issues and compromises, so the IETF or the relevant standards body 185 must be free to define the specifics of each lower layer congestion 186 notification scheme. Nonetheless, if the guidelines are followed, 187 congestion notification should interwork between different 188 technologies, using IP in its role as a 'portability layer'. 190 Therefore, the capitalised term 'SHOULD' or 'SHOULD NOT' are often 191 used in preference to 'MUST' or 'MUST NOT', because it is difficult 192 to know the compromises that will be necessary in each protocol 193 design. If a particular protocol design chooses to contradict a 194 'SHOULD (NOT)' given in the advice below, it MUST include a sound 195 justification. 197 It has not been possible to give common guidelines for all lower 198 layer technologies, because they do not all fit a common pattern. 199 Instead they have been divided into a few distinct modes of 200 operation: feed-forward-and-upward; feed-upward-and-forward; feed- 201 backward; and null mode. These modes are described in Section 4, 202 then in the following sections separate guidelines are given for each 203 mode. 205 This document updates the advice to subnetwork designers about ECN in 206 Section 13 of [RFC3819]. 208 1.1. Scope 210 This document only concerns wire protocol processing of explicit 211 notification of congestion and makes no changes or recommendations 212 concerning algorithms for congestion marking or for congestion 213 response (algorithm issues should be independent of the layer the 214 algorithm operates in). 216 The question of congestion notification signals with different 217 semantics to those of ECN in IP is touched on in a couple of specific 218 cases (e.g. QCN [IEEE802.1Qau]) and with schemes with multiple 219 severity levels such as PCN [RFC6660]). However, no attempt is made 220 to give guidelines about schemes with different semantics that are 221 yet to be invented. 223 The semantics of congestion signals can be relative to the traffic 224 class. Therefore correct propagation of congestion signals could 225 depend on correct propagation of any traffic class field between the 226 layers. In this document, correct propagation of traffic class 227 information is assumed, while what 'correct' means and how it is 228 achieved is covered elsewhere (e.g. [RFC2983]) and is outside the 229 scope of the present document. 231 Note that these guidelines do not require the subnet wire protocol to 232 be changed to accommodate congestion notification. Another way to 233 add congestion notification without consuming header space in the 234 subnet protocol might be to use a parallel control plane protocol. 236 This document focuses on the congestion notification interface 237 between IP and lower layer protocols that can encapsulate IP, where 238 the term 'IP' includes v4 or v6, unicast, multicast or anycast. 239 However, it is likely that the guidelines will also be useful when a 240 lower layer protocol or tunnel encapsulates itself (e.g. Ethernet 241 MAC in MAC [IEEE802.1Qah]) or when it encapsulates other protocols. 242 In the feed-backward mode, propagation of congestion signals for 243 multicast and anycast packets is out-of-scope (because it would be so 244 complicated that it is hoped no-one would attempt such an 245 abomination). 247 2. Terminology 249 The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", 250 "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this 251 document are to be interpreted as described in RFC 2119 [RFC2119]. 253 Further terminology used within this document: 255 Protocol data unit (PDU): Information that is delivered as a unit 256 among peer entities of a layered network consisting of protocol 257 control information (typically a header) and possibly user data 258 (payload) of that layer. The scope of this document includes 259 layer 2 and layer 3 networks, where the PDU is respectively termed 260 a frame or a packet (or a cell in ATM). PDU is a general term for 261 any of these. This definition also includes a payload with a shim 262 header lying somewhere between layer 2 and 3. 264 Transport: The end-to-end transmission control function, 265 conventionally considered at layer-4 in the OSI reference model. 266 Given the audience for this document will often use the word 267 transport to mean low level bit carriage, whenever the term is 268 used it will be qualified, e.g. 'L4 transport'. 270 Encapsulator: The link or tunnel endpoint function that adds an 271 outer header to a PDU (also termed the 'link ingress', the 'subnet 272 ingress', the 'ingress tunnel endpoint' or just the 'ingress' 273 where the context is clear). 275 Decapsulator: The link or tunnel endpoint function that removes an 276 outer header from a PDU (also termed the 'link egress', the 277 'subnet egress', the 'egress tunnel endpoint' or just the 'egress' 278 where the context is clear). 280 Incoming header: The header of an arriving PDU before encapsulation. 282 Outer header: The header added to encapsulate a PDU. 284 Inner header: The header encapsulated by the outer header. 286 Outgoing header: The header forwarded by the decapsulator. 288 CE: Congestion Experienced [RFC3168] 290 ECT: ECN-Capable Transport [RFC3168] 292 Not-ECT: Not ECN-Capable Transport [RFC3168] 294 Load Regulator: For each flow of PDUs, the transport function that 295 is capable of controlling the data rate. Typically located at the 296 data source, but in-path nodes can regulate load in some 297 congestion control arrangements (e.g. admission control, policing 298 nodes or transport circuit-breakers [RFC8084]). Note the term "a 299 function capable of controlling the load" deliberately includes a 300 transport that doesn't actually control the load responsively but 301 ideally it ought to (e.g. a sending application without 302 congestion control that uses UDP). 304 ECN-PDU: A PDU that is part of a feedback loop within which all the 305 nodes that need to propagate explicit congestion notifications 306 back to the Load Regulator are ECN-capable. An IP packet with a 307 non-zero ECN field implies that the endpoints are ECN-capable, so 308 this would be an ECN-PDU. However, ECN-PDU is intended to be a 309 general term for a PDU at any layer, not just IP. 311 Not-ECN-PDU: A PDU that is part of a feedback-loop within which some 312 nodes necessary to propagate explicit congestion notifications 313 back to the load regulator are not ECN-capable. 315 Congestion Baseline: The location of the function on the path that 316 initialised the values of all congestion notification fields in a 317 sequence of packets, before any are set to the congestion 318 experienced (CE) codepoint if they experience congestion further 319 downstream. Typically the original data source at layer-4. 321 3. Guidelines in All Cases 323 RFC 3168 specifies that the ECN field in the IP header is intended to 324 be marked by active queue management algorithms. Any congestion 325 notification from an algorithm that does not conform to the 326 recommendations in [RFC7567] MUST NOT be propagated from a lower 327 layer into the ECN field in IP (see also [RFC4774] on alternate uses 328 of the ECN field). 330 4. Modes of Operation 332 This section sets down the different modes by which congestion 333 information is passed between the lower layer and the higher one. It 334 acts as a reference framework for the following sections, which give 335 normative guidelines for designers of explicit congestion 336 notification protocols, taking each mode in turn: 338 Feed-Forward-and-Up: Nodes feed forward congestion notification 339 towards the egress within the lower layer then up and along the 340 layers towards the end-to-end destination at the transport layer. 341 The following local optimisation is possible: 343 Feed-Up-and-Forward: A lower layer switch feeds-up congestion 344 notification directly into the ECN field in the higher layer 345 (e.g. IP) header, irrespective of whether the node is at the 346 egress of a subnet. 348 Feed-Backward: Nodes feed back congestion signals towards the 349 ingress of the lower layer and (optionally) attempt to control 350 congestion within their own layer. 352 Null: Nodes cannot experience congestion at the lower layer except 353 at ingress nodes (which are IP-aware or equivalently higher-layer- 354 aware). 356 4.1. Feed-Forward-and-Up Mode 358 Like IP and MPLS, many subnet technologies are based on self- 359 contained protocol data units (PDUs) or frames sent unreliably. They 360 provide no feedback channel at the subnetwork layer, instead relying 361 on higher layers (e.g. TCP) to feed back loss signals. 363 In these cases, ECN may best be supported by standardising explicit 364 notification of congestion into the lower layer protocol that carries 365 the data forwards. It will then also be necessary to define how the 366 egress of the lower layer subnet propagates this explicit signal into 367 the forwarded upper layer (IP) header. It can then continue forwards 368 until it finally reaches the destination transport (at L4). Then 369 typically the destination will feed this congestion notification back 370 to the source transport using an end-to-end protocol (e.g. TCP). 371 This is the arrangement that has already been used to add ECN to IP- 372 in-IP tunnels [RFC6040], IP-in-MPLS and MPLS-in-MPLS [RFC5129]. 374 This mode is illustrated in Figure 1. Along the middle of the 375 figure, layers 2, 3 and 4 of the protocol stack are shown, and one 376 packet is shown along the bottom as it progresses across the network 377 from source to destination, crossing two subnets connected by a 378 router, and crossing two switches on the path across each subnet. 379 Congestion at the output of the first switch (shown as *) leads to a 380 congestion marking in the L2 header (shown as C in the illustration 381 of the packet). The chevrons show the progress of the resulting 382 congestion indication. It is propagated from link to link across the 383 subnet in the L2 header, then when the router removes the marked L2 384 header, it propagates the marking up into the L3 (IP) header. The 385 router forwards the marked L3 header into subnet 2, and when it adds 386 a new L2 header it copies the L3 marking into the L2 header as well, 387 as shown by the 'C's in both layers (assuming the technology of 388 subnet 2 also supports explicit congestion marking). 390 Note that there is no implication that each 'C' marking is encoded 391 the same; a different encoding might be used for the 'C' marking in 392 each protocol. 394 Finally, for completeness, we show the L3 marking arriving at the 395 destination, where the host transport protocol (e.g. TCP) feeds it 396 back to the source in the L4 acknowledgement (the 'C' at L4 in the 397 packet at the top of the diagram). 399 _ _ _ 400 /_______ | | |C| ACK Packet (V) 401 \ |_|_|_| 402 +---+ layer: 2 3 4 header +---+ 403 | <|<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<< Packet V <<<<<<<<<<<<<|<< |L4 404 | | +---+ | ^ | 405 | | . . . . . . Packet U. . | >>|>>> Packet U >>>>>>>>>>>>|>^ |L3 406 | | +---+ +---+ | ^ | +---+ +---+ | | 407 | | | *|>>>>>|>>>|>>>>>|>^ | | | | | | |L2 408 |___|_____|___|_____|___|_____|___|_____|___|_____|___|_____|___| 409 source subnet A router subnet B dest 410 __ _ _ _ __ _ _ _ __ _ _ __ _ _ _ 411 | | | | | | | | |C| | | |C| | | |C|C| Data________\ 412 |__|_|_|_| |__|_|_|_| |__|_|_| |__|_|_|_| Packet (U) / 413 layer: 4 3 2A 4 3 2A 4 3 4 3 2B 414 header 416 Figure 1: Feed-Forward-and-Up Mode 418 Of course, modern networks are rarely as simple as this text-book 419 example, often involving multiple nested layers. For example, a 3GPP 420 mobile network may have two IP-in-IP (GTP) tunnels in series and an 421 MPLS backhaul between the base station and the first router. 422 Nonetheless, the example illustrates the general idea of feeding 423 congestion notification forward then upward whenever a header is 424 removed at the egress of a subnet. 426 Note that the FECN (forward ECN) bit in Frame Relay and the explicit 427 forward congestion indication (EFCI [ITU-T.I.371]) bit in ATM user 428 data cells follow a feed-forward pattern. However, in ATM, this 429 arrangement is only part of a feed-forward-and-backward pattern at 430 the lower layer, not feed-forward-and-up out of the lower layer--the 431 intention was never to interface to IP ECN at the subnet egress. To 432 our knowledge, Frame Relay FECN is solely used to detect where more 433 capacity should be provisioned [Buck00]. 435 4.2. Feed-Up-and-Forward Mode 437 Ethernet is particularly difficult to extend incrementally to support 438 explicit congestion notification. One way to support ECN in such 439 cases has been to use so called 'layer-3 switches'. These are 440 Ethernet switches that bury into the Ethernet payload to find an IP 441 header and manipulate or act on certain IP fields (specifically 442 Diffserv & ECN). For instance, in Data Center TCP [DCTCP], layer-3 443 switches are configured to mark the ECN field of the IP header within 444 the Ethernet payload when their output buffer becomes congested. 445 With respect to switching, a layer-3 switch acts solely on the 446 addresses in the Ethernet header; it doesn't use IP addresses, and it 447 doesn't decrement the TTL field in the IP header. 449 _ _ _ 450 /_______ | | |C| ACK packet (V) 451 \ |_|_|_| 452 +---+ layer: 2 3 4 header +---+ 453 | <|<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<< Packet V <<<<<<<<<<<<<|<< |L4 454 | | +---+ | ^ | 455 | | . . . >>>> Packet U >>>|>>>|>>> Packet U >>>>>>>>>>>>|>^ |L3 456 | | +--^+ +---+ | | +---+ +---+ | | 457 | | | *| | | | | | | | | | |L2 458 |___|_____|___|_____|___|_____|___|_____|___|_____|___|_____|___| 459 source subnet E router subnet F dest 460 __ _ _ _ __ _ _ _ __ _ _ __ _ _ _ 461 | | | | | | | |C| | | | |C| | | |C|C| data________\ 462 |__|_|_|_| |__|_|_|_| |__|_|_| |__|_|_|_| packet (U) / 463 layer: 4 3 2 4 3 2 4 3 4 3 2 464 header 466 Figure 2: Feed-Up-and-Forward Mode 468 By comparing Figure 2 with Figure 1, it can be seen that subnet E 469 (perhaps a subnet of layer-3 Ethernet switches) works in feed-up-and- 470 forward mode by notifying congestion directly into L3 at the point of 471 congestion, even though the congested switch does not otherwise act 472 at L3. In this example, the technology in subnet F (e.g. MPLS) does 473 support ECN natively, so when the router adds the layer-2 header it 474 copies the ECN marking from L3 to L2 as well. 476 4.3. Feed-Backward Mode 478 In some layer 2 technologies, explicit congestion notification has 479 been defined for use internally within the subnet with its own 480 feedback and load regulation, but typically the interface with IP for 481 ECN has not been defined. 483 For instance, for the available bit-rate (ABR) service in ATM, the 484 relative rate mechanism was one of the more popular mechanisms for 485 managing traffic, tending to supersede earlier designs. In this 486 approach ATM switches send special resource management (RM) cells in 487 both the forward and backward directions to control the ingress rate 488 of user data into a virtual circuit. If a switch buffer is 489 approaching congestion or is congested it sends an RM cell back 490 towards the ingress with respectively the No Increase (NI) or 491 Congestion Indication (CI) bit set in its message type field 492 [ATM-TM-ABR]. The ingress then holds or decreases its sending bit- 493 rate accordingly. 495 _ _ _ 496 /_______ | | |C| ACK packet (X) 497 \ |_|_|_| 498 +---+ layer: 2 3 4 header +---+ 499 | <|<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<< Packet X <<<<<<<<<<<<<|<< |L4 500 | | +---+ | ^ | 501 | | | *|>>> Packet W >>>>>>>>>>>>|>^ |L3 502 | | +---+ +---+ | | +---+ +---+ | | 503 | | | | | | | <|<<<<<|<<<|<(V)<|<<<| | |L2 504 | | . . | . |Packet U | . . | . | . . | . | . . | .*| . . | |L2 505 |___|_____|___|_____|___|_____|___|_____|___|_____|___|_____|___| 506 source subnet G router subnet H dest 507 __ _ _ _ __ _ _ _ __ _ _ __ _ _ _ later 508 | | | | | | | | | | | | | | | | |C| | data________\ 509 |__|_|_|_| |__|_|_|_| |__|_|_| |__|_|_|_| packet (W) / 510 4 3 2 4 3 2 4 3 4 3 2 511 _ 512 /__ |C| Feedback control 513 \ |_| cell/frame (V) 514 2 515 __ _ _ _ __ _ _ _ __ _ _ __ _ _ _ earlier 516 | | | | | | | | | | | | | | | | | | | data________\ 517 |__|_|_|_| |__|_|_|_| |__|_|_| |__|_|_|_| packet (U) / 518 layer: 4 3 2 4 3 2 4 3 4 3 2 519 header 521 Figure 3: Feed-Backward Mode 523 ATM's feed-backward approach doesn't fit well when layered beneath 524 IP's feed-forward approach--unless the initial data source is the 525 same node as the ATM ingress. Figure 3 shows the feed-backward 526 approach being used in subnet H. If the final switch on the path is 527 congested (*), it doesn't feed-forward any congestion indications on 528 packet (U). Instead it sends a control cell (V) back to the router 529 at the ATM ingress. 531 However, the backward feedback doesn't reach the original data source 532 directly because IP doesn't support backward feedback (and subnet G 533 is independent of subnet H). Instead, the router in the middle 534 throttles down its sending rate but the original data sources don't 535 reduce their rates. The resulting rate mismatch causes the middle 536 router's buffer at layer 3 to back up until it becomes congested, 537 which it signals forwards on later data packets at layer 3 (e.g. 538 packet W). Note that the forward signal from the middle router is 539 not triggered directly by the backward signal. Rather, it is 540 triggered by congestion resulting from the middle router's mismatched 541 rate response to the backward signal. 543 In response to this later forward signalling, end-to-end feedback at 544 layer-4 finally completes the tortuous path of congestion indications 545 back to the origin data source, as before. 547 4.4. Null Mode 549 Often link and physical layer resources are 'non-blocking' by design. 550 In these cases congestion notification may be implemented but it does 551 not need to be deployed at the lower layer; ECN in IP would be 552 sufficient. 554 A degenerate example is a point-to-point Ethernet link. Excess 555 loading of the link merely causes the queue from the higher layer to 556 back up, while the lower layer remains immune to congestion. Even a 557 whole meshed subnetwork can be made immune to interior congestion by 558 limiting ingress capacity and sufficient sizing of interior links, 559 e.g. a non-blocking fat-tree network. An alternative to fat links 560 near the root is numerous thin links with multi-path routing to 561 ensure even worst-case patterns of load cannot congest any link, e.g. 562 a Clos network. 564 5. Feed-Forward-and-Up Mode: Guidelines for Adding Congestion 565 Notification 567 Feed-forward-and-up is the mode already used for signalling ECN up 568 the layers through MPLS into IP [RFC5129] and through IP-in-IP 569 tunnels [RFC6040]. These RFCs take a consistent approach and the 570 following guidelines are designed to ensure this consistency 571 continues as ECN support is added to other protocols that encapsulate 572 IP. The guidelines are also designed to ensure compliance with the 573 more general best current practice for the design of alternate ECN 574 schemes given in [RFC4774]. 576 The rest of this section is structured as follows: 578 o Section 5.1 addresses the most straightforward cases, where 579 [RFC6040] can be applied directly to add ECN to tunnels that are 580 effectively the same as IP-in-IP tunnels. 582 o The subsequent sections give guidelines for adding ECN to a subnet 583 technology that uses feed-forward-and-up mode like IP, but it is 584 not so similar to IP that [RFC6040] rules can be applied directly. 585 Specifically: 587 * Sections 5.2, 5.3 and 5.4 respectively address how to add ECN 588 support to the wire protocol and to the encapsulators and 589 decapsulators at the ingress and egress of the subnet. 591 * Section 5.5 deals with the special, but common, case of 592 sequences of tunnels or subnets that all use the same 593 technology 595 * Section 5.6 deals with the question of reframing when IP 596 packets do not map 1:1 into lower layer frames. 598 5.1. IP-in-IP Tunnels with Tightly Coupled Shim Headers 600 A common pattern for many tunnelling protocols is to encapsulate an 601 inner IP header with shim header(s) then an outer IP header. In many 602 cases the shim header(s) and the outer IP header are always added (or 603 removed) as part of the same process. We call this a tightly coupled 604 shim header. Processing the shim and outer together is often 605 necessary because the shim(s) are not sufficient for packet 606 forwarding in their own right; not unless complemented by an outer 607 header. 609 For all such tightly coupled shim headers (such as those listed in 610 the Introduction), the rules in [RFC6040] for propagating the ECN 611 field can be applied directly between the inner and outer IP headers. 612 [I-D.ietf-tsvwg-rfc6040update-shim] clarifies that RFC 6040 is just 613 as applicable when there is a tightly-coupled shim between two IP 614 headers as wehen there is not. 616 5.2. Wire Protocol Design: Indication of ECN Support 618 This section is intended to guide the redesign of any lower layer 619 protocol that encapsulate IP to add native ECN support at the lower 620 layer. It reflects the approaches used in [RFC6040] and in 621 [RFC5129]. Therefore IP-in-IP tunnels or IP-in-MPLS or MPLS-in-MPLS 622 encapsulations that already comply with [RFC6040] or [RFC5129] will 623 already satisfy this guidance. 625 A lower layer (or subnet) congestion notification system: 627 1. SHOULD NOT apply explicit congestion notifications to PDUs that 628 are destined for legacy layer-4 transport implementations that 629 will not understand ECN, and 631 2. SHOULD NOT apply explicit congestion notifications to PDUs if the 632 egress of the subnet might not propagate congestion notifications 633 onward into the higher layer. 635 We use the term ECN-PDUs for a PDU on a feedback loop that will 636 propagate congestion notification properly because it meets both 637 the above criteria. And a Not-ECN-PDU is a PDU on a feedback 638 loop that does not meet both criteria, and will therefore not 639 propagate congestion notification properly. A corollary of the 640 above is that a lower layer congestion notification protocol: 642 3. SHOULD be able to distinguish ECN-PDUs from Not-ECN-PDUs. 644 Note that there is no need for all interior nodes within a subnet to 645 be able to mark congestion explicitly. A mix of ECN and drop signals 646 from different nodes is fine. However, if _any_ interior nodes might 647 generate ECN markings, guideline 2 above says that all relevant 648 egress node(s) SHOULD be able to propagate those markings up to the 649 higher layer. 651 In IP, if the ECN field in each PDU is cleared to the Not-ECT (not 652 ECN-capable transport) codepoint, it indicates that the L4 transport 653 will not understand congestion markings. A congested buffer must not 654 mark these Not-ECT PDUs, and therefore drops them instead. 656 The mechanism a lower layer uses to distinguish the ECN-capability of 657 PDUs need not mimic that of IP. The above guidelines merely say that 658 the lower layer system, as a whole, should achieve the same outcome. 659 For instance, ECN-capable feedback loops might use PDUs that are 660 identified by a particular set of labels or tags. Alternatively, 661 logical link protocols that use flow state might determine whether a 662 PDU can be congestion marked by checking for ECN-support in the flow 663 state. Other protocols might depend on out-of-band control signals. 665 The per-domain checking of ECN support in MPLS [RFC5129] is a good 666 example of a way to avoid sending congestion markings to transports 667 that will not understand them, without using any header space in the 668 subnet protocol. 670 In MPLS, header space is extremely limited, therefore RFC5129 does 671 not provide a field in the MPLS header to indicate whether the PDU is 672 an ECN-PDU or a Not-ECN-PDU. Instead, interior nodes in a domain are 673 allowed to set explicit congestion indications without checking 674 whether the PDU is destined for a transport that will understand 675 them. Nonetheless, this is made safe by requiring that the network 676 operator upgrades all decapsulating edges of a whole domain at once, 677 as soon as even one switch within the domain is configured to mark 678 rather than drop during congestion. Therefore, any edge node that 679 might decapsulate a packet will be capable of checking whether the 680 higher layer transport is ECN-capable. When decapsulating a CE- 681 marked packet, if the decapsulator discovers that the higher layer 682 (inner header) indicates the transport is not ECN-capable, it drops 683 the packet--effectively on behalf of the earlier congested node (see 684 Decapsulation Guideline 1 in Section 5.4). 686 It was only appropriate to define such an incremental deployment 687 strategy because MPLS is targeted solely at professional operators, 688 who can be expected to ensure that a whole subnetwork is consistently 689 configured. This strategy might not be appropriate for other link 690 technologies targeted at zero-configuration deployment or deployment 691 by the general public (e.g. Ethernet). For such 'plug-and-play' 692 environments it will be necessary to invent a failsafe approach that 693 ensures congestion markings will never fall into black holes, no 694 matter how inconsistently a system is put together. Alternatively, 695 congestion notification relying on correct system configuration could 696 be confined to flavours of Ethernet intended only for professional 697 network operators, such as IEEE 802.1ah Provider Backbone Bridges 698 (PBB). 700 ECN support in TRILL [I-D.ietf-trill-ecn-support] provides a good 701 example of how to add ECN to a lower layer protocol without relying 702 on careful and consistent operator configuration. TRILL provides an 703 extension header word with space for flags of different categories 704 depending on whether logic to understand the extension is critical. 705 The congestion experienced marking has been defined as a 'critical 706 ingress-to-egress' flag. So if a transit RBridge sets this flag and 707 an egress RBridge does not have any logic to process it, it will drop 708 it; which is the desired default action anyway. Therefore TRILL 709 RBridges can be updated with support for ECN in no particular order 710 and, at the egress of the TRILL campus, congestion notification will 711 be propagated to IP as ECN whenever ECN logic has been implemented, 712 and as drop otherwise. 714 QCN [IEEE802.1Qau] provides another example of how to indicate to 715 lower layer devices that the end-points will not understand ECN. An 716 operator can define certain 802.1p classes of service to indicate 717 non-QCN frames and an ingress bridge is required to map arriving not- 718 QCN-capable IP packets to one of these non-QCN 802.1p classes. 720 5.3. Encapsulation Guidelines 722 This section is intended to guide the redesign of any node that 723 encapsulates IP with a lower layer header when adding native ECN 724 support to the lower layer protocol. It reflects the approaches used 725 in [RFC6040] and in [RFC5129]. Therefore IP-in-IP tunnels or IP-in- 726 MPLS or MPLS-in-MPLS encapsulations that already comply with 727 [RFC6040] or [RFC5129] will already satisfy this guidance. 729 1. Egress Capability Check: A subnet ingress needs to be sure that 730 the corresponding egress of a subnet will propagate any 731 congestion notification added to the outer header across the 732 subnet. This is necessary in addition to checking that an 733 incoming PDU indicates an ECN-capable (L4) transport. Examples 734 of how this guarantee might be provided include: 736 * by configuration (e.g. if any label switches in a domain 737 support ECN marking, [RFC5129] requires all egress nodes to 738 have been configured to propagate ECN) 740 * by the ingress explicitly checking that the egress propagates 741 ECN (e.g. TRILL uses IS-IS to check path capabilities before 742 using critical options [RFC7780]) 744 * by inherent design of the protocol (e.g. by encoding ECN 745 marking on the outer header in such a way that a legacy egress 746 that does not understand ECN will consider the PDU corrupt and 747 discard it, thus at least propagating a form of congestion 748 signal). 750 2. Egress Fails Capability Check: If the ingress cannot guarantee 751 that the egress will propagate congestion notification, the 752 ingress SHOULD disable ECN when it forwards the PDU at the lower 753 layer. An example of how the ingress might disable ECN at the 754 lower layer would be by setting the outer header of the PDU to 755 identify it as a Not-ECN-PDU, assuming the subnet technology 756 supports such a concept. 758 3. Standard Congestion Monitoring Baseline: Once the ingress to a 759 subnet has established that the egress will correctly propagate 760 ECN, on encapsulation it SHOULD encode the same level of 761 congestion in outer headers as is arriving in incoming headers. 762 For example it might copy any incoming congestion notification 763 into the outer header of the lower layer protocol. 765 This ensures that all outer headers reflect congestion 766 accumulated along the whole upstream path since the Load 767 Regulator, not just since the ingress of the subnet. A node that 768 is not the Load Regulator SHOULD NOT re-initialise the level of 769 CE markings in the outer to zero. 771 This guideline is intended to ensure that any bulk congestion 772 monitoring of outer headers (e.g. by a network management node 773 monitoring ECN in passing frames) is most meaningful. For 774 instance, if an operator measures CE in 0.4% of passing outer 775 headers, this information is only useful if the operator knows 776 where the proportion of CE markings was last initialised to 0% 777 (the Congestion Baseline). Such monitoring information will not 778 be useful if some subnet ingress nodes reset all outer CE 779 markings while others copy incoming CE markings into the outer. 781 Most information can be extracted if the Congestion Baseline is 782 standardised at the node that is regulating the load (the Load 783 Regulator--typically the data source). Then the operator can 784 measure both congestion since the Load Regulator, and congestion 785 since the subnet ingress. The latter might be measurable by 786 subtracting the level of CE markings on inner headers from that 787 on outer headers (see Appendix C of [RFC6040]). 789 5.4. Decapsulation Guidelines 791 This section is intended to guide the redesign of any node that 792 decapsulates IP from within a lower layer header when adding native 793 ECN support to the lower layer protocol. It reflects the approaches 794 used in [RFC6040] and in [RFC5129]. Therefore IP-in-IP tunnels or 795 IP-in-MPLS or MPLS-in-MPLS encapsulations that already comply with 796 [RFC6040] or [RFC5129] will already satisfy this guidance. 798 A subnet egress SHOULD NOT simply copy congestion notification from 799 outer headers to the forwarded header. It SHOULD calculate the 800 outgoing congestion notification field from the inner and outer 801 headers using the following guidelines. If there is any conflict, 802 rules earlier in the list take precedence over rules later in the 803 list: 805 1. If the arriving inner header is a Not-ECN-PDU it implies the L4 806 transport will not understand explicit congestion markings. 807 Then: 809 * If the outer header carries an explicit congestion marking, 810 drop is the only indication of congestion that the L4 811 transport will understand. If the congestion marking is the 812 most severe possible, the packet MUST be dropped. However, if 813 congestion can be marked with multiple levels severity and the 814 packet's marking is not the most severe, the packet MAY be 815 forwarded, but it SHOULD be dropped. 817 * If the outer is an ECN-PDU that carries no indication of 818 congestion or a Not-ECN-PDU the PDU SHOULD be forwarded, but 819 still as a Not-ECN-PDU. 821 2. If the outer header does not support explicit congestion 822 notification (a Not-ECN-PDU), but the inner header does (an ECN- 823 PDU), the inner header SHOULD be forwarded unchanged. 825 3. In some lower layer protocols congestion may be signalled as a 826 numerical level, such as in the control frames of quantised 827 congestion notification [IEEE802.1Qau]. If such a multi-bit 828 encoding encapsulates an ECN-capable IP data packet, a function 829 will be needed to convert the quantised congestion level into the 830 frequency of congestion markings in outgoing IP packets. 832 4. Congestion indications may be encoded by a severity level. For 833 instance increasing levels of congestion might be encoded by 834 numerically increasing indications, e.g. pre-congestion 835 notification (PCN) can be encoded in each PDU at three severity 836 levels in IP or MPLS [RFC6660]. 838 If the arriving inner header is an ECN-PDU, where the inner and 839 outer headers carry indications of congestion of different 840 severity, the more severe indication SHOULD be forwarded in 841 preference to the less severe. 843 5. The inner and outer headers might carry a combination of 844 congestion notification fields that should not be possible given 845 any currently used protocol transitions. For instance, if 846 Encapsulation Guideline 3 in Section 5.3 had been followed, it 847 should not be possible to have a less severe indication of 848 congestion in the outer than in the inner. It MAY be appropriate 849 to log unexpected combinations of headers and possibly raise an 850 alarm. 852 If a safe outgoing codepoint can be defined for such a PDU, the 853 PDU SHOULD be forwarded rather than dropped. Some implementers 854 discard PDUs with currently unused combinations of headers just 855 in case they represent an attack. However, an approach using 856 alarms and policy-mediated drop is preferable to hard-coded drop, 857 so that operators can keep track of possible attacks but 858 currently unused combinations are not precluded from future use 859 through new standards actions. 861 5.5. Sequences of Similar Tunnels or Subnets 863 In some deployments, particularly in 3GPP networks, an IP packet may 864 traverse two or more IP-in-IP tunnels in sequence that all use 865 identical technology (e.g. GTP). 867 In such cases, it would be sufficient for every encapsulation and 868 decapsulation in the chain to comply with RFC 6040. Alternatively, 869 as an optimisation, a node that decapsulates a packet and immediately 870 re-encapsulates it for the next tunnel MAY copy the incoming outer 871 ECN field directly to the outgoing outer and the incoming inner ECN 872 field directly to the outgoing inner. Then the overall behavior 873 across the sequence of tunnel segments would still be consistent with 874 RFC 6040. 876 Appendix C of RFC6040 describes how a tunnel egress can monitor how 877 much congestion has been introduced within a tunnel. A network 878 operator might want to monitor how much congestion had been 879 introduced within a whole sequence of tunnels. Using the technique 880 in Appendix C of RFC6040 at the final egress, the operator could 881 monitor the whole sequence of tunnels, but only if the above 882 optimisation were used consistently along the sequence of tunnels, in 883 order to make it appear as a single tunnel. Therefore, tunnel 884 endpoint implementations SHOULD allow the operator to configure 885 whether this optimisation is enabled. 887 When ECN support is added to a subnet technology, consideration 888 SHOULD be given to a similar optimisation between subnets in sequence 889 if they all use the same technology. 891 5.6. Reframing and Congestion Markings 893 The guidance in this section is worded in terms of framing 894 boundaries, but it applies equally whether the protocol data units 895 are frames, cells or packets. 897 Where framing boundaries are different between two layers, congestion 898 indications SHOULD be propagated on the basis that a congestion 899 indication on a PDU applies to all the octets in the PDU. On 900 average, an encapsulator or decapsulator SHOULD approximately 901 preserve the number of marked octets arriving and leaving (counting 902 the size of inner headers, but not added encapsulating headers). 904 The next departing frame SHOULD be immediately marked even if only 905 enough incoming marked octets have arrived for part of the departing 906 frame. This ensures that any outstanding congestion marked octets 907 are propagated immediately, rather than held back waiting for a frame 908 no bigger than the outstanding marked octets--which might involve a 909 long wait. 911 For instance, an algorithm for marking departing frames could 912 maintain a counter representing the balance of arriving marked octets 913 minus departing marked octets. It adds the size of every marked 914 frame that arrives and if the counter is positive it marks the next 915 frame to depart and subtracts its size from the counter. This will 916 often leave a negative remainder in the counter, which is deliberate. 918 6. Feed-Up-and-Forward Mode: Guidelines for Adding Congestion 919 Notification 921 The guidance in this section is applicable, for example, when IP 922 packets: 924 o are encapsulated in Ethernet headers, which have no support for 925 ECN; 927 o are forwarded by the eNode-B (base station) of a 3GPP radio access 928 network, which is required to apply ECN marking during congestion, 929 [LTE-RA], [UTRAN], but the Packet Data Convergence Protocol (PDCP) 930 that encapsulates the IP header over the radio access has no 931 support for ECN. 933 This guidance also generalises to encapsulation by other subnet 934 technologies with no native support for explicit congestion 935 notification at the lower layer, but with support for finding and 936 processing an IP header. It is unlikely to be applicable or 937 necessary for IP-in-IP encapsulation, where feed-forward-and-up mode 938 based on [RFC6040] would be more appropriate. 940 Marking the IP header while switching at layer-2 (by using a layer-3 941 switch) or while forwarding in a radio access network seems to 942 represent a layering violation. However, it can be considered as a 943 benign optimisation if the guidelines below are followed. Feed-up- 944 and-forward is certainly not a general alternative to implementing 945 feed-forward congestion notification in the lower layer, because: 947 o IPv4 and IPv6 are not the only layer-3 protocols that might be 948 encapsulated by lower layer protocols 950 o Link-layer encryption might be in use, making the layer-2 payload 951 inaccessible 953 o Many Ethernet switches do not have 'layer-3 switch' capabilities 954 so they cannot read or modify an IP payload 956 o It might be costly to find an IP header (v4 or v6) when it may be 957 encapsulated by more than one lower layer header, e.g. Ethernet 958 MAC in MAC [IEEE802.1Qah]. 960 Nonetheless, configuring lower layer equipment to look for an ECN 961 field in an encapsulated IP header is a useful optimisation. If the 962 implementation follows the guidelines below, this optimisation does 963 not have to be confined to a controlled environment such as within a 964 data centre; it could usefully be applied on any network--even if the 965 operator is not sure whether the above issues will never apply: 967 1. If a native lower-layer congestion notification mechanism exists 968 for a subnet technology, it is safe to mix feed-up-and-forward 969 with feed-forward-and-up on other switches in the same subnet. 970 However, it will generally be more efficient to use the native 971 mechanism. 973 2. The depth of the search for an IP header SHOULD be limited. If 974 an IP header is not found soon enough, or an unrecognised or 975 unreadable header is encountered, the switch SHOULD resort to an 976 alternative means of signalling congestion (e.g. drop, or the 977 native lower layer mechanism if available). 979 3. It is sufficient to use the first IP header found in the stack; 980 the egress of the relevant tunnel can propagate congestion 981 notification upwards to any more deeply encapsulated IP headers 982 later. 984 7. Feed-Backward Mode: Guidelines for Adding Congestion Notification 986 It can be seen from Section 4.3 that congestion notification in a 987 subnet using feed-backward mode has generally not been designed to be 988 directly coupled with IP layer congestion notification. The subnet 989 attempts to minimise congestion internally, and if the incoming load 990 at the ingress exceeds the capacity somewhere through the subnet, the 991 layer 3 buffer into the ingress backs up. Thus, a feed-backward mode 992 subnet is in some sense similar to a null mode subnet, in that there 993 is no need for any direct interaction between the subnet and higher 994 layer congestion notification. Therefore no detailed protocol design 995 guidelines are appropriate. Nonetheless, a more general guideline is 996 appropriate: 998 A subnetwork technology intended to eventually interface to IP 999 SHOULD NOT be designed using only the feed-backward mode, which is 1000 certainly best for a stand-alone subnet, but would need to be 1001 modified to work efficiently as part of the wider Internet, 1002 because IP uses feed-forward-and-up mode. 1004 The feed-backward approach at least works beneath IP, where the term 1005 'works' is used only in a narrow functional sense because feed- 1006 backward can result in very inefficient and sluggish congestion 1007 control--except if it is confined to the subnet directly connected to 1008 the original data source, when it is faster than feed-forward. It 1009 would be valid to design a protocol that could work in feed-backward 1010 mode for paths that only cross one subnet, and in feed-forward-and-up 1011 mode for paths that cross subnets. 1013 In the early days of TCP/IP, a similar feed-backward approach was 1014 tried for explicit congestion signalling, using source-quench (SQ) 1015 ICMP control packets. However, SQ fell out of favour and is now 1016 formally deprecated [RFC6633]. The main problem was that it is hard 1017 for a data source to tell the difference between a spoofed SQ message 1018 and a quench request from a genuine buffer on the path. It is also 1019 hard for a lower layer buffer to address an SQ message to the 1020 original source port number, which may be buried within many layers 1021 of headers, and possibly encrypted. 1023 Quantised congestion notification (QCN--also known as backward 1024 congestion notification or BCN) [IEEE802.1Qau] uses a feed-backward 1025 mode structurally similar to ATM's relative rate mechanism. However, 1026 QCN confines its applicability to scenarios such as some data centres 1027 where all endpoints are directly attached by the same Ethernet 1028 technology. If a QCN subnet were later connected into a wider IP- 1029 based internetwork (e.g. when attempting to interconnect multiple 1030 data centres) it would suffer the inefficiency shown Figure 3. 1032 8. IANA Considerations (to be removed by RFC Editor) 1034 This memo includes no request to IANA. 1036 9. Security Considerations 1038 If a lower layer wire protocol is redesigned to include explicit 1039 congestion signalling in-band in the protocol header, care SHOULD be 1040 take to ensure that the field used is specified as mutable during 1041 transit. Otherwise interior nodes signalling congestion would 1042 invalidate any authentication protocol applied to the lower layer 1043 header--by altering a header field that had been assumed as 1044 immutable. 1046 The redesign of protocols that encapsulate IP in order to propagate 1047 congestion signals between layers raises potential signal integrity 1048 concerns. Experimental or proposed approaches exist for assuring the 1049 end-to-end integrity of in-band congestion signals, e.g.: 1051 o Congestion exposure (ConEx ) for networks to audit that their 1052 congestion signals are not being suppressed by other networks or 1053 by receivers, and for networks to police that senders are 1054 responding sufficiently to the signals, irrespective of the 1055 transport protocol used [RFC7713]. 1057 o The ECN nonce [RFC3540] for a TCP sender to detect whether a 1058 network or the receiver is suppressing congestion signals. 1060 o A test with the same goals as the ECN nonce, but without the need 1061 for the receiver to co-operate with the protocol 1062 [I-D.moncaster-tcpm-rcv-cheat]. 1064 Given these end-to-end approaches are already being specified, it 1065 would make little sense to attempt to design hop-by-hop congestion 1066 signal integrity into a new lower layer protocol, because end-to-end 1067 integrity inherently achieves hop-by-hop integrity. 1069 10. Conclusions 1071 Following the guidance in the document enables ECN support to be 1072 extended to numerous protocols that encapsulate IP (v4 & v6) in a 1073 consistent way, so that IP continues to fulfil its role as an end-to- 1074 end interoperability layer. This includes: 1076 o A wide range of tunnelling protocols with various forms of shim 1077 header between two IP headers; 1079 o A wide range of subnet technologies, particularly those that work 1080 in the same 'feed-forward-and-up' mode that is used to support ECN 1081 in IP and MPLS. 1083 Guidelines have been defined for supporting propagation of ECN 1084 between Ethernet and IP on so-called Layer-3 Ethernet switches, using 1085 a 'feed-up-an-forward' mode. This approach could enable other subnet 1086 technologies to pass ECN signals into the IP layer, even if they do 1087 not support ECN natively. 1089 Finally, attempting to add ECN to a subnet technology in feed- 1090 backward mode is deprecated except in special cases, due to its 1091 likely sluggish response to congestion. 1093 11. Acknowledgements 1095 Thanks to Gorry Fairhurst for extensive reviews. Thanks also to the 1096 following reviewers: Richard Scheffenegger, Ingemar Johansson, Piers 1097 O'Hanlon and Michael Welzl, who pointed out that lower layer 1098 congestion notification signals may have different semantics to those 1099 in IP. Thanks are also due to the tsvwg chairs, TSV ADs and IETF 1100 liaison people such as Eric Gray, Dan Romascanu and Gonzalo Camarillo 1101 for helping with the liaisons with the IEEE and 3GPP. And thanks to 1102 Georg Mayer and particularly to Erik Guttman for the extensive search 1103 and categorisation of any 3GPP specifications that cite ECN 1104 specifications. 1106 Bob Briscoe was part-funded by the European Community under its 1107 Seventh Framework Programme through the Trilogy project (ICT-216372) 1108 for initial drafts and through the Reducing Internet Transport 1109 Latency (RITE) project (ICT-317700) subsequently. The views 1110 expressed here are solely those of the authors. 1112 12. Comments Solicited 1114 Comments and questions are encouraged and very welcome. They can be 1115 addressed to the IETF Transport Area working group mailing list 1116 , and/or to the authors. 1118 13. References 1120 13.1. Normative References 1122 [RFC2119] Bradner, S., "Key words for use in RFCs to Indicate 1123 Requirement Levels", BCP 14, RFC 2119, 1124 DOI 10.17487/RFC2119, March 1997, 1125 . 1127 [RFC3168] Ramakrishnan, K., Floyd, S., and D. Black, "The Addition 1128 of Explicit Congestion Notification (ECN) to IP", 1129 RFC 3168, DOI 10.17487/RFC3168, September 2001, 1130 . 1132 [RFC3819] Karn, P., Ed., Bormann, C., Fairhurst, G., Grossman, D., 1133 Ludwig, R., Mahdavi, J., Montenegro, G., Touch, J., and L. 1134 Wood, "Advice for Internet Subnetwork Designers", BCP 89, 1135 RFC 3819, DOI 10.17487/RFC3819, July 2004, 1136 . 1138 [RFC4774] Floyd, S., "Specifying Alternate Semantics for the 1139 Explicit Congestion Notification (ECN) Field", BCP 124, 1140 RFC 4774, DOI 10.17487/RFC4774, November 2006, 1141 . 1143 [RFC5129] Davie, B., Briscoe, B., and J. Tay, "Explicit Congestion 1144 Marking in MPLS", RFC 5129, DOI 10.17487/RFC5129, January 1145 2008, . 1147 [RFC6040] Briscoe, B., "Tunnelling of Explicit Congestion 1148 Notification", RFC 6040, DOI 10.17487/RFC6040, November 1149 2010, . 1151 13.2. Informative References 1153 [ATM-TM-ABR] 1154 Cisco, "Understanding the Available Bit Rate (ABR) Service 1155 Category for ATM VCs", Design Technote 10415, June 2005. 1157 [Buck00] Buckwalter, J., "Frame Relay: Technology and Practice", 1158 Pub. Addison Wesley ISBN-13: 978-0201485240, 2000. 1160 [DCTCP] Alizadeh, M., Greenberg, A., Maltz, D., Padhye, J., Patel, 1161 P., Prabhakar, B., Sengupta, S., and M. Sridharan, "Data 1162 Center TCP (DCTCP)", ACM SIGCOMM CCR 40(4)63--74, October 1163 2010, . 1165 [GTPv1] 3GPP, "GPRS Tunnelling Protocol (GTP) across the Gn and Gp 1166 interface", Technical Specification TS 29.060. 1168 [GTPv1-U] 3GPP, "General Packet Radio System (GPRS) Tunnelling 1169 Protocol User Plane (GTPv1-U)", Technical Specification TS 1170 29.281. 1172 [GTPv2-C] 3GPP, "Evolved General Packet Radio Service (GPRS) 1173 Tunnelling Protocol for Control plane (GTPv2-C)", 1174 Technical Specification TS 29.274. 1176 [I-D.ietf-nvo3-geneve] 1177 Gross, J., Ganga, I., and T. Sridhar, "Geneve: Generic 1178 Network Virtualization Encapsulation", draft-ietf- 1179 nvo3-geneve-04 (work in progress), March 2017. 1181 [I-D.ietf-nvo3-gue] 1182 Herbert, T., Yong, L., and O. Zia, "Generic UDP 1183 Encapsulation", draft-ietf-nvo3-gue-05 (work in progress), 1184 October 2016. 1186 [I-D.ietf-trill-ecn-support] 1187 Eastlake, D. and B. Briscoe, "TRILL: ECN (Explicit 1188 Congestion Notification) Support", draft-ietf-trill-ecn- 1189 support-02 (work in progress), March 2017. 1191 [I-D.ietf-tsvwg-rfc6040update-shim] 1192 Briscoe, B., "Propagating Explicit Congestion Notification 1193 Across IP Tunnel Headers Separated by a Shim", draft-ietf- 1194 tsvwg-rfc6040update-shim-00 (work in progress), November 1195 2016. 1197 [I-D.moncaster-tcpm-rcv-cheat] 1198 Moncaster, T., Briscoe, B., and A. Jacquet, "A TCP Test to 1199 Allow Senders to Identify Receiver Non-Compliance", draft- 1200 moncaster-tcpm-rcv-cheat-03 (work in progress), July 2014. 1202 [IEEE802.1Qah] 1203 IEEE, "IEEE Standard for Local and Metropolitan Area 1204 Networks--Virtual Bridged Local Area Networks--Amendment 1205 6: Provider Backbone Bridges", IEEE Std 802.1Qah-2008, 1206 August 2008, 1207 . 1209 (Access Controlled link within page) 1211 [IEEE802.1Qau] 1212 Finn, N., Ed., "IEEE Standard for Local and Metropolitan 1213 Area Networks--Virtual Bridged Local Area Networks - 1214 Amendment 13: Congestion Notification", IEEE Std 802.1Qau- 1215 2010, March 2010, . 1218 (Access Controlled link within page) 1220 [ITU-T.I.371] 1221 ITU-T, "Traffic Control and Congestion Control in B-ISDN", 1222 ITU-T Rec. I.371 (03/04), March 2004, 1223 . 1226 [LTE-RA] 3GPP, "Evolved Universal Terrestrial Radio Access (E-UTRA) 1227 and Evolved Universal Terrestrial Radio Access Network 1228 (E-UTRAN); Overall description; Stage 2", Technical 1229 Specification TS 36.300. 1231 [RFC1323] Jacobson, V., Braden, R., and D. Borman, "TCP Extensions 1232 for High Performance", RFC 1323, DOI 10.17487/RFC1323, May 1233 1992, . 1235 [RFC1701] Hanks, S., Li, T., Farinacci, D., and P. Traina, "Generic 1236 Routing Encapsulation (GRE)", RFC 1701, 1237 DOI 10.17487/RFC1701, October 1994, 1238 . 1240 [RFC2003] Perkins, C., "IP Encapsulation within IP", RFC 2003, 1241 DOI 10.17487/RFC2003, October 1996, 1242 . 1244 [RFC2637] Hamzeh, K., Pall, G., Verthein, W., Taarud, J., Little, 1245 W., and G. Zorn, "Point-to-Point Tunneling Protocol 1246 (PPTP)", RFC 2637, DOI 10.17487/RFC2637, July 1999, 1247 . 1249 [RFC2661] Townsley, W., Valencia, A., Rubens, A., Pall, G., Zorn, 1250 G., and B. Palter, "Layer Two Tunneling Protocol "L2TP"", 1251 RFC 2661, DOI 10.17487/RFC2661, August 1999, 1252 . 1254 [RFC2784] Farinacci, D., Li, T., Hanks, S., Meyer, D., and P. 1255 Traina, "Generic Routing Encapsulation (GRE)", RFC 2784, 1256 DOI 10.17487/RFC2784, March 2000, 1257 . 1259 [RFC2884] Hadi Salim, J. and U. Ahmed, "Performance Evaluation of 1260 Explicit Congestion Notification (ECN) in IP Networks", 1261 RFC 2884, DOI 10.17487/RFC2884, July 2000, 1262 . 1264 [RFC2983] Black, D., "Differentiated Services and Tunnels", 1265 RFC 2983, DOI 10.17487/RFC2983, October 2000, 1266 . 1268 [RFC3540] Spring, N., Wetherall, D., and D. Ely, "Robust Explicit 1269 Congestion Notification (ECN) Signaling with Nonces", 1270 RFC 3540, DOI 10.17487/RFC3540, June 2003, 1271 . 1273 [RFC4301] Kent, S. and K. Seo, "Security Architecture for the 1274 Internet Protocol", RFC 4301, DOI 10.17487/RFC4301, 1275 December 2005, . 1277 [RFC6633] Gont, F., "Deprecation of ICMP Source Quench Messages", 1278 RFC 6633, DOI 10.17487/RFC6633, May 2012, 1279 . 1281 [RFC6660] Briscoe, B., Moncaster, T., and M. Menth, "Encoding Three 1282 Pre-Congestion Notification (PCN) States in the IP Header 1283 Using a Single Diffserv Codepoint (DSCP)", RFC 6660, 1284 DOI 10.17487/RFC6660, July 2012, 1285 . 1287 [RFC7348] Mahalingam, M., Dutt, D., Duda, K., Agarwal, P., Kreeger, 1288 L., Sridhar, T., Bursell, M., and C. Wright, "Virtual 1289 eXtensible Local Area Network (VXLAN): A Framework for 1290 Overlaying Virtualized Layer 2 Networks over Layer 3 1291 Networks", RFC 7348, DOI 10.17487/RFC7348, August 2014, 1292 . 1294 [RFC7567] Baker, F., Ed. and G. Fairhurst, Ed., "IETF 1295 Recommendations Regarding Active Queue Management", 1296 BCP 197, RFC 7567, DOI 10.17487/RFC7567, July 2015, 1297 . 1299 [RFC7713] Mathis, M. and B. Briscoe, "Congestion Exposure (ConEx) 1300 Concepts, Abstract Mechanism, and Requirements", RFC 7713, 1301 DOI 10.17487/RFC7713, December 2015, 1302 . 1304 [RFC7780] Eastlake 3rd, D., Zhang, M., Perlman, R., Banerjee, A., 1305 Ghanwani, A., and S. Gupta, "Transparent Interconnection 1306 of Lots of Links (TRILL): Clarifications, Corrections, and 1307 Updates", RFC 7780, DOI 10.17487/RFC7780, February 2016, 1308 . 1310 [RFC8084] Fairhurst, G., "Network Transport Circuit Breakers", 1311 BCP 208, RFC 8084, DOI 10.17487/RFC8084, March 2017, 1312 . 1314 [UTRAN] 3GPP, "UTRAN Overall Description", Technical 1315 Specification TS 25.401. 1317 Appendix A. Outstanding Document Issues 1319 1. [GF] Concern that certain guidelines warrant a MUST (NOT) rather 1320 than a SHOULD (NOT). Given the guidelines say that if any SHOULD 1321 (NOT)s are not followed, a strong justification will be needed, 1322 they have been left as SHOULD (NOT) pending further list 1323 discussion. In particular: 1325 * If inner is a Not-ECN-PDU and Outer is CE (or highest severity 1326 congestion level), MUST (not SHOULD) drop? 1328 This issue has been addressed by explaining when SHOULD or 1329 MUST is appropriate. 1331 2. Consider whether an IETF Standard Track doc will be needed to 1332 Update the IP-in-IP protocols listed in Section 5.1--at least 1333 those that the IETF controls--and which Area it should sit under. 1335 This issue has been addressed by the production of 1336 [I-D.ietf-tsvwg-rfc6040update-shim], but this text is left 1337 outstanding until that draft is adopted. 1339 Appendix B. Changes in This Version (to be removed by RFC Editor) 1341 From ietf-05 to ietf-06: 1343 * Introduction: Added GUE and Geneve as examples of tightly 1344 coupled shims between IP headers that cite RFC 6040. And added 1345 VXLAN to list of those that do not. 1347 * Replaced normative text about tightly coupled shims between IP 1348 headers, with reference to new draft-ietf-tsvwg-rfc6040update- 1349 shim 1351 * Wire Protocol Design: Indication of ECN Support: Added TRILL as 1352 an example of a well-design protocol that does not need an 1353 indication of ECN support in the wire protocol. 1355 * Encapsulation Guidelines: In the case of a Not-ECN-PDU with a 1356 CE outer, replaced SHOULD be dropped, with explanations of when 1357 SHOULD or MUST are appropriate. 1359 * Feed-Up-and-Forward Mode: Explained examples more carefully, 1360 referred to PDCP and cited UTRAN spec as well as E-UTRAN. 1362 * Added the people involved in liaisons to the acknowledgements. 1364 * Updated references. 1366 * Marked open issues as resolved, but did not delete Open Issues 1367 Appendix (yet). 1369 From ietf-04 to ietf-05: 1371 * Explained why tightly coupled shim headers only "SHOULD" comply 1372 with RFC 6040, not "MUST". 1374 * Updated references 1376 From ietf-03 to ietf-04: 1378 * Addressed Richard Scheffenegger's review comments: primarily 1379 editorial corrections, and addition of examples for clarity. 1381 From ietf-02 to ietf-03: 1383 * Updated references, ad cited RFC4774. 1385 From ietf-01 to ietf-02: 1387 * Added Section for guidelines that are applicable in all cases. 1389 * Updated references. 1391 From ietf-00 to ietf-01: Updated references. 1393 From briscoe-04 to ietf-00: Changed filename following tsvwg 1394 adoption. 1396 From briscoe-03 to 04: 1398 * Re-arranged the introduction to describe the purpose of the 1399 document first before introducing ECN in more depth. And 1400 clarified the introduction throughout. 1402 * Added applicability to 3GPP TS 36.300. 1404 From briscoe-02 to 03: 1406 * Scope section: 1408 + Added dependence on correct propagation of traffic class 1409 information 1411 + For the feed-backward mode, deemed multicast and anycast out 1412 of scope 1414 * Ensured all guidelines referring to subnet technologies also 1415 refer to tunnels and vice versa by adding applicability 1416 sentences at the start of sections 4.1, 4.2, 4.3, 4.4, 4.6 and 1417 5. 1419 * Added Security Considerations on ensuring congestion signal 1420 fields are classed as immutable and on using end-to-end 1421 congestion signal integrity technologies rather than hop-by- 1422 hop. 1424 From briscoe-01 to 02: 1426 * Added authors: JK & PT 1428 * Added 1430 + Section 4.1 "IP-in-IP Tunnels with Tightly Coupled Shim 1431 Headers" 1433 + Section 4.5 "Sequences of Similar Tunnels or Subnets" 1435 + roadmap at the start of Section 4, given the subsections 1436 have become quite fragmented. 1438 + Section 9 "Conclusions" 1440 * Clarified why transports are starting to be able to saturate 1441 interior links 1443 * Under Section 1.1, addressed the question of alternative signal 1444 semantics and included multicast & anycast. 1446 * Under Section 3.1, included a 3GPP example. 1448 * Section 4.2. "Wire Protocol Design": 1450 + Altered guideline 2. to make it clear that it only applies 1451 to the immediate subnet egress, not later ones 1453 + Added a reminder that it is only necessary to check that ECN 1454 propagates at the egress, not whether interior nodes mark 1455 ECN 1457 + Added example of how QCN uses 802.1p to indicate support for 1458 QCN. 1460 * Added references to Appendix C of RFC6040, about monitoring the 1461 amount of congestion signals introduced within a tunnel 1463 * Appendix A: Added more issues to be addressed, including plan 1464 to produce a standards track update to IP-in-IP tunnel 1465 protocols. 1467 * Updated acks and references 1469 From briscoe-00 to 01: 1471 * Intended status: BCP (was Informational) & updates 3819 added. 1473 * Briefer Introduction: Introductory para justifying benefits of 1474 ECN. Moved all but a brief enumeration of modes of operation 1475 to their own new section (from both Intro & Scope). Introduced 1476 incr. deployment as most tricky part. 1478 * Tightened & added to terminology section 1480 * Structured with Modes of Operation, then Guidelines section for 1481 each mode. 1483 * Tightened up guideline text to remove vagueness / passive voice 1484 / ambiguity and highlight main guidelines as numbered items. 1486 * Added Outstanding Document Issues Appendix 1488 * Updated references 1490 Authors' Addresses 1492 Bob Briscoe 1493 Simula Research Laboratory 1494 UK 1496 EMail: ietf@bobbriscoe.net 1497 URI: http://bobbriscoe.net/ 1499 John Kaippallimalil 1500 Huawei 1501 5340 Legacy Drive, Suite 175 1502 Plano, Texas 75024 1503 USA 1505 EMail: john.kaippallimalil@huawei.com 1506 Pat Thaler 1507 Broadcom Corporation 1508 5025 Keane Drive 1509 Carmichael, CA 95608 1510 USA 1512 EMail: pthaler@broadcom.com