idnits 2.17.1 draft-ietf-quic-applicability-14.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 (21 January 2022) is 818 days in the past. Is this intentional? Checking references for intended status: Informational ---------------------------------------------------------------------------- == Outdated reference: A later version (-14) exists of draft-ietf-quic-version-negotiation-05 == Outdated reference: A later version (-10) exists of draft-ietf-quic-datagram-08 == Outdated reference: A later version (-18) exists of draft-ietf-quic-manageability-13 == Outdated reference: A later version (-19) exists of draft-ietf-taps-arch-12 == Outdated reference: A later version (-19) exists of draft-ietf-quic-load-balancers-10 -- Obsolete informational reference (is this intentional?): RFC 5077 (Obsoleted by RFC 8446) Summary: 0 errors (**), 0 flaws (~~), 6 warnings (==), 2 comments (--). Run idnits with the --verbose option for more detailed information about the items above. -------------------------------------------------------------------------------- 2 Network Working Group M. Kuehlewind 3 Internet-Draft Ericsson 4 Intended status: Informational B. Trammell 5 Expires: 25 July 2022 Google 6 21 January 2022 8 Applicability of the QUIC Transport Protocol 9 draft-ietf-quic-applicability-14 11 Abstract 13 This document discusses the applicability of the QUIC transport 14 protocol, focusing on caveats impacting application protocol 15 development and deployment over QUIC. Its intended audience is 16 designers of application protocol mappings to QUIC, and implementors 17 of these application protocols. 19 Status of This Memo 21 This Internet-Draft is submitted in full conformance with the 22 provisions of BCP 78 and BCP 79. 24 Internet-Drafts are working documents of the Internet Engineering 25 Task Force (IETF). Note that other groups may also distribute 26 working documents as Internet-Drafts. The list of current Internet- 27 Drafts is at https://datatracker.ietf.org/drafts/current/. 29 Internet-Drafts are draft documents valid for a maximum of six months 30 and may be updated, replaced, or obsoleted by other documents at any 31 time. It is inappropriate to use Internet-Drafts as reference 32 material or to cite them other than as "work in progress." 34 This Internet-Draft will expire on 25 July 2022. 36 Copyright Notice 38 Copyright (c) 2022 IETF Trust and the persons identified as the 39 document authors. All rights reserved. 41 This document is subject to BCP 78 and the IETF Trust's Legal 42 Provisions Relating to IETF Documents (https://trustee.ietf.org/ 43 license-info) in effect on the date of publication of this document. 44 Please review these documents carefully, as they describe your rights 45 and restrictions with respect to this document. Code Components 46 extracted from this document must include Revised BSD License text as 47 described in Section 4.e of the Trust Legal Provisions and are 48 provided without warranty as described in the Revised BSD License. 50 Table of Contents 52 1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . 2 53 2. The Necessity of Fallback . . . . . . . . . . . . . . . . . . 3 54 3. Zero RTT . . . . . . . . . . . . . . . . . . . . . . . . . . 4 55 3.1. Replay Attacks . . . . . . . . . . . . . . . . . . . . . 4 56 3.2. Session resumption versus Keep-alive . . . . . . . . . . 5 57 4. Use of Streams . . . . . . . . . . . . . . . . . . . . . . . 7 58 4.1. Stream versus Flow Multiplexing . . . . . . . . . . . . . 8 59 4.2. Prioritization . . . . . . . . . . . . . . . . . . . . . 9 60 4.3. Ordered and Reliable Delivery . . . . . . . . . . . . . . 9 61 4.4. Flow Control Deadlocks . . . . . . . . . . . . . . . . . 10 62 4.5. Stream Limit Commitments . . . . . . . . . . . . . . . . 11 63 5. Packetization and Latency . . . . . . . . . . . . . . . . . . 12 64 6. Error Handling . . . . . . . . . . . . . . . . . . . . . . . 13 65 7. Acknowledgment Efficiency . . . . . . . . . . . . . . . . . . 14 66 8. Port Selection and Application Endpoint Discovery . . . . . . 14 67 8.1. Source Port Selection . . . . . . . . . . . . . . . . . . 15 68 9. Connection Migration . . . . . . . . . . . . . . . . . . . . 16 69 10. Connection Termination . . . . . . . . . . . . . . . . . . . 16 70 11. Information Exposure and the Connection ID . . . . . . . . . 17 71 11.1. Server-Generated Connection ID . . . . . . . . . . . . . 18 72 11.2. Mitigating Timing Linkability with Connection ID 73 Migration . . . . . . . . . . . . . . . . . . . . . . . 18 74 11.3. Using Server Retry for Redirection . . . . . . . . . . . 19 75 12. Quality of Service (QoS) and DSCP . . . . . . . . . . . . . . 19 76 13. Use of Versions and Cryptographic Handshake . . . . . . . . . 20 77 14. Enabling New Versions . . . . . . . . . . . . . . . . . . . . 20 78 15. Unreliable Datagram Service over QUIC . . . . . . . . . . . . 21 79 16. IANA Considerations . . . . . . . . . . . . . . . . . . . . . 22 80 17. Security Considerations . . . . . . . . . . . . . . . . . . . 22 81 18. Contributors . . . . . . . . . . . . . . . . . . . . . . . . 22 82 19. Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . 23 83 20. References . . . . . . . . . . . . . . . . . . . . . . . . . 23 84 20.1. Normative References . . . . . . . . . . . . . . . . . . 23 85 20.2. Informative References . . . . . . . . . . . . . . . . . 23 86 Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . 27 88 1. Introduction 90 QUIC [QUIC] is a new transport protocol providing a number of 91 advanced features. While initially designed for the HTTP use case, 92 it provides capabilities that can be used with a much wider variety 93 of applications. QUIC is encapsulated in UDP. QUIC version 1 94 integrates TLS 1.3 [TLS13] to encrypt all payload data and most 95 control information. The version of HTTP that uses QUIC is known as 96 HTTP/3 [QUIC-HTTP]. 98 This document provides guidance for application developers that want 99 to use the QUIC protocol without implementing it on their own. This 100 includes general guidance for applications operating over HTTP/3 or 101 directly over QUIC. 103 In the following sections we discuss specific caveats to QUIC's 104 applicability, and issues that application developers must consider 105 when using QUIC as a transport for their application. 107 2. The Necessity of Fallback 109 QUIC uses UDP as a substrate. This enables userspace implementation 110 and permits traversal of network middleboxes (including NAT) without 111 requiring updates to existing network infrastructure. 113 While recent measurements have shown no evidence of a widespread, 114 systematic disadvantage of UDP traffic compared to TCP in the 115 Internet [Edeline16], somewhere between three [Trammell16] and five 116 [Swett16] percent of networks block all UDP traffic. All 117 applications running on top of QUIC must therefore either be prepared 118 to accept connectivity failure on such networks or be engineered to 119 fall back to some other transport protocol. In the case of HTTP, 120 this fallback is TLS over TCP. 122 The IETF TAPS specifications [I-D.ietf-taps-arch] describe a system 123 with a common API for multiple protocols. This is particularly 124 relevant for QUIC as it addresses the implications of fallback among 125 multiple protocols. 127 Specifically, fallback to insecure protocols or to weaker versions of 128 secure protocols needs to be avoided. In general, a application that 129 implements fallback needs to consider the security consequences. A 130 fallback to TCP and TLS exposes control information to modification 131 and manipulation in the network. Further, downgrades to older TLS 132 versions than 1.3, which is used in QUIC version 1, might result in 133 significantly weaker cryptographic protection. For example, the 134 results of protocol negotiation [RFC7301] only have confidentiality 135 protection if TLS 1.3 is used. 137 These applications must operate, perhaps with impaired functionality, 138 in the absence of features provided by QUIC not present in the 139 fallback protocol. For fallback to TLS over TCP, the most obvious 140 difference is that TCP does not provide stream multiplexing and 141 therefore stream multiplexing would need to be implemented in the 142 application layer if needed. Further, TCP implementations and 143 network paths often do not support the Fast Open option [RFC7413], 144 which enables sending of payload data together with the first control 145 packet of a new connection as also provided by 0-RTT session 146 resumption in QUIC. Note that there is some evidence of middleboxes 147 blocking SYN data even if TFO was successfully negotiated (see 148 [PaaschNanog]). And even if Fast Open successfully operates end-to- 149 end, it is limited to a single packet of TLS handshake and 150 application data, unlike QUIC 0-RTT. 152 Moreover, while encryption (in this case TLS) is inseparably 153 integrated with QUIC, TLS negotiation over TCP can be blocked. If 154 TLS over TCP cannot be supported, the connection should be aborted, 155 and the application then ought to present a suitable prompt to the 156 user that secure communication is unavailable. 158 In summary, any fallback mechanism is likely to impose a degradation 159 of performance and can degrade security; however, fallback must not 160 silently violate the application's expectation of confidentiality or 161 integrity of its payload data. 163 3. Zero RTT 165 QUIC provides for 0-RTT connection establishment. Though the same 166 facility exists in TLS 1.3 with TCP, 0-RTT presents opportunities and 167 challenges for applications using QUIC. 169 A transport protocol that provides 0-RTT connection establishment is 170 qualitatively different than one that does not from the point of view 171 of the application using it. Relative trade-offs between the cost of 172 closing and reopening a connection and trying to keep it open are 173 different; see Section 3.2. 175 An application needs to deliberately choose to use 0-RTT, as 0-RTT 176 carries a risk of replay attack. Application protocols that use 177 0-RTT require a profile that describes the types of information that 178 can be safely sent. For HTTP, this profile is described in 179 [HTTP-REPLAY]. 181 3.1. Replay Attacks 183 Retransmission or (malicious) replay of data contained in 0-RTT 184 packets could cause the server side to receive multiple copies of the 185 same data. 187 Application data sent by the client in 0-RTT packets could be 188 processed more than once if it is replayed. Applications need to be 189 aware of what is safe to send in 0-RTT. Application protocols that 190 seek to enable the use of 0-RTT need a careful analysis and a 191 description of what can be sent in 0-RTT; see Section 5.6 of 192 [QUIC-TLS]. 194 In some cases, it might be sufficient to limit application data sent 195 in 0-RTT to that which only causes actions at a server that are known 196 to be free of lasting effect. Initiating data retrieval or 197 establishing configuration are examples of actions that could be 198 safe. Idempotent operations - those for which repetition has the 199 same net effect as a single operation - might be safe. However, it 200 is also possible to combine individually idempotent operations into a 201 non-idempotent sequence of operations. 203 Once a server accepts 0-RTT data there is no means of selectively 204 discarding data that is received. However, protocols can define ways 205 to reject individual actions that might be unsafe if replayed. 207 Some TLS implementations and deployments might be able to provide 208 partial or even complete replay protection, which could be used to 209 manage replay risk. 211 3.2. Session resumption versus Keep-alive 213 Because QUIC is encapsulated in UDP, applications using QUIC must 214 deal with short network idle timeouts. Deployed stateful middleboxes 215 will generally establish state for UDP flows on the first packet 216 sent, and keep state for much shorter idle periods than for TCP. 217 [RFC5382] suggests a TCP idle period of at least 124 minutes, though 218 there is no evidence of widespread implementation of this guideline 219 in the literature. Short network timeout for UDP, however, is well- 220 documented. According to a 2010 study ([Hatonen10]), UDP 221 applications can assume that any NAT binding or other state entry can 222 expire after just thirty seconds of inactivity. Section 3.5 of 223 [RFC8085] further discusses keep-alive intervals for UDP: it requires 224 a minimum value of 15 seconds, but recommends larger values, or 225 omitting keep-alive entirely. 227 By using a connection ID, QUIC is designed to be robust to NAT 228 address rebinding after a timeout. However, this only helps if one 229 endpoint maintains availability at the address its peer uses, and the 230 peer is the one to send after the timeout occurs. 232 Some QUIC connections might not be robust to NAT rebinding because 233 the routing infrastructure (in particular, load balancers) uses the 234 address/port four-tuple to direct traffic. Furthermore, middleboxes 235 with functions other than address translation could still affect the 236 path. In particular, some firewalls do not admit server traffic for 237 which the firewall has no recent state for a corresponding packet 238 sent from the client. 240 QUIC applications can adjust idle periods to manage the risk of 241 timeout. Idle periods and the network idle timeout are distinct from 242 the connection idle timeout, which is defined as the minimum of 243 either endpoint's idle timeout parameter; see Section 10.1 of 244 [QUIC]). There are three options: 246 * Ignore the issue, if the application-layer protocol consists only 247 of interactions with no or very short idle periods, or the 248 protocol's resistance to NAT rebinding is sufficient. 250 * Ensure there are no long idle periods. 252 * Resume the session after a long idle period, using 0-RTT 253 resumption when appropriate. 255 The first strategy is the easiest, but it only applies to certain 256 applications. 258 Either the server or the client in a QUIC application can send PING 259 frames as keep-alives, to prevent the connection and any on-path 260 state from timing out. Recommendations for the use of keep-alives 261 are application-specific, mainly depending on the latency 262 requirements and message frequency of the application. In this case, 263 the application mapping must specify whether the client or server is 264 responsible for keeping the application alive. While [Hatonen10] 265 suggests that 30 seconds might be a suitable value for the public 266 Internet when a NAT is on path, larger values are preferable if the 267 deployment can consistently survive NAT rebinding or is known to be 268 in a controlled environment (e.g. data centres) in order to lower 269 network and computational load. 271 Sending PING frames more frequently than every 30 seconds over long 272 idle periods may result in excessive unproductive traffic in some 273 situations, and to unacceptable power usage for power-constrained 274 (mobile) devices. Additionally, timeouts shorter than 30 seconds can 275 make it harder to handle transient network interruptions, such as VM 276 migration or coverage loss during mobilty. See [RFC8085], especially 277 Section 3.5. 279 Alternatively, the client (but not the server) can use session 280 resumption instead of sending keepalive traffic. In this case, a 281 client that wants to send data to a server over a connection that has 282 been idle longer than the server's idle timeout (available from the 283 idle_timeout transport parameter) can simply reconnect. When 284 possible, this reconnection can use 0-RTT session resumption, 285 reducing the latency involved with restarting the connection. Of 286 course, this approach is only valid in cases in which it is safe to 287 use 0-RTT and when the client is the restarting peer. 289 The tradeoffs between resumption and keep-alives need to be evaluated 290 on a per-application basis. In general, applications should use 291 keep-alives only in circumstances where continued communication is 292 highly likely; [QUIC-HTTP], for instance, recommends using keep- 293 alives only when a request is outstanding. 295 4. Use of Streams 297 QUIC's stream multiplexing feature allows applications to run 298 multiple streams over a single connection, without head-of-line 299 blocking between streams. Stream data is carried within frames, 300 where one QUIC packet on the wire can carry one or multiple stream 301 frames. 303 Streams can be unidirectional or bidirectional, and a stream may be 304 initiated either by client or server. Only the initiator of a 305 unidirectional stream can send data on it. 307 Streams and connections can each carry a maximum of 2^62-1 bytes in 308 each direction, due to encoding limitations on stream offsets and 309 connection flow control limits. In the presently unlikely event that 310 this limit is reached by an application, a new connection would need 311 to be established. 313 Streams can be independently opened and closed, gracefully or 314 abruptly. An application can gracefully close the egress direction 315 of a stream by instructing QUIC to send a FIN bit in a STREAM frame. 316 It cannot gracefully close the ingress direction without a peer- 317 generated FIN, much like in TCP. However, an endpoint can abruptly 318 close the egress direction or request that its peer abruptly close 319 the ingress direction; these actions are fully independent of each 320 other. 322 QUIC does not provide an interface for exceptional handling of any 323 stream. If a stream that is critical for an application is closed, 324 the application can generate error messages on the application layer 325 to inform the other end and/or the higher layer, which can eventually 326 terminate the QUIC connection. 328 Mapping of application data to streams is application-specific and 329 described for HTTP/3 in [QUIC-HTTP]. There are a few general 330 principles to apply when designing an application's use of streams: 332 * A single stream provides ordering. If the application requires 333 certain data to be received in order, that data should be sent on 334 the same stream. There is no guarantee of transmission, 335 reception, or delivery order across streams. 337 * Multiple streams provide concurrency. Data that can be processed 338 independently, and therefore would suffer from head of line 339 blocking if forced to be received in order, should be transmitted 340 over separate streams. 342 * Streams can provide message orientation, and allow messages to be 343 cancelled. If one message is mapped to a single stream, resetting 344 the stream to expire an unacknowledged message can be used to 345 emulate partial reliability for that message. 347 If a QUIC receiver has opened the maximum allowed concurrent streams, 348 and the sender indicates that more streams are needed, it does not 349 automatically lead to an increase of the maximum number of streams by 350 the receiver. Therefore, an application can use the maximum number 351 of allowed, currently open, and currently used streams when 352 determining how to map data to streams. 354 QUIC assigns a numerical identifier to each stream, called the stream 355 ID. While the relationship between these identifiers and stream 356 types is clearly defined in version 1 of QUIC, future versions might 357 change this relationship for various reasons. QUIC implementations 358 should expose the properties of each stream (which endpoint initiated 359 the stream, whether the stream is unidirectional or bidirectional, 360 the stream ID used for the stream); applications should query for 361 these properties rather than attempting to infer them from the stream 362 ID. 364 The method of allocating stream identifiers to streams opened by the 365 application might vary between transport implementations. Therefore, 366 an application should not assume a particular stream ID will be 367 assigned to a stream that has not yet been allocated. For example, 368 HTTP/3 uses stream IDs to refer to streams that have already been 369 opened, but makes no assumptions about future stream IDs or the way 370 in which they are assigned Section 6 of [QUIC-HTTP]). 372 4.1. Stream versus Flow Multiplexing 374 Streams are meaningful only to the application; since stream 375 information is carried inside QUIC's encryption boundary, a given 376 packet exposes no information about which stream(s) are carried 377 within the packet. Therefore, stream multiplexing is not intended to 378 be used for differentiating streams in terms of network treatment. 379 Application traffic requiring different network treatment should 380 therefore be carried over different five-tuples (i.e. multiple QUIC 381 connections). Given QUIC's ability to send application data in the 382 first RTT of a connection (if a previous connection to the same host 383 has been successfully established to provide the necessary 384 credentials), the cost of establishing another connection is 385 extremely low. 387 4.2. Prioritization 389 Stream prioritization is not exposed to either the network or the 390 receiver. Prioritization is managed by the sender, and the QUIC 391 transport should provide an interface for applications to prioritize 392 streams [QUIC]. Applications can implement their own prioritization 393 scheme on top of QUIC: an application protocol that runs on top of 394 QUIC can define explicit messages for signaling priority, such as 395 those defined in [I-D.draft-ietf-httpbis-priority] for HTTP; it can 396 define rules that allow an endpoint to determine priority based on 397 context; or it can provide a higher level interface and leave the 398 determination to the application on top. 400 Priority handling of retransmissions can be implemented by the sender 401 in the transport layer. [QUIC] recommends retransmitting lost data 402 before new data, unless indicated differently by the application. 403 When a QUIC endpoint uses fully reliable streams for transmission, 404 prioritization of retransmissions will be beneficial in most cases, 405 filling in gaps and freeing up the flow control window. For 406 partially reliable or unreliable streams, priority scheduling of 407 retransmissions over data of higher-priority streams might not be 408 desirable. For such streams, QUIC could either provide an explicit 409 interface to control prioritization, or derive the prioritization 410 decision from the reliability level of the stream. 412 4.3. Ordered and Reliable Delivery 414 QUIC streams enable ordered and reliable delivery. Though it is 415 possible for an implementation to provide options that use streams 416 for partial reliability or out-of-order delivery, most 417 implementations will assume that data is reliably delivered in order. 419 Under this assumption, an endpoint that receives stream data might 420 not make forward progress until data that is contiguous with the 421 start of a stream is available. In particular, a receiver might 422 withhold flow control credit until contiguous data is delivered to 423 the application; see Section 2.2 of [QUIC]. To support this receive 424 logic, an endpoint will send stream data until it is acknowledged, 425 ensuring that data at the start of the stream is sent and 426 acknowledged first. 428 An endpoint that uses a different sending behavior and does not 429 negotiate that change with its peer might encounter performance 430 issues or deadlocks. 432 4.4. Flow Control Deadlocks 434 QUIC flow control Section 4 of [QUIC] provides a means of managing 435 access to the limited buffers endpoints have for incoming data. This 436 mechanism limits the amount of data that can be in buffers in 437 endpoints or in transit on the network. However, there are several 438 ways in which limits can produce conditions that can cause a 439 connection to either perform suboptimally or deadlock. 441 Deadlocks in flow control are possible for any protocol that uses 442 QUIC, though whether they become a problem depends on how 443 implementations consume data and provide flow control credit. 444 Understanding what causes deadlocking might help implementations 445 avoid deadlocks. 447 The size and rate of transport flow control credit updates can affect 448 performance. Applications that use QUIC often have a data consumer 449 that reads data from transport buffers. Some implementations might 450 have independent transport-layer and application-layer receive 451 buffers. Consuming data does not always imply it is immediately 452 processed. However, a common flow control implementation technique 453 is to extend credit to the sender, by emitting MAX_DATA and/or 454 MAX_STREAM_DATA frames, as data is consumed. Delivery of these 455 frames is affected by the latency of the back channel from the 456 receiver to the data sender. If credit is not extended in a timely 457 manner, the sending application can be blocked, effectively 458 throttling the sender. 460 Large application messages can produce deadlocking if the recipient 461 does not read data from the transport incrementally. If the message 462 is larger than the flow control credit available and the recipient 463 does not release additional flow control credit until the entire 464 message is received and delivered, a deadlock can occur. This is 465 possible even where stream flow control limits are not reached 466 because connection flow control limits can be consumed by other 467 streams. 469 A length-prefixed message format makes it easier for a data consumer 470 to leave data unread in the transport buffer and thereby withhold 471 flow control credit. If flow control limits prevent the remainder of 472 a message from being sent, a deadlock will result. A length prefix 473 might also enable the detection of this sort of deadlock. Where 474 application protocols have messages that might be processed as a 475 single unit, reserving flow control credit for the entire message 476 atomically makes this style of deadlock less likely. 478 A data consumer can eagerly read all data as it becomes available, in 479 order to make the receiver extend flow control credit and reduce the 480 chances of a deadlock. However, such a data consumer might need 481 other means for holding a peer accountable for the additional state 482 it keeps for partially processed messages. 484 Deadlocking can also occur if data on different streams is 485 interdependent. Suppose that data on one stream arrives before the 486 data on a second stream on which it depends. A deadlock can occur if 487 the first stream is left unread, preventing the receiver from 488 extending flow control credit for the second stream. To reduce the 489 likelihood of deadlock for interdependent data, the sender should 490 ensure that dependent data is not sent until the data it depends on 491 has been accounted for in both stream- and connection- level flow 492 control credit. 494 Some deadlocking scenarios might be resolved by cancelling affected 495 streams with STOP_SENDING or RESET_STREAM. Cancelling some streams 496 results in the connection being terminated in some protocols. 498 4.5. Stream Limit Commitments 500 QUIC endpoints are responsible for communicating the cumulative limit 501 of streams they would allow to be opened by their peer. Initial 502 limits are advertised using the initial_max_streams_bidi and 503 initial_max_streams_uni transport parameters. As streams are opened 504 and closed they are consumed and the cumulative total is incremented. 505 Limits can be increased using the MAX_STREAMS frame but there is no 506 mechanism to reduce limits. Once stream limits are reached, no more 507 streams can be opened, which prevents applications using QUIC from 508 making further progress. At this stage connections can be terminated 509 via idle timeout or explicit close; see Section 10). 511 An application that uses QUIC and communicated a cumulative stream 512 limit might require the connection to be closed before the limit is 513 reached. For example, to stop the server to perform scheduled 514 maintenance. Immediate connection close causes abrupt closure of 515 actively used streams. Depending on how an application uses QUIC 516 streams, this could be undesirable or detrimental to behavior or 517 performance. 519 A more graceful closure technique is to stop sending increases to 520 stream limits and allow the connection to naturally terminate once 521 remaining streams are consumed. However, the period of time it takes 522 to do so is dependent on the peer and an unpredictable closing period 523 might not fit application or operational needs. Applications using 524 QUIC can be conservative with open stream limits in order to reduce 525 the commitment and indeterminism. However, being overly conservative 526 with stream limits affects stream concurrency. Balancing these 527 aspects can be specific to applications and their deployments. 529 Instead of relying on stream limits to avoid abrupt closure, an 530 application-layer graceful close mechanism can be used to communicate 531 the intention to explicitly close the connection at some future 532 point. HTTP/3 provides such a mechanism using the GOAWAY frame. In 533 HTTP/3, when the GOAWAY frame is received by a client, it stops 534 opening new streams even if the cumulative stream limit would allow. 535 Instead, the client would create a new connection on which to open 536 further streams. Once all streams are closed on the old connection, 537 it can be terminated safely by a connection close or after expiration 538 of the idle time out (see also Section 10). 540 5. Packetization and Latency 542 QUIC exposes an interface that provides multiple streams to the 543 application; however, the application usually cannot control how data 544 transmitted over those streams is mapped into frames or how those 545 frames are bundled into packets. 547 By default, many implementations will try to maximally pack QUIC 548 packets DATA frames from one or more streams to minimize bandwidth 549 consumption and computational costs (see Section 13 of [QUIC]). If 550 there is not enough data available to fill a packet, an 551 implementation might wait for a short time, to optimize bandwidth 552 efficiency instead of latency. This delay can either be pre- 553 configured or dynamically adjusted based on the observed sending 554 pattern of the application. 556 If the application requires low latency, with only small chunks of 557 data to send, it may be valuable to indicate to QUIC that all data 558 should be sent out immediately. Alternatively, if the application 559 expects to use a specific sending pattern, it can also provide a 560 suggested delay to QUIC for how long to wait before bundle frames 561 into a packet. 563 Similarly, an application has usually no control about the length of 564 a QUIC packet on the wire. QUIC provides the ability to add a 565 PADDING frame to arbitrarily increase the size of packets. Padding 566 is used by QUIC to ensure that the path is capable of transferring 567 datagrams of at least a certain size, during the handshake (see 568 Sections 8.1 and 14.1 of [QUIC]) and for path validation after 569 connection migration (see Section 8.2 of [QUIC]) as well as for 570 Datagram Packetization Layer PMTU Discovery (DPLMTUD) (see 571 Section 14.3 of [QUIC]). 573 Padding can also be used by an application to reduce leakage of 574 information about the data that is sent. A QUIC implementation can 575 expose an interface that allows an application layer to specify how 576 to apply padding. 578 6. Error Handling 580 QUIC recommends that endpoints signal any detected errors to the 581 peer. Errors can occur at the transport level and the application 582 level. Transport errors, such as a protocol violation, affect the 583 entire connection. Applications that use QUIC can define their own 584 error detection and signaling (see, for example, Section 8 of 585 [QUIC-HTTP]). Application errors can affect an entire connection or 586 a single stream. 588 QUIC defines an error code space that is used for error handling at 589 the transport layer. QUIC encourages endpoints to use the most 590 specific code, although any applicable code is permitted, including 591 generic ones. 593 Applications using QUIC define an error code space that is 594 independent from QUIC or other applications (see, for example, 595 Section 8.1 of [QUIC-HTTP]). The values in an application error code 596 space can be reused across connection-level and stream-level errors. 598 Connection errors lead to connection termination. They are signaled 599 using a CONNECTION_CLOSE frame, which contains an error code and a 600 reason field that can be zero length. Different types of 601 CONNECTION_CLOSE frame are used to signal transport and application 602 errors. 604 Stream errors lead to stream termination. These are signaled using 605 STOP_SENDING or RESET_STREAM frames, which contain only an error 606 code. 608 7. Acknowledgment Efficiency 610 QUIC version 1 without extensions uses an acknowledgment strategy 611 adopted from TCP Section 13.2 of [QUIC]). That is, it recommends 612 every other packet is acknowledged. However, generating and 613 processing QUIC acknowledgments consumes resources at a sender and 614 receiver. Acknowledgments also incur forwarding costs and contribute 615 to link utilization, which can impact performance over some types of 616 network. Applications might be able to improve overall performance 617 by using alternative strategies that reduce the rate of 618 acknowledgments. 620 8. Port Selection and Application Endpoint Discovery 622 In general, port numbers serve two purposes: "first, they provide a 623 demultiplexing identifier to differentiate transport sessions between 624 the same pair of endpoints, and second, they may also identify the 625 application protocol and associated service to which processes 626 connect" [RFC6335]. The assumption that an application can be 627 identified in the network based on the port number is less true today 628 due to encapsulation, mechanisms for dynamic port assignments, and 629 NATs. 631 As QUIC is a general-purpose transport protocol, there are no 632 requirements that servers use a particular UDP port for QUIC. For 633 applications with a fallback to TCP that do not already have an 634 alternate mapping to UDP, usually the registration (if necessary) and 635 use of the UDP port number corresponding to the TCP port already 636 registered for the application is appropriate. For example, the 637 default port for HTTP/3 [QUIC-HTTP] is UDP port 443, analogous to 638 HTTP/1.1 or HTTP/2 over TLS over TCP. 640 Given the prevalence of the assumption in network management practice 641 that a port number maps unambiguously to an application, the use of 642 ports that cannot easily be mapped to a registered service name might 643 lead to blocking or other changes to the forwarding behavior by 644 network elements such as firewalls that use the port number for 645 application identification. 647 Applications could define an alternate endpoint discovery mechanism 648 to allow the usage of ports other than the default. For example, 649 HTTP/3 (Sections 3.2 and 3.3 of [QUIC-HTTP]) specifies the use of 650 HTTP Alternative Services [RFC7838] for an HTTP origin to advertise 651 the availability of an equivalent HTTP/3 endpoint on a certain UDP 652 port by using the "h3" Application-Layer Protocol Negotiation (ALPN) 653 [RFC7301] token. 655 ALPN permits the client and server to negotiate which of several 656 protocols will be used on a given connection. Therefore, multiple 657 applications might be supported on a single UDP port based on the 658 ALPN token offered. Applications using QUIC are required to register 659 an ALPN token for use in the TLS handshake. 661 As QUIC version 1 deferred defining a complete version negotiation 662 mechanism, HTTP/3 requires QUIC version 1 and defines the ALPN token 663 ("h3") to only apply to that version. So far no single approach has 664 been selected for managing the use of different QUIC versions, 665 neither in HTTP/3 nor in general. Application protocols that use 666 QUIC need to consider how the protocol will manage different QUIC 667 versions. Decisions for those protocols might be informed by choices 668 made by other protocols, like HTTP/3. 670 8.1. Source Port Selection 672 Some UDP protocols are vulnerable to reflection attacks, where an 673 attacker is able to direct traffic to a third party as a denial of 674 service. For example, these source ports are associated with 675 applications known to be vulnerable to reflection attacks, often due 676 to server misconfiguration: 678 * port 53 - DNS [RFC1034] 680 * port 123 - NTP [RFC5905] 682 * port 1900 - SSDP [SSDP] 684 * port 5353 - mDNS [RFC6762] 686 * port 11211 - memcached 688 Services might block source ports associated with protocols known to 689 be vulnerable to reflection attacks, to avoid the overhead of 690 processing large numbers of packets. However, this practice has 691 negative effects on clients: not only does it require establishment 692 of a new connection, but in some instances, might cause the client to 693 avoid using QUIC for that service for a period of time, downgrading 694 to a non-UDP protocol (see Section 2). 696 As a result, client implementations are encouraged to avoid using 697 source ports associated with protocols known to be vulnerable to 698 reflection attacks. Note that the list above is not exhaustive; 699 other source ports might be considered reflection vectors as well. 701 9. Connection Migration 703 QUIC supports connection migration by the client. If an IP address 704 changes, a QUIC endpoint can still associate packets with an existing 705 transport connection using the Destination Connection ID field (see 706 also Section 11) in the QUIC header. This supports cases where 707 address information changes, such as NAT rebinding, intentional 708 change of the local interface, or based on an indication in the 709 handshake of the server for a preferred address to be used. 711 Use of a non-zero-length connection ID for the server is strongly 712 recommended if any clients are behind a NAT or could be. A non-zero- 713 length connection ID is also strongly recommended when active 714 migration is supported. If a connection is intentionally migrated to 715 new path, a new connection ID is used to minimize linkability by 716 network observers. The other QUIC endpoint uses the connection ID to 717 link different addresses to the same connection and entity if a non- 718 zero-length connection ID is provided. 720 The base specification of QUIC version 1 only supports the use of a 721 single network path at a time, which enables failover use cases. 722 Path validation is required so that endpoints validate paths before 723 use to avoid address spoofing attacks. Path validation takes at 724 least one RTT and congestion control will also be reset after path 725 migration. Therefore, migration usually has a performance impact. 727 QUIC probing packets, which can be sent on multiple paths at once, 728 are used to perform address validation as well as measure path 729 characteristics. Probing packets cannot carry application data but 730 likely contain padding frames. Endpoints can use information about 731 their receipt as input to congestion control for that path. 732 Applications could use information learned from probing to inform a 733 decision to switch paths. 735 Only the client can actively migrate in version 1 of QUIC. However, 736 servers can indicate during the handshake that they prefer to 737 transfer the connection to a different address after the handshake. 738 For instance, this could be used to move from an address that is 739 shared by multiple servers to an address that is unique to the server 740 instance. The server can provide an IPv4 and an IPv6 address in a 741 transport parameter during the TLS handshake and the client can 742 select between the two if both are provided. See also Section 9.6 of 743 [QUIC]. 745 10. Connection Termination 747 QUIC connections are terminated in one of three ways: implicit idle 748 timeout, explicit immediate close, or explicit stateless reset. 750 QUIC does not provide any mechanism for graceful connection 751 termination; applications using QUIC can define their own graceful 752 termination process (see, for example, Section 5.2 of [QUIC-HTTP]). 754 QUIC idle timeout is enabled via transport parameters. Client and 755 server announce a timeout period and the effective value for the 756 connection is the minimum of the two values. After the timeout 757 period elapses, the connection is silently closed. An application 758 therefore should be able to configure its own maximum value, as well 759 as have access to the computed minimum value for this connection. An 760 application may adjust the maximum idle timeout for new connections 761 based on the number of open or expected connections, since shorter 762 timeout values may free-up resources more quickly. 764 Application data exchanged on streams or in datagrams defers the QUIC 765 idle timeout. Applications that provide their own keep-alive 766 mechanisms will therefore keep a QUIC connection alive. Applications 767 that do not provide their own keep-alive can use transport-layer 768 mechanisms (see Section 10.1.2 of [QUIC], and Section 3.2). However, 769 QUIC implementation interfaces for controlling such transport 770 behavior can vary, affecting the robustness of such approaches. 772 An immediate close is signaled by a CONNECTION_CLOSE frame (see 773 Section 6). Immediate close causes all streams to become immediately 774 closed, which may affect applications; see Section 4.5. 776 A stateless reset is an option of last resort for an endpoint that 777 does not have access to connection state. Receiving a stateless 778 reset is an indication of an unrecoverable error distinct from 779 connection errors in that there is no application-layer information 780 provided. 782 11. Information Exposure and the Connection ID 784 QUIC exposes some information to the network in the unencrypted part 785 of the header, either before the encryption context is established or 786 because the information is intended to be used by the network. For 787 more information on manageability of QUIC see also 788 [I-D.ietf-quic-manageability]. QUIC has a long header that exposes 789 some additional information (the version and the source connection 790 ID), while the short header exposes only the destination connection 791 ID. In QUIC version 1, the long header is used during connection 792 establishment, while the short header is used for data transmission 793 in an established connection. 795 The connection ID can be zero length. Zero length connection IDs can 796 be chosen on each endpoint individually, on any packet except the 797 first packets sent by clients during connection establishment. 799 An endpoint that selects a zero-length connection ID will receive 800 packets with a zero-length destination connection ID. The endpoint 801 needs to use other information, such as the source and destination IP 802 address and port number to identify which connection is referred to. 803 This could mean that the endpoint is unable to match datagrams to 804 connections successfully if these values change, making the 805 connection effectively unable to survive NAT rebinding or migrate to 806 a new path. 808 11.1. Server-Generated Connection ID 810 QUIC supports a server-generated connection ID, transmitted to the 811 client during connection establishment (see Section 7.2 of [QUIC]). 812 Servers behind load balancers may need to change the connection ID 813 during the handshake, encoding the identity of the server or 814 information about its load balancing pool, in order to support 815 stateless load balancing. 817 Server deployments with load balancers and other routing 818 infrastructure need to ensure that this infrastructure consistently 819 routes packets to the server instance that has the connection state, 820 even if addresses, ports, and/or connection IDs change. This might 821 require coordination between servers and infrastructure. One method 822 of achieving this involves encoding routing information into the 823 connection ID. For an example of this technique, see [QUIC-LB]. 825 11.2. Mitigating Timing Linkability with Connection ID Migration 827 QUIC requires that endpoints generate fresh connection IDs for use on 828 new network paths. Choosing values that are unlinkable to an outside 829 observer ensures that activity on different paths cannot be trivially 830 correlated using the connection ID. 832 While sufficiently robust connection ID generation schemes will 833 mitigate linkability issues, they do not provide full protection. 834 Analysis of the lifetimes of six-tuples (source and destination 835 addresses as well as the migrated CID) may expose these links anyway. 837 In the limit where connection migration in a server pool is rare, it 838 is trivial for an observer to associate two connection IDs. 839 Conversely, in the opposite limit where every server handles multiple 840 simultaneous migrations, even an exposed server mapping may be 841 insufficient information. 843 The most efficient mitigations for these attacks are through network 844 design and/or operational practice, by using a load balancing 845 architecture that loads more flows onto a single server-side address, 846 by coordinating the timing of migrations in an attempt to increase 847 the number of simultaneous migrations at a given time, or through 848 other means. 850 11.3. Using Server Retry for Redirection 852 QUIC provides a Retry packet that can be sent by a server in response 853 to the client Initial packet. The server may choose a new connection 854 ID in that packet and the client will retry by sending another client 855 Initial packet with the server-selected connection ID. This 856 mechanism can be used to redirect a connection to a different server, 857 e.g., due to performance reasons or when servers in a server pool are 858 upgraded gradually, and therefore may support different versions of 859 QUIC. 861 In this case, it is assumed that all servers belonging to a certain 862 pool are served in cooperation with load balancers that forward the 863 traffic based on the connection ID. A server can choose the 864 connection ID in the Retry packet such that the load balancer will 865 redirect the next Initial packet to a different server in that pool. 866 Alternatively the load balancer can directly offer a Retry service as 867 further described in [QUIC-LB]. 869 Section 4 of [RFC5077] describes an example approach for constructing 870 TLS resumption tickets that can be also applied for validation 871 tokens, however, the use of more modern cryptographic algorithms is 872 highly recommended. 874 12. Quality of Service (QoS) and DSCP 876 QUIC, as defined in [QUIC], has a single congestion controller and 877 recovery handler. This design assumes that all packets of a QUIC 878 connection, or at least with the same 5-tuple {dest addr, source 879 addr, protocol, dest port, source port}, that have the same DiffServ 880 Code Point (DSCP) [RFC2475] will receive similar network treatment 881 since feedback about loss or delay of each packet is used as input to 882 the congestion controller. Therefore, packets belonging to the same 883 connection should use a single DSCP. Section 5.1 of [RFC7657] 884 provides a discussion of DiffServ interactions with datagram 885 transport protocols [RFC7657] (in this respect the interactions with 886 QUIC resemble those of SCTP). 888 When multiplexing multiple flows over a single QUIC connection, the 889 selected DSCP value should be the one associated with the highest 890 priority requested for all multiplexed flows. 892 If differential network treatment is desired, e.g., by the use of 893 different DSCPs, multiple QUIC connections to the same server may be 894 used. However, in general it is recommended to minimize the number 895 of QUIC connections to the same server, to avoid increased overhead 896 and, more importantly, competing congestion control. 898 As in other uses of DiffServ, when a packet enters a network segment 899 that does not support the DSCP value, this could result in the 900 connection not receiving the network treatment it expects. The DSCP 901 value in this packet could also be remarked as the packet travels 902 along the network path, changing the requested treatment. 904 13. Use of Versions and Cryptographic Handshake 906 Versioning in QUIC may change the protocol's behavior completely, 907 except for the meaning of a few header fields that have been declared 908 to be invariant [QUIC-INVARIANTS]. A version of QUIC with a higher 909 version number will not necessarily provide a better service, but 910 might simply provide a different feature set. As such, an 911 application needs to be able to select which versions of QUIC it 912 wants to use. 914 A new version could use an encryption scheme other than TLS 1.3 or 915 higher. [QUIC] specifies requirements for the cryptographic 916 handshake as currently realized by TLS 1.3 and described in a 917 separate specification [QUIC-TLS]. This split is performed to enable 918 light-weight versioning with different cryptographic handshakes. 920 14. Enabling New Versions 922 QUIC version 1 does not specify a version negotation mechanism in the 923 base spec but [I-D.draft-ietf-quic-version-negotiation] proposes an 924 extension. This process assumes that the set of versions that a 925 server supports is fixed. This complicates the process for deploying 926 new QUIC versions or disabling old versions when servers operate in 927 clusters. 929 A server that rolls out a new version of QUIC can do so in three 930 stages. Each stage is completed across all server instances before 931 moving to the next stage. 933 In the first stage of deployment, all server instances start 934 accepting new connections with the new version. The new version can 935 be enabled progressively across a deployment, which allows for 936 selective testing. This is especially useful when the new version is 937 compatible with an old version, because the new version is more 938 likely to be used. 940 While enabling the new version, servers do not advertise the new 941 version in any Version Negotiation packets they send. This prevents 942 clients that receive a Version Negotiation packet from attempting to 943 connect to server instances that might not have the new version 944 enabled. 946 During the initial deployment, some clients will have received 947 Version Negotiation packets that indicate that the server does not 948 support the new version. Other clients might have successfully 949 connected with the new version and so will believe that the server 950 supports the new version. Therefore, servers need to allow for this 951 ambiguity when validating the negotiated version. 953 The second stage of deployment commences once all server instances 954 are able to accept new connections with the new version. At this 955 point, all servers can start sending the new version in Version 956 Negotiation packets. 958 During the second stage, the server still allows for the possibility 959 that some clients believe the new version to be available and some do 960 not. This state will persist only for as long as any Version 961 Negotiation packets take to be transmitted and responded to. So the 962 third stage can follow after a relatively short delay. 964 The third stage completes the process by enabling authentication of 965 the negotiated version with the assumption that the new version is 966 fully available. 968 The process for disabling an old version or rolling back the 969 introduction of a new version uses the same process in reverse. 970 Servers disable validation of the old version, stop sending the old 971 version in Version Negotiation packets, then the old version is no 972 longer accepted. 974 15. Unreliable Datagram Service over QUIC 976 [I-D.ietf-quic-datagram] specifies a QUIC extension to enable sending 977 and receiving unreliable datagrams over QUIC. Unlike operating 978 directly over UDP, applications that use the QUIC datagram service do 979 not need to implement their own congestion control, per [RFC8085], as 980 QUIC datagrams are congestion controlled. 982 QUIC datagrams are not flow-controlled, and as such data chunks may 983 be dropped if the receiver is overloaded. While the reliable 984 transmission service of QUIC provides a stream-based interface to 985 send and receive data in order over multiple QUIC streams, the 986 datagram service has an unordered message-based interface. If 987 needed, an application layer framing can be used on top to allow 988 separate flows of unreliable datagrams to be multiplexed on one QUIC 989 connection. 991 16. IANA Considerations 993 This document has no actions for IANA; however, note that Section 8 994 recommends that application bindings to QUIC for applications using 995 TCP register UDP ports analogous to their existing TCP registrations. 997 17. Security Considerations 999 See the security considerations in [QUIC] and [QUIC-TLS]; the 1000 security considerations for the underlying transport protocol are 1001 relevant for applications using QUIC, as well. Considerations on 1002 linkability, replay attacks, and randomness discussed in [QUIC-TLS] 1003 should be taken into account when deploying and using QUIC. 1005 Further, migration to an new address exposes a linkage between client 1006 addresses to the server and may expose this linkage also to the path 1007 if the connection ID cannot be changed or flows can otherwise be 1008 correlated. When migration is supported, this needs to be considered 1009 with respective to user privacy. 1011 Application developers should note that any fallback they use when 1012 QUIC cannot be used due to network blocking of UDP should guarantee 1013 the same security properties as QUIC; if this is not possible, the 1014 connection should fail to allow the application to explicitly handle 1015 fallback to a less-secure alternative. See Section 2. 1017 Further, [QUIC-HTTP] provides security considerations specific to 1018 HTTP. However, discussions such as on cross-protocol attacks, 1019 traffic analysis and padding, or migration might be relevant for 1020 other applications using QUIC as well. 1022 18. Contributors 1024 The following people have contributed significant text to and/or 1025 feedback on this document: 1027 * Gorry Fairhurst 1029 * Ian Swett 1030 * Igor Lubashev 1032 * Lucas Pardue 1034 * Mike Bishop 1036 * Mark Nottingham 1038 * Martin Duke 1040 * Martin Thomson 1042 * Sean Turner 1044 * Tommy Pauly 1046 19. Acknowledgments 1048 This work was partially supported by the European Commission under 1049 Horizon 2020 grant agreement no. 688421 Measurement and Architecture 1050 for a Middleboxed Internet (MAMI), and by the Swiss State Secretariat 1051 for Education, Research, and Innovation under contract no. 15.0268. 1052 This support does not imply endorsement. 1054 20. References 1056 20.1. Normative References 1058 [QUIC] Iyengar, J., Ed. and M. Thomson, Ed., "QUIC: A UDP-Based 1059 Multiplexed and Secure Transport", RFC 9000, 1060 DOI 10.17487/RFC9000, May 2021, 1061 . 1063 [QUIC-INVARIANTS] 1064 Thomson, M., "Version-Independent Properties of QUIC", 1065 RFC 8999, DOI 10.17487/RFC8999, May 2021, 1066 . 1068 [QUIC-TLS] Thomson, M., Ed. and S. Turner, Ed., "Using TLS to Secure 1069 QUIC", RFC 9001, DOI 10.17487/RFC9001, May 2021, 1070 . 1072 20.2. Informative References 1074 [Edeline16] 1075 Edeline, K., Kuehlewind, M., Trammell, B., Aben, E., and 1076 B. Donnet, "Using UDP for Internet Transport Evolution 1077 (arXiv preprint 1612.07816)", 22 December 2016, 1078 . 1080 [Hatonen10] 1081 Hatonen, S., Nyrhinen, A., Eggert, L., Strowes, S., 1082 Sarolahti, P., and M. Kojo, "An experimental study of home 1083 gateway characteristics (Proc. ACM IMC 2010)", October 1084 2010. 1086 [HTTP-REPLAY] 1087 Thomson, M., Nottingham, M., and W. Tarreau, "Using Early 1088 Data in HTTP", RFC 8470, DOI 10.17487/RFC8470, September 1089 2018, . 1091 [I-D.draft-ietf-httpbis-priority] 1092 Oku, K. and L. Pardue, "Extensible Prioritization Scheme 1093 for HTTP", Work in Progress, Internet-Draft, draft-ietf- 1094 httpbis-priority-12, 17 January 2022, 1095 . 1098 [I-D.draft-ietf-quic-version-negotiation] 1099 Schinazi, D. and E. Rescorla, "Compatible Version 1100 Negotiation for QUIC", Work in Progress, Internet-Draft, 1101 draft-ietf-quic-version-negotiation-05, 25 October 2021, 1102 . 1105 [I-D.ietf-quic-datagram] 1106 Pauly, T., Kinnear, E., and D. Schinazi, "An Unreliable 1107 Datagram Extension to QUIC", Work in Progress, Internet- 1108 Draft, draft-ietf-quic-datagram-08, 14 January 2022, 1109 . 1112 [I-D.ietf-quic-manageability] 1113 Kuehlewind, M. and B. Trammell, "Manageability of the QUIC 1114 Transport Protocol", Work in Progress, Internet-Draft, 1115 draft-ietf-quic-manageability-13, 2 September 2021, 1116 . 1119 [I-D.ietf-taps-arch] 1120 Pauly, T., Trammell, B., Brunstrom, A., Fairhurst, G., and 1121 C. Perkins, "An Architecture for Transport Services", Work 1122 in Progress, Internet-Draft, draft-ietf-taps-arch-12, 3 1123 January 2022, . 1126 [PaaschNanog] 1127 Paasch, C., "Network Support for TCP Fast Open (NANOG 67 1128 presentation)", 13 June 2016, 1129 . 1132 [QUIC-HTTP] 1133 Bishop, M., "Hypertext Transfer Protocol Version 3 1134 (HTTP/3)", Work in Progress, Internet-Draft, draft-ietf- 1135 quic-http-34, 2 February 2021, 1136 . 1139 [QUIC-LB] Duke, M. and N. Banks, "QUIC-LB: Generating Routable QUIC 1140 Connection IDs", Work in Progress, Internet-Draft, draft- 1141 ietf-quic-load-balancers-10, 4 January 2022, 1142 . 1145 [RFC1034] Mockapetris, P., "Domain names - concepts and facilities", 1146 STD 13, RFC 1034, DOI 10.17487/RFC1034, November 1987, 1147 . 1149 [RFC2475] Blake, S., Black, D., Carlson, M., Davies, E., Wang, Z., 1150 and W. Weiss, "An Architecture for Differentiated 1151 Services", RFC 2475, DOI 10.17487/RFC2475, December 1998, 1152 . 1154 [RFC5077] Salowey, J., Zhou, H., Eronen, P., and H. Tschofenig, 1155 "Transport Layer Security (TLS) Session Resumption without 1156 Server-Side State", RFC 5077, DOI 10.17487/RFC5077, 1157 January 2008, . 1159 [RFC5382] Guha, S., Ed., Biswas, K., Ford, B., Sivakumar, S., and P. 1160 Srisuresh, "NAT Behavioral Requirements for TCP", BCP 142, 1161 RFC 5382, DOI 10.17487/RFC5382, October 2008, 1162 . 1164 [RFC5905] Mills, D., Martin, J., Ed., Burbank, J., and W. Kasch, 1165 "Network Time Protocol Version 4: Protocol and Algorithms 1166 Specification", RFC 5905, DOI 10.17487/RFC5905, June 2010, 1167 . 1169 [RFC6335] Cotton, M., Eggert, L., Touch, J., Westerlund, M., and S. 1170 Cheshire, "Internet Assigned Numbers Authority (IANA) 1171 Procedures for the Management of the Service Name and 1172 Transport Protocol Port Number Registry", BCP 165, 1173 RFC 6335, DOI 10.17487/RFC6335, August 2011, 1174 . 1176 [RFC6762] Cheshire, S. and M. Krochmal, "Multicast DNS", RFC 6762, 1177 DOI 10.17487/RFC6762, February 2013, 1178 . 1180 [RFC7301] Friedl, S., Popov, A., Langley, A., and E. Stephan, 1181 "Transport Layer Security (TLS) Application-Layer Protocol 1182 Negotiation Extension", RFC 7301, DOI 10.17487/RFC7301, 1183 July 2014, . 1185 [RFC7413] Cheng, Y., Chu, J., Radhakrishnan, S., and A. Jain, "TCP 1186 Fast Open", RFC 7413, DOI 10.17487/RFC7413, December 2014, 1187 . 1189 [RFC7657] Black, D., Ed. and P. Jones, "Differentiated Services 1190 (Diffserv) and Real-Time Communication", RFC 7657, 1191 DOI 10.17487/RFC7657, November 2015, 1192 . 1194 [RFC7838] Nottingham, M., McManus, P., and J. Reschke, "HTTP 1195 Alternative Services", RFC 7838, DOI 10.17487/RFC7838, 1196 April 2016, . 1198 [RFC8085] Eggert, L., Fairhurst, G., and G. Shepherd, "UDP Usage 1199 Guidelines", BCP 145, RFC 8085, DOI 10.17487/RFC8085, 1200 March 2017, . 1202 [SSDP] Donoho, A., Roe, B., Bodlaender, M., Gildred, J., Messer, 1203 A., Kim, Y., Fairman, B., and J. Tourzan, "UPnP Device 1204 Architecture 2.0", 17 April 2020, 1205 . 1208 [Swett16] Swett, I., "QUIC Deployment Experience at Google (IETF96 1209 QUIC BoF presentation)", 20 July 2016, 1210 . 1213 [TLS13] Rescorla, E., "The Transport Layer Security (TLS) Protocol 1214 Version 1.3", RFC 8446, DOI 10.17487/RFC8446, August 2018, 1215 . 1217 [Trammell16] 1218 Trammell, B. and M. Kuehlewind, "Internet Path 1219 Transparency Measurements using RIPE Atlas (RIPE72 MAT 1220 presentation)", 25 May 2016, . 1223 Authors' Addresses 1225 Mirja Kuehlewind 1226 Ericsson 1228 Email: mirja.kuehlewind@ericsson.com 1230 Brian Trammell 1231 Google 1232 Gustav-Gull-Platz 1 1233 CH- 8004 Zurich 1234 Switzerland 1236 Email: ietf@trammell.ch