idnits 2.17.1 draft-ietf-lwig-tcp-constrained-node-networks-06.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 document seems to lack an IANA Considerations section. (See Section 2.2 of https://www.ietf.org/id-info/checklist for how to handle the case when there are no actions for IANA.) ** There are 32 instances of too long lines in the document, the longest one being 90 characters in excess of 72. Miscellaneous warnings: ---------------------------------------------------------------------------- == The copyright year in the IETF Trust and authors Copyright Line does not match the current year == The document doesn't use any RFC 2119 keywords, yet has text resembling RFC 2119 boilerplate text. -- The document date (March 27, 2019) is 1854 days in the past. Is this intentional? Checking references for intended status: Informational ---------------------------------------------------------------------------- == Missing Reference: 'RFC 7228' is mentioned on line 779, but not defined ** Obsolete normative reference: RFC 793 (Obsoleted by RFC 9293) ** Obsolete normative reference: RFC 2460 (Obsoleted by RFC 8200) == Outdated reference: A later version (-17) exists of draft-ietf-tcpm-rto-consider-08 -- Obsolete informational reference (is this intentional?): RFC 7230 (Obsoleted by RFC 9110, RFC 9112) -- Obsolete informational reference (is this intentional?): RFC 7540 (Obsoleted by RFC 9113) Summary: 4 errors (**), 0 flaws (~~), 4 warnings (==), 3 comments (--). Run idnits with the --verbose option for more detailed information about the items above. -------------------------------------------------------------------------------- 2 LWIG Working Group C. Gomez 3 Internet-Draft UPC 4 Intended status: Informational J. Crowcroft 5 Expires: September 28, 2019 University of Cambridge 6 M. Scharf 7 Hochschule Esslingen 8 March 27, 2019 10 TCP Usage Guidance in the Internet of Things (IoT) 11 draft-ietf-lwig-tcp-constrained-node-networks-06 13 Abstract 15 This document provides guidance on how to implement and use the 16 Transmission Control Protocol (TCP) in Constrained-Node Networks 17 (CNNs), which are a characterstic of the Internet of Things (IoT). 18 Such environments require a lightweight TCP implementation and may 19 not make use of optional functionality. This document explains a 20 number of known and deployed techniques to simplify a TCP stack as 21 well as corresponding tradeoffs. The objective is to help embedded 22 developers with decisions on which TCP features to use. 24 Status of This Memo 26 This Internet-Draft is submitted in full conformance with the 27 provisions of BCP 78 and BCP 79. 29 Internet-Drafts are working documents of the Internet Engineering 30 Task Force (IETF). Note that other groups may also distribute 31 working documents as Internet-Drafts. The list of current Internet- 32 Drafts is at https://datatracker.ietf.org/drafts/current/. 34 Internet-Drafts are draft documents valid for a maximum of six months 35 and may be updated, replaced, or obsoleted by other documents at any 36 time. It is inappropriate to use Internet-Drafts as reference 37 material or to cite them other than as "work in progress." 39 This Internet-Draft will expire on September 28, 2019. 41 Copyright Notice 43 Copyright (c) 2019 IETF Trust and the persons identified as the 44 document authors. All rights reserved. 46 This document is subject to BCP 78 and the IETF Trust's Legal 47 Provisions Relating to IETF Documents 48 (https://trustee.ietf.org/license-info) in effect on the date of 49 publication of this document. Please review these documents 50 carefully, as they describe your rights and restrictions with respect 51 to this document. Code Components extracted from this document must 52 include Simplified BSD License text as described in Section 4.e of 53 the Trust Legal Provisions and are provided without warranty as 54 described in the Simplified BSD License. 56 Table of Contents 58 1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . 3 59 2. Conventions used in this document . . . . . . . . . . . . . . 4 60 3. Characteristics of CNNs relevant for TCP . . . . . . . . . . 4 61 3.1. Network and link properties . . . . . . . . . . . . . . . 4 62 3.2. Usage scenarios . . . . . . . . . . . . . . . . . . . . . 5 63 3.3. Communication and traffic patterns . . . . . . . . . . . 6 64 4. TCP implementation and configuration in CNNs . . . . . . . . 6 65 4.1. Path properties . . . . . . . . . . . . . . . . . . . . . 6 66 4.1.1. Maximum Segment Size (MSS) . . . . . . . . . . . . . 7 67 4.1.2. Explicit Congestion Notification (ECN) . . . . . . . 8 68 4.1.3. Explicit loss notifications . . . . . . . . . . . . . 9 69 4.2. TCP guidance for single-MSS windows and buffers . . . . . 9 70 4.2.1. Single-MSS stacks - benefits and issues . . . . . . . 9 71 4.2.2. TCP options for single-MSS stacks . . . . . . . . . . 9 72 4.2.3. Delayed Acknowledgments for single-MSS stacks . . . . 10 73 4.2.4. RTO estimation for single-MSS stacks . . . . . . . . 10 74 4.3. General recommendations for TCP in CNNs . . . . . . . . . 11 75 4.3.1. Loss recovery and congestion/flow control . . . . . . 11 76 4.3.1.1. Selective Acknowledgments (SACK) . . . . . . . . 11 77 4.3.2. Delayed Acknowledgments . . . . . . . . . . . . . . . 12 78 5. TCP usage recommendations in CNNs . . . . . . . . . . . . . . 12 79 5.1. TCP connection initiation . . . . . . . . . . . . . . . . 13 80 5.2. Number of concurrent connections . . . . . . . . . . . . 13 81 5.3. TCP connection lifetime . . . . . . . . . . . . . . . . . 13 82 6. Security Considerations . . . . . . . . . . . . . . . . . . . 15 83 7. Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . 15 84 8. Annex. TCP implementations for constrained devices . . . . . 16 85 8.1. uIP . . . . . . . . . . . . . . . . . . . . . . . . . . . 16 86 8.2. lwIP . . . . . . . . . . . . . . . . . . . . . . . . . . 17 87 8.3. RIOT . . . . . . . . . . . . . . . . . . . . . . . . . . 17 88 8.4. TinyOS . . . . . . . . . . . . . . . . . . . . . . . . . 17 89 8.5. FreeRTOS . . . . . . . . . . . . . . . . . . . . . . . . 18 90 8.6. uC/OS . . . . . . . . . . . . . . . . . . . . . . . . . . 18 91 8.7. Summary . . . . . . . . . . . . . . . . . . . . . . . . . 18 92 9. Annex. Changes compared to previous versions . . . . . . . . 20 93 9.1. Changes between -00 and -01 . . . . . . . . . . . . . . . 20 94 9.2. Changes between -01 and -02 . . . . . . . . . . . . . . . 20 95 9.3. Changes between -02 and -03 . . . . . . . . . . . . . . . 20 96 9.4. Changes between -03 and -04 . . . . . . . . . . . . . . . 21 97 9.5. Changes between -04 and -05 . . . . . . . . . . . . . . . 21 98 9.6. Changes between -05 and -06 . . . . . . . . . . . . . . . 21 99 10. References . . . . . . . . . . . . . . . . . . . . . . . . . 21 100 10.1. Normative References . . . . . . . . . . . . . . . . . . 21 101 10.2. Informative References . . . . . . . . . . . . . . . . . 23 102 Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . 26 104 1. Introduction 106 The Internet Protocol suite is being used for connecting Constrained- 107 Node Networks (CNNs) to the Internet, enabling the so-called Internet 108 of Things (IoT) [RFC7228]. In order to meet the requirements that 109 stem from CNNs, the IETF has produced a suite of new protocols 110 specifically designed for such environments (see e.g. [RFC8352]). 111 New IETF protocol stack components include the IPv6 over Low-power 112 Wireless Personal Area Networks (6LoWPAN) adaptation layer, the IPv6 113 Routing Protocol for Low-power and lossy networks (RPL) routing 114 protocol, and the Constrained Application Protocol (CoAP). 116 As of the writing, the main current transport layer protocols in IP- 117 based IoT scenarios are UDP and TCP. However, TCP has been 118 criticized (often, unfairly) as a protocol for the IoT. In fact, 119 some TCP features are not optimal for IoT scenarios, such as 120 relatively long header size, unsuitability for multicast, and always- 121 confirmed data delivery. However, many typical claims on TCP 122 unsuitability for IoT (e.g. a high complexity, connection-oriented 123 approach incompatibility with radio duty-cycling, and spurious 124 congestion control activation in wireless links) are not valid, can 125 be solved, or are also found in well accepted IoT end-to-end 126 reliability mechanisms (see [IntComp] for a detailed analysis). 128 At the application layer, CoAP was developed over UDP [RFC7252]. 129 However, the integration of some CoAP deployments with existing 130 infrastructure is being challenged by middleboxes such as firewalls, 131 which may limit and even block UDP-based communications. This the 132 main reason why a CoAP over TCP specification has been developed 133 [RFC8323]. 135 Other application layer protocols not specifically designed for CNNs 136 are also being considered for the IoT space. Some examples include 137 HTTP/2 and even HTTP/1.1, both of which run over TCP by default 138 [RFC7230] [RFC7540], and the Extensible Messaging and Presence 139 Protocol (XMPP) [RFC6120]. TCP is also used by non-IETF application- 140 layer protocols in the IoT space such as the Message Queue Telemetry 141 Transport (MQTT) and its lightweight variants. 143 TCP is a sophisticated transport protocol that includes optional 144 functionality (e.g. TCP options) that may improve performance in 145 some environments. However, many optional TCP extensions require 146 complex logic inside the TCP stack and increase the codesize and the 147 RAM requirements. Many TCP extensions are not required for 148 interoperability with other standard-compliant TCP endpoints. Given 149 the limited resources on constrained devices, careful "tuning" of the 150 TCP implementation can make an implementation more lightweight. 152 This document provides guidance on how to implement and use TCP in 153 CNNs. The overarching goal is to offer simple measures to allow for 154 lightweight TCP implementation and suitable operation in such 155 environments. A TCP implementation following the guidance in this 156 document is intended to be compatible with a TCP endpoint that is 157 compliant to the TCP standards, albeit possibly with a lower 158 performance. This implies that such a TCP client would always be 159 able to connect with a standard-compliant TCP server, and a 160 corresponding TCP server would always be able to connect with a 161 standard-compliant TCP client. 163 This document assumes that the reader is familiar with TCP. A 164 comprehensive survey of the TCP standards can be found in [RFC7414]. 165 Similar guidance regarding the use of TCP in special environments has 166 been published before, e.g., for cellular wireless networks 167 [RFC3481]. 169 2. Conventions used in this document 171 The key words "MUST", "MUST NOT", "REQUIRED", "SHALL","SHALL NOT", 172 "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this 173 document are to be interpreted as described in [RFC2119]. 175 3. Characteristics of CNNs relevant for TCP 177 3.1. Network and link properties 179 CNNs are defined in [RFC7228] as networks whose characteristics are 180 influenced by being composed of a significant portion of constrained 181 nodes. The latter are characterized by significant limitations on 182 processing, memory, and energy resources, among others [RFC7228]. 183 The first two dimensions pose constraints on the complexity and on 184 the memory footprint of the protocols that constrained nodes can 185 support. The latter requires techniques to save energy, such as 186 radio duty-cycling in wireless devices [RFC8352], as well as 187 minimization of the number of messages transmitted/received (and 188 their size). 190 [RFC7228] lists typical network constraints in CNN, including low 191 achievable bitrate/throughput, high packet loss and high variability 192 of packet loss, highly asymmetric link characteristics, severe 193 penalties for using larger packets, limits on reachability over time, 194 etc. CNN may use wireless or wired technologies (e.g., Power Line 195 Communication), and the transmission rates are typically low (e.g. 196 below 1 Mbps). 198 For use of TCP, one challenge is that not all technologies in CNN may 199 be aligned with typical Internet subnetwork design principles 200 [RFC3819]. For instance, constrained nodes often use physical/link 201 layer technologies that have been characterized as 'lossy', i.e., 202 exhibit a relatively high bit error rate. Dealing with corruption 203 loss is one of the open issues in the Internet [RFC6077]. 205 3.2. Usage scenarios 207 There are different deployment and usage scenarios for CNNs. Some 208 CNNs follow the star topology, whereby one or several hosts are 209 linked to a central device that acts as a router connecting the CNN 210 to the Internet. CNNs may also follow the multihop topology 211 [RFC6606]. One key use case for the use of TCP is a model where 212 constrained devices connect to unconstrained servers in the Internet. 213 But it is also possible that both TCP endpoints run on constrained 214 devices. 216 In constrained environments, there can be different types of devices 217 [RFC7228]. For example, there can be devices with single combined 218 send/receive buffer, devices with a separate send and receive buffer, 219 or devices with a pool of multiple send/receive buffers. In the 220 latter case, it is possible that buffers also be shared for other 221 protocols. 223 When a CNN comprising one or more constrained devices and an 224 unconstrained device communicate over the Internet using TCP, the 225 communication possibly has to traverse a middlebox (e.g. a firewall, 226 NAT, etc.). Figure 1 illustrates such scenario. Note that the 227 scenario is asymmetric, as the unconstrained device will typically 228 not suffer the severe constraints of the constrained device. The 229 unconstrained device is expected to be mains-powered, to have high 230 amount of memory and processing power, and to be connected to a 231 resource-rich network. 233 Assuming that a majority of constrained devices will correspond to 234 sensor nodes, the amount of data traffic sent by constrained devices 235 (e.g. sensor node measurements) is expected to be higher than the 236 amount of data traffic in the opposite direction. Nevertheless, 237 constrained devices may receive requests (to which they may respond), 238 commands (for configuration purposes and for constrained devices 239 including actuators) and relatively infrequent firmware/software 240 updates. 242 +---------------+ 243 o o <-------- TCP communication -----> | | 244 o o | | 245 o o | Unconstrained | 246 o o +-----------+ | device | 247 o o o ------ | Middlebox | ------- | | 248 o o +-----------+ | (e.g. cloud) | 249 o o o | | 250 +---------------+ 251 constrained devices 253 Figure 1: TCP communication between a constrained device and an 254 unconstrained device, traversing a middlebox. 256 3.3. Communication and traffic patterns 258 IoT applications are characterized by a number of different 259 communication patterns. The following non-comprehensive list 260 explains some typical examples: 262 o Unidirectional transfers: An IoT device (e.g. a sensor) can send 263 (repeatedly) updates to the other endpoint. Not in every case 264 there is a need for an application response back to the IoT 265 device. 267 o Request-response patterns: An IoT device receiving a request from 268 the other endpoint, which triggers a response from the IoT device. 270 o Bulk data transfers: A typical example for a long file transfer 271 would be an IoT device firmware update. 273 A typical communication pattern is that a constrained device 274 communicates with an unconstrained device (cf. Figure 1). But it is 275 also possible that constrained devices communicate amongst 276 themselves. 278 4. TCP implementation and configuration in CNNs 280 This section explains how a TCP stack can deal with typical 281 constraints in CNN. The guidance in this section relates to the TCP 282 implementation and its configuration. 284 4.1. Path properties 285 4.1.1. Maximum Segment Size (MSS) 287 For the sake of lightweight implementation and operation, unless 288 applications require handling large data units (i.e. leading to an 289 IPv6 datagram size greater than 1280 bytes), it may be desirable to 290 limit the MTU to 1280 bytes in order to avoid the need to support 291 Path MTU Discovery [RFC8201]. 293 An IPv6 datagram size exceeding 1280 bytes can be avoided by setting 294 the TCP MSS not larger than 1220 bytes. (Note: IP version 6 is 295 assumed.) This assumes that the remote sender will use no TCP 296 options, aside from possibly the MSS option, which is only used in 297 the initial TCP SYN packet. In order to accommodate unrequested TCP 298 options that may be used by some TCP implementations, a constrained 299 device may advertise an MSS not larger than 1200 bytes. 301 Note that setting the MTU to 1280 bytes is possible for link layer 302 technologies in the CNN space, even if some of them are characterized 303 by a short data unit payload size, e.g. up to a few tens or hundreds 304 of bytes. For example, the maximum frame size in IEEE 802.15.4 is 305 127 bytes. 6LoWPAN defined an adaptation layer to support IPv6 over 306 IEEE 802.15.4 networks. The adaptation layer includes a 307 fragmentation mechanism, since IPv6 requires the layer below to 308 support an MTU of 1280 bytes [RFC2460], while IEEE 802.15.4 lacked 309 fragmentation mechanisms. 6LoWPAN defines an IEEE 802.15.4 link MTU 310 of 1280 bytes [RFC4944]. Other technologies, such as Bluetooth LE 311 [RFC7668], ITU-T G.9959 [RFC7428] or DECT-ULE [RFC8105], also use 312 6LoWPAN-based adaptation layers in order to enable IPv6 support. 313 These technologies do support link layer fragmentation. By 314 exploiting this functionality, the adaptation layers that enable IPv6 315 over such technologies also define an MTU of 1280 bytes. 317 On the other hand, there exist technologies also used in the CNN 318 space, such as Master Slave / Token Passing (TP) [RFC8163], 319 Narrowband IoT (NB-IoT) [RFC8376] or IEEE 802.11ah 320 [I-D.delcarpio-6lo-wlanah], that do not suffer the same degree of 321 frame size limitations as the technologies mentioned above. The MTU 322 for MS/TP is recommended to be 1500 bytes [RFC8163], the MTU in NB- 323 IoT is 1600 bytes, and the maximum frame payload size for IEEE 324 802.11ah is 7991 bytes. 326 Finally, note that using larger MSS (to a suitable extent) may be 327 beneficial, especially when transferring large payloads, as it 328 reduces the number of packets (and packet headers) required for a 329 given payload. 331 4.1.2. Explicit Congestion Notification (ECN) 333 Explicit Congestion Notification (ECN) [RFC3168] ECN allows a router 334 to signal in the IP header of a packet that congestion is arising, 335 for example when a queue size reaches a certain threshold. An ECN- 336 enabled TCP receiver will echo back the congestion signal to the TCP 337 sender by setting a flag in its next TCP ACK. The sender triggers 338 congestion control measures as if a packet loss had happened. 340 The document [RFC8087] outlines the principal gains in terms of 341 increased throughput, reduced delay, and other benefits when ECN is 342 used over a network path that includes equipment that supports 343 Congestion Experienced (CE) marking. In the context of CNNs, a 344 remarkable feature of ECN is that congestion can be signalled without 345 incurring packet drops (which will lead to retransmissions and 346 consumption of limited resources such as energy and bandwitdh). 348 ECN can further reduce packet losses since congestion control 349 measures can be applied earlier [RFC2884]. Less lost packets implies 350 that the number of retransmitted segments decreases, which is 351 particularly beneficial in CNNs, where energy and bandwidth resources 352 are typically limited. Also, it makes sense to try to avoid packet 353 drops for transactional workloads with small data sizes, which are 354 typical for CNNs. In such traffic patterns, it is more difficult to 355 detect packet loss without retransmission timeouts (e.g., as there 356 may be no three duplicate ACKs). Any retransmission timeout slows 357 down the data transfer significantly. In addition, if the 358 constrained device uses power saving techniques, a retransmission 359 timeout will incur a wake-up action, in contrast to ACK clock- 360 triggered sending. When the congestion window of a TCP sender has a 361 size of one segment, the TCP sender resets the retransmit timer, and 362 the sender will only be able to send a new packet when the retransmit 363 timer expires [RFC3168]. Effectively, the TCP sender reduces at that 364 moment its sending rate from 1 segment per Round Trip Time (RTT) to 1 365 segment per RTO, which can result in a very low throughput. In 366 addition to better throughput, ECN can also help reducing latency and 367 jitter. 369 ECN can be incrementally deployed in the Internet. Guidance on 370 configuration and usage of ECN is provided in [RFC7567]. Given the 371 benefits, more and more TCP stacks in the Internet support ECN, and 372 it specifically makes sense to leverage ECN in controlled 373 environments such as CNNs. Note, however, that supporting ECN 374 increases implementation complexity. 376 4.1.3. Explicit loss notifications 378 There has been a significant body of research on solutions capable of 379 explicitly indicating whether a TCP segment loss is due to 380 corruption, in order to avoid activation of congestion control 381 mechanisms [ETEN] [RFC2757]. While such solutions may provide 382 significant improvement, they have not been widely deployed and 383 remain as experimental work. In fact, as of today, the IETF has not 384 standardized any such solution. 386 4.2. TCP guidance for single-MSS windows and buffers 388 This section discusses TCP stacks that focus on transferring a single 389 MSS. More general guidance is provided in Section 4.3. 391 4.2.1. Single-MSS stacks - benefits and issues 393 A TCP stack can reduce the RAM requirements by advertising a TCP 394 window size of one MSS, and also transmit at most one MSS of 395 unacknowledged data. In that case, both congestion and flow control 396 implementation is quite simple. Such a small receive and send window 397 may be sufficient for simple message exchanges in the CNN space. 398 However, only using a window of one MSS can significantly affect 399 performance. A stop-and-wait operation results in low throughput for 400 transfers that exceed the lengths of one MSS, e.g., a firmware 401 download. 403 If CoAP is used over TCP with the default setting for NSTART in 404 [RFC7252], a CoAP endpoint is not allowed to send a new message to a 405 destination until a response for the previous message sent to that 406 destination has been received. This is equivalent to an application- 407 layer window size of 1. For this use of CoAP, a maximum TCP window 408 of one MSS will be sufficient. 410 4.2.2. TCP options for single-MSS stacks 412 A TCP implementation needs to support, at a minimum, TCP options 2, 1 413 and 0. These are, respectively, the Maximum Segment Size (MSS) 414 option, the No-Operation option, and the End Of Option List marker 415 [RFC0793]. None of these are a substantial burden to support. These 416 options are sufficient for interoperability with a standard-compliant 417 TCP endpoint, albeit many TCP stacks support additional options and 418 can negotiate their use. A TCP implementation is permitted to 419 silently ignore all other TCP options. 421 A TCP implementation for a constrained device that uses a single-MSS 422 TCP receive or transmit window size may not benefit from supporting 423 the following TCP options: Window scale [RFC7323], TCP Timestamps 425 [RFC7323], Selective Acknowledgments (SACK) and SACK-Permitted 426 [RFC2018]. Also other TCP options may not be required on a 427 constrained device with a very lightweight implementation. With 428 regard to the Window scale option, note that it is only useful if a 429 window size greater than 64 kB is needed. 431 One potentially relevant TCP option in the context of CNNs is TCP 432 Fast Open (TFO) [RFC7413]. As described in Section 5.3, TFO can be 433 used to address the problem of traversing middleboxes that perform 434 early filter state record deletion. 436 4.2.3. Delayed Acknowledgments for single-MSS stacks 438 TCP Delayed Acknowledgments are meant to reduce the number of ACKs 439 sent within a TCP connection, thus reducing network overhead, but 440 they may increase the time until a sender may receive an ACK. In 441 general, usefulness of Delayed ACKs depends heavily on the usage 442 scenario. There can be interactions with stacks that use single-MSS 443 windows. 445 A device that advertises a single-MSS receive window should avoid use 446 of Delayed ACKs in order to avoid contributing unnecessary delay (of 447 up to 500 ms) to the RTT [RFC5681], which limits the throughput and 448 can increase the data delivery time. 450 A device that can send at most one MSS of data is significantly 451 affected if the receiver uses Delayed ACKs, e.g., if a TCP server or 452 receiver is outside the CNN. One known workaround is to split the 453 data to be sent into two segments of smaller size. A standard 454 compliant TCP receiver will then immediately acknowledge the second 455 segment, which can improve throughput. This "split hack" works if 456 the TCP receiver uses Delayed ACKs, but the downside is the overhead 457 of sending two IP packets instead of one. 459 Similar issues happen when the sender uses the Nagle algorithm. 460 Disabling the algorithm will not have impact if the sender can only 461 handle stop-and-wait operation. 463 4.2.4. RTO estimation for single-MSS stacks 465 The Retransmission Timeout (RTO) estimation is one of the fundamental 466 TCP algorithms. There is a fundamental trade-off: A short, 467 aggressive RTO behavior reduces wait time before retransmissions, but 468 it also increases the probability of spurious timeouts. The latter 469 lead to unnecessary waste of potentially scarce resources in CNNs 470 such as energy and bandwidth. In contrast, a conservative timeout 471 can result in long error recovery times and thus needlessly delay 472 data delivery. 474 [RFC6298] describes the standard TCP RTO algorithm. If a TCP sender 475 uses very small window size, and it cannot use Fast Retransmit/Fast 476 Recovery or SACK, the Retransmission Timeout (RTO) algorithm has a 477 larger impact on performance than for a more powerful TCP stack. In 478 that case, RTO algorithm tuning may be considered, although careful 479 assessment of possible drawbacks is recommended 480 [I-D.ietf-tcpm-rto-consider]. 482 As an example, an adaptive RTO algorithm for CoAP over UDP has been 483 defined [I-D.ietf-core-cocoa] that has been found to perform well in 484 CNN scenarios [Commag]. 486 4.3. General recommendations for TCP in CNNs 488 This section summarizes some widely used techniques to improve TCP, 489 with a focus on their use in CNNs. The TCP extensions discussed here 490 are useful in a wide range of network scenarios, including CNNs. 491 This section is not comprehensive. A comprehensive survey of TCP 492 extensions is published in [RFC7414]. 494 4.3.1. Loss recovery and congestion/flow control 496 Devices that have enough memory to allow larger TCP window size can 497 leverage a more efficient loss recovery using Fast Retransmit and 498 Fast Recovery [RFC5681], at the expense of slightly greater 499 complexity and TCB size. Assuming that Delayed ACKs are used by the 500 receiver, the mentioned algorithms work efficiently for window sizes 501 of at least 5 MSS: If in a given TCP transmission of segments 502 1,2,3,4,5, and 6 the segment 2 gets lost, the sender should get an 503 ACK for segment 1 when 3 arrives and duplicate acknowledgements when 504 4, 5, and 6 arrive. It will retransmit segment 2 when the third 505 duplicate ACK arrives. In order to have segment 2, 3, 4, 5, and 6 506 sent, the window has to be at least five. With an MSS of 1220 byte, 507 a buffer of the size of 5 MSS would require 6100 bytes. 509 For bulk data transfers further TCP improvements may also be useful, 510 such as limited transmit [RFC3042]. 512 4.3.1.1. Selective Acknowledgments (SACK) 514 If a device with less severe memory and processing constraints can 515 afford advertising a TCP window size of several MSS, it makes sense 516 to support the SACK option to improve performance. SACK allows a 517 data receiver to inform the data sender of non-contiguous data blocks 518 received, thus a sender (having previously sent the SACK-Permitted 519 option) can avoid performing unnecessary retransmissions, saving 520 energy and bandwidth, as well as reducing latency. SACK is 521 particularly useful for bulk data transfers. The receiver supporting 522 SACK will need to manage the reception of possible out-of-order 523 received segments, requiring sufficient buffer space. SACK adds 524 8*n+2 bytes to the TCP header, where n denotes the number of data 525 blocks received, up to 4 blocks. For a low number of out-of-order 526 segments, the header overhead penalty of SACK is compensated by 527 avoiding unnecessary retransmissions. 529 4.3.2. Delayed Acknowledgments 531 For certain traffic patterns, Delayed ACKs may have a detrimental 532 effect, as already noted in Section 4.2.3. Advanced TCP stacks may 533 use heuristics to determine the maximum delay for an ACK. For CNNs, 534 the recommendation depends on the expected communication patterns. 536 When traffic over a CNN is expected to mostly be unidirectional 537 messages with a size typically up to one MSS, and the time between 538 two consecutive message transmissions is greater than the delayed ACK 539 timeout, it may make sense to use a small timeout or disable delayed 540 ACKs at the receiver. This avoids incurring additional delay, as 541 well as the energy consumption of the sender (which might e.g. keep 542 its radio interface in receive mode) during that time. Note that 543 disabling delayed ACKs may only be possible if the peer device is 544 administered by the same entity managing the constrained device. For 545 request-response traffic, enabling delayed ACKs is recommended, in 546 order to allow combining a response with the ACK into a single 547 segment, thus increasing efficiency. 549 In contrast, Delayed ACKs allow to reduce the number of ACKs in bulk 550 transfer type of traffic, e.g. for firmware/software updates or for 551 transferring larger data units containing a batch of sensor readings. 553 Note that, in many scenarios, the peer that a constrained device 554 communicates with will be a general purpose system that communicates 555 with both constrained and unconstrained devices. Since delayed ACKs 556 are often configured through system-wide parameters, delayed ACKs 557 behavior at the peer will be the same regardless of the nature of the 558 endpoints it talks to. Such a peer will typically have delayed ACKs 559 enabled. 561 5. TCP usage recommendations in CNNs 563 This section discusses how a TCP stack can be used by applications 564 that are developed for CNN scenarios. These remarks are by and large 565 independent of how TCP is exactly implemented. 567 5.1. TCP connection initiation 569 In the constrained device to unconstrained device scenario 570 illustrated above, a TCP connection is typically initiated by the 571 constrained device, in order for this device to support possible 572 sleep periods to save energy. 574 5.2. Number of concurrent connections 576 TCP endpoints with a small amount of RAM may only support a small 577 number of connections. Each TCP connection requires storing a number 578 of variables in the Transmission Control Block (TCB). Depending on 579 the internal TCP implementation, each connection may result in 580 further memory overhead, and connections may compete for scarce 581 resources (e.g. further memory overhead for send and receive buffers, 582 etc). 584 A careful application design may try to keep the number of concurrent 585 connections as small as possible. A client can for instance limit 586 the number of simultaneous open connections that it maintains to a 587 given server. Multiple connections could for instance be used to 588 avoid the "head-of-line blocking" problem in an application transfer. 589 However, in addition to comsuming resources, using multiple 590 connections can also cause undesirable side effects in congested 591 networks. For example, the HTTP/1.1 specification encourages clients 592 to be conservative when opening multiple connections [RFC7230]. 593 Furthermore, each new connection will start with a 3-way handshake, 594 therefore increasing message overhead. 596 Being conservative when opening multiple TCP connections is of 597 particular importance in Constrained-Node Networks. 599 5.3. TCP connection lifetime 601 In order to minimize message overhead, it makes sense to keep a TCP 602 connection open as long as the two TCP endpoints have more data to 603 send. If applications exchange data rather infrequently, i.e., if 604 TCP connections would stay idle for a long time, the idle time can 605 result in problems. For instance, certain middleboxes such as 606 firewalls or NAT devices are known to delete state records after an 607 inactivity interval typically in the order of a few minutes 608 [RFC6092]. The timeout duration used by a middlebox implementation 609 may not be known to the TCP endpoints. 611 In CNNs, such middleboxes may e.g. be present at the boundary between 612 the CNN and other networks. If the middlebox can be optimized for 613 CNN use cases, it makes sense to increase the initial value for 614 filter state inactivity timers to avoid problems with idle 615 connections. Apart from that, this problem can be dealt with by 616 different connection handling strategies, each having pros and cons. 618 One approach for infrequent data transfer is to use short-lived TCP 619 connections. Instead of trying to maintain a TCP connection for long 620 time, possibly short-lived connections can be opened between two 621 endpoints, which are closed if no more data needs to be exchanged. 622 For use cases that can cope with the additional messages and the 623 latency resulting from starting new connections, it is recommended to 624 use a sequence of short-lived connections, instead of maintaining a 625 single long-lived connection. 627 The message and latency overhead that stems from using a sequence of 628 short-lived connections could be reduced by TCP Fast Open (TFO) 629 [RFC7413], which is an experimental TCP extension, at the expense of 630 increased implementation complexity and increased TCP Control Block 631 (TCB) size. TFO allows data to be carried in SYN (and SYN-ACK) 632 segments, and to be consumed immediately by the receiving endpoint. 633 This reduces the message and latency overhead compared to the 634 traditional three-way handshake to establish a TCP connection. For 635 security reasons, the connection initiator has to request a TFO 636 cookie from the other endpoint. The cookie, with a size of 4 or 16 637 bytes, is then included in SYN packets of subsequent connections. 638 The cookie needs to be refreshed (and obtained by the client) after a 639 certain amount of time. Nevertheless, TFO is more efficient than 640 frequently opening new TCP connections with the traditional three-way 641 handshake, as long as the cookie can be reused in subsequent 642 connections. However, as stated in RFC 7413, TFO deviates from the 643 standard TCP semantics, since the data in the SYN could be replayed 644 to an application in some rare circumstances. Applications should 645 not use TFO unless they can tolerate this issue, e.g., by using 646 Transport Layer Security (TLS) [RFC7413]. A comprehensive discussion 647 on TFO can be found at RFC 7413. 649 Another approach is to use long-lived TCP connections with 650 application-layer heartbeat messages. Various application protocols 651 support such heartbeat messages. Periodic heartbeats requires 652 transmission of packets, but they also allow aliveness checks at 653 application level. In addition, they can prevent early filter state 654 record deletion in middleboxes. In general, it makes sense realize 655 aliveness checks at the highest protocol layer possible that is 656 meaningful to the application, in order to maximize the depth of the 657 aliveness check. In addition, timely detection of a dead peer may 658 allow savings in terms of TCB memory use. 660 A TCP implementation may also be able to send "keep-alive" segments 661 to test a TCP connection. According to [RFC1122], "keep-alives" are 662 an optional TCP mechanism that is turned off by default, i.e., an 663 application must explicitly enable it for a TCP connection. The 664 interval between "keep-alive" messages must be configurable and it 665 must default to no less than two hours. With this large timeout, TCP 666 keep-alive messages are not very useful to avoid deletion of filter 667 state records in middleboxes such as firewalls. However, sending TCP 668 keep-alive probes more frequently risks draining power on energy- 669 constrained devices. 671 6. Security Considerations 673 Best current practise for securing TCP and TCP-based communication 674 also applies to CNN. As example, use of Transport Layer Security 675 (TLS) is strongly recommended if it is applicable. 677 There are also TCP options which can improve TCP security. One 678 example is the TCP Authentication Option (TCP-AO) [RFC5925]. 679 However, this option adds overhead and complexity. TCP-AO typically 680 has a size of 16-20 bytes. 682 For the mechanisms discussed in this document, the corresponding 683 considerations apply. For instance, if TFO is used, the security 684 considerations of [RFC7413] apply. 686 Constrained devices are expected to support smaller TCP window sizes 687 than less limited devices. In such conditions, segment 688 retransmission triggered by RTO expiration is expected to be 689 relatively frequent, due to lack of (enough) duplicate ACKs, 690 especially when a constrained device uses a single-MSS window size. 691 For this reason, constrained devices running TCP may appear as 692 particularly appealing victims of the so-called "shrew" Denial of 693 Service (DoS) attack [shrew], whereby one or more sources generate a 694 packet spike targetted to coincide with consecutive RTO-expiration- 695 triggered retry attempts of a victim node. Note that the attack may 696 be performed by Internet-connected devices, including constrained 697 devices in the same CNN as the victim, as well as remote ones. 698 Mitigation techniques include RTO randomization and attack blocking 699 by routers able to detect shrew attacks based on their traffic 700 pattern. 702 7. Acknowledgments 704 Carles Gomez has been funded in part by the Spanish Government 705 (Ministerio de Educacion, Cultura y Deporte) through the Jose 706 Castillejo grants CAS15/00336 and and CAS18/00170, and by European 707 Regional Development Fund (ERDF) and the Spanish Government through 708 project TEC2016-79988-P, AEI/FEDER, UE. Part of his contribution to 709 this work has been carried out during his stays as a visiting scholar 710 at the Computer Laboratory of the University of Cambridge. 712 The authors appreciate the feedback received for this document. The 713 following folks provided comments that helped improve the document: 714 Carsten Bormann, Zhen Cao, Wei Genyu, Ari Keranen, Abhijan 715 Bhattacharyya, Andres Arcia-Moret, Yoshifumi Nishida, Joe Touch, Fred 716 Baker, Nik Sultana, Kerry Lynn, Erik Nordmark, Markku Kojo, Hannes 717 Tschofenig, David Black, Yoshifumi Nishida, Ilpo Jarvinen, Emmanuel 718 Baccelli, and Stuart Cheshire. Simon Brummer provided details, and 719 kindly performed RAM and ROM usage measurements, on the RIOT TCP 720 implementation. Xavi Vilajosana provided details on the OpenWSN TCP 721 implementation. Rahul Jadhav kindly performed code size measurements 722 on the Contiki-NG and lwIP 2.1.2 TCP implementations. He also 723 provided details on the uIP TCP implementation. 725 8. Annex. TCP implementations for constrained devices 727 This section overviews the main features of TCP implementations for 728 constrained devices. The survey is limited to open source stacks 729 with small footprint. It is not meant to be all-encompassing. For 730 more powerful embedded systems (e.g., with 32-bit processors), there 731 are further stacks that comprehensively implement TCP. On the other 732 hand, please be aware that this Annex is based on information 733 available as of the writing. 735 8.1. uIP 737 uIP is a TCP/IP stack, targetted for 8 and 16-bit microcontrollers, 738 which pioneered TCP/IP implementations for constrained devices. uIP 739 has been deployed with Contiki and the Arduino Ethernet shield. A 740 code size of ~5 kB (which comprises checksumming, IP, ICMP and TCP) 741 has been reported for uIP [Dunk]. 743 uIP uses the same global buffer for both incoming and outgoing 744 traffic, which has a size of a single packet. In case of a 745 retransmission, an application must be able to reproduce the same 746 user data that had been transmitted. Multiple connections are 747 supported, but need to share the global buffer. 749 The MSS is announced via the MSS option on connection establishment 750 and the receive window size (of one MSS) is not modified during a 751 connection. Stop-and-wait operation is used for sending data. Among 752 other optimizations, this allows to avoid sliding window operations, 753 which use 32-bit arithmetic extensively and are expensive on 8-bit 754 CPUs. 756 Contiki uses the "split hack" technique (see Section 4.2.3) to avoid 757 Delayed ACKs for senders using a single segment. 759 The code size of the TCP implementation in Contiki-NG has been 760 measured to be of 3.2 kB on CC2538DK, cross-compiling on Linux. 762 8.2. lwIP 764 lwIP is a TCP/IP stack, targetted for 8- and 16-bit microcontrollers. 765 lwIP has a total code size of ~14 kB to ~22 kB (which comprises 766 memory management, checksumming, network interfaces, IP, ICMP and 767 TCP), and a TCP code size of ~9 kB to ~14 kB [Dunk]. 769 In contrast with uIP, lwIP decouples applications from the network 770 stack. lwIP supports a TCP transmission window greater than a single 771 segment, as well as buffering of incoming and outcoming data. Other 772 implemented mechanisms comprise slow start, congestion avoidance, 773 fast retransmit and fast recovery. SACK and Window Scale support has 774 been recently added to lwIP. 776 8.3. RIOT 778 The RIOT TCP implementation (called GNRC TCP) has been designed for 779 Class 1 devices [RFC 7228]. The main target platforms are 8- and 780 16-bit microcontrollers, with 32-bit platforms also supported. GNRC 781 TCP offers a similar function set as uIP, but it provides and 782 maintains an independent receive buffer for each connection. In 783 contrast to uIP, retransmission is also handled by GNRC TCP. GNRC 784 TCP uses a single-MSS window size, which simplifies the 785 implementation. The application programmer does not need to know 786 anything about the TCP internals, therefore GNRC TCP can be seen as a 787 user-friendly uIP TCP implementation. 789 The MSS is set on connections establishment and cannot be changed 790 during connection lifetime. GNRC TCP allows multiple connections in 791 parallel, but each TCB must be allocated somewhere in the system. By 792 default there is only enough memory allocated for a single TCP 793 connection, but it can be increased at compile time if the user needs 794 multiple parallel connections. 796 The RIOT TCP implementation offers an optional POSIX socket wrapper 797 that enables POSIX compliance, if needed. 799 Further details on RIOT and GNRC can be found in the literature 800 [RIOT], [GNRC]. 802 8.4. TinyOS 804 TinyOS was important as platform for early constrained devices. 805 TinyOS has an experimental TCP stack that uses a simple nonblocking 806 library-based implementation of TCP, which provides a subset of the 807 socket interface primitives. The application is responsible for 808 buffering. The TCP library does not do any receive-side buffering. 809 Instead, it will immediately dispatch new, in-order data to the 810 application and otherwise drop the segment. A send buffer is 811 provided by the application. Multiple TCP connections are possible. 812 Recently there has been little further work on the stack. 814 8.5. FreeRTOS 816 FreeRTOS is a real-time operating system kernel for embedded devices 817 that is supported by 16- and 32-bit microprocessors. Its TCP 818 implementation is based on multiple-segment window size, although a 819 'Tiny-TCP' option, which is a single-MSS variant, can be enabled. 820 Delayed ACKs are supported, with a 20-ms Delayed ACK timer as a 821 technique intended 'to gain performance'. 823 8.6. uC/OS 825 uC/OS is a real-time operating system kernel for embedded devices, 826 which is maintained by Micrium. uC/OS is intended for 8-, 16- and 827 32-bit microprocessors. The uC/OS TCP implementation supports a 828 multiple-segment window size. 830 8.7. Summary 831 +---+---------+--------+----+------+--------+-----+ 832 |uIP|lwIP orig|lwIP 2.1|RIOT|TinyOS|FreeRTOS|uC/OS| 833 +------+-------------+---+---------+--------+----+------+--------+-----+ 834 |Memory|Code size(kB)| <5|~9 to ~14| 38 | <7 | N/A | <9.2 | N/A | 835 | | |(a)| (T1) | (T4) |(T3)| | (T2) | | 836 +------+-------------+---+---------+--------+----+------+--------+-----+ 837 | | Single-Segm.|Yes| No | No | Yes| No | No | No | 838 | +-------------+---+---------+--------+----+------+--------+-----+ 839 | | Slow start | No| Yes | Yes | No | Yes | No | Yes | 840 | T +-------------+---+---------+--------+----+------+--------+-----+ 841 | C |Fast rec/retx| No| Yes | Yes | No | Yes | No | Yes | 842 | P +-------------+---+---------+--------+----+------+--------+-----+ 843 | | Keep-alive | No| No | Yes | No | No | Yes | Yes | 844 | +-------------+---+---------+--------+----+------+--------+-----+ 845 | f | Win. Scale | No| No | Yes | No | No | Yes | No | 846 | e +-------------+---+---------+--------+----+------+--------+-----+ 847 | a | TCP timest.| No| No | Yes | No | No | Yes | No | 848 | t +-------------+---+---------+--------+----+------+--------+-----+ 849 | u | SACK | No| No | Yes | No | No | Yes | No | 850 | r +-------------+---+---------+--------+----+------+--------+-----+ 851 | e | Del. ACKs | No| Yes | Yes | No | No | Yes | Yes | 852 | s +-------------+---+---------+--------+----+------+--------+-----+ 853 | | Socket | No| No |Optional|(I) |Subset| Yes | Yes | 854 | +-------------+---+---------+--------+----+------+--------+-----+ 855 | |Concur. Conn.|Yes| Yes | Yes | Yes| Yes | Yes | Yes | 856 +------+-------------+---+---------+--------+----+------+--------+-----+ 857 | TLS supported | No| No | Yes | Yes| Yes | Yes | Yes | 858 +--------------------+---+---------+--------+----+------+--------+-----+ 860 (T1) = TCP-only, on x86 and AVR platforms 861 (T2) = TCP-only, on ARM Cortex-M platform 862 (T3) = TCP-only, on ARM Cortex-M0+ platform (NOTE: RAM usage for the same platform 863 is ~2.5 kB for one TCP connection plus ~1.2 kB for each additional connection) 864 (T4) = TCP-only, on CC2538DK, cross-compiling on Linux 865 (a) = includes IP, ICMP and TCP on x86 and AVR platforms. The Contiki-NG TCP implementation has a code size of 3.2 kB on CC2538DK, cross-compiling on Linux 866 (I) = optional POSIX socket wrapper which enables POSIX compliance if needed 867 Mult. = Multiple 868 N/A = Not Available 870 Figure 2: Summary of TCP features for differrent lightweight TCP 871 implementations. None of the implementations considered in this 872 Annex support ECN or TFO. 874 9. Annex. Changes compared to previous versions 876 RFC Editor: To be removed prior to publication 878 9.1. Changes between -00 and -01 880 o Changed title and abstract 882 o Clarification that communcation with standard-compliant TCP 883 endpoints is required, based on feedback from Joe Touch 885 o Additional discussion on communication patters 887 o Numerous changes to address a comprehensive review from Hannes 888 Tschofenig 890 o Reworded security considerations 892 o Additional references and better distinction between normative and 893 informative entries 895 o Feedback from Rahul Jadhav on the uIP TCP implementation 897 o Basic data for the TinyOS TCP implementation added, based on 898 source code analysis 900 9.2. Changes between -01 and -02 902 o Added text to the Introduction section, and a reference, on 903 traditional bad perception of TCP for IoT 905 o Added sections on FreeRTOS and uC/OS 907 o Updated TinyOS section 909 o Updated summary table 911 o Reorganized Section 4 (single-MSS vs multiple-MSS window size), 912 some content now also in new Section 5 914 9.3. Changes between -02 and -03 916 o Rewording to better explain the benefit of ECN 918 o Additional context information on the surveyed implementations 920 o Added details, but removed "Data size" raw, in the summary table 921 o Added discussion on shrew attacks 923 9.4. Changes between -03 and -04 925 o Addressing the remaining TODOs 927 o Alignment of the wording on TCP "keep-alives" with related 928 discussions in the IETF transport area 930 o Added further discussion on delayed ACKs 932 o Removed OpenWSN subsection from the Annex 934 9.5. Changes between -04 and -05 936 o Addressing comments by Yoshifumi Nishida 938 o Removed mentioning MD5 as an example (comment by David Black) 940 o Added memory footprint details of TCP implementations (Contiki-NG 941 and lwIP 2.1.2) provided by Rahul Jadhav in the Annex 943 o Addressed comments by Ilpo Jarvinen throughout the whole document 945 o Improved the RIOT section in the Annex, based on feedback from 946 Emmanuel Baccelli 948 9.6. Changes between -05 and -06 950 o Incorporated suggestions by Stuart Cheshire 952 10. References 954 10.1. Normative References 956 [RFC0793] Postel, J., "Transmission Control Protocol", STD 7, 957 RFC 793, DOI 10.17487/RFC0793, September 1981, 958 . 960 [RFC1122] Braden, R., Ed., "Requirements for Internet Hosts - 961 Communication Layers", STD 3, RFC 1122, 962 DOI 10.17487/RFC1122, October 1989, 963 . 965 [RFC2018] Mathis, M., Mahdavi, J., Floyd, S., and A. Romanow, "TCP 966 Selective Acknowledgment Options", RFC 2018, 967 DOI 10.17487/RFC2018, October 1996, 968 . 970 [RFC2119] Bradner, S., "Key words for use in RFCs to Indicate 971 Requirement Levels", BCP 14, RFC 2119, 972 DOI 10.17487/RFC2119, March 1997, 973 . 975 [RFC2460] Deering, S. and R. Hinden, "Internet Protocol, Version 6 976 (IPv6) Specification", RFC 2460, DOI 10.17487/RFC2460, 977 December 1998, . 979 [RFC3042] Allman, M., Balakrishnan, H., and S. Floyd, "Enhancing 980 TCP's Loss Recovery Using Limited Transmit", RFC 3042, 981 DOI 10.17487/RFC3042, January 2001, 982 . 984 [RFC3168] Ramakrishnan, K., Floyd, S., and D. Black, "The Addition 985 of Explicit Congestion Notification (ECN) to IP", 986 RFC 3168, DOI 10.17487/RFC3168, September 2001, 987 . 989 [RFC3819] Karn, P., Ed., Bormann, C., Fairhurst, G., Grossman, D., 990 Ludwig, R., Mahdavi, J., Montenegro, G., Touch, J., and L. 991 Wood, "Advice for Internet Subnetwork Designers", BCP 89, 992 RFC 3819, DOI 10.17487/RFC3819, July 2004, 993 . 995 [RFC5681] Allman, M., Paxson, V., and E. Blanton, "TCP Congestion 996 Control", RFC 5681, DOI 10.17487/RFC5681, September 2009, 997 . 999 [RFC5925] Touch, J., Mankin, A., and R. Bonica, "The TCP 1000 Authentication Option", RFC 5925, DOI 10.17487/RFC5925, 1001 June 2010, . 1003 [RFC6298] Paxson, V., Allman, M., Chu, J., and M. Sargent, 1004 "Computing TCP's Retransmission Timer", RFC 6298, 1005 DOI 10.17487/RFC6298, June 2011, 1006 . 1008 [RFC7228] Bormann, C., Ersue, M., and A. Keranen, "Terminology for 1009 Constrained-Node Networks", RFC 7228, 1010 DOI 10.17487/RFC7228, May 2014, 1011 . 1013 [RFC7323] Borman, D., Braden, B., Jacobson, V., and R. 1014 Scheffenegger, Ed., "TCP Extensions for High Performance", 1015 RFC 7323, DOI 10.17487/RFC7323, September 2014, 1016 . 1018 [RFC7413] Cheng, Y., Chu, J., Radhakrishnan, S., and A. Jain, "TCP 1019 Fast Open", RFC 7413, DOI 10.17487/RFC7413, December 2014, 1020 . 1022 10.2. Informative References 1024 [Commag] A. Betzler, C. Gomez, I. Demirkol, J. Paradells, "CoAP 1025 Congestion Control for the Internet of Things", IEEE 1026 Communications Magazine, June 2016. 1028 [Dunk] A. Dunkels, "Full TCP/IP for 8-Bit Architectures", 2003. 1030 [ETEN] R. Krishnan et al, "Explicit transport error notification 1031 (ETEN) for error-prone wireless and satellite networks", 1032 Computer Networks 2004. 1034 [GNRC] M. Lenders et al., "Connecting the World of Embedded 1035 Mobiles: The RIOTApproach to Ubiquitous Networking for the 1036 IoT", 2018. 1038 [I-D.delcarpio-6lo-wlanah] 1039 Vega, L., Robles, I., and R. Morabito, "IPv6 over 1040 802.11ah", draft-delcarpio-6lo-wlanah-01 (work in 1041 progress), October 2015. 1043 [I-D.ietf-core-cocoa] 1044 Bormann, C., Betzler, A., Gomez, C., and I. Demirkol, 1045 "CoAP Simple Congestion Control/Advanced", draft-ietf- 1046 core-cocoa-03 (work in progress), February 2018. 1048 [I-D.ietf-tcpm-rto-consider] 1049 Allman, M., "Retransmission Timeout Requirements", draft- 1050 ietf-tcpm-rto-consider-08 (work in progress), February 1051 2019. 1053 [IntComp] C. Gomez, A. Arcia-Moret, J. Crowcroft, "TCP in the 1054 Internet of Things: from ostracism to prominence", IEEE 1055 Internet Computing, January-February 2018. 1057 [RFC2757] Montenegro, G., Dawkins, S., Kojo, M., Magret, V., and N. 1058 Vaidya, "Long Thin Networks", RFC 2757, 1059 DOI 10.17487/RFC2757, January 2000, 1060 . 1062 [RFC2884] Hadi Salim, J. and U. Ahmed, "Performance Evaluation of 1063 Explicit Congestion Notification (ECN) in IP Networks", 1064 RFC 2884, DOI 10.17487/RFC2884, July 2000, 1065 . 1067 [RFC3481] Inamura, H., Ed., Montenegro, G., Ed., Ludwig, R., Gurtov, 1068 A., and F. Khafizov, "TCP over Second (2.5G) and Third 1069 (3G) Generation Wireless Networks", BCP 71, RFC 3481, 1070 DOI 10.17487/RFC3481, February 2003, 1071 . 1073 [RFC4944] Montenegro, G., Kushalnagar, N., Hui, J., and D. Culler, 1074 "Transmission of IPv6 Packets over IEEE 802.15.4 1075 Networks", RFC 4944, DOI 10.17487/RFC4944, September 2007, 1076 . 1078 [RFC6077] Papadimitriou, D., Ed., Welzl, M., Scharf, M., and B. 1079 Briscoe, "Open Research Issues in Internet Congestion 1080 Control", RFC 6077, DOI 10.17487/RFC6077, February 2011, 1081 . 1083 [RFC6092] Woodyatt, J., Ed., "Recommended Simple Security 1084 Capabilities in Customer Premises Equipment (CPE) for 1085 Providing Residential IPv6 Internet Service", RFC 6092, 1086 DOI 10.17487/RFC6092, January 2011, 1087 . 1089 [RFC6120] Saint-Andre, P., "Extensible Messaging and Presence 1090 Protocol (XMPP): Core", RFC 6120, DOI 10.17487/RFC6120, 1091 March 2011, . 1093 [RFC6606] Kim, E., Kaspar, D., Gomez, C., and C. Bormann, "Problem 1094 Statement and Requirements for IPv6 over Low-Power 1095 Wireless Personal Area Network (6LoWPAN) Routing", 1096 RFC 6606, DOI 10.17487/RFC6606, May 2012, 1097 . 1099 [RFC7230] Fielding, R., Ed. and J. Reschke, Ed., "Hypertext Transfer 1100 Protocol (HTTP/1.1): Message Syntax and Routing", 1101 RFC 7230, DOI 10.17487/RFC7230, June 2014, 1102 . 1104 [RFC7252] Shelby, Z., Hartke, K., and C. Bormann, "The Constrained 1105 Application Protocol (CoAP)", RFC 7252, 1106 DOI 10.17487/RFC7252, June 2014, 1107 . 1109 [RFC7414] Duke, M., Braden, R., Eddy, W., Blanton, E., and A. 1110 Zimmermann, "A Roadmap for Transmission Control Protocol 1111 (TCP) Specification Documents", RFC 7414, 1112 DOI 10.17487/RFC7414, February 2015, 1113 . 1115 [RFC7428] Brandt, A. and J. Buron, "Transmission of IPv6 Packets 1116 over ITU-T G.9959 Networks", RFC 7428, 1117 DOI 10.17487/RFC7428, February 2015, 1118 . 1120 [RFC7540] Belshe, M., Peon, R., and M. Thomson, Ed., "Hypertext 1121 Transfer Protocol Version 2 (HTTP/2)", RFC 7540, 1122 DOI 10.17487/RFC7540, May 2015, 1123 . 1125 [RFC7567] Baker, F., Ed. and G. Fairhurst, Ed., "IETF 1126 Recommendations Regarding Active Queue Management", 1127 BCP 197, RFC 7567, DOI 10.17487/RFC7567, July 2015, 1128 . 1130 [RFC7668] Nieminen, J., Savolainen, T., Isomaki, M., Patil, B., 1131 Shelby, Z., and C. Gomez, "IPv6 over BLUETOOTH(R) Low 1132 Energy", RFC 7668, DOI 10.17487/RFC7668, October 2015, 1133 . 1135 [RFC8087] Fairhurst, G. and M. Welzl, "The Benefits of Using 1136 Explicit Congestion Notification (ECN)", RFC 8087, 1137 DOI 10.17487/RFC8087, March 2017, 1138 . 1140 [RFC8105] Mariager, P., Petersen, J., Ed., Shelby, Z., Van de Logt, 1141 M., and D. Barthel, "Transmission of IPv6 Packets over 1142 Digital Enhanced Cordless Telecommunications (DECT) Ultra 1143 Low Energy (ULE)", RFC 8105, DOI 10.17487/RFC8105, May 1144 2017, . 1146 [RFC8163] Lynn, K., Ed., Martocci, J., Neilson, C., and S. 1147 Donaldson, "Transmission of IPv6 over Master-Slave/Token- 1148 Passing (MS/TP) Networks", RFC 8163, DOI 10.17487/RFC8163, 1149 May 2017, . 1151 [RFC8201] McCann, J., Deering, S., Mogul, J., and R. Hinden, Ed., 1152 "Path MTU Discovery for IP version 6", STD 87, RFC 8201, 1153 DOI 10.17487/RFC8201, July 2017, 1154 . 1156 [RFC8323] Bormann, C., Lemay, S., Tschofenig, H., Hartke, K., 1157 Silverajan, B., and B. Raymor, Ed., "CoAP (Constrained 1158 Application Protocol) over TCP, TLS, and WebSockets", 1159 RFC 8323, DOI 10.17487/RFC8323, February 2018, 1160 . 1162 [RFC8352] Gomez, C., Kovatsch, M., Tian, H., and Z. Cao, Ed., 1163 "Energy-Efficient Features of Internet of Things 1164 Protocols", RFC 8352, DOI 10.17487/RFC8352, April 2018, 1165 . 1167 [RFC8376] Farrell, S., Ed., "Low-Power Wide Area Network (LPWAN) 1168 Overview", RFC 8376, DOI 10.17487/RFC8376, May 2018, 1169 . 1171 [RIOT] E. Baccelli et al., "RIOT: an Open Source Operating 1172 Systemfor Low-end Embedded Devices in the IoT", 2018. 1174 [shrew] A. Kuzmanovic, E. Knightly, "Low-Rate TCP-Targeted Denial 1175 of Service Attacks", SIGCOMM'03 2003. 1177 Authors' Addresses 1179 Carles Gomez 1180 UPC 1181 C/Esteve Terradas, 7 1182 Castelldefels 08860 1183 Spain 1185 Email: carlesgo@entel.upc.edu 1187 Jon Crowcroft 1188 University of Cambridge 1189 JJ Thomson Avenue 1190 Cambridge, CB3 0FD 1191 United Kingdom 1193 Email: jon.crowcroft@cl.cam.ac.uk 1195 Michael Scharf 1196 Hochschule Esslingen 1197 Flandernstr. 101 1198 Esslingen 73732 1199 Germany 1201 Email: michael.scharf@hs-esslingen.de