idnits 2.17.1 draft-ietf-tcpimpl-cong-control-05.txt: Checking boilerplate required by RFC 5378 and the IETF Trust (see https://trustee.ietf.org/license-info): ---------------------------------------------------------------------------- ** Looks like you're using RFC 2026 boilerplate. This must be updated to follow RFC 3978/3979, as updated by RFC 4748. Checking nits according to https://www.ietf.org/id-info/1id-guidelines.txt: ---------------------------------------------------------------------------- ** Missing expiration date. The document expiration date should appear on the first and last page. ** The document seems to lack a 1id_guidelines paragraph about Internet-Drafts being working documents. ** The document seems to lack a 1id_guidelines paragraph about 6 months document validity -- however, there's a paragraph with a matching beginning. Boilerplate error? ** The document seems to lack a 1id_guidelines paragraph about the list of current Internet-Drafts. ** The document seems to lack a 1id_guidelines paragraph about the list of Shadow Directories. == No 'Intended status' indicated for this document; assuming Proposed Standard 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.) ** The document seems to lack separate sections for Informative/Normative References. All references will be assumed normative when checking for downward references. ** There are 5 instances of lines with control characters in the document. Miscellaneous warnings: ---------------------------------------------------------------------------- -- The document seems to lack a disclaimer for pre-RFC5378 work, but may have content which was first submitted before 10 November 2008. If you have contacted all the original authors and they are all willing to grant the BCP78 rights to the IETF Trust, then this is fine, and you can ignore this comment. If not, you may need to add the pre-RFC5378 disclaimer. (See the Legal Provisions document at https://trustee.ietf.org/license-info for more information.) -- Couldn't find a document date in the document -- date freshness check skipped. Checking references for intended status: Proposed Standard ---------------------------------------------------------------------------- (See RFCs 3967 and 4897 for information about using normative references to lower-maturity documents in RFCs) == Missing Reference: 'MM6b' is mentioned on line 434, but not defined == Unused Reference: 'Flo94' is defined on line 508, but no explicit reference was found in the text ** Obsolete normative reference: RFC 2414 (ref. 'AFP98') (Obsoleted by RFC 3390) ** Obsolete normative reference: RFC 813 (ref. 'Cla82') (Obsoleted by RFC 7805) -- Possible downref: Non-RFC (?) normative reference: ref. 'FF96' == Outdated reference: A later version (-02) exists of draft-ietf-tcpimpl-newreno-00 ** Downref: Normative reference to an Experimental draft: draft-ietf-tcpimpl-newreno (ref. 'FH98') -- Possible downref: Non-RFC (?) normative reference: ref. 'Flo94' -- Possible downref: Non-RFC (?) normative reference: ref. 'Hoe96' -- Possible downref: Normative reference to a draft: ref. 'HTH98' -- Possible downref: Non-RFC (?) normative reference: ref. 'Jac88' -- Possible downref: Non-RFC (?) normative reference: ref. 'Jac90' -- Possible downref: Non-RFC (?) normative reference: ref. 'MM96a' -- Possible downref: Non-RFC (?) normative reference: ref. 'MM96b' -- Possible downref: Non-RFC (?) normative reference: ref. 'Pax97' ** Obsolete normative reference: RFC 793 (ref. 'Pos81') (Obsoleted by RFC 9293) -- Possible downref: Non-RFC (?) normative reference: ref. 'Ste94' ** Obsolete normative reference: RFC 2001 (ref. 'Ste97') (Obsoleted by RFC 2581) -- Possible downref: Non-RFC (?) normative reference: ref. 'WS95' Summary: 14 errors (**), 0 flaws (~~), 4 warnings (==), 13 comments (--). Run idnits with the --verbose option for more detailed information about the items above. -------------------------------------------------------------------------------- 1 TCP Implementation Working Group M. Allman 2 INTERNET DRAFT NASA Lewis/Sterling Software 3 File: draft-ietf-tcpimpl-cong-control-05.txt V. Paxson 4 LBNL 5 W. Stevens 6 Consultant 7 February, 1999 9 TCP Congestion Control 11 Status of this Memo 13 This document is an Internet-Draft and is in full conformance with 14 all provisions of Section 10 of RFC2026. Internet-Draft. 15 Internet-Drafts are working documents of the Internet Engineering 16 Task Force (IETF), its areas, and its working groups. Note that 17 other groups may also distribute working documents as 18 Internet-Drafts. 20 Internet-Drafts are draft documents valid for a maximum of six 21 months and may be updated, replaced, or obsoleted by other documents 22 at any time. It is inappropriate to use Internet-Drafts as 23 reference material or to cite them other than as ``work in 24 progress.'' 26 To view the entire list of current Internet-Drafts, please check the 27 "1id-abstracts.txt" listing contained in the Internet-Drafts Shadow 28 Directories on ftp.is.co.za (Africa), ftp.nordu.net (Northern 29 Europe), ftp.nis.garr.it (Southern Europe), munnari.oz.au (Pacific 30 Rim), ftp.ietf.org (US East Coast), or ftp.isi.edu (US West Coast). 32 Abstract 34 This document defines TCP's four intertwined congestion control 35 algorithms: slow start, congestion avoidance, fast retransmit, and 36 fast recovery. In addition, the document specifies how TCP should 37 begin transmission after a relatively long idle period, as well as 38 discussing various acknowledgment generation methods. 40 1 Introduction 42 This document specifies four TCP [Pos81] congestion control 43 algorithms: slow start, congestion avoidance, fast retransmit and 44 fast recovery. These algorithms were devised in [Jac88] and 45 [Jac90]. Their use with TCP is standardized in [Bra89]. 47 This document is an update of [Ste97]. In addition to specifying 48 the congestion control algorithms, this document specifies what TCP 49 connections should do after a relatively long idle period, as well 50 as specifying and clarifying some of the issues pertaining to TCP 51 ACK generation. 53 Note that [Ste94] provides examples of these algorithms in action 54 and [WS95] provides an explanation of the source code for the BSD 55 implementation of these algorithms. 57 This document is organized as follows. Section 2 provides various 58 definitions which will be used throughout the document. Section 3 59 provides a specification of the congestion control algorithms. 60 Section 4 outlines concerns related to the congestion control 61 algorithms and finally, section 5 outlines security considerations. 63 The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", 64 "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this 65 document are to be interpreted as described in [Bra97]. 67 2 Definitions 69 This section provides the definition of several terms that will be 70 used throughout the remainder of this document. 72 SEGMENT: 73 A segment is ANY TCP/IP data or acknowledgment packet (or both). 75 SENDER MAXIMUM SEGMENT SIZE (SMSS): 76 The SMSS is the size of the largest segment that the sender can 77 transmit. This value can be based on the maximum transmission 78 unit of the network, the path MTU discovery [MD90] algorithm, 79 RMSS (see next item), or other factors. The size does not 80 include the TCP/IP headers and options. 82 RECEIVER MAXIMUM SEGMENT SIZE (RMSS): 83 The RMSS is the size of the largest segment the receiver is 84 willing to accept. This is the value specified in the MSS 85 option sent by the receiver during connection startup. Or, if 86 the MSS option is not used, 536 bytes [Bra89]. The size does 87 not include the TCP/IP headers and options. 89 FULL-SIZED SEGMENT: 90 A segment that contains the maximum number of data bytes 91 permitted (i.e., a segment containing SMSS bytes of data). 93 RECEIVER WINDOW (rwnd) 94 The most recently advertised receiver window. 96 CONGESTION WINDOW (cwnd): 97 A TCP state variable that limits the amount of data a TCP can 98 send. At any given time, a TCP MUST NOT send data with a 99 sequence number higher than the sum of the highest acknowledged 100 sequence number and the minimum of cwnd and rwnd. 102 INITIAL WINDOW (IW): 103 The initial window is the size of the sender's congestion window 104 after the three-way handshake is completed. 106 LOSS WINDOW (LW): 107 The loss window is the size of the congestion window after a TCP 108 sender detects loss using its retransmission timer. 110 RESTART WINDOW (RW): 111 The restart window is the size of the congestion window after a 112 TCP restarts transmission after an idle period (if the slow 113 start algorithm is used; see section 4.1 for more discussion). 115 FLIGHT SIZE: 116 The amount of data that has been sent but not yet acknowledged. 118 3 Congestion Control Algorithms 120 This section defines the four congestion control algorithms: slow 121 start, congestion avoidance, fast retransmit and fast recovery, 122 developed in [Jac88] and [Jac90]. In some situations it may be 123 beneficial for a TCP sender to be more conservative than the 124 algorithms allow, however a TCP MUST NOT be more aggressive than the 125 following algorithms allow (that is, MUST NOT send data when the 126 value of cwnd computed by the following algorithms would not allow 127 the data to be sent). 129 3.1 Slow Start and Congestion Avoidance 131 The slow start and congestion avoidance algorithms MUST be used by a 132 TCP sender to control the amount of outstanding data being injected 133 into the network. To implement these algorithms, two variables are 134 added to the TCP per-connection state. The congestion window (cwnd) 135 is a sender-side limit on the amount of data the sender can transmit 136 into the network before receiving an acknowledgment (ACK), while the 137 receiver's advertised window (rwnd) is a receiver-side limit on the 138 amount of outstanding data. The minimum of cwnd and rwnd governs 139 data transmission. 141 Another state variable, the slow start threshold (ssthresh), is used 142 to determine whether the slow start or congestion avoidance 143 algorithm is used to control data transmission, as discussed below. 145 Beginning transmission into a network with unknown conditions 146 requires TCP to slowly probe the network to determine the available 147 capacity, in order to avoid congesting the network with an 148 inappropriately large burst of data. The slow start algorithm is 149 used for this purpose at the beginning of a transfer, or after 150 repairing loss detected by the retransmission timer. 152 IW, the initial value of cwnd, MUST be less than or equal to 2*SMSS 153 bytes and MUST NOT be more than 2 segments. 155 We note that a non-standard, experimental TCP extension allows that 156 a TCP MAY use a larger initial window (IW), as defined in equation 1 157 [AFP98]: 159 IW = min (4*SMSS, max (2*SMSS, 4380 bytes)) (1) 161 With this extension, a TCP sender MAY use a 3 or 4 segment initial 162 window, provided the combined size of the segments does not exceed 163 4380 bytes. We do NOT allow this change as part of the standard 164 defined by this document. However, we include discussion of (1) in 165 the remainder of this document as a guideline for those 166 experimenting with the change, rather than conforming to the present 167 standards for TCP congestion control. 169 The initial value of ssthresh MAY be arbitrarily high (for example, 170 some implementations use the size of the advertised window), but it 171 may be reduced in response to congestion. The slow start algorithm 172 is used when cwnd < ssthresh, while the congestion avoidance 173 algorithm is used when cwnd > ssthresh. When cwnd and ssthresh are 174 equal the sender may use either slow start or congestion avoidance. 176 During slow start, a TCP increments cwnd by at most SMSS bytes for 177 each ACK received that acknowledges new data. Slow start ends when 178 cwnd exceeds ssthresh (or, optionally, when it reaches it, as noted 179 above) or when congestion is observed. 181 During congestion avoidance, cwnd is incremented by 1 full-sized 182 segment per round-trip time (RTT). Congestion avoidance continues 183 until cwnd congestion is detected. One formula commonly used to 184 update cwnd during congestion avoidance is given in equation 2: 186 cwnd += SMSS*SMSS/cwnd (2) 188 This adjustment is executed on every incoming non-duplicate ACK. 189 Equation (2) provides an acceptable approximation to the underlying 190 principle of increasing cwnd by 1 full-sized segment per RTT. (Note 191 that for a connection in which the receiver acknowledges every data 192 segment, (2) proves slightly more aggressive than 1 segment per RTT, 193 and for a receiver acknowledging every-other packet, (2) is less 194 aggressive.) 196 Implementation Note: Since integer arithmetic is usually used in TCP 197 implementations, the formula given in equation 2 can fail to 198 increase cwnd when the congestion window is very large (larger than 199 SMSS*SMSS). If the above formula yields 0, the result SHOULD be 200 rounded up to 1 byte. 202 Implementation Note: older implementations have an additional 203 additive constant on the right-hand side of equation (2). This is 204 incorrect and can actually lead to diminished performance [PAD+98]. 206 Another acceptable way to increase cwnd during congestion avoidance 207 is to count the number of bytes that have been acknowledged by ACKs 208 for new data. (A drawback of this implementation is that it 209 requires maintaining an additional state variable.) When the number 210 of bytes acknowledged reaches cwnd, then cwnd can be incremented by 211 up to SMSS bytes. Note that during congestion avoidance, cwnd MUST 212 NOT be increased by more than the larger of either 1 full-sized 213 segment per RTT, or the value computed using equation 2. 215 Implementation Note: some implementations maintain cwnd in units of 216 bytes, while others in units of full-sized segments. The latter 217 will find equation (2) difficult to use, and may prefer to use the 218 counting approach discussed in the previous paragraph. 220 When a TCP sender detects segment loss using the retransmission 221 timer, the value of ssthresh MUST be set to no more than the value 222 given in equation 3: 224 ssthresh = max (FlightSize / 2, 2*SMSS) (3) 226 As discussed above, FlightSize is the amount of outstanding data in 227 the network. 229 Implementation Note: an easy mistake to make is to simply use cwnd, 230 rather than FlightSize, which in some implementations may 231 incidentally increase well beyond rwnd. 233 Furthermore, upon a timeout cwnd MUST be set to no more than the 234 loss window, LW, which equals 1 full-sized segment (regardless of 235 the value of IW). Therefore, after retransmitting the dropped 236 segment the TCP sender uses the slow start algorithm to increase the 237 window from 1 full-sized segment to the new value of ssthresh, at 238 which point congestion avoidance again takes over. 240 3.2 Fast Retransmit/Fast Recovery 242 A TCP receiver SHOULD send an immediate duplicate ACK when an 243 out-of-order segment arrives. The purpose of this ACK is to inform 244 the sender that a segment was received out-of-order and which 245 sequence number is expected. From the sender's perspective, 246 duplicate ACKs can be caused by a number of network problems. 247 First, they can be caused by dropped segments. In this case, all 248 segments after the dropped segment will trigger duplicate ACKs. 249 Second, duplicate ACKs can be caused by the re-ordering of data 250 segments by the network (not a rare event along some network paths 251 [Pax97]). Finally, duplicate ACKs can be caused by replication of 252 ACK or data segments by the network. In addition, a TCP receiver 253 SHOULD send an immediate ACK when the incoming segment fills in all 254 or part of a gap in the sequence space. This will generate more 255 timely information for a sender recovering from a loss through a 256 retransmission timeout, a fast retransmit, or an experimental loss 257 recovery algorithm, such as NewReno [FH98]. 259 The TCP sender SHOULD use the "fast retransmit" algorithm to detect 260 and repair loss, based on incoming duplicate ACKs. The fast 261 retransmit algorithm uses the arrival of 3 duplicate ACKs (4 262 identical ACKs without the arrival of any other intervening packets) 263 as an indication that a segment has been lost. After receiving 3 264 duplicate ACKs, TCP performs a retransmission of what appears to be 265 the missing segment, without waiting for the retransmission timer to 266 expire. 268 After the fast retransmit algorithm sends what appears to be the 269 missing segment, the "fast recovery" algorithm governs the 270 transmission of new data until a non-duplicate ACK arrives. The 271 reason for not performing slow start is that the receipt of the 272 duplicate ACKs not only indicates that a segment has been lost, but 273 also that segments are most likely leaving the network (although a 274 massive segment duplication by the network can invalidate this 275 conclusion). In other words, since the receiver can only generate a 276 duplicate ACK when a segment has arrived, that segment has left the 277 network and is in the receiver's buffer, so we know it is no longer 278 consuming network resources. Furthermore, since the ACK "clock" 279 [Jac88] is preserved, the TCP sender can continue to transmit new 280 segments (although transmission must continue using a reduced cwnd). 282 The fast retransmit and fast recovery algorithms are usually 283 implemented together as follows. 285 1. When the third duplicate ACK is received, set ssthresh to no 286 more than the value given in equation 3. 288 2. Retransmit the lost segment and set cwnd to ssthresh plus 289 3*SMSS. This artificially "inflates" the congestion window by 290 the number of segments (three) that have left the network and 291 which the receiver has buffered. 293 3. For each additional duplicate ACK received, increment cwnd by 294 SMSS. This artificially inflates the congestion window in order 295 to reflect the additional segment that has left the network. 297 4. Transmit a segment, if allowed by the new value of cwnd and the 298 receiver's advertised window. 300 5. When the next ACK arrives that acknowledges new data, set cwnd 301 to ssthresh (the value set in step 1). This is termed 302 "deflating" the window. 304 This ACK should be the acknowledgment elicited by the 305 retransmission from step 1, one RTT after the retransmission 306 (though it may arrive sooner in the presence of significant 307 out-of-order delivery of data segments at the receiver). 308 Additionally, this ACK should acknowledge all the intermediate 309 segments sent between the lost segment and the receipt of the 310 third duplicate ACK, if none of these were lost. 312 Note: This algorithm is known to generally not recover very 313 efficiently from multiple losses in a single flight of packets 314 [FF96]. One proposed set of modifications to it to address this 315 problem can be found in [FH98]. 317 4 Additional Considerations 319 4.1 Re-starting Idle Connections 321 A known problem with the TCP congestion control algorithms described 322 above is that they allow a potentially inappropriate burst of 323 traffic to be transmitted after TCP has been idle for a relatively 324 long period of time. After an idle period, TCP cannot use the ACK 325 clock to strobe new segments into the network, as all the ACKs have 326 drained from the network. Therefore, as specified above, TCP can 327 potentially send a cwnd-size line-rate burst into the network after 328 an idle period. 330 [Jac88] recommends that a TCP use slow start to restart transmission 331 after a relatively long idle period. Slow start serves to restart 332 the ACK clock, just as it does at the beginning of a transfer. This 333 mechanism has been widely deployed in the following manner. When 334 TCP has not received a segment for more than one retransmission 335 timeout, cwnd is reduced to the value of the restart window (RW) 336 before transmission begins. 338 For the purposes of this standard, we define RW = IW. 340 We note that the non-standard experimental extension to TCP defined 341 in [AFP98] defines RW = min(IW, cwnd), with the definition of IW 342 adjusted per equation (1) above. 344 Using the last time a segment was received to determine whether or 345 not to decrease cwnd fails to deflate cwnd in the common case of 346 persistent HTTP connections [HTH98]. In this case, a WWW server 347 receives a request before transmitting data to the WWW browser. The 348 reception of the request makes the test for an idle connection fail, 349 and allows the TCP to begin transmission with a possibly 350 inappropriately large cwnd. 352 Therefore, a TCP SHOULD set cwnd to no more than RW before beginning 353 transmission if the TCP has not sent data in an interval exceeding 354 the retransmission timeout. 356 4.2 Generating Acknowledgments 358 The delayed ACK algorithm specified in [Bra89] SHOULD be used by a 359 TCP receiver. When used, a TCP receiver MUST NOT excessively delay 360 acknowledgments. Specifically, an ACK SHOULD be generated for at 361 least every second full-sized segment, and MUST be generated within 362 500 ms of the arrival of the first unacknowledged packet. 364 The requirement that an ACK "SHOULD" be generated for at least every 365 second full-sized segment is listed in [Bra89] in one place as a 366 SHOULD and another as a MUST. Here we unambiguously state it is a 367 SHOULD. We also emphasize that this is a SHOULD, meaning that an 368 implementor should indeed only deviate from this requirement after 369 careful consideration of the implications. See the discussion of 370 "Stretch ACK violation" in [PAD+98] and the references therein for a 371 discussion of the possible performance problems with generating ACKs 372 less frequently than every second full-sized segment. 374 In some cases, the sender and receiver may not agree on what 375 constitutes a full-sized segment. An implementation is deemed to 376 comply with this requirement if it sends at least one acknowledgment 377 every time it receives 2*RMSS bytes of new data from the sender, 378 where RMSS is the Maximum Segment Size specified by the receiver to 379 the sender (or the default value of 536 bytes, per [Bra89], if the 380 receiver does not specify an MSS option during connection 381 establishment). The sender may be forced to use a segment size less 382 than RMSS due to the maximum transmission unit (MTU), the path MTU 383 discovery algorithm or other factors. For instance, consider the 384 case when the receiver announces an RMSS of X bytes but the sender 385 ends up using a segment size of Y bytes (Y < X) due to path MTU 386 discovery (or the sender's MTU size). The receiver will generate 387 stretch ACKs if it waits for 2*X bytes to arrive before an ACK is 388 sent. Clearly this will take more than 2 segments of size Y bytes. 389 Therefore, while a specific algorithm is not defined, it is 390 desirable for receivers to attempt to prevent this situation, for 391 example by acknowledging at least every second segment, regardless 392 of size. Finally, we repeat that an ACK MUST NOT be delayed for 393 more than 500 ms waiting on a second full-sized segment to arrive. 395 Out-of-order data segments SHOULD be acknowledged immediately, in 396 order to accelerate loss recovery. To trigger the fast retransmit 397 algorithm, the receiver SHOULD send an immediate duplicate ACK when 398 it receives a data segment above a gap in the sequence space. To 399 provide feedback to senders recovering from losses, the receiver 400 SHOULD send an immediate ACK when it receives a data segment that 401 fills in all or part of a gap in the sequence space. 403 A TCP receiver MUST NOT generate more than one ACK for every 404 incoming segment, other than to update the offered window as the 405 receiving application consumes new data [page 42, Pos81][Cla82]. 407 4.3 Loss Recovery Mechanisms 409 A number of loss recovery algorithms that augment fast retransmit 410 and fast recovery have been suggested by TCP researchers. While 411 some of these algorithms are based on the TCP selective 412 acknowledgment (SACK) option [MMFR96], such as [FF96,MM96a,MM96b], 413 others do not require SACKs [Hoe96,FF96,FH98]. The non-SACK 414 algorithms use "partial acknowledgments" (ACKs which cover new data, 415 but not all the data outstanding when loss was detected) to trigger 416 retransmissions. While this document does not standardize any of 417 the specific algorithms that may improve fast retransmit/fast 418 recovery, these enhanced algorithms are implicitly allowed, as long 419 as they follow the general principles of the basic four algorithms 420 outlined above. 422 Therefore, when the first loss in a window of data is detected, 423 ssthresh MUST be set to no more than the value given by equation 424 (3). Second, until all lost segments in the window of data in 425 question are repaired, the number of segments transmitted in each 426 RTT MUST be no more than half the number of outstanding segments 427 when the loss was detected. Finally, after all loss in the given 428 window of segments has been successfully retransmitted, cwnd MUST be 429 set to no more than ssthresh and congestion avoidance MUST be used 430 to further increase cwnd. Loss in two successive windows of data, 431 or the loss of a retransmission, should be taken as two indications 432 of congestion and, therefore, cwnd (and ssthresh) MUST be lowered 433 twice in this case. The algorithms outlined in 434 [Hoe96,FF96,MM96a,MM6b] follow the principles of the basic four 435 congestion control algorithms outlined in this document. 437 5. Security Considerations 439 This document requires a TCP to diminish its sending rate in the 440 presence of retransmission timeouts and the arrival of duplicate 441 acknowledgments. An attacker can therefore impair the performance 442 of a TCP connection by either causing data packets or their 443 acknowledgments to be lost, or by forging excessive duplicate 444 acknowledgments. Causing two congestion control events back-to-back 445 will often cut ssthresh to its minimum value of 2*SMSS, causing the 446 connection to immediately enter the slower-performing congestion 447 avoidance phase. 449 The Internet to a considerable degree relies on the correct 450 implementation of these algorithms in order to preserve network 451 stability and avoid congestion collapse. An attacker could cause 452 TCP endpoints to respond more aggressively in the face of congestion 453 by forging excessive duplicate acknowledgments or excessive 454 acknowledgments for new data. Conceivably, such an attack could 455 drive a portion of the network into congestion collapse. 457 6. Changes Relative to RFC 2001 459 This document has been extensively rewritten editorially and it is 460 not feasible to itemize the list of changes between the two 461 documents. The intention of this document is not to change any of 462 the recommendations given in RFC 2001, but to further clarify cases 463 that were not discussed in detail in 2001. Specifically, this 464 document suggests what TCP connections should do after a relatively 465 long idle period, as well as specifying and clarifying some of the 466 issues pertaining to TCP ACK generation. Finally, the allowable 467 upper bound for the initial congestion window has also been raised 468 from one to two segments. 470 Acknowledgments 472 The four algorithms that are described were developed by Van 473 Jacobson. 475 Some of the text from this document is taken from "TCP/IP 476 Illustrated, Volume 1: The Protocols" by W. Richard Stevens 477 (Addison-Wesley, 1994) and "TCP/IP Illustrated, Volume 2: The 478 Implementation" by Gary R. Wright and W. Richard Stevens 479 (Addison-Wesley, 1995). This material is used with the permission 480 of Addison-Wesley. 482 Neal Cardwell, Sally Floyd, Craig Partridge and Joe Touch 483 contributed a number of helpful suggestions. 485 References 487 [AFP98] M. Allman, S. Floyd, C. Partridge, Increasing TCP's Initial 488 Window Size, September 1998. RFC 2414. 490 [Bra89] B. Braden, ed., Requirements for Internet Hosts -- 491 Communication Layers, RFC 1122, Oct. 1989. 493 [Bra97] S. Bradner, Key words for use in RFCs to Indicate 494 Requirement Levels, BCP 14, RFC 2119, March 1997. 496 [Cla82] D. Clark, Window and Acknowledgment Strategy in TCP, RFC 497 813. July 1982. 499 [FF96] K. Fall, S. Floyd. Simulation-based Comparisons of Tahoe, 500 Reno and SACK TCP. Computer Communication Review, July 1996. 501 ftp://ftp.ee.lbl.gov/papers/sacks.ps.Z. 503 [FH98] S. Floyd, T. Henderson. The NewReno Modification to TCP's 504 Fast Recovery Algorithm. Internet-Draft 505 draft-ietf-tcpimpl-newreno-00.txt, November 1998. (Work in 506 progress). 508 [Flo94] S. Floyd, TCP and Successive Fast Retransmits. Technical 509 report, October 1994. 510 ftp://ftp.ee.lbl.gov/papers/fastretrans.ps. 512 [Hoe96] J. Hoe, Improving the Start-up Behavior of a Congestion 513 Control Scheme for TCP. In ACM SIGCOMM, August 1996. 515 [HTH98] A. Hughes, J. Touch, J. Heidemann. Issues in TCP Slow-Start 516 Restart After Idle. Internet-Draft 517 draft-ietf-tcpimpl-restart-00.txt, March 1998. (Work in 518 progress). 520 [Jac88] V. Jacobson, Congestion Avoidance and Control, Computer 521 Communication Review, vol. 18, no. 4, pp. 314-329, Aug. 1988. 522 ftp://ftp.ee.lbl.gov/papers/congavoid.ps.Z. 524 [Jac90] V. Jacobson, Modified TCP Congestion Avoidance Algorithm, 525 end2end-interest mailing list, April 30, 1990. 526 ftp://ftp.isi.edu/end2end/end2end-interest-1990.mail. 528 [MD90] J. Mogul, S. Deering. Path MTU Discovery, November 1990. 529 RFC 1191. 531 [MM96a] M. Mathis, J. Mahdavi, Forward Acknowledgment: Refining TCP 532 Congestion Control, Proceedings of SIGCOMM'96, August, 1996, 533 Stanford, CA. Available from 534 http://www.psc.edu/networking/papers/papers.html 536 [MM96b] M. Mathis, J. Mahdavi, TCP Rate-Halving with Bounding 537 Parameters. Technical report. Available from 538 http://www.psc.edu/networking/papers/FACKnotes/current. 540 [MMFR96] M. Mathis, J. Mahdavi, S. Floyd, A. Romanow, TCP Selective 541 Acknowledgement Options, October 1996. RFC 2018. 543 [PAD+98] V. Paxson, M. Allman, S. Dawson, W. Fenner, J. Griner, 544 I. Heavens, K. Lahey, J. Semke, B. Volz. Known TCP 545 Implementation Problems. Internet-Draft 546 draft-ietf-tcpimpl-prob-05.txt, November 1998. (Work in 547 progress). 549 [Pax97] V. Paxson, End-to-End Internet Packet Dynamics, 550 Proceedings of SIGCOMM '97, Cannes, France, Sep. 1997. 552 [Pos81] J. Postel, Transmission Control Protocol, September 1981. 553 RFC 793. 555 [Ste94] W. R. Stevens, TCP/IP Illustrated, Volume 1: The 556 Protocols, Addison-Wesley, 1994. 558 [Ste97] W. R. Stevens, "TCP Slow Start, Congestion Avoidance, Fast 559 Retransmit, and Fast Recovery Algorithms", January 1997. RFC 560 2001. 562 [WS95] G. R. Wright, W. R. Stevens, TCP/IP Illustrated, Volume 2: 563 The Implementation, Addison-Wesley, 1995. 565 Author's Address: 567 Mark Allman 568 NASA Lewis Research Center/Sterling Software 569 21000 Brookpark Rd. MS 54-2 570 Cleveland, OH 44135 571 216-433-6586 572 mallman@lerc.nasa.gov 573 http://roland.lerc.nasa.gov/~mallman 575 Vern Paxson 576 Network Research Group 577 Lawrence Berkeley National Laboratory 578 Berkeley, CA 94720 579 USA 580 510-486-7504 581 vern@ee.lbl.gov 583 W. Richard Stevens 584 1202 E. Paseo del Zorro 585 Tucson, AZ 85718 586 520-297-9416 587 rstevens@kohala.com 588 http://www.kohala.com/~rstevens