idnits 2.17.1 draft-ietf-tsvwg-initwin-01.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: ---------------------------------------------------------------------------- ** 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? == 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 Introduction section. ** 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 8 instances of too long lines in the document, the longest one being 11 characters in excess of 72. ** The abstract seems to contain references ([RFC2119]), which it shouldn't. Please replace those with straight textual mentions of the documents in question. 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: 'RFC2481' is mentioned on line 564, but not defined ** Obsolete undefined reference: RFC 2481 (Obsoleted by RFC 3168) == Missing Reference: 'MMFR96' is mentioned on line 576, but not defined == Unused Reference: 'FF96' is defined on line 455, but no explicit reference was found in the text == Unused Reference: 'FJ93' is defined on line 464, but no explicit reference was found in the text == Unused Reference: 'Flo96' is defined on line 471, but no explicit reference was found in the text == Unused Reference: 'RFC2309' is defined on line 510, but no explicit reference was found in the text == Unused Reference: 'RFC3168' is defined on line 532, but no explicit reference was found in the text -- Possible downref: Non-RFC (?) normative reference: ref. 'AHO98' -- Possible downref: Non-RFC (?) normative reference: ref. 'All97a' -- Possible downref: Non-RFC (?) normative reference: ref. 'All97b' -- Possible downref: Non-RFC (?) normative reference: ref. 'All00' -- Possible downref: Non-RFC (?) normative reference: ref. 'FF96' -- Possible downref: Non-RFC (?) normative reference: ref. 'FF98' -- Possible downref: Non-RFC (?) normative reference: ref. 'FJ93' -- Possible downref: Non-RFC (?) normative reference: ref. 'Flo94' -- Possible downref: Non-RFC (?) normative reference: ref. 'Flo96' -- Possible downref: Non-RFC (?) normative reference: ref. 'Flo97' -- Possible downref: Non-RFC (?) normative reference: ref. 'KAGT98' -- Possible downref: Non-RFC (?) normative reference: ref. 'Mor97' -- Possible downref: Non-RFC (?) normative reference: ref. 'Nic97' ** Obsolete normative reference: RFC 821 (ref. 'Pos82') (Obsoleted by RFC 2821) ** Downref: Normative reference to an Informational RFC: RFC 1945 ** Obsolete normative reference: RFC 2068 (Obsoleted by RFC 2616) ** Obsolete normative reference: RFC 2309 (Obsoleted by RFC 7567) ** Downref: Normative reference to an Informational RFC: RFC 2415 ** Downref: Normative reference to an Informational RFC: RFC 2416 ** Obsolete normative reference: RFC 2581 (Obsoleted by RFC 5681) ** Obsolete normative reference: RFC 2988 (Obsoleted by RFC 6298) Summary: 16 errors (**), 0 flaws (~~), 8 warnings (==), 15 comments (--). Run idnits with the --verbose option for more detailed information about the items above. -------------------------------------------------------------------------------- 1 Internet Engineering Task Force Mark Allman 2 INTERNET DRAFT BBN/NASA GRC 3 File: draft-ietf-tsvwg-initwin-01.txt February, 2002 4 Expires: August, 2002 5 Sally Floyd 6 ICIR 7 Craig Partridge 8 BBN Technologies 10 Increasing TCP's Initial Window 12 Status of this Memo 14 This document is an Internet-Draft and is in full conformance with 15 all provisions of Section 10 of RFC2026. 17 Internet-Drafts are working documents of the Internet Engineering 18 Task Force (IETF), its areas, and its working groups. Note that 19 other groups may also distribute working documents as 20 Internet-Drafts. 22 Internet-Drafts are draft documents valid for a maximum of six 23 months and may be updated, replaced, or obsoleted by other documents 24 at any time. It is inappropriate to use Internet- Drafts as 25 reference material or to cite them other than as "work in progress." 27 The list of current Internet-Drafts can be accessed at 28 http://www.ietf.org/ietf/1id-abstracts.txt 30 The list of Internet-Draft Shadow Directories can be accessed at 31 http://www.ietf.org/shadow.html. 33 Abstract 35 This document specifies an increase in the permitted initial window 36 for TCP from one segment to roughly 4K bytes. This document also 37 discusses the advantages and disadvantages of the change, outlining 38 experimental results that indicate the costs and benefits. 40 Terminology 42 The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", 43 "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this 44 document are to be interpreted as described in RFC 2119 [RFC2119]. 46 1. TCP Modification 48 This document specifies an increase in the permitted upper bound for 49 TCP's initial window from one segment to between two and four 50 segments. In most cases, this change results in an upper bound on 51 the initial window of roughly 4K bytes (although given a large 52 segment size, the permitted initial window of two segments may be 53 significantly larger than 4K bytes). The upper bound for the 54 initial window is given more precisely in (1): 56 min (4*MSS, max (2*MSS, 4380 bytes)) (1) 58 Equivalently, the upper bound for the initial window size is based 59 on the maximum segment size (MSS), as follows: 61 If (MSS <= 1095 bytes) 62 then win <= 4 * MSS; 63 If (1095 bytes < MSS < 2190 bytes) 64 then win <= 4380; 65 If (2190 bytes <= MSS) 66 then win <= 2 * MSS; 68 This increased initial window is optional: that a TCP MAY start with 69 a larger initial window, not that it SHOULD. 71 This upper bound for the initial window size represents a change 72 from RFC 2581 [RFC2581], which specified that the congestion window 73 be initialized to one or two segments. 75 This change applies to the initial window of the connection in the 76 first round trip time (RTT) of data transmission following the TCP three- 77 way handshake. Neither the SYN/ACK nor its acknowledgment (ACK) in 78 the three-way handshake should increase the initial window size 79 above that outlined in equation (1). If the SYN or SYN/ACK is lost, 80 the initial window used by a sender after a correctly transmitted 81 SYN MUST be one segment consisting of MSS bytes. 83 TCP implementations use slow start in as many as three different 84 ways: (1) to start a new connection (the initial window); (2) to 85 restart transmission after a long idle period (the restart window); 86 and (3) to restart transmission after a retransmit timeout (the loss 87 window). The change specified in this document affects the value of 88 the initial window. Optionally, a TCP MAY set the restart window to 89 the minimum of the value used for the initial window and the current 90 value of cwnd (in other words, using a larger value for the restart 91 window should never increase the size of cwnd). These changes do 92 NOT change the loss window, which must remain 1 segment of MSS bytes 93 (to permit the lowest possible window size in the case of severe 94 congestion). 96 2. Implementation Issues 98 When larger initial windows are implemented along with Path MTU 99 Discovery [RFC1191], and the MSS being used is found to be too large, 100 the congestion window `cwnd' SHOULD be reduced to prevent large 101 bursts of smaller segments. Specifically, `cwnd' SHOULD be reduced 102 by the ratio of the old segment size to the new segment size. 104 When larger initial windows are implemented along with Path MTU 105 Discovery [RFC1191], alternatives are to set the "Don't Fragment" 106 (DF) bit in all segments in the initial window, or to set the "Don't 107 Fragment" (DF) bit in one of the segments. It is an open question 108 which of these two alternatives is best; we would hope that 109 implementation experiences will shed light on this question. In the 110 first case of setting the DF bit in all segments, if the initial 111 packets are too large, then all of the initial packets will be 112 dropped in the network. In the second case of setting the DF bit in 113 only one segment, if the initial packets are too large, then all but 114 one of the initial packets will be fragmented in the network. When 115 the second case is followed, setting the DF bit in the last segment 116 in the initial window provides the least chance for needless 117 retransmissions when the initial segment size is found to be too 118 large, because it minimizes the chances of duplicate ACKs triggering 119 a Fast Retransmit. However, more attention needs to be paid to the 120 interaction between larger initial windows and Path MTU Discovery. 122 The larger initial window specified in this document is not intended 123 as encouragement for web browsers to open multiple simultaneous 124 TCP connections all with large initial windows. When web browsers 125 open simultaneous TCP connections to the same destination, this 126 works against TCP's congestion control mechanisms [FF98], regardless 127 of the size of the initial window. Combining this behavior with 128 larger initial windows further increases the unfairness to other 129 traffic in the network. 131 3. Advantages of Larger Initial Windows 133 1. When the initial window is one segment, a receiver employing 134 delayed ACKs [RFC1122] is forced to wait for a timeout before 135 generating an ACK. With an initial window of at least two 136 segments, the receiver will generate an ACK after the second 137 data segment arrives. This eliminates the wait on the timeout 138 (often up to 200 msec, and possibly up to 500 msec [RFC1122]). 140 2. For connections transmitting only a small amount of data, a 141 larger initial window reduces the transmission time (assuming at 142 most moderate segment drop rates). For many email (SMTP 143 [Pos82]) and web page (HTTP [RFC1945, RFC2068]) transfers that 144 are less than 4K bytes, the larger initial window would reduce 145 the data transfer time to a single RTT. 147 3. For connections that will be able to use large congestion 148 windows, this modification eliminates up to three RTTs and a 149 delayed ACK timeout during the initial slow-start phase. This 150 will be of particular benefit for high-bandwidth large- 151 propagation-delay TCP connections, such as those over satellite 152 links. 154 4. Disadvantages of Larger Initial Windows for the Individual 155 Connection 157 In high-congestion environments, particularly for routers that have 158 a bias against bursty traffic (as in the typical Drop Tail router 159 queues), a TCP connection can sometimes be better off starting with 160 an initial window of one segment. There are scenarios where a TCP 161 connection slow-starting from an initial window of one segment might 162 not have segments dropped, while a TCP connection starting with an 163 initial window of four segments might experience unnecessary 164 retransmits due to the inability of the router to handle small 165 bursts. This could result in an unnecessary retransmit timeout. 166 For a large-window connection that is able to recover without a 167 retransmit timeout, this could result in an unnecessarily-early 168 transition from the slow-start to the congestion-avoidance phase of 169 the window increase algorithm. These premature segment drops are 170 unlikely to occur in uncongested networks with sufficient buffering 171 or in moderately-congested networks where the congested router uses 172 active queue management (such as Random Early Detection [FJ93, 173 RFC2309]). 175 Some TCP connections will receive better performance with the larger 176 initial window even if the burstiness of the initial window results 177 in premature segment drops. This will be true if (1) the TCP 178 connection recovers from the segment drop without a retransmit 179 timeout, and (2) the TCP connection is ultimately limited to a small 180 congestion window by either network congestion or by the receiver's 181 advertised window. 183 5. Disadvantages of Larger Initial Windows for the Network 185 In terms of the potential for congestion collapse, we consider two 186 separate potential dangers for the network. The first danger would 187 be a scenario where a large number of segments on congested links 188 were duplicate segments that had already been received at the 189 receiver. The second danger would be a scenario where a large 190 number of segments on congested links were segments that would be 191 dropped later in the network before reaching their final 192 destination. 194 In terms of the negative effect on other traffic in the network, a 195 potential disadvantage of larger initial windows would be that they 196 increase the general packet drop rate in the network. We discuss 197 these three issues below. 199 Duplicate segments: 201 As described in the previous section, the larger initial window 202 could occasionally result in a segment dropped from the initial 203 window, when that segment might not have been dropped if the 204 sender had slow-started from an initial window of one segment. 205 However, Appendix A shows that even in this case, the larger 206 initial window would not result in the transmission of a large 207 number of duplicate segments. 209 Segments dropped later in the network: 211 How much would the larger initial window for TCP increase the 212 number of segments on congested links that would be dropped 213 before reaching their final destination? This is a problem that 214 can only occur for connections with multiple congested links, 215 where some segments might use scarce bandwidth on the first 216 congested link along the path, only to be dropped later along 217 the path. 219 First, many of the TCP connections will have only one congested 220 link along the path. Segments dropped from these connections do 221 not "waste" scarce bandwidth, and do not contribute to 222 congestion collapse. 224 However, some network paths will have multiple congested links, 225 and segments dropped from the initial window could use scarce 226 bandwidth along the earlier congested links before ultimately 227 being dropped on subsequent congested links. To the extent that 228 the drop rate is independent of the initial window used by TCP 229 segments, the problem of congested links carrying segments that 230 will be dropped before reaching their destination will be 231 similar for TCP connections that start by sending four segments 232 or one segment. 234 An increased packet drop rate: 236 For a network with a high segment drop rate, increasing the TCP 237 initial window could increase the segment drop rate even 238 further. This is in part because routers with Drop Tail queue 239 management have difficulties with bursty traffic in times of 240 congestion. However, given uncorrelated arrivals for TCP 241 connections, the larger TCP initial window should not 242 significantly increase the segment drop rate. Simulation-based 243 explorations of these issues are discussed in Section 7.2. 245 These potential dangers for the network are explored in simulations 246 and experiments described in the section below. Our judgment is that 247 while there are dangers of congestion collapse in the current 248 Internet (see [FF98] for a discussion of the dangers of congestion 249 collapse from an increased deployment of UDP connections without 250 end-to-end congestion control), there is no such danger to the 251 network from increasing the TCP initial window to 4K bytes. 253 6. Interactions with the Retransmission Timer 255 Using a larger initial burst of data can exacerbate existing 256 problems with spurious retransmit timeouts on low-bandwidth paths, 257 assuming the standard algorithm for determining the TCP 258 retransmission timeout (RTO) [RFC2988]. The problem is that across 259 low-bandwidth network paths on which the transmission time of a 260 packet is a large portion of the round-trip time, the small packets 261 used to establish a TCP connection do not seed the RTO estimator appropriately. 262 When the first window of data packets is transmitted, the sender's 263 retransmit timer could expire before the acknowledgments for those 264 packets are received. As each acknowledgment arrives, the 265 retransmit timer is generally reset. Thus, the retransmit timer 266 will not expire as long as an acknowledgment arrives at least once 267 a second, given the one-second minimum on the RTO recommended in RFC 268 2988. 270 For instance, consider a 9.6 Kbps link. The initial RTT measurement 271 will be on the order of 67 msec, if we simply consider the 272 transmission time of 2 packets (the SYN and SYN-ACK) each consisting 273 of 40 bytes. Using the RTO estimator given in [RFC2988], this 274 yields an initial RTO of 201 msec (67 + 4*(67/2)). However, we 275 round the RTO to 1 second as specified in RFC 2988. Then assume we 276 send an initial window of one or more 1500-byte packets (1460 data 277 bytes plus overhead). Each packet will take on the order of 1.25 278 seconds to transmit. Clearly the RTO will fire before the ACK for 279 the first packet returns, causing a spurious timeout. In this case, 280 a larger initial window of three or four packets exacerbates the 281 problems caused by this spurious timeout. 283 One way to deal with this problem is to make the RTO algorithm more 284 conservative. During the initial window of data, for instance, we 285 could update the RTO for each acknowledgment received. In 286 addition, if the retransmit timer expires for some packet lost in 287 the first window of data, we could leave the exponential-backoff of 288 the retransmit timer engaged until at least one valid RTT measurement is 289 received that involves a data packet. 291 Another method would be to refrain from taking a RTT sample during 292 connection establishment, leaving the default RTO in place until TCP 293 takes a sample from a data segment and the corresponding ACK. While 294 this method likely helps prevent spurious retransmits it also slows 295 the data transfer down if loss occurs before the RTO is seeded. 297 This specification leaves the decision about what to do (if 298 anything) with regards to the RTO when using a larger initial window 299 to the implementer. 301 7. Typical Levels of Burstiness for TCP Traffic. 303 Larger TCP initial windows would not dramatically increase the 304 burstiness of TCP traffic in the Internet today, because such 305 traffic is already fairly bursty. Bursts of two and three segments 306 are already typical of TCP [Flo97]; A delayed ACK (covering two 307 previously unacknowledged segments) received during congestion 308 avoidance causes the congestion window to slide and two segments to 309 be sent. The same delayed ACK received during slow start causes the 310 window to slide by two segments and then be incremented by one 311 segment, resulting in a three-segment burst. While not necessarily 312 typical, bursts of four and five segments for TCP are not rare. 313 Assuming delayed ACKs, a single dropped ACK causes the subsequent 314 ACK to cover four previously unacknowledged segments. During 315 congestion avoidance this leads to a four-segment burst and during 316 slow start a five-segment burst is generated. 318 There are also changes in progress that reduce the performance 319 problems posed by moderate traffic bursts. One such change is the 320 deployment of higher-speed links in some parts of the network, where 321 a burst of 4K bytes can represent a small quantity of data. A 322 second change, for routers with sufficient buffering, is the 323 deployment of queue management mechanisms such as RED, which is 324 designed to be tolerant of transient traffic bursts. 326 8. Simulations and Experimental Results 328 8.1 Studies of TCP Connections using that Larger Initial Window 330 This section surveys simulations and experiments that have been used 331 to explore the effect of larger initial windows on TCP 332 connections. The first set of experiments 333 explores performance over satellite links. Larger initial windows 334 have been shown to improve performance of TCP connections over 335 satellite channels [All97b]. In this study, an initial window of 336 four segments (512 byte MSS) resulted in throughput improvements of 337 up to 30% (depending upon transfer size). [KAGT98] shows that the 338 use of larger initial windows results in a decrease in transfer time 339 in HTTP tests over the ACTS satellite system. A study involving 340 simulations of a large number of HTTP transactions over hybrid fiber 341 coax (HFC) indicates that the use of larger initial windows 342 decreases the time required to load WWW pages [Nic97]. 344 A second set of experiments has explored TCP performance over dialup 345 modem links. In experiments over a 28.8 bps dialup channel [All97a, 346 AHO98], a four-segment initial window decreased the transfer time of 347 a 16KB file by roughly 10%, with no accompanying increase in the 348 drop rate. A particular area of concern has been TCP performance 349 over low speed tail circuits (e.g., dialup modem links) with routers 350 with small buffers. A simulation study [RFC2416] investigated the 351 effects of using a larger initial window on a host connected by a 352 slow modem link and a router with a 3 packet buffer. The study 353 concluded that for the scenario investigated, the use of larger 354 initial windows was not harmful to TCP performance. Questions have 355 been raised concerning the effects of larger initial windows on the 356 transfer time for short transfers in this environment, but these 357 effects have not been quantified. A question has also been raised 358 concerning the possible effect on existing TCP connections sharing 359 the link. 361 Finally, [All00] illustrates that the percentage of connections at a 362 particular web server that experience loss in the initial window of 363 data transmission increases with the size of the initial congestion 364 window. However, the increase is in line with what would be 365 expected from sending a larger burst into the network. 367 8.2 Studies of Networks using Larger Initial Windows 369 This section surveys simulations and experiments investigating the 370 impact of the larger window on other TCP connections sharing the 371 path. Experiments in [All97a, AHO98] show that for 16 KB transfers 372 to 100 Internet hosts, four-segment initial windows resulted in a 373 small increase in the drop rate of 0.04 segments/transfer. While 374 the drop rate increased slightly, the transfer time was reduced by 375 roughly 25% for transfers using the four-segment (512 byte MSS) 376 initial window when compared to an initial window of one segment. 378 One scenario of concern is heavily loaded links. For instance, 379 several years ago one of the trans-Atlantic links was so heavily 380 loaded that the correct congestion window size for each connection was 381 about one segment. In this environment, new connections using 382 larger initial windows would be starting with windows that were four 383 times too big. What would the effects be? Do connections thrash? 385 A simulation study in [RFC2415] explores the impact of a larger initial 386 window on competing network traffic. In this investigation, HTTP 387 and FTP flows share a single congested gateway (where the number of 388 HTTP and FTP flows varies from one simulation set to another). For 389 each simulation set, the paper examines aggregate link utilization 390 and packet drop rates, median web page delay, and network power for 391 the FTP transfers. The larger initial window generally resulted in 392 increased throughput, slightly-increased packet drop rates, and an 393 increase in overall network power. With the exception of one 394 scenario, the larger initial window resulted in an increase in the 395 drop rate of less than 1% above the loss rate experienced when using 396 a one-segment initial window; in this scenario, the drop rate 397 increased from 3.5% with one-segment initial windows, to 4.5% with 398 four-segment initial windows. The overall conclusions were that 399 increasing the TCP initial window to three packets (or 4380 bytes) 400 helps to improve perceived performance. 402 Morris [Mor97] investigated larger initial windows in a very 403 congested network with transfers of size 20K. The loss rate in 404 networks where all TCP connections use an initial window of four 405 segments is shown to be 1-2% greater than in a network where all 406 connections use an initial window of one segment. This relationship 407 held in scenarios where the loss rates with one-segment initial 408 windows ranged from 1% to 11%. In addition, in networks where 409 connections used an initial window of four segments, TCP connections 410 spent more time waiting for the retransmit timer (RTO) to expire to 411 resend a segment than was spent when using an initial window of one 412 segment. The time spent waiting for the RTO timer to expire 413 represents idle time when no useful work was being accomplished for 414 that connection. These results show that in a very congested 415 environment, where each connection's share of the bottleneck 416 bandwidth is close to one segment, using a larger initial window can 417 cause a perceptible increase in both loss rates and retransmit 418 timeouts. 420 9. Security Considerations 422 This document discusses the initial congestion window permitted for 423 TCP connections. Changing this value does not raise any known new 424 security issues with TCP. 426 10. Conclusion 428 This document specifies a small change to TCP that will likely be beneficial 429 to short-lived TCP connections and those over links with long RTTs 430 (saving several RTTs during the initial slow-start phase). 432 11. Acknowledgments 433 We would like to acknowledge Vern Paxson, Tim Shepard, members of 434 the End-to-End-Interest Mailing List, and members of the IETF TCP 435 Implementation Working Group for continuing discussions of these 436 issues for discussions and feedback on this document. 438 12. References 440 [AHO98] Mark Allman, Chris Hayes, and Shawn Ostermann, An Evaluation 441 of TCP with Larger Initial Windows, March 1998. Submitted to 442 ACM Computer Communication Review. URL: 443 "http://roland.lerc.nasa.gov/~mallman/papers/initwin.ps". 445 [All97a] Mark Allman. An Evaluation of TCP with Larger Initial 446 Windows. 40th IETF Meeting -- TCP Implementations WG. 447 December, 1997. Washington, DC. 449 [All97b] Mark Allman. Improving TCP Performance Over Satellite 450 Channels. Master's thesis, Ohio University, June 1997. 452 [All00] Mark Allman. A Web Server's View of the Transport Layer. ACM 453 Computer Communication Review, 30(5), October 2000. 455 [FF96] Fall, K., and Floyd, S., Simulation-based Comparisons of 456 Tahoe, Reno, and SACK TCP. Computer Communication Review, 457 26(3), July 1996. 459 [FF98] Sally Floyd, Kevin Fall. Promoting the Use of End-to-End 460 Congestion Control in the Internet. Submitted to IEEE 461 Transactions on Networking. URL "http://www- 462 nrg.ee.lbl.gov/floyd/end2end-paper.html". 464 [FJ93] Floyd, S., and Jacobson, V., Random Early Detection gateways 465 for Congestion Avoidance. IEEE/ACM Transactions on Networking, 466 V.1 N.4, August 1993, p. 397-413. 468 [Flo94] Floyd, S., TCP and Explicit Congestion Notification. 469 Computer Communication Review, 24(5):10-23, October 1994. 471 [Flo96] Floyd, S., Issues of TCP with SACK. Technical report, 472 January 1996. Available from http://www-nrg.ee.lbl.gov/floyd/. 474 [Flo97] Floyd, S., Increasing TCP's Initial Window. Viewgraphs, 475 40th IETF Meeting - TCP Implementations WG. December, 1997. URL 476 "ftp://ftp.ee.lbl.gov/talks/sf-tcp-ietf97.ps". 478 [KAGT98] Hans Kruse, Mark Allman, Jim Griner, Diepchi Tran. HTTP 479 Page Transfer Rates Over Geo-Stationary Satellite Links. March 480 1998. Proceedings of the Sixth International Conference on 481 Telecommunication Systems. URL 482 "http://roland.lerc.nasa.gov/~mallman/papers/nash98.ps". 484 [Mor97] Robert Morris. Private communication, 1997. Cited for 485 acknowledgement purposes only. 487 [Nic97] Kathleen Nichols. Improving Network Simulation with 488 Feedback. Com21, Inc. Technical Report. Available from 489 http://www.com21.com/pages/papers/068.pdf. 491 [Pos82] Postel, J., "Simple Mail Transfer Protocol", STD 10, RFC 492 821, August 1982. 494 [RFC1122] Braden, R., "Requirements for Internet Hosts -- 495 Communication Layers", STD 3, RFC 1122, October 1989. 497 [RFC1191] Mogul, J., and S. Deering, "Path MTU Discovery", RFC 1191, 498 November 1990. 500 [RFC1945] Berners-Lee, T., Fielding, R., and H. Nielsen, "Hypertext 501 Transfer Protocol -- HTTP/1.0", RFC 1945, May 1996. 503 [RFC2068] Fielding, R., Mogul, J., Gettys, J., Frystyk, H., and T. 504 Berners-Lee, "Hypertext Transfer Protocol -- HTTP/1.1", RFC 505 2068, January 1997. 507 [RFC2119] Bradner, S., "Key words for use in RFCs to Indicate 508 Requirement Levels", BCP 14, RFC 2119, March 1997. 510 [RFC2309] Braden, B., Clark, D., Crowcroft, J., Davie, B., Deering, 511 S., Estrin, D., Floyd, S., Jacobson, V., Minshall, G., 512 Partridge, C., Peterson, L., Ramakrishnan, K., Shenker, S., 513 Wroclawski, J., and L. Zhang, "Recommendations on Queue 514 Management and Congestion Avoidance in the Internet", RFC 2309, 515 April 1998. 517 [RFC2415] Poduri, K., and K. Nichols, "Simulation Studies of 518 Increased Initial TCP Window Size", RFC 2415, September 1998. 520 [RFC2416] Shepard, T., and C. Partridge, "When TCP Starts Up With 521 Four Packets Into Only Three Buffers", RFC 2416, September 1998. 523 [RFC2581] Mark Allman, Vern Paxson, W. Richard Stevens. TCP 524 Congestion Control, April 1999. RFC 2581. 526 [RFC2988] Vern Paxson, Mark Allman. Computing TCP's Retransmission 527 Timer, November 2000. RFC 2988. 529 [RFC3042] M. Allman, H. Balakrishnan, and S. Floyd, Enhancing TCP's 530 Loss Recovery Using Limited Transmit, RFC 3042, January 2001. 532 [RFC3168] Ramakrishnan, K.K., Floyd, S., and Black, D., "The 533 Addition of Explicit Congestion Notification (ECN) to IP", RFC 534 3168, September 2001. 536 13. Author's Addresses 538 Mark Allman 539 BBN Technologies/NASA Glenn Research Center 540 21000 Brookpark Road 541 MS 54-5 542 Cleveland, OH 44135 543 EMail: mallman@bbn.com 544 http://roland.lerc.nasa.gov/~mallman/ 546 Sally Floyd 547 ICSI Center for Internet Research 548 1947 Center St, Suite 600 549 Berkeley, CA 94704 550 Phone: +1 (510) 666-2989 551 EMail: floyd@icir.org 552 http://www.icir.org/floyd/ 554 Craig Partridge 555 BBN Technologies 556 10 Moulton Street 557 Cambridge, MA 02138 559 EMail: craig@bbn.com 561 13. Appendix - Duplicate Segments 563 In the current environment (without Explicit Congestion Notification 564 [Flo94] [RFC2481]), all TCPs use segment drops as indications from 565 the network about the limits of available bandwidth. We argue here 566 that the change to a larger initial window should not result in the 567 sender retransmitting a large number of duplicate segments that have 568 already arrived at the receiver. 570 If one segment is dropped from the initial window, there are three 571 different ways for TCP to recover: (1) Slow-starting from a window 572 of one segment, as is done after a retransmit timeout, or after Fast 573 Retransmit in Tahoe TCP; (2) Fast Recovery without selective 574 acknowledgments (SACK), as is done after three duplicate ACKs in 575 Reno TCP; and (3) Fast Recovery with SACK, for TCP where both the 576 sender and the receiver support the SACK option [MMFR96]. In all 577 three cases, if a single segment is dropped from the initial window, 578 no duplicate segments (i.e., segments that have already been 579 received at the receiver) are transmitted. Note that for a TCP 580 sending four 512-byte segments in the initial window, a single 581 segment drop will not require a retransmit timeout, but can be 582 recovered from using the Fast Retransmit algorithm (unless the 583 retransmit timer expires prematurely). In addition, a single 584 segment dropped from an initial window of three segments might be 585 repaired using the fast retransmit algorithm, depending on which 586 segment is dropped and whether or not delayed ACKs are used. For 587 example, dropping the first segment of a three segment initial 588 window will always require waiting for a timeout, in the absence of 589 Limited Transmit [RFC3042]. However, dropping the third segment 590 will always allow recovery via the fast retransmit algorithm, as 591 long as no ACKs are lost. 593 Next we consider scenarios where the initial window contains two to 594 four segments, and at least two of those segments are dropped. If 595 all segments in the initial window are dropped, then clearly no 596 duplicate segments are retransmitted, as the receiver has not yet 597 received any segments. (It is still a possibility that these 598 dropped segments used scarce bandwidth on the way to their drop 599 point; this issue was discussed in Section 5.) 601 When two segments are dropped from an initial window of three 602 segments, the sender will only send a duplicate segment if the first 603 two of the three segments were dropped, and the sender does not 604 receive a packet with the SACK option acknowledging the third 605 segment. 607 When two segments are dropped from an initial window of four 608 segments, an examination of the six possible scenarios (which we 609 don't go through here) shows that, depending on the position of the 610 dropped packets, in the absence of SACK the sender might send one 611 duplicate segment. There are no scenarios in which the sender sends 612 two duplicate segments. 614 When three segments are dropped from an initial window of four 615 segments, then, in the absence of SACK, it is possible that one 616 duplicate segment will be sent, depending on the position of the 617 dropped segments. 619 The summary is that in the absence of SACK, there are some scenarios 620 with multiple segment drops from the initial window where one 621 duplicate segment will be transmitted. There are no scenarios where 622 more that one duplicate segment will be transmitted. Our conclusion 623 is that the number of duplicate segments transmitted as a result of 624 a larger initial window should be small.