idnits 2.17.1 draft-rhee-tcpm-cubic-02.txt: Checking boilerplate required by RFC 5378 and the IETF Trust (see https://trustee.ietf.org/license-info): ---------------------------------------------------------------------------- ** It looks like you're using RFC 3978 boilerplate. You should update this to the boilerplate described in the IETF Trust License Policy document (see https://trustee.ietf.org/license-info), which is required now. -- Found old boilerplate from RFC 3978, Section 5.1 on line 18. -- Found old boilerplate from RFC 3978, Section 5.5, updated by RFC 4748 on line 635. -- Found old boilerplate from RFC 3979, Section 5, paragraph 1 on line 646. -- Found old boilerplate from RFC 3979, Section 5, paragraph 2 on line 653. -- Found old boilerplate from RFC 3979, Section 5, paragraph 3 on line 659. Checking nits according to https://www.ietf.org/id-info/1id-guidelines.txt: ---------------------------------------------------------------------------- No issues found here. Checking nits according to https://www.ietf.org/id-info/checklist : ---------------------------------------------------------------------------- No issues found here. Miscellaneous warnings: ---------------------------------------------------------------------------- == The copyright year in the IETF Trust Copyright Line does not match the current year -- The document seems to lack a disclaimer for pre-RFC5378 work, but may have content which was first submitted before 10 November 2008. If you have contacted all the original authors and they are all willing to grant the BCP78 rights to the IETF Trust, then this is fine, and you can ignore this comment. If not, you may need to add the pre-RFC5378 disclaimer. (See the Legal Provisions document at https://trustee.ietf.org/license-info for more information.) -- The document date (August 26, 2008) is 5721 days in the past. Is this intentional? Checking references for intended status: Experimental ---------------------------------------------------------------------------- ** Obsolete normative reference: RFC 2581 (Obsoleted by RFC 5681) ** Obsolete normative reference: RFC 3448 (Obsoleted by RFC 5348) ** Obsolete normative reference: RFC 3782 (Obsoleted by RFC 6582) ** Obsolete normative reference: RFC 4960 (Obsoleted by RFC 9260) Summary: 5 errors (**), 0 flaws (~~), 1 warning (==), 7 comments (--). Run idnits with the --verbose option for more detailed information about the items above. -------------------------------------------------------------------------------- 2 Network Working Group I. Rhee 3 Internet-Draft NCSU 4 Intended status: Experimental L. Xu 5 Expires: February 27, 2009 UNL 6 S. Ha 7 NCSU 8 August 26, 2008 10 CUBIC for Fast Long-Distance Networks 11 draft-rhee-tcpm-cubic-02.txt 13 Status of this Memo 15 By submitting this Internet-Draft, each author represents that any 16 applicable patent or other IPR claims of which he or she is aware 17 have been or will be disclosed, and any of which he or she becomes 18 aware will be disclosed, in accordance with Section 6 of BCP 79. 20 Internet-Drafts are working documents of the Internet Engineering 21 Task Force (IETF), its areas, and its working groups. Note that 22 other groups may also distribute working documents as Internet- 23 Drafts. 25 Internet-Drafts are draft documents valid for a maximum of six months 26 and may be updated, replaced, or obsoleted by other documents at any 27 time. It is inappropriate to use Internet-Drafts as reference 28 material or to cite them other than as "work in progress." 30 The list of current Internet-Drafts can be accessed at 31 http://www.ietf.org/ietf/1id-abstracts.txt. 33 The list of Internet-Draft Shadow Directories can be accessed at 34 http://www.ietf.org/shadow.html. 36 This Internet-Draft will expire on February 27, 2009. 38 Abstract 40 CUBIC is an extension to the current TCP standards. The protocol 41 differs from the current TCP standards only in the congestion window 42 adjustment function in the sender side. In particular, it uses a 43 cubic function instead of a linear window increase of the current TCP 44 standards to improve scalability and stability under fast and long 45 distance networks. BIC-TCP, a predecessor of CUBIC, has been a 46 default TCP adopted by Linux since year 2005 and has already been 47 deployed globally and in use for several years by the Internet 48 community at large. CUBIC is using a similar window growth function 49 as BIC-TCP and is designed to be less aggressive and fairer to TCP in 50 bandwidth usage than BIC-TCP while maintaining the strengths of BIC- 51 TCP such as stability, window scalability and RTT fairness. Through 52 extensive testing in various Internet scenarios, we believe that 53 CUBIC is safe for deployment and testing in the global Internet. The 54 intent of this document is to provide the protocol specification of 55 CUBIC for a third party implementation and solicit the community 56 feedback through experimentation on the performance of CUBIC. We 57 expect this document to be eventually published as an experimental 58 RFC. 60 Table of Contents 62 1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . 4 63 2. Conventions . . . . . . . . . . . . . . . . . . . . . . . . . 7 64 3. CUBIC Congestion Control . . . . . . . . . . . . . . . . . . . 8 65 3.1. Window growth function . . . . . . . . . . . . . . . . . . 8 66 3.2. TCP-friendly region . . . . . . . . . . . . . . . . . . . 8 67 3.3. Concave region . . . . . . . . . . . . . . . . . . . . . . 9 68 3.4. Convex region . . . . . . . . . . . . . . . . . . . . . . 9 69 3.5. Multiplicative decrease . . . . . . . . . . . . . . . . . 9 70 3.6. Fast convergence . . . . . . . . . . . . . . . . . . . . . 10 71 4. Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . 11 72 4.1. Fairness to standard TCP . . . . . . . . . . . . . . . . . 11 73 4.2. Using Spare Capacity . . . . . . . . . . . . . . . . . . . 13 74 4.3. Difficult Environments . . . . . . . . . . . . . . . . . . 14 75 4.4. Investigating a Range of Environments . . . . . . . . . . 14 76 4.5. Protection against Congestion Collapse . . . . . . . . . . 14 77 4.6. Fairness within the Alternative Congestion Control 78 Algorithm. . . . . . . . . . . . . . . . . . . . . . . . . 14 79 4.7. Performance with Misbehaving Nodes and Outside 80 Attackers . . . . . . . . . . . . . . . . . . . . . . . . 15 81 4.8. Responses to Sudden or Transient Events . . . . . . . . . 15 82 4.9. Incremental Deployment . . . . . . . . . . . . . . . . . . 15 83 5. Security Considerations . . . . . . . . . . . . . . . . . . . 16 84 6. IANA Considerations . . . . . . . . . . . . . . . . . . . . . 17 85 7. References . . . . . . . . . . . . . . . . . . . . . . . . . . 18 86 7.1. Normative References . . . . . . . . . . . . . . . . . . . 18 87 7.2. Informative References . . . . . . . . . . . . . . . . . . 18 88 Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . . 20 89 Intellectual Property and Copyright Statements . . . . . . . . . . 21 91 1. Introduction 93 The low utilization problem of TCP in fast long-distance networks is 94 well documented in [K03][RFC3649]. This problem arises from a slow 95 increase of congestion window following a congestion event in a 96 network with a large bandwidth delay product (BDP). Our experience 97 [H+06] indicates that this problem is frequently observed even in the 98 range of congestion window sizes over several hundreds of packets 99 (each packet is sized around 1000 bytes) especially under a network 100 path with over 100ms round-trip times (RTTs). This problem is 101 equally applicable to all Reno style TCP standards and their 102 variants, including TCP-RENO [RFC2581], TCP-NewReno [RFC3782], TCP- 103 SACK [RFC2018], SCTP [RFC4960], TFRC [RFC3448] that use the same 104 linear increase function for window growth, which we refer to 105 collectively as Standard TCP below. 107 CUBIC [H+08] is a modification to the congestion control mechanism of 108 Standard TCP, in particular, to the window increase function of 109 Standard TCP senders, to remedy this problem. It uses a cubic 110 increase function in terms of the elapsed time from the last 111 congestion event. While most alternative algorithms to Standard TCP 112 uses a convex increase function where after a loss event, the window 113 increment is always increasing, CUBIC uses both the concave and 114 convex profiles of a cubic function for window increase. After a 115 window reduction following a loss event, it registers the window size 116 where it got the loss event as W_max and performs a multiplicative 117 decrease of congestion window and the regular fast recovery and 118 retransmit of Standard TCP. After it enters into congestion 119 avoidance from fast recovery, it starts to increase the window using 120 the concave profile of the cubic function. The cubic function is set 121 to have its plateau at W_max so the concave growth continues until 122 the window size becomes W_max. After that, the cubic function turns 123 into a convex profile and the convex window growth begins. This 124 style of window adjustment (concave and then convex) improves 125 protocol and network stability while maintaining high network 126 utilization [C+07]. This is because the window size remains almost 127 constant, forming a plateau around W_max where network utilization is 128 deemed highest and under steady state, most window size samples of 129 CUBIC are close to W_max, thus promoting high network utilization and 130 protocol stability. Note that protocols with convex increase 131 functions have the maximum increments around W_max and introduces a 132 large number of packet bursts around the saturation point of the 133 network, likely causing frequent global loss synchronizations. 135 Another notable feature of CUBIC is that its window increase rate is 136 mostly independent of RTT, and follows a (cubic) function of the 137 elapsed time since the last loss event. This feature promotes per- 138 flow fairness to Standard TCP as well as RTT-fairness. Note that 139 Standard TCP performs well under short RTT and small bandwidth (or 140 small BDP) networks. Only in a large long RTT and large bandwidth 141 (or large BDP) networks, it has the scalability problem. An 142 alternative protocol to Standard TCP designed to be friendly to 143 Standard TCP at a per-flow basis must operate must increase its 144 window much less aggressively in small BDP networks than in large BDP 145 networks. In CUBIC, its window growth rate is slowest around the 146 inflection point of the cubic function and this function does not 147 depend on RTT. In a smaller BDP network where Standard TCP flows are 148 working well, the absolute amount of the window decrease at a loss 149 event is always smaller because of the multiplicative decrease. 150 Therefore, in CUBIC, the starting window size after a loss event from 151 which the window starts to increase, is smaller in a smaller BDP 152 network, thus falling nearer to the plateau of the cubic function 153 where the growth rate is slowest. By setting appropriate values of 154 the cubic function parameters, CUBIC sets its growth rate always no 155 faster than Standard TCP around its inflection point. When the cubic 156 function grows slower than the window of Standard TCP, CUBIC simply 157 follows the window size of Standard TCP to ensure fairness to 158 Standard TCP in a small BDP network. We call this region where CUBIC 159 behaves like Standard TCP, the TCP-friendly region. 161 CUBIC maintains the same window growth rate independent of RTTs 162 outside of the TCP-friendly region, and flows with different RTTs 163 have the similar window sizes under steady state when they operate 164 outside the TCP-friendly region. This ensures CUBIC flows with 165 different RTTs to have their bandwidth shares linearly proportional 166 to the inverse of their RTT ratio (the longer RTT, the smaller the 167 share). This behavior is the same as that of Standard TCP under high 168 statistical multiplexing environments where packet losses are 169 independent of individual flow rates. However, under low statistical 170 multiplexing environments, the bandwidth share ratio of Standard TCP 171 flows with different RTTs is squarely proportional to the inverse of 172 their RTT ratio [XHR04]. CUBIC always ensures the linear ratio 173 independent of the levels of statistical multiplexing. This is an 174 improvement over Standard TCP. While there is no consensus on a 175 particular bandwidth share ratios of different RTT flows, we believe 176 that under wired Internet, use of the linear share notion seems more 177 reasonable than equal share or a higher order shares. HTCP [LS08] 178 currently uses the equal share. 180 CUBIC sets the multiplicative window decrease factor to 0.2 while 181 Standard TCP uses 0.5. While this improves the scalability of the 182 protocol, a side effect of this decision is slower convergence 183 especially under low statistical multiplexing environments. This 184 design choice is following the observation that the author of HSTCP 185 [RFC3649] has made along with other researchers (e.g., [GV02]): the 186 current Internet becomes more asynchronous with less frequent loss 187 synchronizations with high statistical multiplexing. Under this 188 environment, even strict MIMD can converge. CUBIC flows with the 189 same RTT always converge to the same share of bandwidth independent 190 of statistical multiplexing, thus achieving intra-protocol fairness. 191 We also find that under the environments with sufficient statistical 192 multiplexing, the convergence speed of CUBIC flows is reasonable. 194 In the ensuing sections, we provide the exact specification of CUBIC 195 and discuss the safety features of CUBIC following the guidelines 196 specified in [RFC5033]. 198 2. Conventions 200 The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", 201 "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this 202 document are to be interpreted as described in [RFC2119]. 204 3. CUBIC Congestion Control 206 3.1. Window growth function 208 CUBIC maintains the acknowledgment (ACK) clocking of Standard TCP by 209 increasing congestion window only at the reception of ACK. The 210 protocol does not make any change to the fast recovery and retransmit 211 of TCP-NewReno [RFC3782] and TCP-SACK [RFC2018]. During congestion 212 avoidance after fast recovery, CUBIC changes the window update 213 algorithm of Standard TCP. Suppose that W_max is the window size 214 before the window is reduced in the last fast retransmit and 215 recovery. 217 The window growth function of CUBIC uses the following function: 219 W(t) = C*(t-K)^3 + W_max (Eq. 1) 221 where C is a constant fixed to determine the aggressiveness of window 222 growth in high BDP networks, t is the elapsed time from the last 223 window reduction,and K is the time period that the above function 224 takes to increase W to W_max when there is no further loss event and 225 is calculated by using the following equation: 227 K = cubic_root(W_max*beta/C) (Eq. 2) 229 where beta is the multiplication decrease factor. We discuss how we 230 set C in the next Section in more details. 232 Upon receiving an ACK during congestion avoidance, CUBIC computes the 233 window growth rate during the next RTT period using Eq. 1. It sets 234 W(t+RTT) as the candidate target value of congestion window. Suppose 235 that the current window size is cwnd. Depending on the value of 236 cwnd, CUBIC runs in three different modes. First, if cwnd is less 237 than the window size that Standard TCP would reach at time t after 238 the last loss event, then CUBIC is in the TCP friendly region (we 239 describe below how to determine this window size of Standard TCP in 240 term of time t). Otherwise, if cwnd is less than W_max, then CUBIC 241 is the concave region, and if cwnd is larger than W_max, CUBIC is in 242 the convex region. Below, we describe the exact actions taken by 243 CUBIC in each region. 245 3.2. TCP-friendly region 247 When receiving an ACK in congestion avoidance, we first check whether 248 the protocol is in the TCP region or not. This is done as follows. 249 We can analyze the window size of Standard TCP in terms of the 250 elapsed time t. Using a simple analysis in [FHP00], we can analyze 251 the average window size of additive increase and multiplicative 252 decrease (AIMD) with an additive factor alpha and a multiplicative 253 factor beta to be the following function: 255 (alpha/2 * (2-beta)/beta * 1/p)^0.5 (Eq. 3) 257 By the same analysis, the average window size of Standard TCP with 258 alpha 1 and beta 0.5 is (3/2 *1/p)^0.5. Thus, for Eq. 3 to be the 259 same as that of Standard TCP, alpha must be equal to 3*beta/(2-beta). 260 As Standard TCP increases its window by alpha per RTT, we can get the 261 window size of Standard TCP in terms of the elapsed time t as 262 follows: 264 W_tcp(t) = W_max*(1-beta) + 3*beta/(2-beta)* t/RTT (Eq. 4) 266 If cwnd is less than W_tcp(t), then the protocol is in the TCP 267 friendly region and cwnd SHOULD be set to W_tcp(t) at each reception 268 of ACK. 270 3.3. Concave region 272 When receiving an ACK in congestion avoidance, if the protocol is not 273 in the TCP-friendly region and cwnd is less than W_max, then the 274 protocol is in the concave region. In this region, cwnd MUST be 275 incremented by (W(t+RTT) - cwnd)/cwnd. 277 3.4. Convex region 279 When the window size of CUBIC is larger than W_max, it passes the 280 plateau of the cubic function after which CUBIC follows the convex 281 profile of the cubic function. Since cwnd is larger than the 282 previous saturation point W_max, this indicates that the network 283 conditions might have been perturbed since the last loss event, 284 possibly implying more available bandwidth after some flow 285 departures. Since the Internet is highly asynchronous, some amount 286 of perturbation is always possible without causing a major change in 287 available bandwidth. In this phase, CUBIC is being very careful by 288 very slowly increasing its window size. The convex profile ensures 289 that the window increases very slowly at the beginning and gradually 290 increases its growth rate. We also call this phase as the maximum 291 probing phase since CUBIC is searching for a new W_max. In this 292 region, cwnd MUST be incremented by (W(t+RTT) - cwnd)/cwnd for each 293 received ACK. 295 3.5. Multiplicative decrease 297 When a packet loss occurs, CUBIC reduces its window size by a factor 298 of beta. Parameter beta SHOULD be set to 0.2. 300 W_max = cwnd; // remember the window size before reduction 301 cwnd = cwnd * (1-beta); // window reduction 303 Figure 1 305 A side effect of setting beta to a smaller value than 0.5 is slower 306 convergence. We believe that while a more adaptive setting of beta 307 could result in faster convergence, it will make the analysis of the 308 protocol much harder. This adaptive adjustment of beta is an item 309 for the next version of CUBIC. 311 3.6. Fast convergence 313 To improve the convergence speed of CUBIC, we add a heuristic in the 314 protocol. When a new flow joins the network, existing flows in the 315 network need to give up their bandwidth shares to allow the flow soem 316 room for growth if the existing flows have been using all the 317 bandwidth of the network. To increase this release of bandwidth by 318 existing flows, the following mechanism called fast convergence 319 SHOULD be implemented. 321 With fast convergence, when a loss event occurs, before a window 322 reduction of congestion window, a flow remembers the last value of 323 W_max before it updates W_max for the current loss event. Let us 324 call the last value of W_max to be W_last_max. 326 if (W_max < W_last_max){ // check downward trend, 327 W_last_max = W_max; // remember the last W_max. 328 W_max = W_max*(2-beta)/2; // further reduce W_max. 329 } else // check upward trend. 330 W_last_max = W_max // remember the last W_max. 332 Figure 2 334 This allows W_max to be slightly less than the original W_max. Since 335 flows spend most of time around their W_max, flows with larger 336 bandwidth shares tend to spend more time around the plateau allowing 337 more time for flows with smaller shares to increase their windows. 339 4. Discussion 341 With a deterministic loss model where the number of packets between 342 two successive lost events is always 1/p, CUBIC always operates with 343 the concave window profile which greatly simplifies the performance 344 analysis of CUBIC. The average window size of CUBIC can be obtained 345 by the following function: 347 (C*(4-beta)/4/beta)^0.25 * RTT^0.75 / p^0.75 (Eq. 5) 349 With beta set to 0.2, the above formula is reduced to: 351 (C*3.8/0.8)^0.25 * RTT^0.75 / p^0.75 (Eq. 6) 353 We will determine the value of C in the following subsection using 354 Eq. 6. 356 4.1. Fairness to standard TCP 358 In environments where standard TCP is able to make reasonable use of 359 the available bandwidth, CUBIC does not significantly change this 360 state. 362 Standard TCP performs well in the following two types of networks: 364 1. networks with a small bandwidth-delay product (BDP). 366 2. networks with a short RTT, but not necessarily a small BDP 368 CUBIC is designed to behave very similarly to standard TCP in the 369 above two types of networks. The following two tables show the 370 average window size of standard TCP, HSTCP, and CUBIC. The average 371 window size of standard TCP and HSTCP is from [RFC3649]. The average 372 window size of CUBIC is calculated by using Eq. 6 and CUBIC TCP 373 friendly mode for three different values of C. 375 +----------+-------+--------+-------------+-------------+-----------+ 376 | Loss | TCP | HSTCP | CUBIC | CUBIC | CUBIC | 377 | Rate P | | | (C=0.04) | (C=0.4) | (C=4) | 378 +----------+-------+--------+-------------+-------------+-----------+ 379 | 10^-2 | 12 | 12 | 12 | 12 | 12 | 380 | | | | | | | 381 | 10^-3 | 38 | 38 | 38 | 38 | 66 | 382 | | | | | | | 383 | 10^-4 | 120 | 263 | 120 | 209 | 371 | 384 | | | | | | | 385 | 10^-5 | 379 | 1795 | 660 | 1174 | 2087 | 386 | | | | | | | 387 | 10^-6 | 1200 | 12279 | 3713 | 6602 | 11740 | 388 | | | | | | | 389 | 10^-7 | 3795 | 83981 | 20878 | 37126 | 66022 | 390 | | | | | | | 391 | 10^-8 | 12000 | 574356 | 117405 | 208780 | 371269 | 392 +----------+-------+--------+-------------+-------------+-----------+ 394 Response function of standard TCP, HSTCP, and CUBIC in networks with 395 RTT = 100ms. The average window size W is in MSS-sized segments. 397 Table 1 399 +--------+-----------+-----------+------------+-----------+---------+ 400 | Loss | Average | Average | CUBIC | CUBIC | CUBIC | 401 | Rate P | TCP W | HSTCP W | (C=0.04) | (C=0.4) | (C=4) | 402 +--------+-----------+-----------+------------+-----------+---------+ 403 | 10^-2 | 12 | 12 | 12 | 12 | 12 | 404 | | | | | | | 405 | 10^-3 | 38 | 38 | 38 | 38 | 38 | 406 | | | | | | | 407 | 10^-4 | 120 | 263 | 120 | 120 | 120 | 408 | | | | | | | 409 | 10^-5 | 379 | 1795 | 379 | 379 | 379 | 410 | | | | | | | 411 | 10^-6 | 1200 | 12279 | 1200 | 1200 | 2087 | 412 | | | | | | | 413 | 10^-7 | 3795 | 83981 | 3795 | 6603 | 11740 | 414 | | | | | | | 415 | 10^-8 | 12000 | 574356 | 20878 | 37126 | 66022 | 416 +--------+-----------+-----------+------------+-----------+---------+ 418 Response function of standard TCP, HSTCP, and CUBIC in networks with 419 RTT = 10ms. The average window size W is in MSS-sized segments. 421 Table 2 423 Both tables show that CUBIC with any of these three C values is more 424 friendly to TCP than HSTCP, especially in networks with a short RTT 425 where TCP performs reasonably well. For example, in a network with 426 RTT = 10ms and p=10^-6, TCP has an average window of 1200 packets. 427 If the packet size is 1500 bytes, then TCP can achieve an average 428 rate of 1.44 Gbps. In this case, CUBIC with C=0.04 or C=0.4 achieves 429 exactly the same rate as Standard TCP, whereas HSTCP is about ten 430 times more aggressive than Standard TCP. 432 We can see that C determines the aggressiveness of CUBIC in competing 433 with other protocols for the bandwidth. CUBIC is more friendly to 434 the Standard TCP, if the value of C is lower. However, we do not 435 recommend to set C to a very low value like 0.04, since CUBIC with a 436 low C cannot efficiently use the bandwidth in long RTT and high 437 bandwidth networks. Based on these observations, we find C=0.4 gives 438 a good balance between TCP-friendliness and aggressiveness of window 439 growth. Therefore, C SHOULD be set to 0.4. With C set to 0.4, Eq. 6 440 is reduced to: 442 1.17 * RTT^0.75 / p^0.75 (Eq. 7) 444 Eq. 7 is then used in the next subsection to show the scalability of 445 CUBIC. 447 4.2. Using Spare Capacity 449 CUBIC uses a more aggressive window growth function than Standard TCP 450 under long RTT and high bandwidth networks. 452 The following table shows that to achieve 10Gbps rate, standard TCP 453 requires a packet loss rate of 2.0e-10, while CUBIC requires a packet 454 loss rate of 3.4e-8. 456 +------------------+-----------+---------+---------+---------+ 457 | Throughput(Mbps) | Average W | TCP P | HSTCP P | CUBIC P | 458 +------------------+-----------+---------+---------+---------+ 459 | 1 | 8.3 | 2.0e-2 | 2.0e-2 | 2.0e-2 | 460 | | | | | | 461 | 10 | 83.3 | 2.0e-4 | 3.9e-4 | 3.3e-4 | 462 | | | | | | 463 | 100 | 833.3 | 2.0e-6 | 2.5e-5 | 1.6e-5 | 464 | | | | | | 465 | 1000 | 8333.3 | 2.0e-8 | 1.5e-6 | 7.3e-7 | 466 | | | | | | 467 | 10000 | 83333.3 | 2.0e-10 | 1.0e-7 | 3.4e-8 | 468 +------------------+-----------+---------+---------+---------+ 470 Required packet loss rate for Standard TCP, HSTCP, and CUBIC to 471 achieve a certain throughput. We use 1500-Byte Packets and a Round- 472 Trip Time of 0.1 Seconds. 474 Table 3 476 Our test results in [H+06] indicate that CUBIC uses the spare 477 bandwidth left unused by existing Standard TCP flows in the same 478 bottleneck link without taking away much bandwidth from the existing 479 flows. 481 4.3. Difficult Environments 483 CUBIC is designed to remedy the poor performance of TCP in fast long- 484 distance networks. It is not designed for wireless networks. 486 4.4. Investigating a Range of Environments 488 CUBIC has been extensively studied by using both NS-2 simulation and 489 test-bed experiments covering a wide range of network environments. 490 More information can be found in [H+06]. 492 4.5. Protection against Congestion Collapse 494 In case that there is congestion collapse, CUBIC behaves likely 495 standard TCP since CUBIC modifies only the window adjustment 496 algorithm of TCP. Thus, it does not modify the ACK clocking and 497 Timeout behaviors of Standard TCP. 499 4.6. Fairness within the Alternative Congestion Control Algorithm. 501 CUBIC ensures convergence of competing CUBIC flows with the same RTT 502 in the same bottleneck links to an equal bandwidth share. When 503 competing flows have different RTTs, their bandwidth shares are 504 linearly proportional to the invese of their RTT ratios. This is 505 true independent of the level of stastistical multiplexing in the 506 link. 508 4.7. Performance with Misbehaving Nodes and Outside Attackers 510 This is not considered in the current CUBIC. 512 4.8. Responses to Sudden or Transient Events 514 In case that there is a sudden congestion, a routing change, or a 515 mobility event, CUBIC behaves the same as Standard TCP. 517 4.9. Incremental Deployment 519 CUBIC requires only the change of TCP senders, and does not require 520 any assistant of routers. 522 5. Security Considerations 524 This proposal makes no changes to the underlying security of TCP. 526 6. IANA Considerations 528 There are no IANA considerations regarding this document. 530 7. References 532 7.1. Normative References 534 [RFC2018] Mathis, M., Mahdavi, J., Floyd, S., and A. Romanow, "TCP 535 Selective Acknowledgment Options", RFC 2018, October 1996. 537 [RFC2119] Bradner, S., "Key words for use in RFCs to Indicate 538 Requirement Levels", RFC 2119, March 1997. 540 [RFC2581] Allman, M., Paxson, V., and W. Stevens, "TCP Congestion 541 Control", RFC 2581, April 1999. 543 [RFC3448] Handley, M., Floyd, S., Padhye, J., and J. Widmer, "TCP 544 Friendly Rate Control (TFRC): Protocol Specification", 545 RFC 3448, January 2003. 547 [RFC3649] Floyd, S., "HighSpeed TCP for Large Congestion Windows", 548 RFC 3649, December 2003. 550 [RFC3782] Floyd, S., Henderson, T., and A. Gurtov, "The NewReno 551 Modification to TCP's Fast Recovery Algorithm", RFC 3782, 552 April 2004. 554 [RFC4960] Stewart, R., "Stream Control Transmission Protocol", 555 RFC 4960, September 2007. 557 [RFC5033] Floyd, S. and M. Allman, "Specifying New Congestion 558 Control Algorithms", RFC 5033, August 2007. 560 7.2. Informative References 562 [C+07] Cai, H., Eun, D., Ha, S., Rhee, I., and L. Xu, "Stochastic 563 Ordering for Internet Congestion Control and its 564 Applications", In Proceedings of IEEE INFOCOM , May 2007. 566 [FHP00] Floyd, S., Handley, M., and J. Padhye, "A Comparison of 567 Equation-Based and AIMD Congestion Control", May 2000. 569 [GV02] Gorinsky, S. and H. Vin, "Extended Analysis of Binary 570 Adjustment Algorithms", Technical Report TR2002-29, 571 Department of Computer Sciences , The University of Texas 572 at Austin , August 2002. 574 [H+06] Ha, S., Kim, Y., Le, L., Rhee, I., and L. Xu, "A Step 575 toward Realistic Performance Evaluation of High-Speed TCP 576 Variants", International Workshop on Protocols for Fast 577 Long-Distance Networks , February 2006. 579 [H+08] Ha, S., Rhee, I., and L. Xu, "CUBIC: A New TCP-Friendly 580 High-Speed TCP Variant", ACM SIGOPS Operating System 581 Review , 2008. 583 [K03] Kelly, T., "Scalable TCP: Improving Performance in 584 HighSpeed Wide Area Networks", ACM SIGCOMM Computer 585 Communication Review , April 2003. 587 [LS08] Leith, D. and R. Shorten, "H-TCP: TCP Congestion Control 588 for High Bandwidth-Delay Product Paths", Internet-draft 589 draft-leith-tcp-htcp-06 , April 2008. 591 [XHR04] Xu, L., Harfoush, K., and I. Rhee, "Binary Increase 592 Congestion Control for Fast, Long Distance Networks", In 593 Proceedings of IEEE INFOCOM , March 2004. 595 Authors' Addresses 597 Injong Rhee 598 North Carolina State University 599 Department of Computer Science 600 Raleigh, NC 27695-7534 601 US 603 Email: rhee@ncsu.edu 605 Lisong Xu 606 University of Nebraska-Lincoln 607 Department of Computer Science and Engineering 608 Lincoln, NE 68588-0115 609 US 611 Email: xu@cse.unl.edu 613 Sangtae Ha 614 North Carolina State University 615 Department of Computer Science 616 Raleigh, NC 27695-7534 617 US 619 Email: sha2@ncsu.edu 621 Full Copyright Statement 623 Copyright (C) The IETF Trust (2008). 625 This document is subject to the rights, licenses and restrictions 626 contained in BCP 78, and except as set forth therein, the authors 627 retain all their rights. 629 This document and the information contained herein are provided on an 630 "AS IS" basis and THE CONTRIBUTOR, THE ORGANIZATION HE/SHE REPRESENTS 631 OR IS SPONSORED BY (IF ANY), THE INTERNET SOCIETY, THE IETF TRUST AND 632 THE INTERNET ENGINEERING TASK FORCE DISCLAIM ALL WARRANTIES, EXPRESS 633 OR IMPLIED, INCLUDING BUT NOT LIMITED TO ANY WARRANTY THAT THE USE OF 634 THE INFORMATION HEREIN WILL NOT INFRINGE ANY RIGHTS OR ANY IMPLIED 635 WARRANTIES OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE. 637 Intellectual Property 639 The IETF takes no position regarding the validity or scope of any 640 Intellectual Property Rights or other rights that might be claimed to 641 pertain to the implementation or use of the technology described in 642 this document or the extent to which any license under such rights 643 might or might not be available; nor does it represent that it has 644 made any independent effort to identify any such rights. Information 645 on the procedures with respect to rights in RFC documents can be 646 found in BCP 78 and BCP 79. 648 Copies of IPR disclosures made to the IETF Secretariat and any 649 assurances of licenses to be made available, or the result of an 650 attempt made to obtain a general license or permission for the use of 651 such proprietary rights by implementers or users of this 652 specification can be obtained from the IETF on-line IPR repository at 653 http://www.ietf.org/ipr. 655 The IETF invites any interested party to bring to its attention any 656 copyrights, patents or patent applications, or other proprietary 657 rights that may cover technology that may be required to implement 658 this standard. Please address the information to the IETF at 659 ietf-ipr@ietf.org.