idnits 2.17.1 draft-ietf-rmcat-sbd-03.txt: Checking boilerplate required by RFC 5378 and the IETF Trust (see https://trustee.ietf.org/license-info): ---------------------------------------------------------------------------- No issues found here. Checking nits according to https://www.ietf.org/id-info/1id-guidelines.txt: ---------------------------------------------------------------------------- No issues found here. Checking nits according to https://www.ietf.org/id-info/checklist : ---------------------------------------------------------------------------- No issues found here. Miscellaneous warnings: ---------------------------------------------------------------------------- == The copyright year in the IETF Trust and authors Copyright Line does not match the current year == Line 80 has weird spacing: '...onse of the...' == Line 81 has weird spacing: '...onse of the...' -- The document date (October 19, 2015) is 3111 days in the past. Is this intentional? Checking references for intended status: Experimental ---------------------------------------------------------------------------- == Unused Reference: 'ITU-Y1540' is defined on line 857, but no explicit reference was found in the text == Unused Reference: 'RFC5481' is defined on line 880, but no explicit reference was found in the text == Outdated reference: A later version (-05) exists of draft-welzl-rmcat-coupled-cc-04 Summary: 0 errors (**), 0 flaws (~~), 6 warnings (==), 1 comment (--). Run idnits with the --verbose option for more detailed information about the items above. -------------------------------------------------------------------------------- 2 RTP Media Congestion Avoidance Techniques D. Hayes, Ed. 3 Internet-Draft University of Oslo 4 Intended status: Experimental S. Ferlin 5 Expires: April 21, 2016 Simula Research Laboratory 6 M. Welzl 7 K. Hiorth 8 University of Oslo 9 October 19, 2015 11 Shared Bottleneck Detection for Coupled Congestion Control for RTP 12 Media. 13 draft-ietf-rmcat-sbd-03 15 Abstract 17 This document describes a mechanism to detect whether end-to-end data 18 flows share a common bottleneck. It relies on summary statistics 19 that are calculated by a data receiver based on continuous 20 measurements and regularly fed to a grouping algorithm that runs 21 wherever the knowledge is needed. This mechanism complements the 22 coupled congestion control mechanism in draft-welzl-rmcat-coupled-cc. 24 Status of This Memo 26 This Internet-Draft is submitted in full conformance with the 27 provisions of BCP 78 and BCP 79. 29 Internet-Drafts are working documents of the Internet Engineering 30 Task Force (IETF). Note that other groups may also distribute 31 working documents as Internet-Drafts. The list of current Internet- 32 Drafts is at http://datatracker.ietf.org/drafts/current/. 34 Internet-Drafts are draft documents valid for a maximum of six months 35 and may be updated, replaced, or obsoleted by other documents at any 36 time. It is inappropriate to use Internet-Drafts as reference 37 material or to cite them other than as "work in progress." 39 This Internet-Draft will expire on April 21, 2016. 41 Copyright Notice 43 Copyright (c) 2015 IETF Trust and the persons identified as the 44 document authors. All rights reserved. 46 This document is subject to BCP 78 and the IETF Trust's Legal 47 Provisions Relating to IETF Documents 48 (http://trustee.ietf.org/license-info) in effect on the date of 49 publication of this document. Please review these documents 50 carefully, as they describe your rights and restrictions with respect 51 to this document. Code Components extracted from this document must 52 include Simplified BSD License text as described in Section 4.e of 53 the Trust Legal Provisions and are provided without warranty as 54 described in the Simplified BSD License. 56 Table of Contents 58 1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . 3 59 1.1. The signals . . . . . . . . . . . . . . . . . . . . . . . 3 60 1.1.1. Packet Loss . . . . . . . . . . . . . . . . . . . . . 3 61 1.1.2. Packet Delay . . . . . . . . . . . . . . . . . . . . 3 62 1.1.3. Path Lag . . . . . . . . . . . . . . . . . . . . . . 4 63 2. Definitions . . . . . . . . . . . . . . . . . . . . . . . . . 4 64 2.1. Parameters and their Effect . . . . . . . . . . . . . . . 6 65 2.2. Recommended Parameter Values . . . . . . . . . . . . . . 7 66 3. Mechanism . . . . . . . . . . . . . . . . . . . . . . . . . . 7 67 3.1. Key metrics and their calculation . . . . . . . . . . . . 9 68 3.1.1. Mean delay . . . . . . . . . . . . . . . . . . . . . 9 69 3.1.2. Skewness Estimate . . . . . . . . . . . . . . . . . . 9 70 3.1.3. Variability Estimate . . . . . . . . . . . . . . . . 10 71 3.1.4. Oscillation Estimate . . . . . . . . . . . . . . . . 11 72 3.1.5. Packet loss . . . . . . . . . . . . . . . . . . . . . 11 73 3.2. Flow Grouping . . . . . . . . . . . . . . . . . . . . . . 12 74 3.2.1. Flow Grouping Algorithm . . . . . . . . . . . . . . . 12 75 3.2.2. Using the flow group signal . . . . . . . . . . . . . 13 76 3.3. Removing Noise from the Estimates . . . . . . . . . . . . 13 77 3.3.1. Oscillation noise . . . . . . . . . . . . . . . . . . 14 78 3.3.2. Clock skew . . . . . . . . . . . . . . . . . . . . . 14 79 3.4. Reducing lag and Improving Responsiveness . . . . 14 80 3.4.1. Improving the response of the skewness estimate . 15 81 3.4.2. Improving the response of the variability estimate 17 82 4. Measuring OWD . . . . . . . . . . . . . . . . . . . . . . . . 17 83 4.1. Time stamp resolution . . . . . . . . . . . . . . . . . . 17 84 5. Implementation status . . . . . . . . . . . . . . . . . . . . 18 85 6. Acknowledgements . . . . . . . . . . . . . . . . . . . . . . 18 86 7. IANA Considerations . . . . . . . . . . . . . . . . . . . . . 18 87 8. Security Considerations . . . . . . . . . . . . . . . . . . . 18 88 9. Change history . . . . . . . . . . . . . . . . . . . . . . . 18 89 10. References . . . . . . . . . . . . . . . . . . . . . . . . . 19 90 10.1. Normative References . . . . . . . . . . . . . . . . . . 19 91 10.2. Informative References . . . . . . . . . . . . . . . . . 19 92 Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . 20 94 1. Introduction 96 In the Internet, it is not normally known if flows (e.g., TCP 97 connections or UDP data streams) traverse the same bottlenecks. Even 98 flows that have the same sender and receiver may take different paths 99 and share a bottleneck or not. Flows that share a bottleneck link 100 usually compete with one another for their share of the capacity. 101 This competition has the potential to increase packet loss and 102 delays. This is especially relevant for interactive applications 103 that communicate simultaneously with multiple peers (such as multi- 104 party video). For RTP media applications such as RTCWEB, 105 [I-D.welzl-rmcat-coupled-cc] describes a scheme that combines the 106 congestion controllers of flows in order to honor their priorities 107 and avoid unnecessary packet loss as well as delay. This mechanism 108 relies on some form of Shared Bottleneck Detection (SBD); here, a 109 measurement-based SBD approach is described. 111 1.1. The signals 113 The current Internet is unable to explicitly inform endpoints as to 114 which flows share bottlenecks, so endpoints need to infer this from 115 whatever information is available to them. The mechanism described 116 here currently utilises packet loss and packet delay, but is not 117 restricted to these. 119 1.1.1. Packet Loss 121 Packet loss is often a relatively rare signal. Therefore, on its own 122 it is of limited use for SBD, however, it is a valuable supplementary 123 measure when it is more prevalent. 125 1.1.2. Packet Delay 127 End-to-end delay measurements include noise from every device along 128 the path in addition to the delay perturbation at the bottleneck 129 device. The noise is often significantly increased if the round-trip 130 time is used. The cleanest signal is obtained by using One-Way-Delay 131 (OWD). 133 Measuring absolute OWD is difficult since it requires both the sender 134 and receiver clocks to be synchronised. However, since the 135 statistics being collected are relative to the mean OWD, a relative 136 OWD measurement is sufficient. Clock skew is not usually significant 137 over the time intervals used by this SBD mechanism (see [RFC6817] A.2 138 for a discussion on clock skew and OWD measurements). However, in 139 circumstances where it is significant, Section 3.3.2 outlines a way 140 of adjusting the calculations to cater for it. 142 Each packet arriving at the bottleneck buffer may experience very 143 different queue lengths, and therefore different waiting times. A 144 single OWD sample does not, therefore, characterize the path well. 145 However, multiple OWD measurements do reflect the distribution of 146 delays experienced at the bottleneck. 148 1.1.3. Path Lag 150 Flows that share a common bottleneck may traverse different paths, 151 and these paths will often have different base delays. This makes it 152 difficult to correlate changes in delay or loss. This technique uses 153 the long term shape of the delay distribution as a base for 154 comparison to counter this. 156 2. Definitions 158 The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", 159 "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this 160 document are to be interpreted as described in RFC 2119 [RFC2119]. 162 Acronyms used in this document: 164 OWD -- One Way Delay 166 MAD -- Mean Absolute Deviation 168 RTT -- Round Trip Time 170 SBD -- Shared Bottleneck Detection 172 Conventions used in this document: 174 T -- the base time interval over which measurements are 175 made. 177 N -- the number of base time, T, intervals used in some 178 calculations. 180 sum_T(...) -- summation of all the measurements of the variable 181 in parentheses taken over the interval T 183 sum(...) -- summation of terms of the variable in parentheses 185 sum_N(...) -- summation of N terms of the variable in parentheses 187 sum_NT(...) -- summation of all measurements taken over the 188 interval N*T 190 E_T(...) -- the expectation or mean of the measurements of the 191 variable in parentheses over T 193 E_N(...) -- the expectation or mean of the last N values of the 194 variable in parentheses 196 E_M(...) -- the expectation or mean of the last M values of the 197 variable in parentheses, where M <= N. 199 max_T(...) -- the maximum recorded measurement of the variable in 200 parentheses taken over the interval T 202 min_T(...) -- the minimum recorded measurement of the variable in 203 parentheses taken over the interval T 205 num_T(...) -- the count of measurements of the variable in 206 parentheses taken in the interval T 208 num_VM(...) -- the count of valid values of the variable in 209 parentheses given M records 211 PB -- a boolean variable indicating the particular flow 212 was identified transiting a bottleneck in the 213 previous interval T (i.e. Previously Bottleneck) 215 skew_est -- a measure of skewness in a OWD distribution. 217 skew_base_T -- a variable used as an intermediate step in 218 calculating skew_est. 220 var_est -- a measure of variability in OWD measurements. 222 var_base_T -- a variable used as an intermediate step in 223 calculating var_est. 225 freq_est -- a measure of low frequency oscillation in the OWD 226 measurements. 228 p_l, p_f, p_mad, c_s, c_h, p_s, p_d, p_v -- various thresholds 229 used in the mechanism 231 M and F -- number of values related to N 233 . 235 2.1. Parameters and their Effect 237 T T should be long enough so that there are enough packets 238 received during T for a useful estimate of short term mean 239 OWD and variation statistics. Making T too large can limit 240 the efficacy of freq_est. It will also increase the response 241 time of the mechanism. Making T too small will make the 242 metrics noisier. 244 N & M N should be large enough to provide a stable estimate of 245 oscillations in OWD. Usually M=N, though having M mean_delay) skew_base_T-- 427 The mean_delay does not include the mean of the current T interval to 428 enable it to be calculated iteratively. 430 skew_est = sum_MT(skew_base_T)/num_MT(OWD) 432 where skew_est is a number between -1 and 1 434 Note: Care must be taken when implementing the comparisons to ensure 435 that rounding does not bias skew_est. It is important that the mean 436 is calculated with a higher precision than the samples. 438 3.1.3. Variability Estimate 440 Mean Absolute Deviation (MAD) delay is a robust variability measure 441 that copes well with different send rates. It can be implemented in 442 an online manner as follows: 444 var_base_T = sum_T(|OWD - E_T(OWD)|) 446 where 448 |x| is the absolute value of x 450 E_T(OWD) is the mean OWD calculated in the previous T 452 var_est = MAD_MT = sum_MT(var_base_T)/num_MT(OWD) 454 For calculation of freq_est p_v=0.7 456 For the grouping threshold p_mad=0.1 458 3.1.4. Oscillation Estimate 460 An estimate of the low frequency oscillation of the delay signal is 461 calculated by counting and normalising the significant mean, 462 E_T(OWD), crossings of mean_delay: 464 freq_est = number_of_crossings / N 466 where we define a significant mean crossing as a crossing that 467 extends p_v * var_est from mean_delay. In our experiments we 468 have found that p_v = 0.7 is a good value. 470 Freq_est is a number between 0 and 1. Freq_est can be approximated 471 incrementally as follows: 473 With each new calculation of E_T(OWD) a decision is made as to 474 whether this value of E_T(OWD) significantly crosses the current 475 long term mean, mean_delay, with respect to the previous 476 significant mean crossing. 478 A cyclic buffer, last_N_crossings, records a 1 if there is a 479 significant mean crossing, otherwise a 0. 481 The counter, number_of_crossings, is incremented when there is a 482 significant mean crossing and decremented when a non-zero value is 483 removed from the last_N_crossings. 485 This approximation of freq_est was not used in [Hayes-LCN14], which 486 calculated freq_est every T using the current E_N(E_T(OWD)). Our 487 tests show that this approximation of freq_est yields results that 488 are almost identical to when the full calculation is performed every 489 T. 491 3.1.5. Packet loss 493 The proportion of packets lost over the period NT is used as a 494 supplementary measure: 496 pkt_loss = sum_NT(lost packets) / sum_NT(total packets) 498 Note: When pkt_loss is small it is very variable, however, when 499 pkt_loss is high it becomes a stable measure for making grouping 500 decisions. 502 3.2. Flow Grouping 504 3.2.1. Flow Grouping Algorithm 506 The following grouping algorithm is RECOMMENDED for SBD in the RMCAT 507 context and is sufficient and efficient for small to moderate numbers 508 of flows. For very large numbers of flows (e.g. hundreds), a more 509 complex clustering algorithm may be substituted. 511 Since no single metric is precise enough to group flows (due to 512 noise), the algorithm uses multiple metrics. Each metric offers a 513 different "view" of the bottleneck link characteristics, and used 514 together they enable a more precise grouping of flows than would 515 otherwise be possible. 517 Flows determined to be transiting a bottleneck are successively 518 divided into groups based on freq_est, var_est, skew_est and 519 pkt_loss. 521 The first step is to determine which flows are transiting a 522 bottleneck. This is important, since if a flow is not transiting a 523 bottleneck its delay based metrics will not describe the bottleneck, 524 but the "noise" from the rest of the path. Skewness, with proportion 525 of packet loss as a supplementary measure, is used to do this: 527 1. Grouping will be performed on flows that are inferred to be 528 traversing a bottleneck by: 530 skew_est < c_s 532 || ( skew_est < c_h & PB ) || pkt_loss > p_l 534 The parameter c_s controls how sensitive the mechanism is in 535 detecting a bottleneck. C_s = 0.0 was used in [Hayes-LCN14]. A 536 value of c_s = 0.05 is a little more sensitive, and c_s = -0.05 is a 537 little less sensitive. C_h controls the hysteresis on flows that 538 were grouped as transiting a bottleneck last time. If the test 539 result is TRUE, PB=TRUE, otherwise PB=FALSE. 541 These flows, flows transiting a bottleneck, are then progressively 542 divided into groups based on the freq_est, var_est, and skew_est 543 summary statistics. The process proceeds according to the following 544 steps: 546 2. Group flows whose difference in sorted freq_est is less than a 547 threshold: 549 diff(freq_est) < p_f 551 3. Group flows whose difference in sorted E_M(var_est) (highest to 552 lowest) is less than a threshold: 554 diff(var_est) < (p_mad * var_est) 556 The threshold, (p_mad * var_est), is with respect to the highest 557 value in the difference. 559 4. Group flows whose difference in sorted skew_est is less than a 560 threshold: 562 diff(skew_est) < p_s 564 5. When packet loss is high enough to be reliable (pkt_loss > p_l), 565 group flows whose difference is less than a threshold 567 diff(pkt_loss) < (p_d * pkt_loss) 569 The threshold, (p_d * pkt_loss), is with respect to the highest 570 value in the difference. 572 This procedure involves sorting estimates from highest to lowest. It 573 is simple to implement, and efficient for small numbers of flows (up 574 to 10-20). 576 3.2.2. Using the flow group signal 578 Grouping decisions can be made every T from the second T, however 579 they will not attain their full design accuracy until after the 580 2*N'th T interval. We recommend that grouping decisions are not made 581 until 2*M T intervals. 583 Network conditions, and even the congestion controllers, can cause 584 bottlenecks to fluctuate. A coupled congestion controller MAY decide 585 only to couple groups that remain stable, say grouped together 90% of 586 the time, depending on its objectives. Recommendations concerning 587 this are beyond the scope of this draft and will be specific to the 588 coupled congestion controllers objectives. 590 3.3. Removing Noise from the Estimates 592 The following describe small changes to the calculation of the key 593 metrics that help remove noise from them. Currently these "tweaks" 594 are described separately to keep the main description succinct. In 595 future revisions of the draft these enhancements may replace the 596 original key metric calculations. 598 3.3.1. Oscillation noise 600 When a path has no bottleneck, var_est will be very small and the 601 recorded significant mean crossings will be the result of path noise. 602 Thus up to N-1 meaningless mean crossings can be a source of error at 603 the point a link becomes a bottleneck and flows traversing it begin 604 to be grouped. 606 To remove this source of noise from freq_est: 608 1. Set the current var_base_T = NaN (a value representing an invalid 609 record, i.e. Not a Number) for flows that are deemed to not be 610 transiting a bottleneck by the first skew_est based grouping test 611 (see Section 3.2.1). 613 2. Then var_est = sum_MT(var_base_T != NaN) / num_MT(OWD) 615 3. For freq_est, only record a significant mean crossing if flow 616 deemed to be transiting a bottleneck. 618 These three changes can help to remove the non-bottleneck noise from 619 freq_est. 621 3.3.2. Clock skew 623 Generally sender and receiver clock skew will be too small to cause 624 significant errors in the estimators. Skew_est is most sensitive to 625 this type of noise. In circumstances where clock skew is high, 626 basing skew_est only on the previous T's mean provides a noisier but 627 reliable signal. 629 A better method is to estimate the effect the clock skew is having on 630 the summary statistics, and then adjust statistics accordingly. A 631 simple online method of doing this based on min_T(OWD) will be 632 described here in a subsequent version of the draft. 634 3.4. Reducing lag and Improving Responsiveness 636 Measurement based shared bottleneck detection makes decisions in the 637 present based on what has been measured in the past. This means that 638 there is always a lag in responding to changing conditions. This 639 mechanism is based on summary statistics taken over (N*T) seconds. 640 This mechanism can be made more responsive to changing conditions by: 642 1. Reducing N and/or M -- but at the expense of having less accurate 643 metrics, and/or 645 2. Exploiting the fact that more recent measurements are more 646 valuable than older measurements and weighting them accordingly. 648 Although more recent measurements are more valuable, older 649 measurements are still needed to gain an accurate estimate of the 650 distribution descriptor we are measuring. Unfortunately, the simple 651 exponentially weighted moving average weights drop off too quickly 652 for our requirements and have an infinite tail. A simple linearly 653 declining weighted moving average also does not provide enough weight 654 to the most recent measurements. We propose a piecewise linear 655 distribution of weights, such that the first section (samples 1:F) is 656 flat as in a simple moving average, and the second section (samples 657 F+1:M) is linearly declining weights to the end of the averaging 658 window. We choose integer weights, which allows incremental 659 calculation without introducing rounding errors. 661 3.4.1. Improving the response of the skewness estimate 663 The weighted moving average for skew_est, based on skew_est in 664 Section 3.1.2, can be calculated as follows: 666 skew_est = ((M-F+1)*sum(skew_base_T(1:F)) 668 + sum([(M-F):1].*skew_base_T(F+1:M))) 670 / ((M-F+1)*sum(numsampT(1:F)) 672 + sum([(M-F):1].*numsampT(F+1:M))) 674 where numsampT is an array of the number of OWD samples in each T 675 (i.e. num_T(OWD)), and numsampT(1) is the most recent; skew_base_T(1) 676 is the most recent calculation of skew_base_T; 1:F refers to the 677 integer values 1 through to F, and [(M-F):1] refers to an array of 678 the integer values (M-F) declining through to 1; and ".*" is the 679 array scalar dot product operator. 681 To calculate this weighted skew_est incrementally: 683 Notation: F_ - flat portion, D_ - declining portion, W_ - weighted 684 component 686 Initialise: sum_skewbase = 0, F_skewbase=0, W_D_skewbase=0 688 skewbase_hist = buffer length M initialize to 0 690 numsampT = buffer length M initialzed to 0 692 Steps per iteration: 694 1. old_skewbase = skewbase_hist(M) 696 2. old_numsampT = numsampT(M) 698 3. cycle(skewbase_hist) 700 4. cycle(numsampT) 702 5. numsampT(1) = num_T(OWD) 704 6. skewbase_hist(1) = skew_base_T 706 7. F_skewbase = F_skewbase + skew_base_T - skewbase_hist(F+1) 708 8. W_D_skewbase = W_D_skewbase + (M-F)*skewbase_hist(F+1) 709 - sum_skewbase 711 9. W_D_numsamp = W_D_numsamp + (M-F)*numsampT(F+1) - sum_numsamp 712 + F_numsamp 714 10. F_numsamp = F_numsamp + numsampT(1) - numsampT(F+1) 716 11. sum_skewbase = sum_skewbase + skewbase_hist(F+1) - old_skewbase 718 12. sum_numsamp = sum_numsamp + numsampT(1) - old_numsampT 720 13. skew_est = ((M-F+1)*F_skewbase + W_D_skewbase) / 721 ((M-F+1)*F_numsamp+W_D_numsamp) 723 Where cycle(....) refers to the operation on a cyclic buffer where 724 the start of the buffer is now the next element in the buffer. 726 3.4.2. Improving the response of the variability estimate 728 Similarly the weighted moving average for var_est can be calculated 729 as follows: 731 var_est = ((M-F+1)*sum(var_base_T(1:F)) 733 + sum([(M-F):1].*var_base_T(F+1:M))) 735 / ((M-F+1)*sum(numsampT(1:F)) 737 + sum([(M-F):1].*numsampT(F+1:M))) 739 where numsampT is an array of the number of OWD samples in each T 740 (i.e. num_T(OWD)), and numsampT(1) is the most recent; skew_base_T(1) 741 is the most recent calculation of skew_base_T; 1:F refers to the 742 integer values 1 through to F, and [(M-F):1] refers to an array of 743 the integer values (M-F) declining through to 1; and ".*" is the 744 array scalar dot product operator. When removing oscillation noise 745 (see Section 3.3.1) this calculation must be adjusted to allow for 746 invalid var_base_T records. 748 Var_est can be calculated incrementally in the same way as skew_est 749 in Section 3.4.1. However, note that the buffer numsampT is used for 750 both calculations so the operations on it should not be repeated. 752 4. Measuring OWD 754 This section discusses the OWD measurements required for this 755 algorithm to detect shared bottlenecks. 757 The SBD mechanism described in this draft relies on differences 758 between OWD measurements to avoid the practical problems with 759 measuring absolute OWD (see [Hayes-LCN14] section IIIC). Since all 760 summary statistics are relative to the mean OWD and sender/receiver 761 clock offsets should be approximately constant over the measurement 762 periods, the offset is subtracted out in the calculation. 764 4.1. Time stamp resolution 766 The SBD mechanism requires timing information precise enough to be 767 able to make comparisons. As a rule of thumb, the time resolution 768 should be less than one hundredth of a typical path's range of 769 delays. In general, the lower the time resolution, the more care 770 that needs to be taken to ensure rounding errors do not bias the 771 skewness calculation. 773 Typical RTP media flows use sub-millisecond timers, which should be 774 adequate in most situations. 776 5. Implementation status 778 The University of Oslo is currently working on an implementation of 779 this in the Chromium browser. 781 6. Acknowledgements 783 This work was part-funded by the European Community under its Seventh 784 Framework Programme through the Reducing Internet Transport Latency 785 (RITE) project (ICT-317700). The views expressed are solely those of 786 the authors. 788 7. IANA Considerations 790 This memo includes no request to IANA. 792 8. Security Considerations 794 The security considerations of RFC 3550 [RFC3550], RFC 4585 795 [RFC4585], and RFC 5124 [RFC5124] are expected to apply. 797 Non-authenticated RTCP packets carrying shared bottleneck indications 798 and summary statistics could allow attackers to alter the bottleneck 799 sharing characteristics for private gain or disruption of other 800 parties communication. 802 9. Change history 804 Changes made to this document: 806 WG-02->WG-03 : Correct misspelled author 808 WG-01->WG-02 : Removed ambiguity associated with the term 809 "congestion". Expanded the description of 810 initialisation messages. Removed PDV metric. 811 Added description of incremental weighted metric 812 calculations for skew_est. Various clarifications 813 based on implementation work. Fixed typos and 814 tuned parameters. 816 WG-00->WG-01 : Moved unbiased skew section to replace skew 817 estimate, more robust variability estimator, the 818 term variance replaced with variability, clock 819 drift term corrected to clock skew, revision to 820 clock skew section with a place holder, description 821 of parameters. 823 02->WG-00 : Fixed missing 0.5 in 3.3.2 and missing brace in 824 3.3.3 826 01->02 : New section describing improvements to the key 827 metric calculations that help to remove noise, 828 bias, and reduce lag. Some revisions to the 829 notation to make it clearer. Some tightening of 830 the thresholds. 832 00->01 : Revisions to terminology for clarity 834 10. References 836 10.1. Normative References 838 [RFC2119] Bradner, S., "Key words for use in RFCs to Indicate 839 Requirement Levels", BCP 14, RFC 2119, DOI 10.17487/ 840 RFC2119, March 1997, 841 . 843 10.2. Informative References 845 [Hayes-LCN14] 846 Hayes, D., Ferlin, S., and M. Welzl, "Practical Passive 847 Shared Bottleneck Detection using Shape Summary 848 Statistics", Proc. the IEEE Local Computer Networks (LCN) 849 p150-158, September 2014, . 852 [I-D.welzl-rmcat-coupled-cc] 853 Welzl, M., Islam, S., and S. Gjessing, "Coupled congestion 854 control for RTP media", draft-welzl-rmcat-coupled-cc-04 855 (work in progress), October 2014. 857 [ITU-Y1540] 858 ITU-T, "Internet Protocol Data Communication Service - IP 859 Packet Transfer and Availability Performance Parameters", 860 Series Y: Global Information Infrastructure, Internet 861 Protocol Aspects and Next-Generation Networks , March 862 2011, . 864 [RFC3550] Schulzrinne, H., Casner, S., Frederick, R., and V. 865 Jacobson, "RTP: A Transport Protocol for Real-Time 866 Applications", STD 64, RFC 3550, DOI 10.17487/RFC3550, 867 July 2003, . 869 [RFC4585] Ott, J., Wenger, S., Sato, N., Burmeister, C., and J. Rey, 870 "Extended RTP Profile for Real-time Transport Control 871 Protocol (RTCP)-Based Feedback (RTP/AVPF)", RFC 4585, DOI 872 10.17487/RFC4585, July 2006, 873 . 875 [RFC5124] Ott, J. and E. Carrara, "Extended Secure RTP Profile for 876 Real-time Transport Control Protocol (RTCP)-Based Feedback 877 (RTP/SAVPF)", RFC 5124, DOI 10.17487/RFC5124, February 878 2008, . 880 [RFC5481] Morton, A. and B. Claise, "Packet Delay Variation 881 Applicability Statement", RFC 5481, DOI 10.17487/RFC5481, 882 March 2009, . 884 [RFC6817] Shalunov, S., Hazel, G., Iyengar, J., and M. Kuehlewind, 885 "Low Extra Delay Background Transport (LEDBAT)", RFC 6817, 886 DOI 10.17487/RFC6817, December 2012, 887 . 889 Authors' Addresses 891 David Hayes (editor) 892 University of Oslo 893 PO Box 1080 Blindern 894 Oslo N-0316 895 Norway 897 Phone: +47 2284 5566 898 Email: davihay@ifi.uio.no 900 Simone Ferlin 901 Simula Research Laboratory 902 P.O.Box 134 903 Lysaker 1325 904 Norway 906 Phone: +47 4072 0702 907 Email: ferlin@simula.no 908 Michael Welzl 909 University of Oslo 910 PO Box 1080 Blindern 911 Oslo N-0316 912 Norway 914 Phone: +47 2285 2420 915 Email: michawe@ifi.uio.no 917 Kristian Hiorth 918 University of Oslo 919 PO Box 1080 Blindern 920 Oslo N-0316 921 Norway 923 Email: kristahi@ifi.uio.no