| 1 |
<pre>Internet Engineering Task Force (IETF) I. Johansson |
| 2 |
Request for Comments: 8298 Z. Sarker |
| 3 |
Category: Experimental Ericsson AB |
| 4 |
ISSN: 2070-1721 December 2017 |
| 5 |
|
| 6 |
|
| 7 |
<span class="h1">Self-Clocked Rate Adaptation for Multimedia</span> |
| 8 |
|
| 9 |
Abstract |
| 10 |
|
| 11 |
This memo describes a rate adaptation algorithm for conversational |
| 12 |
media services such as interactive video. The solution conforms to |
| 13 |
the packet conservation principle and uses a hybrid loss-and-delay- |
| 14 |
based congestion control algorithm. The algorithm is evaluated over |
| 15 |
both simulated Internet bottleneck scenarios as well as in a Long |
| 16 |
Term Evolution (LTE) system simulator and is shown to achieve both |
| 17 |
low latency and high video throughput in these scenarios. |
| 18 |
|
| 19 |
Status of This Memo |
| 20 |
|
| 21 |
This document is not an Internet Standards Track specification; it is |
| 22 |
published for examination, experimental implementation, and |
| 23 |
evaluation. |
| 24 |
|
| 25 |
This document defines an Experimental Protocol for the Internet |
| 26 |
community. This document is a product of the Internet Engineering |
| 27 |
Task Force (IETF). It represents the consensus of the IETF |
| 28 |
community. It has received public review and has been approved for |
| 29 |
publication by the Internet Engineering Steering Group (IESG). Not |
| 30 |
all documents approved by the IESG are a candidate for any level of |
| 31 |
Internet Standard; see <a href="./rfc7841#section-2">Section 2 of RFC 7841</a>. |
| 32 |
|
| 33 |
Information about the current status of this document, any errata, |
| 34 |
and how to provide feedback on it may be obtained at |
| 35 |
<a href="https://www.rfc-editor.org/info/rfc8298">https://www.rfc-editor.org/info/rfc8298</a>. |
| 36 |
|
| 37 |
|
| 38 |
|
| 39 |
|
| 40 |
|
| 41 |
|
| 42 |
|
| 43 |
|
| 44 |
|
| 45 |
|
| 46 |
|
| 47 |
|
| 48 |
|
| 49 |
|
| 50 |
|
| 51 |
|
| 52 |
<span class="grey">Johansson & Sarker Experimental [Page 1]</span></pre> |
| 53 |
<hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-2" ></span> |
| 54 |
<span class="grey"><a href="./rfc8298">RFC 8298</a> SCReAM December 2017</span> |
| 55 |
|
| 56 |
|
| 57 |
Copyright Notice |
| 58 |
|
| 59 |
Copyright (c) 2017 IETF Trust and the persons identified as the |
| 60 |
document authors. All rights reserved. |
| 61 |
|
| 62 |
This document is subject to <a href="https://www.rfc-editor.org/bcp/bcp78">BCP 78</a> and the IETF Trust's Legal |
| 63 |
Provisions Relating to IETF Documents |
| 64 |
(<a href="https://trustee.ietf.org/license-info">https://trustee.ietf.org/license-info</a>) in effect on the date of |
| 65 |
publication of this document. Please review these documents |
| 66 |
carefully, as they describe your rights and restrictions with respect |
| 67 |
to this document. Code Components extracted from this document must |
| 68 |
include Simplified BSD License text as described in Section 4.e of |
| 69 |
the Trust Legal Provisions and are provided without warranty as |
| 70 |
described in the Simplified BSD License. |
| 71 |
|
| 72 |
|
| 73 |
|
| 74 |
|
| 75 |
|
| 76 |
|
| 77 |
|
| 78 |
|
| 79 |
|
| 80 |
|
| 81 |
|
| 82 |
|
| 83 |
|
| 84 |
|
| 85 |
|
| 86 |
|
| 87 |
|
| 88 |
|
| 89 |
|
| 90 |
|
| 91 |
|
| 92 |
|
| 93 |
|
| 94 |
|
| 95 |
|
| 96 |
|
| 97 |
|
| 98 |
|
| 99 |
|
| 100 |
|
| 101 |
|
| 102 |
|
| 103 |
|
| 104 |
|
| 105 |
|
| 106 |
|
| 107 |
|
| 108 |
<span class="grey">Johansson & Sarker Experimental [Page 2]</span></pre> |
| 109 |
<hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-3" ></span> |
| 110 |
<span class="grey"><a href="./rfc8298">RFC 8298</a> SCReAM December 2017</span> |
| 111 |
|
| 112 |
|
| 113 |
Table of Contents |
| 114 |
|
| 115 |
<a href="#section-1">1</a>. Introduction . . . . . . . . . . . . . . . . . . . . . . . . <a href="#page-4">4</a> |
| 116 |
<a href="#section-1.1">1.1</a>. Wireless (LTE) Access Properties . . . . . . . . . . . . <a href="#page-4">4</a> |
| 117 |
<a href="#section-1.2">1.2</a>. Why is it a self-clocked algorithm? . . . . . . . . . . . <a href="#page-5">5</a> |
| 118 |
<a href="#section-2">2</a>. Requirements Language . . . . . . . . . . . . . . . . . . . . <a href="#page-5">5</a> |
| 119 |
<a href="#section-3">3</a>. Overview of SCReAM Algorithm . . . . . . . . . . . . . . . . <a href="#page-6">6</a> |
| 120 |
<a href="#section-3.1">3.1</a>. Network Congestion Control . . . . . . . . . . . . . . . <a href="#page-8">8</a> |
| 121 |
<a href="#section-3.2">3.2</a>. Sender Transmission Control . . . . . . . . . . . . . . . <a href="#page-9">9</a> |
| 122 |
<a href="#section-3.3">3.3</a>. Media Rate Control . . . . . . . . . . . . . . . . . . . <a href="#page-9">9</a> |
| 123 |
<a href="#section-4">4</a>. Detailed Description of SCReAM . . . . . . . . . . . . . . . <a href="#page-10">10</a> |
| 124 |
<a href="#section-4.1">4.1</a>. SCReAM Sender . . . . . . . . . . . . . . . . . . . . . . <a href="#page-10">10</a> |
| 125 |
<a href="#section-4.1.1">4.1.1</a>. Constants and Parameter Values . . . . . . . . . . . <a href="#page-10">10</a> |
| 126 |
<a href="#section-4.1.1.1">4.1.1.1</a>. Constants . . . . . . . . . . . . . . . . . . . . <a href="#page-11">11</a> |
| 127 |
<a href="#section-4.1.1.2">4.1.1.2</a>. State Variables . . . . . . . . . . . . . . . . . <a href="#page-12">12</a> |
| 128 |
<a href="#section-4.1.2">4.1.2</a>. Network Congestion Control . . . . . . . . . . . . . <a href="#page-14">14</a> |
| 129 |
<a href="#section-4.1.2.1">4.1.2.1</a>. Reaction to Packet Loss and ECN . . . . . . . . . <a href="#page-17">17</a> |
| 130 |
<a href="#section-4.1.2.2">4.1.2.2</a>. Congestion Window Update . . . . . . . . . . . . <a href="#page-17">17</a> |
| 131 |
<a href="#section-4.1.2.3">4.1.2.3</a>. Competing Flows Compensation . . . . . . . . . . <a href="#page-20">20</a> |
| 132 |
<a href="#section-4.1.2.4">4.1.2.4</a>. Lost Packet Detection . . . . . . . . . . . . . . <a href="#page-22">22</a> |
| 133 |
<a href="#section-4.1.2.5">4.1.2.5</a>. Send Window Calculation . . . . . . . . . . . . . <a href="#page-23">23</a> |
| 134 |
<a href="#section-4.1.2.6">4.1.2.6</a>. Packet Pacing . . . . . . . . . . . . . . . . . . <a href="#page-24">24</a> |
| 135 |
<a href="#section-4.1.2.7">4.1.2.7</a>. Resuming Fast Increase Mode . . . . . . . . . . . <a href="#page-24">24</a> |
| 136 |
<a href="#section-4.1.2.8">4.1.2.8</a>. Stream Prioritization . . . . . . . . . . . . . . <a href="#page-24">24</a> |
| 137 |
<a href="#section-4.1.3">4.1.3</a>. Media Rate Control . . . . . . . . . . . . . . . . . <a href="#page-25">25</a> |
| 138 |
<a href="#section-4.2">4.2</a>. SCReAM Receiver . . . . . . . . . . . . . . . . . . . . . <a href="#page-28">28</a> |
| 139 |
<a href="#section-4.2.1">4.2.1</a>. Requirements on Feedback Elements . . . . . . . . . . <a href="#page-28">28</a> |
| 140 |
<a href="#section-4.2.2">4.2.2</a>. Requirements on Feedback Intensity . . . . . . . . . <a href="#page-30">30</a> |
| 141 |
<a href="#section-5">5</a>. Discussion . . . . . . . . . . . . . . . . . . . . . . . . . <a href="#page-31">31</a> |
| 142 |
<a href="#section-6">6</a>. Suggested Experiments . . . . . . . . . . . . . . . . . . . . <a href="#page-31">31</a> |
| 143 |
<a href="#section-7">7</a>. IANA Considerations . . . . . . . . . . . . . . . . . . . . . <a href="#page-32">32</a> |
| 144 |
<a href="#section-8">8</a>. Security Considerations . . . . . . . . . . . . . . . . . . . <a href="#page-32">32</a> |
| 145 |
<a href="#section-9">9</a>. References . . . . . . . . . . . . . . . . . . . . . . . . . <a href="#page-33">33</a> |
| 146 |
<a href="#section-9.1">9.1</a>. Normative References . . . . . . . . . . . . . . . . . . <a href="#page-33">33</a> |
| 147 |
<a href="#section-9.2">9.2</a>. Informative References . . . . . . . . . . . . . . . . . <a href="#page-34">34</a> |
| 148 |
Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . <a href="#page-36">36</a> |
| 149 |
Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . <a href="#page-36">36</a> |
| 150 |
|
| 151 |
|
| 152 |
|
| 153 |
|
| 154 |
|
| 155 |
|
| 156 |
|
| 157 |
|
| 158 |
|
| 159 |
|
| 160 |
|
| 161 |
|
| 162 |
|
| 163 |
|
| 164 |
<span class="grey">Johansson & Sarker Experimental [Page 3]</span></pre> |
| 165 |
<hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-4" ></span> |
| 166 |
<span class="grey"><a href="./rfc8298">RFC 8298</a> SCReAM December 2017</span> |
| 167 |
|
| 168 |
|
| 169 |
<span class="h2"><a class="selflink" id="section-1" href="#section-1">1</a>. Introduction</span> |
| 170 |
|
| 171 |
Congestion in the Internet occurs when the transmitted bitrate is |
| 172 |
higher than the available capacity over a given transmission path. |
| 173 |
Applications that are deployed in the Internet have to employ |
| 174 |
congestion control to achieve robust performance and to avoid |
| 175 |
congestion collapse in the Internet. Interactive real-time |
| 176 |
communication imposes a lot of requirements on the transport; |
| 177 |
therefore, a robust, efficient rate adaptation for all access types |
| 178 |
is an important part of interactive real-time communications, as the |
| 179 |
transmission channel bandwidth can vary over time. Wireless access |
| 180 |
such as LTE, which is an integral part of the current Internet, |
| 181 |
increases the importance of rate adaptation as the channel bandwidth |
| 182 |
of a default LTE bearer [<a href="#ref-QoS-3GPP" title=""Policy and charging control architecture"">QoS-3GPP</a>] can change considerably in a very |
| 183 |
short time frame. Thus, a rate adaptation solution for interactive |
| 184 |
real-time media, such as WebRTC [<a href="./rfc7478" title=""Web Real- Time Communication Use Cases and Requirements"">RFC7478</a>], should be both quick and |
| 185 |
be able to operate over a large range in channel capacity. This memo |
| 186 |
describes Self-Clocked Rate Adaptation for Multimedia (SCReAM), a |
| 187 |
solution that implements congestion control for RTP streams |
| 188 |
[<a href="./rfc3550" title=""RTP: A Transport Protocol for Real-Time Applications"">RFC3550</a>]. While SCReAM was originally devised for WebRTC, it can |
| 189 |
also be used for other applications where congestion control of RTP |
| 190 |
streams is necessary. SCReAM is based on the self-clocking principle |
| 191 |
of TCP and uses techniques similar to what is used in the rate |
| 192 |
adaptation algorithm based on Low Extra Delay Background Transport |
| 193 |
(LEDBAT) [<a href="./rfc6817" title=""Low Extra Delay Background Transport (LEDBAT)"">RFC6817</a>]. SCReAM is not entirely self-clocked as it |
| 194 |
augments self-clocking with pacing and a minimum send rate. SCReAM |
| 195 |
can take advantage of Explicit Congestion Notification (ECN) in cases |
| 196 |
where ECN is supported by the network and the hosts. However, ECN is |
| 197 |
not required for the basic congestion control functionality in |
| 198 |
SCReAM. |
| 199 |
|
| 200 |
<span class="h3"><a class="selflink" id="section-1.1" href="#section-1.1">1.1</a>. Wireless (LTE) Access Properties</span> |
| 201 |
|
| 202 |
[<a id="ref-WIRELESS-TESTS">WIRELESS-TESTS</a>] describes the complications that can be observed in |
| 203 |
wireless environments. Wireless access such as LTE typically cannot |
| 204 |
guarantee a given bandwidth; this is true especially for default |
| 205 |
bearers. The network throughput can vary considerably, for instance, |
| 206 |
in cases where the wireless terminal is moving around. Even though |
| 207 |
LTE can support bitrates well above 100 Mbps, there are cases when |
| 208 |
the available bitrate can be much lower; examples are situations with |
| 209 |
high network load and poor coverage. An additional complication is |
| 210 |
that the network throughput can drop for short time intervals (e.g., |
| 211 |
at handover); these short glitches are initially very difficult to |
| 212 |
distinguish from more permanent reductions in throughput. |
| 213 |
|
| 214 |
Unlike wireline bottlenecks with large statistical multiplexing, it |
| 215 |
is not possible to try to maintain a given bitrate when congestion is |
| 216 |
detected with the hope that other flows will yield. This is because |
| 217 |
|
| 218 |
|
| 219 |
|
| 220 |
<span class="grey">Johansson & Sarker Experimental [Page 4]</span></pre> |
| 221 |
<hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-5" ></span> |
| 222 |
<span class="grey"><a href="./rfc8298">RFC 8298</a> SCReAM December 2017</span> |
| 223 |
|
| 224 |
|
| 225 |
there are generally few other flows competing for the same |
| 226 |
bottleneck. Each user gets its own variable throughput bottleneck, |
| 227 |
where the throughput depends on factors like channel quality, network |
| 228 |
load, and historical throughput. The bottom line is, if the |
| 229 |
throughput drops, the sender has no other option than to reduce the |
| 230 |
bitrate. Once the radio scheduler has reduced the resource |
| 231 |
allocation for a bearer, a flow (which is using RTP Media Congestion |
| 232 |
Avoidance Techniques (RMCAT)) in that bearer aims to reduce the |
| 233 |
sending rate quite quickly (within one RTT) in order to avoid |
| 234 |
excessive queuing delay or packet loss. |
| 235 |
|
| 236 |
<span class="h3"><a class="selflink" id="section-1.2" href="#section-1.2">1.2</a>. Why is it a self-clocked algorithm?</span> |
| 237 |
|
| 238 |
Self-clocked congestion control algorithms provide a benefit over |
| 239 |
their rate-based counterparts in that the former consists of two |
| 240 |
adaptation mechanisms: |
| 241 |
|
| 242 |
o A congestion window computation that evolves over a longer |
| 243 |
timescale (several RTTs) especially when the congestion window |
| 244 |
evolution is dictated by estimated delay (to minimize |
| 245 |
vulnerability to, e.g., short-term delay variations). |
| 246 |
|
| 247 |
o A fine-grained congestion control given by the self-clocking; it |
| 248 |
operates on a shorter time scale (1 RTT). The benefits of self- |
| 249 |
clocking are also elaborated upon in [<a href="#ref-TFWC" title=""Fairer TCP-Friendly Congestion Control Protocol for Multimedia Streaming Applications"">TFWC</a>]. |
| 250 |
|
| 251 |
A rate-based congestion control algorithm typically adjusts the rate |
| 252 |
based on delay and loss. The congestion detection needs to be done |
| 253 |
with a certain time lag to avoid overreaction to spurious congestion |
| 254 |
events such as delay spikes. Despite the fact that there are two or |
| 255 |
more congestion indications, the outcome is that there is still only |
| 256 |
one mechanism to adjust the sending rate. This makes it difficult to |
| 257 |
reach the goals of high throughput and prompt reaction to congestion. |
| 258 |
|
| 259 |
<span class="h2"><a class="selflink" id="section-2" href="#section-2">2</a>. Requirements Language</span> |
| 260 |
|
| 261 |
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", |
| 262 |
"SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and |
| 263 |
"OPTIONAL" in this document are to be interpreted as described in |
| 264 |
<a href="https://www.rfc-editor.org/bcp/bcp14">BCP 14</a> [<a href="./rfc2119" title=""Key words for use in RFCs to Indicate Requirement Levels"">RFC2119</a>] [<a href="./rfc8174" title=""Ambiguity of Uppercase vs Lowercase in RFC 2119 Key Words"">RFC8174</a>] when, and only when, they appear in all |
| 265 |
capitals, as shown here. |
| 266 |
|
| 267 |
|
| 268 |
|
| 269 |
|
| 270 |
|
| 271 |
|
| 272 |
|
| 273 |
|
| 274 |
|
| 275 |
|
| 276 |
<span class="grey">Johansson & Sarker Experimental [Page 5]</span></pre> |
| 277 |
<hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-6" ></span> |
| 278 |
<span class="grey"><a href="./rfc8298">RFC 8298</a> SCReAM December 2017</span> |
| 279 |
|
| 280 |
|
| 281 |
<span class="h2"><a class="selflink" id="section-3" href="#section-3">3</a>. Overview of SCReAM Algorithm</span> |
| 282 |
|
| 283 |
The core SCReAM algorithm has similarities to the concepts of self- |
| 284 |
clocking used in TCP-friendly window-based congestion control [<a href="#ref-TFWC" title=""Fairer TCP-Friendly Congestion Control Protocol for Multimedia Streaming Applications"">TFWC</a>] |
| 285 |
and follows the packet conservation principle. The packet |
| 286 |
conservation principle is described as a key factor behind the |
| 287 |
protection of networks from congestion [<a href="#ref-Packet-conservation">Packet-conservation</a>]. |
| 288 |
|
| 289 |
In SCReAM, the receiver of the media echoes a list of received RTP |
| 290 |
packets and the timestamp of the RTP packet with the highest sequence |
| 291 |
number back to the sender in feedback packets. The sender keeps a |
| 292 |
list of transmitted packets, their respective sizes, and the time |
| 293 |
they were transmitted. This information is used to determine the |
| 294 |
number of bytes that can be transmitted at any given time instant. A |
| 295 |
congestion window puts an upper limit on how many bytes can be in |
| 296 |
flight, i.e., transmitted but not yet acknowledged. |
| 297 |
|
| 298 |
The congestion window is determined in a way similar to LEDBAT |
| 299 |
[<a href="./rfc6817" title=""Low Extra Delay Background Transport (LEDBAT)"">RFC6817</a>]. LEDBAT is a congestion control algorithm that uses send |
| 300 |
and receive timestamps to estimate the queuing delay (from now on |
| 301 |
denoted "qdelay") along the transmission path. This information is |
| 302 |
used to adjust the congestion window. The use of LEDBAT ensures that |
| 303 |
the end-to-end latency is kept low. [<a href="#ref-LEDBAT-delay-impact">LEDBAT-delay-impact</a>] shows that |
| 304 |
LEDBAT has certain inherent issues that make it counteract its |
| 305 |
purpose of achieving low delay. The general problem described in the |
| 306 |
paper is that the base delay is offset by LEDBAT's own queue buildup. |
| 307 |
The big difference with using LEDBAT in the SCReAM context lies in |
| 308 |
the facts that the source is rate limited and that the RTP queue must |
| 309 |
be kept short (preferably empty). In addition, the output from a |
| 310 |
video encoder is rarely constant bitrate; static content (talking |
| 311 |
heads, for instance) gives almost zero video bitrate. This yields |
| 312 |
two useful properties when LEDBAT is used with SCReAM; they help to |
| 313 |
avoid the issues described in [<a href="#ref-LEDBAT-delay-impact">LEDBAT-delay-impact</a>]: |
| 314 |
|
| 315 |
1. There is always a certain probability that SCReAM is short of |
| 316 |
data to transmit; this means that the network queue will become |
| 317 |
empty every once in a while. |
| 318 |
|
| 319 |
2. The max video bitrate can be lower than the link capacity. If |
| 320 |
the max video bitrate is 5 Mbps and the capacity is 10 Mbps, then |
| 321 |
the network queue will become empty. |
| 322 |
|
| 323 |
It is sufficient that any of the two conditions above is fulfilled to |
| 324 |
make the base delay update properly. Furthermore, |
| 325 |
[<a href="#ref-LEDBAT-delay-impact">LEDBAT-delay-impact</a>] describes an issue with short-lived competing |
| 326 |
flows. In SCReAM, these short-lived flows will cause the self- |
| 327 |
clocking to slow down, thereby building up the RTP queue; in turn, |
| 328 |
this results in a reduced media video bitrate. Thus, SCReAM slows |
| 329 |
|
| 330 |
|
| 331 |
|
| 332 |
<span class="grey">Johansson & Sarker Experimental [Page 6]</span></pre> |
| 333 |
<hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-7" ></span> |
| 334 |
<span class="grey"><a href="./rfc8298">RFC 8298</a> SCReAM December 2017</span> |
| 335 |
|
| 336 |
|
| 337 |
the bitrate more when there are competing short-lived flows than the |
| 338 |
traditional use of LEDBAT does. The basic functionality in the use |
| 339 |
of LEDBAT in SCReAM is quite simple; however, there are a few steps |
| 340 |
in order to make the concept work with conversational media: |
| 341 |
|
| 342 |
o Congestion window validation techniques. These are similar to the |
| 343 |
method described in [<a href="./rfc7661" title=""Updating TCP to Support Rate-Limited Traffic"">RFC7661</a>]. Congestion window validation |
| 344 |
ensures that the congestion window is limited by the actual number |
| 345 |
bytes in flight; this is important especially in the context of |
| 346 |
rate-limited sources such as video. Lack of congestion window |
| 347 |
validation would lead to a slow reaction to congestion as the |
| 348 |
congestion window does not properly reflect the congestion state |
| 349 |
in the network. The allowed idle period in this memo is shorter |
| 350 |
than in [<a href="./rfc7661" title=""Updating TCP to Support Rate-Limited Traffic"">RFC7661</a>]; this to avoid excessive delays in the cases |
| 351 |
where, e.g., wireless throughput has decreased during a period |
| 352 |
where the output bitrate from the media coder has been low (for |
| 353 |
instance, due to inactivity). Furthermore, this memo allows for |
| 354 |
more relaxed rules for when the congestion window is allowed to |
| 355 |
grow; this is necessary as the variable output bitrate generally |
| 356 |
means that the congestion window is often underutilized. |
| 357 |
|
| 358 |
o Fast increase mode makes the bitrate increase faster when no |
| 359 |
congestion is detected. It makes the media bitrate ramp up within |
| 360 |
5 to 10 seconds. The behavior is similar to TCP slowstart. Fast |
| 361 |
increase mode is exited when congestion is detected. However, |
| 362 |
fast increase mode can resume if the congestion level is low; this |
| 363 |
enables a reasonably quick rate increase in case link throughput |
| 364 |
increases. |
| 365 |
|
| 366 |
o A qdelay trend is computed for earlier detection of incipient |
| 367 |
congestion; as a result, it reduces jitter. |
| 368 |
|
| 369 |
o Addition of a media rate control function. |
| 370 |
|
| 371 |
o Use of inflection points in the media rate calculation to achieve |
| 372 |
reduced jitter. |
| 373 |
|
| 374 |
o Adjustment of qdelay target for better performance when competing |
| 375 |
with other loss-based congestion-controlled flows. |
| 376 |
|
| 377 |
The above-mentioned features will be described in more detail in |
| 378 |
Sections <a href="#section-3.1">3.1</a> to <a href="#section-3.3">3.3</a>. The full details are described in <a href="#section-4">Section 4</a>. |
| 379 |
|
| 380 |
|
| 381 |
|
| 382 |
|
| 383 |
|
| 384 |
|
| 385 |
|
| 386 |
|
| 387 |
|
| 388 |
<span class="grey">Johansson & Sarker Experimental [Page 7]</span></pre> |
| 389 |
<hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-8" ></span> |
| 390 |
<span class="grey"><a href="./rfc8298">RFC 8298</a> SCReAM December 2017</span> |
| 391 |
|
| 392 |
|
| 393 |
+---------------------------+ |
| 394 |
| Media encoder | |
| 395 |
+---------------------------+ |
| 396 |
^ | |
| 397 |
| |(1) |
| 398 |
|(3) RTP |
| 399 |
| V |
| 400 |
| +-----------+ |
| 401 |
+---------+ | | |
| 402 |
| Media | (2) | Queue | |
| 403 |
| rate |<------| | |
| 404 |
| control | |RTP packets| |
| 405 |
+---------+ | | |
| 406 |
+-----------+ |
| 407 |
| |
| 408 |
|(4) |
| 409 |
RTP |
| 410 |
| |
| 411 |
v |
| 412 |
+------------+ +--------------+ |
| 413 |
| Network | (7) | Sender | |
| 414 |
+-->| congestion |------>| Transmission | |
| 415 |
| | control | | Control | |
| 416 |
| +------------+ +--------------+ |
| 417 |
| | |
| 418 |
|-------------RTCP----------| |(5) |
| 419 |
(6) | RTP |
| 420 |
| v |
| 421 |
+------------+ |
| 422 |
| UDP | |
| 423 |
| socket | |
| 424 |
+------------+ |
| 425 |
|
| 426 |
Figure 1: SCReAM Sender Functional View |
| 427 |
|
| 428 |
The SCReAM algorithm consists of three main parts: network congestion |
| 429 |
control, sender transmission control, and media rate control. All of |
| 430 |
these parts reside at the sender side. Figure 1 shows the functional |
| 431 |
overview of a SCReAM sender. The receiver-side algorithm is very |
| 432 |
simple in comparison, as it only generates feedback containing |
| 433 |
acknowledgements of received RTP packets and an ECN count. |
| 434 |
|
| 435 |
<span class="h3"><a class="selflink" id="section-3.1" href="#section-3.1">3.1</a>. Network Congestion Control</span> |
| 436 |
|
| 437 |
The network congestion control sets an upper limit on how much data |
| 438 |
can be in the network (bytes in flight); this limit is called CWND |
| 439 |
(congestion window) and is used in the sender transmission control. |
| 440 |
|
| 441 |
|
| 442 |
|
| 443 |
|
| 444 |
<span class="grey">Johansson & Sarker Experimental [Page 8]</span></pre> |
| 445 |
<hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-9" ></span> |
| 446 |
<span class="grey"><a href="./rfc8298">RFC 8298</a> SCReAM December 2017</span> |
| 447 |
|
| 448 |
|
| 449 |
The SCReAM congestion control method uses techniques similar to |
| 450 |
LEDBAT [<a href="./rfc6817" title=""Low Extra Delay Background Transport (LEDBAT)"">RFC6817</a>] to measure the qdelay. As is the case with LEDBAT, |
| 451 |
it is not necessary to use synchronized clocks in the sender and |
| 452 |
receiver in order to compute the qdelay. However, it is necessary |
| 453 |
that they use the same clock frequency, or that the clock frequency |
| 454 |
at the receiver can be inferred reliably by the sender. Failure to |
| 455 |
meet this requirement leads to malfunction in the SCReAM congestion |
| 456 |
control algorithm due to incorrect estimation of the network queue |
| 457 |
delay. |
| 458 |
|
| 459 |
The SCReAM sender calculates the congestion window based on the |
| 460 |
feedback from the SCReAM receiver. The congestion window is allowed |
| 461 |
to increase if the qdelay is below a predefined qdelay target; |
| 462 |
otherwise, the congestion window decreases. The qdelay target is |
| 463 |
typically set to 50-100 ms. This ensures that the queuing delay is |
| 464 |
kept low. The reaction to loss or ECN events leads to an instant |
| 465 |
reduction of CWND. Note that the source rate-limited nature of real- |
| 466 |
time media, such as video, typically means that the queuing delay |
| 467 |
will mostly be below the given delay target. This is contrary to the |
| 468 |
case where large files are transmitted using LEDBAT congestion |
| 469 |
control and the queuing delay will stay close to the delay target. |
| 470 |
|
| 471 |
<span class="h3"><a class="selflink" id="section-3.2" href="#section-3.2">3.2</a>. Sender Transmission Control</span> |
| 472 |
|
| 473 |
The sender transmission control limits the output of data, given by |
| 474 |
the relation between the number of bytes in flight and the congestion |
| 475 |
window. Packet pacing is used to mitigate issues with ACK |
| 476 |
compression that MAY cause increased jitter and/or packet loss in the |
| 477 |
media traffic. Packet pacing limits the packet transmission rate |
| 478 |
given by the estimated link throughput. Even if the send window |
| 479 |
allows for the transmission of a number of packets, these packets are |
| 480 |
not transmitted immediately; rather, they are transmitted in |
| 481 |
intervals given by the packet size and the estimated link throughput. |
| 482 |
|
| 483 |
<span class="h3"><a class="selflink" id="section-3.3" href="#section-3.3">3.3</a>. Media Rate Control</span> |
| 484 |
|
| 485 |
The media rate control serves to adjust the media bitrate to ramp up |
| 486 |
quickly enough to get a fair share of the system resources when link |
| 487 |
throughput increases. |
| 488 |
|
| 489 |
The reaction to reduced throughput MUST be prompt in order to avoid |
| 490 |
getting too much data queued in the RTP packet queue(s) in the |
| 491 |
sender. The media bitrate is decreased if the RTP queue size exceeds |
| 492 |
a threshold. |
| 493 |
|
| 494 |
In cases where the sender's frame queues increase rapidly, such as in |
| 495 |
the case of a Radio Access Type (RAT) handover, the SCReAM sender MAY |
| 496 |
implement additional actions, such as discarding of encoded media |
| 497 |
|
| 498 |
|
| 499 |
|
| 500 |
<span class="grey">Johansson & Sarker Experimental [Page 9]</span></pre> |
| 501 |
<hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-10" ></span> |
| 502 |
<span class="grey"><a href="./rfc8298">RFC 8298</a> SCReAM December 2017</span> |
| 503 |
|
| 504 |
|
| 505 |
frames or frame skipping in order to ensure that the RTP queues are |
| 506 |
drained quickly. Frame skipping results in the frame rate being |
| 507 |
temporarily reduced. Which method to use is a design choice and is |
| 508 |
outside the scope of this algorithm description. |
| 509 |
|
| 510 |
<span class="h2"><a class="selflink" id="section-4" href="#section-4">4</a>. Detailed Description of SCReAM</span> |
| 511 |
|
| 512 |
<span class="h3"><a class="selflink" id="section-4.1" href="#section-4.1">4.1</a>. SCReAM Sender</span> |
| 513 |
|
| 514 |
This section describes the sender-side algorithm in more detail. It |
| 515 |
is split between the network congestion control, sender transmission |
| 516 |
control, and media rate control. |
| 517 |
|
| 518 |
A SCReAM sender implements media rate control and an RTP queue for |
| 519 |
each media type or source, where RTP packets containing encoded media |
| 520 |
frames are temporarily stored for transmission. Figure 1 shows the |
| 521 |
details when a single media source (or stream) is used. A |
| 522 |
transmission scheduler (not shown in the figure) is added to support |
| 523 |
multiple streams. The transmission scheduler can enforce differing |
| 524 |
priorities between the streams and act like a coupled congestion |
| 525 |
controller for multiple flows. Support for multiple streams is |
| 526 |
implemented in [<a href="#ref-SCReAM-CPP-implementation">SCReAM-CPP-implementation</a>]. |
| 527 |
|
| 528 |
Media frames are encoded and forwarded to the RTP queue (1) in |
| 529 |
Figure 1. The media rate adaptation adapts to the size of the RTP |
| 530 |
queue (2) and provides a target rate for the media encoder (3). The |
| 531 |
RTP packets are picked from the RTP queue (4), for multiple flows |
| 532 |
from each RTP queue based on some defined priority order or simply in |
| 533 |
a round-robin fashion, by the sender transmission controller. The |
| 534 |
sender transmission controller (in case of multiple flows a |
| 535 |
transmission scheduler) sends the RTP packets to the UDP socket (5). |
| 536 |
In the general case, all media SHOULD go through the sender |
| 537 |
transmission controller and is limited so that the number of bytes in |
| 538 |
flight is less than the congestion window. RTCP packets are received |
| 539 |
(6) and the information about the bytes in flight and congestion |
| 540 |
window is exchanged between the network congestion control and the |
| 541 |
sender transmission control (7). |
| 542 |
|
| 543 |
<span class="h4"><a class="selflink" id="section-4.1.1" href="#section-4.1.1">4.1.1</a>. Constants and Parameter Values</span> |
| 544 |
|
| 545 |
Constants and state variables are listed in this section. Temporary |
| 546 |
variables are not listed; instead, they are appended with '_t' in the |
| 547 |
pseudocode to indicate their local scope. |
| 548 |
|
| 549 |
|
| 550 |
|
| 551 |
|
| 552 |
|
| 553 |
|
| 554 |
|
| 555 |
|
| 556 |
<span class="grey">Johansson & Sarker Experimental [Page 10]</span></pre> |
| 557 |
<hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-11" ></span> |
| 558 |
<span class="grey"><a href="./rfc8298">RFC 8298</a> SCReAM December 2017</span> |
| 559 |
|
| 560 |
|
| 561 |
<span class="h5"><a class="selflink" id="section-4.1.1.1" href="#section-4.1.1.1">4.1.1.1</a>. Constants</span> |
| 562 |
|
| 563 |
The RECOMMENDED values, within parentheses "()", for the constants |
| 564 |
are deduced from experiments. |
| 565 |
|
| 566 |
QDELAY_TARGET_LO (0.1 s) |
| 567 |
Target value for the minimum qdelay. |
| 568 |
|
| 569 |
QDELAY_TARGET_HI (0.4 s) |
| 570 |
Target value for the maximum qdelay. This parameter provides an |
| 571 |
upper limit to how much the target qdelay (qdelay_target) can be |
| 572 |
increased in order to cope with competing loss-based flows. |
| 573 |
However, the target qdelay does not have to be initialized to this |
| 574 |
high value, as it would increase end-to-end delay and also make the |
| 575 |
rate control and congestion control loops sluggish. |
| 576 |
|
| 577 |
QDELAY_WEIGHT (0.1) |
| 578 |
Averaging factor for qdelay_fraction_avg. |
| 579 |
|
| 580 |
QDELAY_TREND_TH (0.2) |
| 581 |
Threshold for the detection of incipient congestion. |
| 582 |
|
| 583 |
MIN_CWND (3000 bytes) |
| 584 |
Minimum congestion window. |
| 585 |
|
| 586 |
MAX_BYTES_IN_FLIGHT_HEAD_ROOM (1.1) |
| 587 |
Headroom for the limitation of CWND. |
| 588 |
|
| 589 |
GAIN (1.0) |
| 590 |
Gain factor for congestion window adjustment. |
| 591 |
|
| 592 |
BETA_LOSS (0.8) |
| 593 |
CWND scale factor due to loss event. |
| 594 |
|
| 595 |
BETA_ECN (0.9) |
| 596 |
CWND scale factor due to ECN event. |
| 597 |
|
| 598 |
BETA_R (0.9) |
| 599 |
Scale factor for target rate due to loss event. |
| 600 |
|
| 601 |
MSS (1000 byte) |
| 602 |
Maximum segment size = Max RTP packet size. |
| 603 |
|
| 604 |
RATE_ADJUST_INTERVAL (0.2 s) |
| 605 |
Interval between media bitrate adjustments. |
| 606 |
|
| 607 |
TARGET_BITRATE_MIN |
| 608 |
Minimum target bitrate in bps (bits per second). |
| 609 |
|
| 610 |
|
| 611 |
|
| 612 |
<span class="grey">Johansson & Sarker Experimental [Page 11]</span></pre> |
| 613 |
<hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-12" ></span> |
| 614 |
<span class="grey"><a href="./rfc8298">RFC 8298</a> SCReAM December 2017</span> |
| 615 |
|
| 616 |
|
| 617 |
TARGET_BITRATE_MAX |
| 618 |
Maximum target bitrate in bps. |
| 619 |
|
| 620 |
RAMP_UP_SPEED (200000 bps/s) |
| 621 |
Maximum allowed rate increase speed. |
| 622 |
|
| 623 |
PRE_CONGESTION_GUARD (0.0..1.0) |
| 624 |
Guard factor against early congestion onset. A higher value gives |
| 625 |
less jitter, possibly at the expense of a lower link utilization. |
| 626 |
This value MAY be subject to tuning depending on e.g., media coder |
| 627 |
characteristics. Experiments with H264 and VP8 indicate that 0.1 |
| 628 |
is a suitable value. See [<a href="#ref-SCReAM-CPP-implementation">SCReAM-CPP-implementation</a>] and |
| 629 |
[<a href="#ref-SCReAM-implementation-experience">SCReAM-implementation-experience</a>] for evaluation of a real |
| 630 |
implementation. |
| 631 |
|
| 632 |
TX_QUEUE_SIZE_FACTOR (0.0..2.0) |
| 633 |
Guard factor against RTP queue buildup. This value MAY be subject |
| 634 |
to tuning depending on, e.g., media coder characteristics. |
| 635 |
Experiments with H264 and VP8 indicate that 1.0 is a suitable |
| 636 |
value. See [<a href="#ref-SCReAM-CPP-implementation">SCReAM-CPP-implementation</a>] and |
| 637 |
[<a href="#ref-SCReAM-implementation-experience">SCReAM-implementation-experience</a>] for evaluation of a real |
| 638 |
implementation. |
| 639 |
|
| 640 |
RTP_QDELAY_TH (0.02 s) RTP queue delay threshold for a target rate |
| 641 |
reduction. |
| 642 |
|
| 643 |
TARGET_RATE_SCALE_RTP_QDELAY (0.95) Scale factor for target rate |
| 644 |
when RTP qdelay threshold exceeds RTP_QDELAY_TH. |
| 645 |
|
| 646 |
QDELAY_TREND_LO (0.2) Threshold value for qdelay_trend. |
| 647 |
|
| 648 |
T_RESUME_FAST_INCREASE (5 s) Time span until fast increase mode can |
| 649 |
be resumed, given that the qdelay_trend is below QDELAY_TREND_LO. |
| 650 |
|
| 651 |
RATE_PACE_MIN (50000 bps) Minimum pacing rate. |
| 652 |
|
| 653 |
<span class="h5"><a class="selflink" id="section-4.1.1.2" href="#section-4.1.1.2">4.1.1.2</a>. State Variables</span> |
| 654 |
|
| 655 |
The values within parentheses "()" indicate initial values. |
| 656 |
|
| 657 |
qdelay_target (QDELAY_TARGET_LO) |
| 658 |
qdelay target, a variable qdelay target is introduced to manage |
| 659 |
cases where a fixed qdelay target would otherwise starve the RMCAT |
| 660 |
flow under such circumstances (e.g., FTP competes for the bandwidth |
| 661 |
over the same bottleneck). The qdelay target is allowed to vary |
| 662 |
between QDELAY_TARGET_LO and QDELAY_TARGET_HI. |
| 663 |
|
| 664 |
|
| 665 |
|
| 666 |
|
| 667 |
|
| 668 |
<span class="grey">Johansson & Sarker Experimental [Page 12]</span></pre> |
| 669 |
<hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-13" ></span> |
| 670 |
<span class="grey"><a href="./rfc8298">RFC 8298</a> SCReAM December 2017</span> |
| 671 |
|
| 672 |
|
| 673 |
qdelay_fraction_avg (0.0) |
| 674 |
Fractional qdelay filtered by the Exponentially Weighted Moving |
| 675 |
Average (EWMA). |
| 676 |
|
| 677 |
qdelay_fraction_hist[20] ({0,..,0}) |
| 678 |
Vector of the last 20 fractional qdelay samples. |
| 679 |
|
| 680 |
qdelay_trend (0.0) |
| 681 |
qdelay trend; indicates incipient congestion. |
| 682 |
|
| 683 |
qdelay_trend_mem (0.0) |
| 684 |
Low-pass filtered version of qdelay_trend. |
| 685 |
|
| 686 |
qdelay_norm_hist[100] ({0,..,0}) |
| 687 |
Vector of the last 100 normalized qdelay samples. |
| 688 |
|
| 689 |
in_fast_increase (true) |
| 690 |
True if in fast increase mode. |
| 691 |
|
| 692 |
cwnd (MIN_CWND) |
| 693 |
Congestion window. |
| 694 |
|
| 695 |
bytes_newly_acked (0) |
| 696 |
The number of bytes that was acknowledged with the last received |
| 697 |
acknowledgement, i.e., bytes acknowledged since the last CWND |
| 698 |
update. |
| 699 |
|
| 700 |
max_bytes_in_flight (0) |
| 701 |
The maximum number of bytes in flight over a sliding time window, |
| 702 |
i.e., transmitted but not yet acknowledged bytes. |
| 703 |
|
| 704 |
send_wnd (0) |
| 705 |
Upper limit to how many bytes can currently be transmitted. |
| 706 |
Updated when cwnd is updated and when RTP packet is transmitted. |
| 707 |
|
| 708 |
target_bitrate (0 bps) |
| 709 |
Media target bitrate. |
| 710 |
|
| 711 |
target_bitrate_last_max (1 bps) |
| 712 |
Inflection point of the media target bitrate, i.e., the last known |
| 713 |
highest target_bitrate. Used to limit bitrate increase speed close |
| 714 |
to the last known congestion point. |
| 715 |
|
| 716 |
rate_transmit (0.0 bps) |
| 717 |
Measured transmit bitrate. |
| 718 |
|
| 719 |
rate_ack (0.0 bps) |
| 720 |
Measured throughput based on received acknowledgements. |
| 721 |
|
| 722 |
|
| 723 |
|
| 724 |
<span class="grey">Johansson & Sarker Experimental [Page 13]</span></pre> |
| 725 |
<hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-14" ></span> |
| 726 |
<span class="grey"><a href="./rfc8298">RFC 8298</a> SCReAM December 2017</span> |
| 727 |
|
| 728 |
|
| 729 |
rate_media (0.0 bps) |
| 730 |
Measured bitrate from the media encoder. |
| 731 |
|
| 732 |
rate_media_median (0.0 bps) |
| 733 |
Median value of rate_media, computed over more than 10 s. |
| 734 |
|
| 735 |
s_rtt (0.0s) |
| 736 |
Smoothed RTT (in seconds), computed with a similar method to that |
| 737 |
described in [<a href="./rfc6298" title=""Computing TCP's Retransmission Timer"">RFC6298</a>]. |
| 738 |
|
| 739 |
rtp_queue_size (0 bits) |
| 740 |
Sum of the sizes of RTP packets in queue. |
| 741 |
|
| 742 |
rtp_size (0 byte) |
| 743 |
Size of the last transmitted RTP packet. |
| 744 |
|
| 745 |
loss_event_rate (0.0) |
| 746 |
The estimated fraction of RTTs with lost packets detected. |
| 747 |
|
| 748 |
<span class="h4"><a class="selflink" id="section-4.1.2" href="#section-4.1.2">4.1.2</a>. Network Congestion Control</span> |
| 749 |
|
| 750 |
This section explains the network congestion control, which performs |
| 751 |
two main functions: |
| 752 |
|
| 753 |
o Computation of congestion window at the sender: This gives an |
| 754 |
upper limit to the number of bytes in flight. |
| 755 |
|
| 756 |
o Calculation of send window at the sender: RTP packets are |
| 757 |
transmitted if allowed by the relation between the number of bytes |
| 758 |
in flight and the congestion window. This is controlled by the |
| 759 |
send window. |
| 760 |
|
| 761 |
SCReAM is a window-based and byte-oriented congestion control |
| 762 |
protocol, where the number of bytes transmitted is inferred from the |
| 763 |
size of the transmitted RTP packets. Thus, a list of transmitted RTP |
| 764 |
packets and their respective transmission times (wall-clock time) |
| 765 |
MUST be kept for further calculation. |
| 766 |
|
| 767 |
The number of bytes in flight (bytes_in_flight) is computed as the |
| 768 |
sum of the sizes of the RTP packets ranging from the RTP packet most |
| 769 |
recently transmitted, down to but not including the acknowledged |
| 770 |
packet with the highest sequence number. This can be translated to |
| 771 |
the difference between the highest transmitted byte sequence number |
| 772 |
and the highest acknowledged byte sequence number. As an example: If |
| 773 |
an RTP packet with sequence number SN is transmitted and the last |
| 774 |
acknowledgement indicates SN-5 as the highest received sequence |
| 775 |
number, then bytes_in_flight is computed as the sum of the size of |
| 776 |
RTP packets with sequence number SN-4, SN-3, SN-2, SN-1, and SN. It |
| 777 |
|
| 778 |
|
| 779 |
|
| 780 |
<span class="grey">Johansson & Sarker Experimental [Page 14]</span></pre> |
| 781 |
<hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-15" ></span> |
| 782 |
<span class="grey"><a href="./rfc8298">RFC 8298</a> SCReAM December 2017</span> |
| 783 |
|
| 784 |
|
| 785 |
does not matter if, for instance, the packet with sequence number |
| 786 |
SN-3 was lost -- the size of RTP packet with sequence number SN-3 |
| 787 |
will still be considered in the computation of bytes_in_flight. |
| 788 |
|
| 789 |
Furthermore, a variable bytes_newly_acked is incremented with a value |
| 790 |
corresponding to how much the highest sequence number has increased |
| 791 |
since the last feedback. As an example: If the previous |
| 792 |
acknowledgement indicated the highest sequence number N and the new |
| 793 |
acknowledgement indicated N+3, then bytes_newly_acked is incremented |
| 794 |
by a value equal to the sum of the sizes of RTP packets with sequence |
| 795 |
number N+1, N+2, and N+3. Packets that are lost are also included, |
| 796 |
which means that even though, e.g., packet N+2 was lost, its size is |
| 797 |
still included in the update of bytes_newly_acked. The |
| 798 |
bytes_newly_acked variable is reset to zero after a CWND update. |
| 799 |
|
| 800 |
The feedback from the receiver is assumed to consist of the following |
| 801 |
elements. |
| 802 |
|
| 803 |
o A list of received RTP packets' sequence numbers. |
| 804 |
|
| 805 |
o The wall-clock timestamp corresponding to the received RTP packet |
| 806 |
with the highest sequence number. |
| 807 |
|
| 808 |
o The accumulated number of ECN-CE-marked packets (n_ECN). Here, |
| 809 |
"CE" refers to "Congestion Experienced". |
| 810 |
|
| 811 |
When the sender receives RTCP feedback, the qdelay is calculated as |
| 812 |
outlined in [<a href="./rfc6817" title=""Low Extra Delay Background Transport (LEDBAT)"">RFC6817</a>]. A qdelay sample is obtained for each received |
| 813 |
acknowledgement. No smoothing of the qdelay is performed; however, |
| 814 |
some smoothing occurs anyway because the CWND computation is a low- |
| 815 |
pass filter function. A number of variables are updated as |
| 816 |
illustrated by the pseudocode below; temporary variables are appended |
| 817 |
with '_t'. As mentioned in <a href="#section-6">Section 6</a>, calculation of the proper |
| 818 |
congestion window and media bitrate may benefit from additional |
| 819 |
optimizations to handle very high and very low bitrates, and from |
| 820 |
additional damping to handle periodic packet bursts. Some such |
| 821 |
optimizations are implemented in [<a href="#ref-SCReAM-CPP-implementation">SCReAM-CPP-implementation</a>], but |
| 822 |
they do not form part of the specification of SCReAM at this time. |
| 823 |
|
| 824 |
|
| 825 |
|
| 826 |
|
| 827 |
|
| 828 |
|
| 829 |
|
| 830 |
|
| 831 |
|
| 832 |
|
| 833 |
|
| 834 |
|
| 835 |
|
| 836 |
<span class="grey">Johansson & Sarker Experimental [Page 15]</span></pre> |
| 837 |
<hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-16" ></span> |
| 838 |
<span class="grey"><a href="./rfc8298">RFC 8298</a> SCReAM December 2017</span> |
| 839 |
|
| 840 |
|
| 841 |
<CODE BEGINS> |
| 842 |
update_variables(qdelay): |
| 843 |
qdelay_fraction_t = qdelay / qdelay_target |
| 844 |
# Calculate moving average |
| 845 |
qdelay_fraction_avg = (1 - QDELAY_WEIGHT) * qdelay_fraction_avg + |
| 846 |
QDELAY_WEIGHT * qdelay_fraction_t |
| 847 |
update_qdelay_fraction_hist(qdelay_fraction_t) |
| 848 |
# Compute the average of the values in qdelay_fraction_hist |
| 849 |
avg_t = average(qdelay_fraction_hist) |
| 850 |
# R is an autocorrelation function of qdelay_fraction_hist, |
| 851 |
# with the mean (DC component) removed, at lag K |
| 852 |
# The subtraction of the scalar avg_t from |
| 853 |
# qdelay_fraction_hist is performed element-wise |
| 854 |
a_t = R(qdelay_fraction_hist-avg_t, 1) / |
| 855 |
R(qdelay_fraction_hist-avg_t, 0) |
| 856 |
# Calculate qdelay trend |
| 857 |
qdelay_trend = min(1.0, max(0.0, a_t * qdelay_fraction_avg)) |
| 858 |
# Calculate a 'peak-hold' qdelay_trend; this gives a memory |
| 859 |
# of congestion in the past |
| 860 |
qdelay_trend_mem = max(0.99 * qdelay_trend_mem, qdelay_trend) |
| 861 |
<CODE ENDS> |
| 862 |
|
| 863 |
The qdelay fraction is sampled every 50 ms, and the last 20 samples |
| 864 |
are stored in a vector (qdelay_fraction_hist). This vector is used |
| 865 |
in the computation of a qdelay trend that gives a value between 0.0 |
| 866 |
and 1.0 depending on the estimated congestion level. The prediction |
| 867 |
coefficient 'a_t' has positive values if qdelay shows an increasing |
| 868 |
or decreasing trend; thus, an indication of congestion is obtained |
| 869 |
before the qdelay target is reached. As a side effect, if qdelay |
| 870 |
decreases, it's taken as a sign of congestion; however, experiments |
| 871 |
have shown that this is beneficial, as increasing or decreasing queue |
| 872 |
delay is an indication that the transmit rate is very close to the |
| 873 |
path capacity. |
| 874 |
|
| 875 |
The autocorrelation function 'R' is defined as follows. Let x be a |
| 876 |
vector constituting N values, the biased autocorrelation function for |
| 877 |
a given lag=k for the vector x is given by. |
| 878 |
|
| 879 |
n=N-k |
| 880 |
R(x,k) = SUM x(n) * x(n + k) |
| 881 |
n=1 |
| 882 |
|
| 883 |
The prediction coefficient is further multiplied with |
| 884 |
qdelay_fraction_avg to reduce sensitivity to increasing qdelay when |
| 885 |
it is very small. The 50 ms sampling is a simplification that could |
| 886 |
have the effect that the same qdelay is sampled several times; |
| 887 |
however, this does not pose any problem, as the vector is only used |
| 888 |
to determine if the qdelay is increasing or decreasing. The |
| 889 |
|
| 890 |
|
| 891 |
|
| 892 |
<span class="grey">Johansson & Sarker Experimental [Page 16]</span></pre> |
| 893 |
<hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-17" ></span> |
| 894 |
<span class="grey"><a href="./rfc8298">RFC 8298</a> SCReAM December 2017</span> |
| 895 |
|
| 896 |
|
| 897 |
qdelay_trend is utilized in the media rate control to indicate |
| 898 |
incipient congestion and to determine when to exit from fast increase |
| 899 |
mode. qdelay_trend_mem is used to enforce a less aggressive rate |
| 900 |
increase after congestion events. The function |
| 901 |
update_qdelay_fraction_hist(..) removes the oldest element and adds |
| 902 |
the latest qdelay_fraction element to the qdelay_fraction_hist |
| 903 |
vector. |
| 904 |
|
| 905 |
<span class="h5"><a class="selflink" id="section-4.1.2.1" href="#section-4.1.2.1">4.1.2.1</a>. Reaction to Packet Loss and ECN</span> |
| 906 |
|
| 907 |
A loss event is indicated if one or more RTP packets are declared |
| 908 |
missing. The loss detection is described in <a href="#section-4.1.2.4">Section 4.1.2.4</a>. Once a |
| 909 |
loss event is detected, further detected lost RTP packets SHOULD be |
| 910 |
ignored for a full smoothed round-trip time; the intention is to |
| 911 |
limit the congestion window decrease to at most once per round trip. |
| 912 |
|
| 913 |
The congestion window back-off due to loss events is deliberately a |
| 914 |
bit less than is the case with TCP Reno, for example. TCP is |
| 915 |
generally used to transmit whole files; the file is then like a |
| 916 |
source with an infinite bitrate until the whole file has been |
| 917 |
transmitted. SCReAM, on the other hand, has a source whose rate is |
| 918 |
limited to a value close to the available transmit rate and often |
| 919 |
below that value; the effect is that SCReAM has less opportunity to |
| 920 |
grab free capacity than a TCP-based file transfer. To compensate for |
| 921 |
this, it is RECOMMENDED to let SCReAM reduce the congestion window |
| 922 |
less than what is the case with TCP when loss events occur. |
| 923 |
|
| 924 |
An ECN event is detected if the n_ECN counter in the feedback report |
| 925 |
has increased since the previous received feedback. Once an ECN |
| 926 |
event is detected, the n_ECN counter is ignored for a full smoothed |
| 927 |
round-trip time; the intention is to limit the congestion window |
| 928 |
decrease to at most once per round trip. The congestion window back- |
| 929 |
off due to an ECN event MAY be smaller than if a loss event occurs. |
| 930 |
This is in line with the idea outlined in [<a href="#ref-ALT-BACKOFF">ALT-BACKOFF</a>] to enable ECN |
| 931 |
marking thresholds lower than the corresponding packet drop |
| 932 |
thresholds. |
| 933 |
|
| 934 |
<span class="h5"><a class="selflink" id="section-4.1.2.2" href="#section-4.1.2.2">4.1.2.2</a>. Congestion Window Update</span> |
| 935 |
|
| 936 |
The update of the congestion window depends on if loss, ECN-marking, |
| 937 |
or neither of the two occurs. The pseudocode below describes the |
| 938 |
actions for each case. |
| 939 |
|
| 940 |
|
| 941 |
|
| 942 |
|
| 943 |
|
| 944 |
|
| 945 |
|
| 946 |
|
| 947 |
|
| 948 |
<span class="grey">Johansson & Sarker Experimental [Page 17]</span></pre> |
| 949 |
<hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-18" ></span> |
| 950 |
<span class="grey"><a href="./rfc8298">RFC 8298</a> SCReAM December 2017</span> |
| 951 |
|
| 952 |
|
| 953 |
<CODE BEGINS> |
| 954 |
on congestion event(qdelay): |
| 955 |
# Either loss or ECN mark is detected |
| 956 |
in_fast_increase = false |
| 957 |
if (is loss) |
| 958 |
# Loss is detected |
| 959 |
cwnd = max(MIN_CWND, cwnd * BETA_LOSS) |
| 960 |
else |
| 961 |
# No loss, so it is then an ECN mark |
| 962 |
cwnd = max(MIN_CWND, cwnd * BETA_ECN) |
| 963 |
end |
| 964 |
adjust_qdelay_target(qdelay) #compensating for competing flows |
| 965 |
calculate_send_window(qdelay, qdelay_target) |
| 966 |
|
| 967 |
# When no congestion event |
| 968 |
on acknowledgement(qdelay): |
| 969 |
update_bytes_newly_acked() |
| 970 |
update_cwnd(bytes_newly_acked) |
| 971 |
adjust_qdelay_target(qdelay) # compensating for competing flows |
| 972 |
calculate_send_window(qdelay, qdelay_target) |
| 973 |
check_to_resume_fast_increase() |
| 974 |
<CODE ENDS> |
| 975 |
|
| 976 |
The methods are described in detail below. |
| 977 |
|
| 978 |
The congestion window update is based on qdelay, except for the |
| 979 |
occurrence of loss events (one or more lost RTP packets in one RTT) |
| 980 |
or ECN events, which were described earlier. |
| 981 |
|
| 982 |
Pseudocode for the update of the congestion window is found below. |
| 983 |
|
| 984 |
|
| 985 |
|
| 986 |
|
| 987 |
|
| 988 |
|
| 989 |
|
| 990 |
|
| 991 |
|
| 992 |
|
| 993 |
|
| 994 |
|
| 995 |
|
| 996 |
|
| 997 |
|
| 998 |
|
| 999 |
|
| 1000 |
|
| 1001 |
|
| 1002 |
|
| 1003 |
|
| 1004 |
<span class="grey">Johansson & Sarker Experimental [Page 18]</span></pre> |
| 1005 |
<hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-19" ></span> |
| 1006 |
<span class="grey"><a href="./rfc8298">RFC 8298</a> SCReAM December 2017</span> |
| 1007 |
|
| 1008 |
|
| 1009 |
<CODE BEGINS> |
| 1010 |
update_cwnd(bytes_newly_acked): |
| 1011 |
# In fast increase mode? |
| 1012 |
if (in_fast_increase) |
| 1013 |
if (qdelay_trend >= QDELAY_TREND_TH) |
| 1014 |
# Incipient congestion detected; exit fast increase mode |
| 1015 |
in_fast_increase = false |
| 1016 |
else |
| 1017 |
# No congestion yet; increase cwnd if it |
| 1018 |
# is sufficiently used |
| 1019 |
# Additional slack of bytes_newly_acked is |
| 1020 |
# added to ensure that CWND growth occurs |
| 1021 |
# even when feedback is sparse |
| 1022 |
if (bytes_in_flight * 1.5 + bytes_newly_acked > cwnd) |
| 1023 |
cwnd = cwnd + bytes_newly_acked |
| 1024 |
end |
| 1025 |
return |
| 1026 |
end |
| 1027 |
end |
| 1028 |
|
| 1029 |
# Not in fast increase mode |
| 1030 |
# off_target calculated as with LEDBAT |
| 1031 |
off_target_t = (qdelay_target - qdelay) / qdelay_target |
| 1032 |
|
| 1033 |
gain_t = GAIN |
| 1034 |
# Adjust congestion window |
| 1035 |
cwnd_delta_t = |
| 1036 |
gain_t * off_target_t * bytes_newly_acked * MSS / cwnd |
| 1037 |
if (off_target_t > 0 && |
| 1038 |
bytes_in_flight * 1.25 + bytes_newly_acked <= cwnd) |
| 1039 |
# No cwnd increase if window is underutilized |
| 1040 |
# Additional slack of bytes_newly_acked is |
| 1041 |
# added to ensure that CWND growth occurs |
| 1042 |
# even when feedback is sparse |
| 1043 |
cwnd_delta_t = 0; |
| 1044 |
end |
| 1045 |
|
| 1046 |
# Apply delta |
| 1047 |
cwnd += cwnd_delta_t |
| 1048 |
# limit cwnd to the maximum number of bytes in flight |
| 1049 |
cwnd = min(cwnd, max_bytes_in_flight * |
| 1050 |
MAX_BYTES_IN_FLIGHT_HEAD_ROOM) |
| 1051 |
cwnd = max(cwnd, MIN_CWND) |
| 1052 |
|
| 1053 |
<CODE ENDS> |
| 1054 |
|
| 1055 |
|
| 1056 |
|
| 1057 |
|
| 1058 |
|
| 1059 |
|
| 1060 |
<span class="grey">Johansson & Sarker Experimental [Page 19]</span></pre> |
| 1061 |
<hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-20" ></span> |
| 1062 |
<span class="grey"><a href="./rfc8298">RFC 8298</a> SCReAM December 2017</span> |
| 1063 |
|
| 1064 |
|
| 1065 |
CWND is updated differently depending on whether or not the |
| 1066 |
congestion control is in fast increase mode, as controlled by the |
| 1067 |
variable in_fast_increase. |
| 1068 |
|
| 1069 |
When in fast increase mode, the congestion window is increased with |
| 1070 |
the number of newly acknowledged bytes as long as the window is |
| 1071 |
sufficiently used. Sparse feedback can potentially limit congestion |
| 1072 |
window growth; therefore, additional slack is added, given by the |
| 1073 |
number of newly acknowledged bytes. |
| 1074 |
|
| 1075 |
The congestion window growth when in_fast_increase is false is |
| 1076 |
dictated by the relation between qdelay and qdelay_target; congestion |
| 1077 |
window growth is limited if the window is not used sufficiently. |
| 1078 |
|
| 1079 |
SCReAM calculates the GAIN in a similar way to what is specified in |
| 1080 |
[<a href="./rfc6817" title=""Low Extra Delay Background Transport (LEDBAT)"">RFC6817</a>]. However, [<a href="./rfc6817" title=""Low Extra Delay Background Transport (LEDBAT)"">RFC6817</a>] specifies that the CWND increase is |
| 1081 |
limited by an additional function controlled by a constant |
| 1082 |
ALLOWED_INCREASE. This additional limitation is removed in this |
| 1083 |
specification. |
| 1084 |
|
| 1085 |
Further, the CWND is limited by max_bytes_in_flight and MIN_CWND. |
| 1086 |
The limitation of the congestion window by the maximum number of |
| 1087 |
bytes in flight over the last 5 seconds (max_bytes_in_flight) avoids |
| 1088 |
possible overestimation of the throughput after, for example, idle |
| 1089 |
periods. An additional MAX_BYTES_IN_FLIGHT_HEAD_ROOM provides slack |
| 1090 |
to allow for a certain amount of variability in the media coder |
| 1091 |
output rate. |
| 1092 |
|
| 1093 |
<span class="h5"><a class="selflink" id="section-4.1.2.3" href="#section-4.1.2.3">4.1.2.3</a>. Competing Flows Compensation</span> |
| 1094 |
|
| 1095 |
It is likely that a flow using the SCReAM algorithm will have to |
| 1096 |
share congested bottlenecks with other flows that use a more |
| 1097 |
aggressive congestion control algorithm (for example, large FTP flows |
| 1098 |
using loss-based congestion control). The worst condition occurs |
| 1099 |
when the bottleneck queues are of tail-drop type with a large buffer |
| 1100 |
size. SCReAM takes care of such situations by adjusting the |
| 1101 |
qdelay_target when loss-based flows are detected, as shown in the |
| 1102 |
pseudocode below. |
| 1103 |
|
| 1104 |
|
| 1105 |
|
| 1106 |
|
| 1107 |
|
| 1108 |
|
| 1109 |
|
| 1110 |
|
| 1111 |
|
| 1112 |
|
| 1113 |
|
| 1114 |
|
| 1115 |
|
| 1116 |
<span class="grey">Johansson & Sarker Experimental [Page 20]</span></pre> |
| 1117 |
<hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-21" ></span> |
| 1118 |
<span class="grey"><a href="./rfc8298">RFC 8298</a> SCReAM December 2017</span> |
| 1119 |
|
| 1120 |
|
| 1121 |
<CODE BEGINS> |
| 1122 |
adjust_qdelay_target(qdelay) |
| 1123 |
qdelay_norm_t = qdelay / QDELAY_TARGET_LOW |
| 1124 |
update_qdelay_norm_history(qdelay_norm_t) |
| 1125 |
# Compute variance |
| 1126 |
qdelay_norm_var_t = VARIANCE(qdelay_norm_history(200)) |
| 1127 |
# Compensation for competing traffic |
| 1128 |
# Compute average |
| 1129 |
qdelay_norm_avg_t = AVERAGE(qdelay_norm_history(50)) |
| 1130 |
# Compute upper limit to target delay |
| 1131 |
new_target_t = qdelay_norm_avg_t + sqrt(qdelay_norm_var_t) |
| 1132 |
new_target_t *= QDELAY_TARGET_LO |
| 1133 |
if (loss_event_rate > 0.002) |
| 1134 |
# Packet losses detected |
| 1135 |
qdelay_target = 1.5 * new_target_t |
| 1136 |
else |
| 1137 |
if (qdelay_norm_var_t < 0.2) |
| 1138 |
# Reasonably safe to set target qdelay |
| 1139 |
qdelay_target = new_target_t |
| 1140 |
else |
| 1141 |
# Check if target delay can be reduced; this helps prevent |
| 1142 |
# the target delay from being locked to high values forever |
| 1143 |
if (new_target_t < QDELAY_TARGET_LO) |
| 1144 |
# Decrease target delay quickly, as measured queuing |
| 1145 |
# delay is lower than target |
| 1146 |
qdelay_target = max(qdelay_target * 0.5, new_target_t) |
| 1147 |
else |
| 1148 |
# Decrease target delay slowly |
| 1149 |
qdelay_target *= 0.9 |
| 1150 |
end |
| 1151 |
end |
| 1152 |
end |
| 1153 |
|
| 1154 |
# Apply limits |
| 1155 |
qdelay_target = min(QDELAY_TARGET_HI, qdelay_target) |
| 1156 |
qdelay_target = max(QDELAY_TARGET_LO, qdelay_target) |
| 1157 |
<CODE ENDS> |
| 1158 |
|
| 1159 |
Two temporary variables are calculated. qdelay_norm_avg_t is the |
| 1160 |
long-term average queue delay, qdelay_norm_var_t is the long-term |
| 1161 |
variance of the queue delay. A high qdelay_norm_var_t indicates that |
| 1162 |
the queue delay changes; this can be an indication that bottleneck |
| 1163 |
bandwidth is reduced or that a competing flow has just entered. |
| 1164 |
Thus, it indicates that it is not safe to adjust the queue delay |
| 1165 |
target. |
| 1166 |
|
| 1167 |
A low qdelay_norm_var_t indicates that the queue delay is relatively |
| 1168 |
stable. The reason could be that the queue delay is low, but it |
| 1169 |
|
| 1170 |
|
| 1171 |
|
| 1172 |
<span class="grey">Johansson & Sarker Experimental [Page 21]</span></pre> |
| 1173 |
<hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-22" ></span> |
| 1174 |
<span class="grey"><a href="./rfc8298">RFC 8298</a> SCReAM December 2017</span> |
| 1175 |
|
| 1176 |
|
| 1177 |
could also be that a competing flow is causing the bottleneck to |
| 1178 |
reach the point that packet losses start to occur, in which case the |
| 1179 |
queue delay will stay relatively high for a longer time. |
| 1180 |
|
| 1181 |
The queue delay target is allowed to be increased if either the loss |
| 1182 |
event rate is above a given threshold or qdelay_norm_var_t is low. |
| 1183 |
Both these conditions indicate that a competing flow may be present. |
| 1184 |
In all other cases, the queue delay target is decreased. |
| 1185 |
|
| 1186 |
The function that adjusts the qdelay_target is simple and could |
| 1187 |
produce false positives and false negatives. The case that self- |
| 1188 |
inflicted congestion by the SCReAM algorithm may be falsely |
| 1189 |
interpreted as the presence of competing loss-based FTP flows is a |
| 1190 |
false positive. The opposite case -- where the algorithm fails to |
| 1191 |
detect the presence of a competing FTP flow -- is a false negative. |
| 1192 |
|
| 1193 |
Extensive simulations have shown that the algorithm performs well in |
| 1194 |
LTE test cases and that it also performs well in simple bandwidth- |
| 1195 |
limited bottleneck test cases with competing FTP flows. However, the |
| 1196 |
potential failure of the algorithm cannot be completely ruled out. A |
| 1197 |
false positive (i.e., when self-inflicted congestion is mistakenly |
| 1198 |
identified as competing flows) is especially problematic when it |
| 1199 |
leads to increasing the target queue delay, which can cause the end- |
| 1200 |
to-end delay to increase dramatically. |
| 1201 |
|
| 1202 |
If it is deemed unlikely that competing flows occur over the same |
| 1203 |
bottleneck, the algorithm described in this section MAY be turned |
| 1204 |
off. One such case is QoS-enabled bearers in 3GPP-based access such |
| 1205 |
as LTE. However, when sending over the Internet, often the network |
| 1206 |
conditions are not known for sure, so in general it is not possible |
| 1207 |
to make safe assumptions on how a network is used and whether or not |
| 1208 |
competing flows share the same bottleneck. Therefore, turning this |
| 1209 |
algorithm off must be considered with caution, as it can lead to |
| 1210 |
basically zero throughput if competing with loss-based traffic. |
| 1211 |
|
| 1212 |
<span class="h5"><a class="selflink" id="section-4.1.2.4" href="#section-4.1.2.4">4.1.2.4</a>. Lost Packet Detection</span> |
| 1213 |
|
| 1214 |
Lost packet detection is based on the received sequence number list. |
| 1215 |
A reordering window SHOULD be applied to prevent packet reordering |
| 1216 |
from triggering loss events. The reordering window is specified as a |
| 1217 |
time unit, similar to the ideas behind Recent ACKnowledgement (RACK) |
| 1218 |
[<a href="#ref-RACK" title=""RACK: a time- based fast loss detection algorithm for TCP"">RACK</a>]. The computation of the reordering window is made possible by |
| 1219 |
means of a lost flag in the list of transmitted RTP packets. This |
| 1220 |
flag is set if the received sequence number list indicates that the |
| 1221 |
given RTP packet is missing. If later feedback indicates that a |
| 1222 |
previously lost marked packet was indeed received, then the |
| 1223 |
reordering window is updated to reflect the reordering delay. The |
| 1224 |
reordering window is given by the difference in time between the |
| 1225 |
|
| 1226 |
|
| 1227 |
|
| 1228 |
<span class="grey">Johansson & Sarker Experimental [Page 22]</span></pre> |
| 1229 |
<hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-23" ></span> |
| 1230 |
<span class="grey"><a href="./rfc8298">RFC 8298</a> SCReAM December 2017</span> |
| 1231 |
|
| 1232 |
|
| 1233 |
event that the packet was marked as lost and the event that it was |
| 1234 |
indicated as successfully received. Loss is detected if a given RTP |
| 1235 |
packet is not acknowledged within a time window (indicated by the |
| 1236 |
reordering window) after an RTP packet with a higher sequence number |
| 1237 |
was acknowledged. |
| 1238 |
|
| 1239 |
<span class="h5"><a class="selflink" id="section-4.1.2.5" href="#section-4.1.2.5">4.1.2.5</a>. Send Window Calculation</span> |
| 1240 |
|
| 1241 |
The basic design principle behind packet transmission in SCReAM is to |
| 1242 |
allow transmission only if the number of bytes in flight is less than |
| 1243 |
the congestion window. There are, however, two reasons why this |
| 1244 |
strict rule will not work optimally: |
| 1245 |
|
| 1246 |
o Bitrate variations: Media sources such as video encoders generally |
| 1247 |
produce frames whose size always vary to a larger or smaller |
| 1248 |
extent. The RTP queue absorbs the natural variations in frame |
| 1249 |
sizes. However, the RTP queue should be as short as possible to |
| 1250 |
prevent the end-to-end delay from increasing. To achieve that, |
| 1251 |
the media rate control takes the RTP queue size into account when |
| 1252 |
the target bitrate for the media is computed. A strict 'send only |
| 1253 |
when bytes in flight is less than the congestion window' rule can |
| 1254 |
cause the RTP queue to grow simply because the send window is |
| 1255 |
limited; in turn, this can cause the target bitrate to be pushed |
| 1256 |
down. The consequence is that the congestion window will not |
| 1257 |
increase, or will increase very slowly, because the congestion |
| 1258 |
window is only allowed to increase when there is a sufficient |
| 1259 |
amount of data in flight. The final effect is that the media |
| 1260 |
bitrate increases very slowly or not at all. |
| 1261 |
|
| 1262 |
o Reverse (feedback) path congestion: Especially in transport over |
| 1263 |
buffer-bloated networks, the one-way delay in the reverse |
| 1264 |
direction can jump due to congestion. The effect is that the |
| 1265 |
acknowledgements are delayed, and the self-clocking is temporarily |
| 1266 |
halted, even though the forward path is not congested. |
| 1267 |
|
| 1268 |
The send window is adjusted depending on qdelay, its relation to the |
| 1269 |
qdelay target, and the relation between the congestion window and the |
| 1270 |
number of bytes in flight. A strict rule is applied when qdelay is |
| 1271 |
higher than qdelay_target, to avoid further queue buildup in the |
| 1272 |
network. For cases when qdelay is lower than the qdelay_target, a |
| 1273 |
more relaxed rule is applied. This allows the bitrate to increase |
| 1274 |
quickly when no congestion is detected while still being able to |
| 1275 |
exhibit stable behavior in congested situations. |
| 1276 |
|
| 1277 |
The send window is given by the relation between the adjusted |
| 1278 |
congestion window and the amount of bytes in flight according to the |
| 1279 |
pseudocode below. |
| 1280 |
|
| 1281 |
|
| 1282 |
|
| 1283 |
|
| 1284 |
<span class="grey">Johansson & Sarker Experimental [Page 23]</span></pre> |
| 1285 |
<hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-24" ></span> |
| 1286 |
<span class="grey"><a href="./rfc8298">RFC 8298</a> SCReAM December 2017</span> |
| 1287 |
|
| 1288 |
|
| 1289 |
<CODE BEGINS> |
| 1290 |
calculate_send_window(qdelay, qdelay_target) |
| 1291 |
# send window is computed differently depending on congestion level |
| 1292 |
if (qdelay <= qdelay_target) |
| 1293 |
send_wnd = cwnd + MSS - bytes_in_flight |
| 1294 |
else |
| 1295 |
send_wnd = cwnd - bytes_in_flight |
| 1296 |
end |
| 1297 |
<CODE ENDS> |
| 1298 |
|
| 1299 |
The send window is updated whenever an RTP packet is transmitted or |
| 1300 |
an RTCP feedback messaged is received. |
| 1301 |
|
| 1302 |
<span class="h5"><a class="selflink" id="section-4.1.2.6" href="#section-4.1.2.6">4.1.2.6</a>. Packet Pacing</span> |
| 1303 |
|
| 1304 |
Packet pacing is used in order to mitigate coalescing, i.e., when |
| 1305 |
packets are transmitted in bursts, with the risks of increased jitter |
| 1306 |
and potentially increased packet loss. Packet pacing also mitigates |
| 1307 |
possible issues with queue overflow due to key-frame generation in |
| 1308 |
video coders. The time interval between consecutive packet |
| 1309 |
transmissions is greater than or equal to t_pace, where t_pace is |
| 1310 |
given by the equations below : |
| 1311 |
|
| 1312 |
<CODE BEGINS> |
| 1313 |
pace_bitrate = max (RATE_PACE_MIN, cwnd * 8 / s_rtt) |
| 1314 |
t_pace = rtp_size * 8 / pace_bitrate |
| 1315 |
<CODE ENDS> |
| 1316 |
|
| 1317 |
rtp_size is the size of the last transmitted RTP packet, and s_rtt is |
| 1318 |
the smoothed round trip time. RATE_PACE_MIN is the minimum pacing |
| 1319 |
rate. |
| 1320 |
|
| 1321 |
<span class="h5"><a class="selflink" id="section-4.1.2.7" href="#section-4.1.2.7">4.1.2.7</a>. Resuming Fast Increase Mode</span> |
| 1322 |
|
| 1323 |
Fast increase mode can resume in order to speed up the bitrate |
| 1324 |
increase if congestion abates. The condition to resume fast increase |
| 1325 |
mode (in_fast_increase = true) is that qdelay_trend is less than |
| 1326 |
QDELAY_TREND_LO for T_RESUME_FAST_INCREASE seconds or more. |
| 1327 |
|
| 1328 |
<span class="h5"><a class="selflink" id="section-4.1.2.8" href="#section-4.1.2.8">4.1.2.8</a>. Stream Prioritization</span> |
| 1329 |
|
| 1330 |
The SCReAM algorithm makes a good distinction between network |
| 1331 |
congestion control and media rate control. This is easily extended |
| 1332 |
to many streams -- RTP packets from two or more RTP queues are |
| 1333 |
scheduled at the rate permitted by the network congestion control. |
| 1334 |
|
| 1335 |
The scheduling can be done by means of a few different scheduling |
| 1336 |
regimes. For example, the method for coupled congestion control |
| 1337 |
|
| 1338 |
|
| 1339 |
|
| 1340 |
<span class="grey">Johansson & Sarker Experimental [Page 24]</span></pre> |
| 1341 |
<hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-25" ></span> |
| 1342 |
<span class="grey"><a href="./rfc8298">RFC 8298</a> SCReAM December 2017</span> |
| 1343 |
|
| 1344 |
|
| 1345 |
specified in [<a href="#ref-COUPLED-CC">COUPLED-CC</a>] can be used. One implementation of SCReAM |
| 1346 |
[<a href="#ref-SCReAM-CPP-implementation">SCReAM-CPP-implementation</a>] uses credit-based scheduling. In credit- |
| 1347 |
based scheduling, credit is accumulated by queues as they wait for |
| 1348 |
service and is spent while the queues are being serviced. For |
| 1349 |
instance, if one queue is allowed to transmit 1000 bytes, then a |
| 1350 |
credit of 1000 bytes is allocated to the other unscheduled queues. |
| 1351 |
This principle can be extended to weighted scheduling, where the |
| 1352 |
credit allocated to unscheduled queues depends on the relative |
| 1353 |
weights. The latter is also implemented in |
| 1354 |
[<a href="#ref-SCReAM-CPP-implementation">SCReAM-CPP-implementation</a>]. |
| 1355 |
|
| 1356 |
<span class="h4"><a class="selflink" id="section-4.1.3" href="#section-4.1.3">4.1.3</a>. Media Rate Control</span> |
| 1357 |
|
| 1358 |
The media rate control algorithm is executed at regular intervals, |
| 1359 |
indicated by RATE_ADJUSTMENT_INTERVAL, with the exception of a prompt |
| 1360 |
reaction to loss events. The media rate control operates based on |
| 1361 |
the size of the RTP packet send queue and observed loss events. In |
| 1362 |
addition, qdelay_trend is also considered in the media rate control |
| 1363 |
in order to reduce the amount of induced network jitter. |
| 1364 |
|
| 1365 |
The role of the media rate control is to strike a reasonable balance |
| 1366 |
between a low amount of queuing in the RTP queue(s) and a sufficient |
| 1367 |
amount of data to send in order to keep the data path busy. Setting |
| 1368 |
the media rate control too cautiously leads to possible |
| 1369 |
underutilization of network capacity; this can cause the flow to |
| 1370 |
become starved out by other more opportunistic traffic. On the other |
| 1371 |
hand, setting it too aggressively leads to increased jitter. |
| 1372 |
|
| 1373 |
The target_bitrate is adjusted depending on the congestion state. |
| 1374 |
The target bitrate can vary between a minimum value |
| 1375 |
(TARGET_BITRATE_MIN) and a maximum value (TARGET_BITRATE_MAX). |
| 1376 |
TARGET_BITRATE_MIN SHOULD be set to a low enough value to prevent RTP |
| 1377 |
packets from becoming queued up when the network throughput is |
| 1378 |
reduced. The sender SHOULD also be equipped with a mechanism that |
| 1379 |
discards RTP packets when the network throughput becomes very low and |
| 1380 |
RTP packets are excessively delayed. |
| 1381 |
|
| 1382 |
For the overall bitrate adjustment, two network throughput estimates |
| 1383 |
are computed : |
| 1384 |
|
| 1385 |
o rate_transmit: The measured transmit bitrate. |
| 1386 |
|
| 1387 |
o rate_ack: The ACKed bitrate, i.e., the volume of ACKed bits per |
| 1388 |
second. |
| 1389 |
|
| 1390 |
Both estimates are updated every 200 ms. |
| 1391 |
|
| 1392 |
|
| 1393 |
|
| 1394 |
|
| 1395 |
|
| 1396 |
<span class="grey">Johansson & Sarker Experimental [Page 25]</span></pre> |
| 1397 |
<hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-26" ></span> |
| 1398 |
<span class="grey"><a href="./rfc8298">RFC 8298</a> SCReAM December 2017</span> |
| 1399 |
|
| 1400 |
|
| 1401 |
The current throughput, current_rate, is computed as the maximum |
| 1402 |
value of rate_transmit and rate_ack. The rationale behind the use of |
| 1403 |
rate_ack in addition to rate_transmit is that rate_transmit is |
| 1404 |
affected also by the amount of data that is available to transmit, |
| 1405 |
thus a lack of data to transmit can be seen as reduced throughput |
| 1406 |
that can cause an unnecessary rate reduction. To overcome this |
| 1407 |
shortcoming, rate_ack is used as well. This gives a more stable |
| 1408 |
throughput estimate. |
| 1409 |
|
| 1410 |
The rate change behavior depends on whether a loss or ECN event has |
| 1411 |
occurred and whether the congestion control is in fast increase mode. |
| 1412 |
|
| 1413 |
<CODE BEGINS> |
| 1414 |
# The target_bitrate is updated at a regular interval according |
| 1415 |
# to RATE_ADJUST_INTERVAL |
| 1416 |
|
| 1417 |
on loss: |
| 1418 |
# Loss event detected |
| 1419 |
target_bitrate = max(BETA_R * target_bitrate, |
| 1420 |
TARGET_BITRATE_MIN) |
| 1421 |
exit |
| 1422 |
on ecn_mark: |
| 1423 |
# ECN event detected |
| 1424 |
target_bitrate = max(BETA_ECN * target_bitrate, |
| 1425 |
TARGET_BITRATE_MIN) |
| 1426 |
exit |
| 1427 |
|
| 1428 |
ramp_up_speed_t = min(RAMP_UP_SPEED, target_bitrate / 2.0) |
| 1429 |
scale_t = (target_bitrate - target_bitrate_last_max) / |
| 1430 |
target_bitrate_last_max |
| 1431 |
scale_t = max(0.2, min(1.0, (scale_t * 4)^2)) |
| 1432 |
# min scale_t value 0.2, as the bitrate should be allowed to |
| 1433 |
# increase slowly. This prevents locking the rate to |
| 1434 |
# target_bitrate_last_max |
| 1435 |
if (in_fast_increase = true) |
| 1436 |
increment_t = ramp_up_speed_t * RATE_ADJUST_INTERVAL |
| 1437 |
increment_t *= scale_t |
| 1438 |
target_bitrate += increment_t |
| 1439 |
else |
| 1440 |
current_rate_t = max(rate_transmit, rate_ack) |
| 1441 |
# Compute a bitrate change |
| 1442 |
delta_rate_t = current_rate_t * (1.0 - PRE_CONGESTION_GUARD * |
| 1443 |
queue_delay_trend) - TX_QUEUE_SIZE_FACTOR * rtp_queue_size |
| 1444 |
# Limit a positive increase if close to target_bitrate_last_max |
| 1445 |
if (delta_rate_t > 0) |
| 1446 |
delta_rate_t *= scale_t |
| 1447 |
delta_rate_t = |
| 1448 |
min(delta_rate_t, ramp_up_speed_t * RATE_ADJUST_INTERVAL) |
| 1449 |
|
| 1450 |
|
| 1451 |
|
| 1452 |
<span class="grey">Johansson & Sarker Experimental [Page 26]</span></pre> |
| 1453 |
<hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-27" ></span> |
| 1454 |
<span class="grey"><a href="./rfc8298">RFC 8298</a> SCReAM December 2017</span> |
| 1455 |
|
| 1456 |
|
| 1457 |
end |
| 1458 |
target_bitrate += delta_rate_t |
| 1459 |
# Force a slight reduction in bitrate if RTP queue |
| 1460 |
# builds up |
| 1461 |
rtp_queue_delay_t = rtp_queue_size / current_rate_t |
| 1462 |
if (rtp_queue_delay_t > RTP_QDELAY_TH) |
| 1463 |
target_bitrate *= TARGET_RATE_SCALE_RTP_QDELAY |
| 1464 |
end |
| 1465 |
end |
| 1466 |
|
| 1467 |
rate_media_limit_t = |
| 1468 |
max(current_rate_t, max(rate_media, rtp_rate_median)) |
| 1469 |
rate_media_limit_t *= (2.0 - qdelay_trend_mem) |
| 1470 |
target_bitrate = min(target_bitrate, rate_media_limit_t) |
| 1471 |
target_bitrate = min(TARGET_BITRATE_MAX, |
| 1472 |
max(TARGET_BITRATE_MIN, target_bitrate)) |
| 1473 |
<CODE ENDS> |
| 1474 |
|
| 1475 |
In case of a loss event, the target_bitrate is updated and the rate |
| 1476 |
change procedure is exited. Otherwise, the rate change procedure |
| 1477 |
continues. The rationale behind the rate reduction due to loss is |
| 1478 |
that a congestion window reduction will take effect, and a rate |
| 1479 |
reduction proactively prevents RTP packets from being queued up when |
| 1480 |
the transmit rate decreases due to the reduced congestion window. A |
| 1481 |
similar rate reduction happens when ECN events are detected. |
| 1482 |
|
| 1483 |
The rate update frequency is limited by RATE_ADJUST_INTERVAL, unless |
| 1484 |
a loss event occurs. The value is based on experimentation with |
| 1485 |
real-life limitations in video coders taken into account |
| 1486 |
[<a href="#ref-SCReAM-CPP-implementation">SCReAM-CPP-implementation</a>]. A too short interval is shown to make |
| 1487 |
the rate control loop in video coders more unstable; a too long |
| 1488 |
interval makes the overall congestion control sluggish. |
| 1489 |
|
| 1490 |
When in fast increase mode (in_fast_increase = true), the bitrate |
| 1491 |
increase is given by the desired ramp-up speed (RAMP_UP_SPEED). The |
| 1492 |
ramp-up speed is limited when the target bitrate is low to avoid rate |
| 1493 |
oscillation at low bottleneck bitrates. The setting of RAMP_UP_SPEED |
| 1494 |
depends on preferences. A high setting such as 1000 kbps/s makes it |
| 1495 |
possible to quickly get high-quality media; however, this is at the |
| 1496 |
expense of increased jitter, which can manifest itself as choppy |
| 1497 |
video rendering, for example. |
| 1498 |
|
| 1499 |
When in_fast_increase is false, the bitrate increase is given by the |
| 1500 |
current bitrate and is also controlled by the estimated RTP queue and |
| 1501 |
the qdelay trend, thus it is sufficient that an increased congestion |
| 1502 |
level is sensed by the network congestion control to limit the |
| 1503 |
bitrate. The target_bitrate_last_max is updated when congestion is |
| 1504 |
detected. |
| 1505 |
|
| 1506 |
|
| 1507 |
|
| 1508 |
<span class="grey">Johansson & Sarker Experimental [Page 27]</span></pre> |
| 1509 |
<hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-28" ></span> |
| 1510 |
<span class="grey"><a href="./rfc8298">RFC 8298</a> SCReAM December 2017</span> |
| 1511 |
|
| 1512 |
|
| 1513 |
Finally, the target_bitrate is within the defined min and max values. |
| 1514 |
|
| 1515 |
The aware reader may notice the dependency on the qdelay in the |
| 1516 |
computation of the target bitrate; this manifests itself in the use |
| 1517 |
of the qdelay_trend. As these parameters are used also in the |
| 1518 |
network congestion control, one may suspect some odd interaction |
| 1519 |
between the media rate control and the network congestion control. |
| 1520 |
This is in fact the case if the parameter PRE_CONGESTION_GUARD is set |
| 1521 |
to a high value. The use of qdelay_trend in the media rate control |
| 1522 |
is solely to reduce jitter; the dependency can be removed by setting |
| 1523 |
PRE_CONGESTION_GUARD=0. The effect is a somewhat larger rate |
| 1524 |
increase after congestion, at the expense of increased jitter in |
| 1525 |
congested situations. |
| 1526 |
|
| 1527 |
<span class="h3"><a class="selflink" id="section-4.2" href="#section-4.2">4.2</a>. SCReAM Receiver</span> |
| 1528 |
|
| 1529 |
The simple task of the SCReAM receiver is to feed back |
| 1530 |
acknowledgements of received packets and total ECN count to the |
| 1531 |
SCReAM sender. In addition, the receive time of the RTP packet with |
| 1532 |
the highest sequence number is echoed back. Upon reception of each |
| 1533 |
RTP packet, the receiver MUST maintain enough information to send the |
| 1534 |
aforementioned values to the SCReAM sender via an RTCP transport- |
| 1535 |
layer feedback message. The frequency of the feedback message |
| 1536 |
depends on the available RTCP bandwidth. The requirements on the |
| 1537 |
feedback elements and the feedback interval are described below. |
| 1538 |
|
| 1539 |
<span class="h4"><a class="selflink" id="section-4.2.1" href="#section-4.2.1">4.2.1</a>. Requirements on Feedback Elements</span> |
| 1540 |
|
| 1541 |
The following feedback elements are REQUIRED for basic functionality |
| 1542 |
in SCReAM. |
| 1543 |
|
| 1544 |
o A list of received RTP packets. This list SHOULD be sufficiently |
| 1545 |
long to cover all received RTP packets. This list can be realized |
| 1546 |
with the Loss RLE (Run Length Encoding) Report Block in [<a href="./rfc3611" title=""RTP Control Protocol Extended Reports (RTCP XR)"">RFC3611</a>]. |
| 1547 |
|
| 1548 |
o A wall-clock timestamp corresponding to the received RTP packet |
| 1549 |
with the highest sequence number is required in order to compute |
| 1550 |
the qdelay. This can be realized by means of the Packet Receipt |
| 1551 |
Times Report Block in [<a href="./rfc3611" title=""RTP Control Protocol Extended Reports (RTCP XR)"">RFC3611</a>]. begin_seq MUST be set to the |
| 1552 |
highest received sequence number (which has possibly wrapped |
| 1553 |
around); end_seq MUST be set to begin_seq+1 modulo 65536. The |
| 1554 |
timestamp clock MAY be set according to [<a href="./rfc3611" title=""RTP Control Protocol Extended Reports (RTCP XR)"">RFC3611</a>], i.e., equal to |
| 1555 |
the RTP timestamp clock. Detailed individual packet receive times |
| 1556 |
are not necessary, as SCReAM does currently not describe how they |
| 1557 |
can be used. |
| 1558 |
|
| 1559 |
|
| 1560 |
|
| 1561 |
|
| 1562 |
|
| 1563 |
|
| 1564 |
<span class="grey">Johansson & Sarker Experimental [Page 28]</span></pre> |
| 1565 |
<hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-29" ></span> |
| 1566 |
<span class="grey"><a href="./rfc8298">RFC 8298</a> SCReAM December 2017</span> |
| 1567 |
|
| 1568 |
|
| 1569 |
The basic feedback needed for SCReAM involves the use of the Loss RLE |
| 1570 |
Report Block and the Packet Receipt Times Report Block as shown in |
| 1571 |
Figure 2. |
| 1572 |
|
| 1573 |
0 1 2 3 |
| 1574 |
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 |
| 1575 |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |
| 1576 |
|V=2|P|reserved | PT=XR=207 | length | |
| 1577 |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |
| 1578 |
| SSRC | |
| 1579 |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |
| 1580 |
| BT=2 | rsvd. | T=0 | block length | |
| 1581 |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |
| 1582 |
| SSRC of source | |
| 1583 |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |
| 1584 |
| begin_seq | end_seq | |
| 1585 |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |
| 1586 |
| chunk 1 | chunk 2 | |
| 1587 |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |
| 1588 |
: ... : |
| 1589 |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |
| 1590 |
| chunk n-1 | chunk n | |
| 1591 |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |
| 1592 |
| BT=3 | rsvd. | T=0 | block length | |
| 1593 |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |
| 1594 |
| SSRC of source | |
| 1595 |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |
| 1596 |
| begin_seq | end_seq | |
| 1597 |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |
| 1598 |
| Receipt time of packet begin_seq | |
| 1599 |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |
| 1600 |
|
| 1601 |
Figure 2: Basic Feedback Message for SCReAM, Based on <a href="./rfc3611">RFC 3611</a> |
| 1602 |
|
| 1603 |
In a typical use case, no more than four Loss RLE chunks are needed, |
| 1604 |
thus the feedback message will be 44 bytes. It is obvious from |
| 1605 |
Figure 2 that there is a lot of redundant information in the feedback |
| 1606 |
message. A more optimized feedback format, including the additional |
| 1607 |
feedback elements listed below, could reduce the feedback message |
| 1608 |
size a bit. |
| 1609 |
|
| 1610 |
An additional feedback element that can improve the performance of |
| 1611 |
SCReAM is: |
| 1612 |
|
| 1613 |
o Accumulated number of ECN-CE-marked packets (n_ECN). For |
| 1614 |
instance, this can be realized with the ECN Feedback Report Format |
| 1615 |
in [<a href="./rfc6679" title=""Explicit Congestion Notification (ECN) for RTP over UDP"">RFC6679</a>]. The given feedback report format is slightly |
| 1616 |
overkill, as SCReAM would do quite well with only a counter that |
| 1617 |
|
| 1618 |
|
| 1619 |
|
| 1620 |
<span class="grey">Johansson & Sarker Experimental [Page 29]</span></pre> |
| 1621 |
<hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-30" ></span> |
| 1622 |
<span class="grey"><a href="./rfc8298">RFC 8298</a> SCReAM December 2017</span> |
| 1623 |
|
| 1624 |
|
| 1625 |
increments by one for each received packet with the ECN-CE |
| 1626 |
codepoint set. The more bulky format could nevertheless be useful |
| 1627 |
for, e.g., ECN black-hole detection. |
| 1628 |
|
| 1629 |
<span class="h4"><a class="selflink" id="section-4.2.2" href="#section-4.2.2">4.2.2</a>. Requirements on Feedback Intensity</span> |
| 1630 |
|
| 1631 |
SCReAM benefits from relatively frequent feedback. It is RECOMMENDED |
| 1632 |
that a SCReAM implementation follows the guidelines below. |
| 1633 |
|
| 1634 |
The feedback interval depends on the media bitrate. At low bitrates, |
| 1635 |
it is sufficient with a feedback interval of 100 to 400 ms; while at |
| 1636 |
high bitrates, a feedback interval of roughly 20 ms is preferred. At |
| 1637 |
very high bitrates, even shorter feedback intervals MAY be needed in |
| 1638 |
order to keep the self-clocking in SCReAM working well. One |
| 1639 |
indication that feedback is too sparse is that the SCReAM |
| 1640 |
implementation cannot reach high bitrates, even in uncongested links. |
| 1641 |
More frequent feedback might solve this issue. |
| 1642 |
|
| 1643 |
The numbers above can be formulated as a feedback interval function |
| 1644 |
that can be useful for the computation of the desired RTCP bandwidth. |
| 1645 |
The following equation expresses the feedback rate: |
| 1646 |
|
| 1647 |
rate_fb = min(50, max(2.5, rate_media / 10000)) |
| 1648 |
|
| 1649 |
rate_media is the RTP media bitrate expressed in bps; rate_fb is the |
| 1650 |
feedback rate expressed in packets/s. Converting to feedback |
| 1651 |
interval, we get: |
| 1652 |
|
| 1653 |
fb_int = 1.0 / min(50, max(2.5, rate_media / 10000)) |
| 1654 |
|
| 1655 |
The transmission interval is not critical. So, in the case of multi- |
| 1656 |
stream handling between two hosts, the feedback for two or more |
| 1657 |
synchronization sources (SSRCs) can be bundled to save UDP/IP |
| 1658 |
overhead. However, the final realized feedback interval SHOULD not |
| 1659 |
exceed 2*fb_int in such cases, meaning that a scheduled feedback |
| 1660 |
transmission event should not be delayed more than fb_int. |
| 1661 |
|
| 1662 |
SCReAM works with AVPF regular mode; immediate or early mode is not |
| 1663 |
required by SCReAM but can nonetheless be useful for RTCP messages |
| 1664 |
not directly related to SCReAM, such as those specified in [<a href="./rfc4585" title=""Extended RTP Profile for Real-time Transport Control Protocol (RTCP)-Based Feedback (RTP/AVPF)"">RFC4585</a>]. |
| 1665 |
It is RECOMMENDED to use reduced-size RTCP [<a href="./rfc5506" title=""Support for Reduced-Size Real-Time Transport Control Protocol (RTCP): Opportunities and Consequences"">RFC5506</a>], where regular |
| 1666 |
full compound RTCP transmission is controlled by trr-int as described |
| 1667 |
in [<a href="./rfc4585" title=""Extended RTP Profile for Real-time Transport Control Protocol (RTCP)-Based Feedback (RTP/AVPF)"">RFC4585</a>]. |
| 1668 |
|
| 1669 |
|
| 1670 |
|
| 1671 |
|
| 1672 |
|
| 1673 |
|
| 1674 |
|
| 1675 |
|
| 1676 |
<span class="grey">Johansson & Sarker Experimental [Page 30]</span></pre> |
| 1677 |
<hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-31" ></span> |
| 1678 |
<span class="grey"><a href="./rfc8298">RFC 8298</a> SCReAM December 2017</span> |
| 1679 |
|
| 1680 |
|
| 1681 |
<span class="h2"><a class="selflink" id="section-5" href="#section-5">5</a>. Discussion</span> |
| 1682 |
|
| 1683 |
This section covers a few discussion points. |
| 1684 |
|
| 1685 |
o Clock drift: SCReAM can suffer from the same issues with clock |
| 1686 |
drift as is the case with LEDBAT [<a href="./rfc6817" title=""Low Extra Delay Background Transport (LEDBAT)"">RFC6817</a>]. However, <a href="./rfc6817#appendix-A.2">Appendix A.2 |
| 1687 |
in [RFC6817]</a> describes ways to mitigate issues with clock drift. |
| 1688 |
|
| 1689 |
o Support for alternate ECN semantics: This specification adopts the |
| 1690 |
proposal in [<a href="#ref-ALT-BACKOFF">ALT-BACKOFF</a>] to reduce the congestion window less |
| 1691 |
when ECN-based congestion events are detected. Future work on Low |
| 1692 |
Loss, Low Latency for Scalable throughput (L4S) may lead to |
| 1693 |
updates in a future document that describes SCReAM support for |
| 1694 |
L4S. |
| 1695 |
|
| 1696 |
o A new transport-layer feedback message (as specified in <a href="./rfc4585">RFC 4585</a>) |
| 1697 |
could be standardized if the use of the already existing RTCP |
| 1698 |
extensions as described in <a href="#section-4.2">Section 4.2</a> is not deemed sufficient. |
| 1699 |
|
| 1700 |
o The target bitrate given by SCReAM is the bitrate including the |
| 1701 |
RTP and Forward Error Correction (FEC) overhead. The media |
| 1702 |
encoder SHOULD take this overhead into account when the media |
| 1703 |
bitrate is set. This means that the media coder bitrate SHOULD be |
| 1704 |
computed as |
| 1705 |
|
| 1706 |
media_rate = target_bitrate - rtp_plus_fec_overhead_bitrate |
| 1707 |
|
| 1708 |
It is not necessary to make a 100% perfect compensation for the |
| 1709 |
overhead, as the SCReAM algorithm will inherently compensate for |
| 1710 |
moderate errors. Under-compensating for the overhead has the |
| 1711 |
effect of increasing jitter, while overcompensating will cause the |
| 1712 |
bottleneck link to become underutilized. |
| 1713 |
|
| 1714 |
<span class="h2"><a class="selflink" id="section-6" href="#section-6">6</a>. Suggested Experiments</span> |
| 1715 |
|
| 1716 |
SCReAM has been evaluated in a number of different ways, mostly in a |
| 1717 |
simulator. The OpenWebRTC implementation work ([<a href="#ref-OpenWebRTC">OpenWebRTC</a>] and |
| 1718 |
[<a href="#ref-SCReAM-implementation">SCReAM-implementation</a>]) involved extensive testing with artificial |
| 1719 |
bottlenecks with varying bandwidths and using two different video |
| 1720 |
coders (OpenH264 and VP9). |
| 1721 |
|
| 1722 |
|
| 1723 |
|
| 1724 |
|
| 1725 |
|
| 1726 |
|
| 1727 |
|
| 1728 |
|
| 1729 |
|
| 1730 |
|
| 1731 |
|
| 1732 |
<span class="grey">Johansson & Sarker Experimental [Page 31]</span></pre> |
| 1733 |
<hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-32" ></span> |
| 1734 |
<span class="grey"><a href="./rfc8298">RFC 8298</a> SCReAM December 2017</span> |
| 1735 |
|
| 1736 |
|
| 1737 |
Preferably, further experiments will be done by means of |
| 1738 |
implementation in real clients and web browsers. RECOMMENDED |
| 1739 |
experiments are: |
| 1740 |
|
| 1741 |
o Trials with various access technologies: EDGE/3G/4G, Wi-Fi, DSL. |
| 1742 |
Some experiments have already been carried out with LTE access; |
| 1743 |
see [<a href="#ref-SCReAM-CPP-implementation">SCReAM-CPP-implementation</a>] and |
| 1744 |
[<a href="#ref-SCReAM-implementation-experience">SCReAM-implementation-experience</a>]. |
| 1745 |
|
| 1746 |
o Trials with different kinds of media: Audio, video, slideshow |
| 1747 |
content. Evaluation of multi-stream handling in SCReAM. |
| 1748 |
|
| 1749 |
o Evaluation of functionality of the compensation mechanism when |
| 1750 |
there are competing flows: Evaluate how SCReAM performs with |
| 1751 |
competing TCP-like traffic and to what extent the compensation for |
| 1752 |
competing flows causes self-inflicted congestion. |
| 1753 |
|
| 1754 |
o Determine proper parameters: A set of default parameters are given |
| 1755 |
that makes SCReAM work over a reasonably large operation range. |
| 1756 |
However, for very low or very high bitrates, it may be necessary |
| 1757 |
to use different values for the RAMP_UP_SPEED, for instance. |
| 1758 |
|
| 1759 |
o Experimentation with further improvements to the congestion window |
| 1760 |
and media bitrate calculation. [<a href="#ref-SCReAM-CPP-implementation">SCReAM-CPP-implementation</a>] |
| 1761 |
implements some optimizations, not described in this memo, that |
| 1762 |
improve performance slightly. Further experiments are likely to |
| 1763 |
lead to more optimizations of the algorithm. |
| 1764 |
|
| 1765 |
<span class="h2"><a class="selflink" id="section-7" href="#section-7">7</a>. IANA Considerations</span> |
| 1766 |
|
| 1767 |
This document does not require any IANA actions. |
| 1768 |
|
| 1769 |
<span class="h2"><a class="selflink" id="section-8" href="#section-8">8</a>. Security Considerations</span> |
| 1770 |
|
| 1771 |
The feedback can be vulnerable to attacks similar to those that can |
| 1772 |
affect TCP. It is therefore RECOMMENDED that the RTCP feedback is at |
| 1773 |
least integrity protected. Furthermore, as SCReAM is self-clocked, a |
| 1774 |
malicious middlebox can drop RTCP feedback packets and thus cause the |
| 1775 |
self-clocking in SCReAM to stall. However, this attack is mitigated |
| 1776 |
by the minimum send rate maintained by SCReAM when no feedback is |
| 1777 |
received. |
| 1778 |
|
| 1779 |
|
| 1780 |
|
| 1781 |
|
| 1782 |
|
| 1783 |
|
| 1784 |
|
| 1785 |
|
| 1786 |
|
| 1787 |
|
| 1788 |
<span class="grey">Johansson & Sarker Experimental [Page 32]</span></pre> |
| 1789 |
<hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-33" ></span> |
| 1790 |
<span class="grey"><a href="./rfc8298">RFC 8298</a> SCReAM December 2017</span> |
| 1791 |
|
| 1792 |
|
| 1793 |
<span class="h2"><a class="selflink" id="section-9" href="#section-9">9</a>. References</span> |
| 1794 |
|
| 1795 |
<span class="h3"><a class="selflink" id="section-9.1" href="#section-9.1">9.1</a>. Normative References</span> |
| 1796 |
|
| 1797 |
[<a id="ref-RFC2119">RFC2119</a>] Bradner, S., "Key words for use in RFCs to Indicate |
| 1798 |
Requirement Levels", <a href="https://www.rfc-editor.org/bcp/bcp14">BCP 14</a>, <a href="./rfc2119">RFC 2119</a>, |
| 1799 |
DOI 10.17487/RFC2119, March 1997, |
| 1800 |
<<a href="https://www.rfc-editor.org/info/rfc2119">https://www.rfc-editor.org/info/rfc2119</a>>. |
| 1801 |
|
| 1802 |
[<a id="ref-RFC3550">RFC3550</a>] Schulzrinne, H., Casner, S., Frederick, R., and V. |
| 1803 |
Jacobson, "RTP: A Transport Protocol for Real-Time |
| 1804 |
Applications", STD 64, <a href="./rfc3550">RFC 3550</a>, DOI 10.17487/RFC3550, |
| 1805 |
July 2003, <<a href="https://www.rfc-editor.org/info/rfc3550">https://www.rfc-editor.org/info/rfc3550</a>>. |
| 1806 |
|
| 1807 |
[<a id="ref-RFC3611">RFC3611</a>] Friedman, T., Ed., Caceres, R., Ed., and A. Clark, Ed., |
| 1808 |
"RTP Control Protocol Extended Reports (RTCP XR)", |
| 1809 |
<a href="./rfc3611">RFC 3611</a>, DOI 10.17487/RFC3611, November 2003, |
| 1810 |
<<a href="https://www.rfc-editor.org/info/rfc3611">https://www.rfc-editor.org/info/rfc3611</a>>. |
| 1811 |
|
| 1812 |
[<a id="ref-RFC4585">RFC4585</a>] Ott, J., Wenger, S., Sato, N., Burmeister, C., and J. Rey, |
| 1813 |
"Extended RTP Profile for Real-time Transport Control |
| 1814 |
Protocol (RTCP)-Based Feedback (RTP/AVPF)", <a href="./rfc4585">RFC 4585</a>, |
| 1815 |
DOI 10.17487/RFC4585, July 2006, |
| 1816 |
<<a href="https://www.rfc-editor.org/info/rfc4585">https://www.rfc-editor.org/info/rfc4585</a>>. |
| 1817 |
|
| 1818 |
[<a id="ref-RFC5506">RFC5506</a>] Johansson, I. and M. Westerlund, "Support for Reduced-Size |
| 1819 |
Real-Time Transport Control Protocol (RTCP): Opportunities |
| 1820 |
and Consequences", <a href="./rfc5506">RFC 5506</a>, DOI 10.17487/RFC5506, April |
| 1821 |
2009, <<a href="https://www.rfc-editor.org/info/rfc5506">https://www.rfc-editor.org/info/rfc5506</a>>. |
| 1822 |
|
| 1823 |
[<a id="ref-RFC6298">RFC6298</a>] Paxson, V., Allman, M., Chu, J., and M. Sargent, |
| 1824 |
"Computing TCP's Retransmission Timer", <a href="./rfc6298">RFC 6298</a>, |
| 1825 |
DOI 10.17487/RFC6298, June 2011, |
| 1826 |
<<a href="https://www.rfc-editor.org/info/rfc6298">https://www.rfc-editor.org/info/rfc6298</a>>. |
| 1827 |
|
| 1828 |
[<a id="ref-RFC6817">RFC6817</a>] Shalunov, S., Hazel, G., Iyengar, J., and M. Kuehlewind, |
| 1829 |
"Low Extra Delay Background Transport (LEDBAT)", <a href="./rfc6817">RFC 6817</a>, |
| 1830 |
DOI 10.17487/RFC6817, December 2012, |
| 1831 |
<<a href="https://www.rfc-editor.org/info/rfc6817">https://www.rfc-editor.org/info/rfc6817</a>>. |
| 1832 |
|
| 1833 |
[<a id="ref-RFC8174">RFC8174</a>] Leiba, B., "Ambiguity of Uppercase vs Lowercase in <a href="./rfc2119">RFC</a> |
| 1834 |
<a href="./rfc2119">2119</a> Key Words", <a href="https://www.rfc-editor.org/bcp/bcp14">BCP 14</a>, <a href="./rfc8174">RFC 8174</a>, DOI 10.17487/RFC8174, |
| 1835 |
May 2017, <<a href="https://www.rfc-editor.org/info/rfc8174">https://www.rfc-editor.org/info/rfc8174</a>>. |
| 1836 |
|
| 1837 |
|
| 1838 |
|
| 1839 |
|
| 1840 |
|
| 1841 |
|
| 1842 |
|
| 1843 |
|
| 1844 |
<span class="grey">Johansson & Sarker Experimental [Page 33]</span></pre> |
| 1845 |
<hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-34" ></span> |
| 1846 |
<span class="grey"><a href="./rfc8298">RFC 8298</a> SCReAM December 2017</span> |
| 1847 |
|
| 1848 |
|
| 1849 |
<span class="h3"><a class="selflink" id="section-9.2" href="#section-9.2">9.2</a>. Informative References</span> |
| 1850 |
|
| 1851 |
[<a id="ref-ALT-BACKOFF">ALT-BACKOFF</a>] |
| 1852 |
Khademi, N., Welzl, M., Armitage, G., and G. Fairhurst, |
| 1853 |
"TCP Alternative Backoff with ECN (ABE)", Work in |
| 1854 |
Progress, <a href="./draft-ietf-tcpm-alternativebackoff-ecn-04">draft-ietf-tcpm-alternativebackoff-ecn-04</a>, |
| 1855 |
November 2017. |
| 1856 |
|
| 1857 |
[<a id="ref-COUPLED-CC">COUPLED-CC</a>] |
| 1858 |
Islam, S., Welzl, M., and S. Gjessing, "Coupled congestion |
| 1859 |
control for RTP media", Work in Progress, <a href="./draft-ietf-rmcat-coupled-cc-07">draft-ietf-</a> |
| 1860 |
<a href="./draft-ietf-rmcat-coupled-cc-07">rmcat-coupled-cc-07</a>, September 2017. |
| 1861 |
|
| 1862 |
[<a id="ref-LEDBAT-delay-impact">LEDBAT-delay-impact</a>] |
| 1863 |
Ros, D. and M. Welzl, "Assessing LEDBAT's Delay Impact", |
| 1864 |
IEEE Communications Letters, Vol. 17, No. 5, |
| 1865 |
DOI 10.1109/LCOMM.2013.040213.130137, May 2013, |
| 1866 |
<<a href="http://home.ifi.uio.no/michawe/research/publications/ledbat-impact-letters.pdf">http://home.ifi.uio.no/michawe/research/publications/</a> |
| 1867 |
<a href="http://home.ifi.uio.no/michawe/research/publications/ledbat-impact-letters.pdf">ledbat-impact-letters.pdf</a>>. |
| 1868 |
|
| 1869 |
[<a id="ref-OpenWebRTC">OpenWebRTC</a>] |
| 1870 |
Ericsson Research, "OpenWebRTC", |
| 1871 |
<<a href="http://www.openwebrtc.org">http://www.openwebrtc.org</a>>. |
| 1872 |
|
| 1873 |
[<a id="ref-Packet-conservation">Packet-conservation</a>] |
| 1874 |
Jacobson, V., "Congestion Avoidance and Control", ACM |
| 1875 |
SIGCOMM Computer Communication Review, |
| 1876 |
DOI 10.1145/52325.52356, August 1988. |
| 1877 |
|
| 1878 |
[<a id="ref-QoS-3GPP">QoS-3GPP</a>] 3GPP, "Policy and charging control architecture", 3GPP TS |
| 1879 |
23.203, July 2017, |
| 1880 |
<<a href="http://www.3gpp.org/ftp/specs/archive/23_series/23.203/">http://www.3gpp.org/ftp/specs/archive/23_series/23.203/</a>>. |
| 1881 |
|
| 1882 |
[<a id="ref-RACK">RACK</a>] Cheng, Y., Cardwell, N., and N. Dukkipati, "RACK: a time- |
| 1883 |
based fast loss detection algorithm for TCP", Work in |
| 1884 |
Progress, <a href="./draft-ietf-tcpm-rack-02">draft-ietf-tcpm-rack-02</a>, March 2017. |
| 1885 |
|
| 1886 |
[<a id="ref-RFC6679">RFC6679</a>] Westerlund, M., Johansson, I., Perkins, C., O'Hanlon, P., |
| 1887 |
and K. Carlberg, "Explicit Congestion Notification (ECN) |
| 1888 |
for RTP over UDP", <a href="./rfc6679">RFC 6679</a>, DOI 10.17487/RFC6679, August |
| 1889 |
2012, <<a href="https://www.rfc-editor.org/info/rfc6679">https://www.rfc-editor.org/info/rfc6679</a>>. |
| 1890 |
|
| 1891 |
[<a id="ref-RFC7478">RFC7478</a>] Holmberg, C., Hakansson, S., and G. Eriksson, "Web Real- |
| 1892 |
Time Communication Use Cases and Requirements", <a href="./rfc7478">RFC 7478</a>, |
| 1893 |
DOI 10.17487/RFC7478, March 2015, |
| 1894 |
<<a href="https://www.rfc-editor.org/info/rfc7478">https://www.rfc-editor.org/info/rfc7478</a>>. |
| 1895 |
|
| 1896 |
|
| 1897 |
|
| 1898 |
|
| 1899 |
|
| 1900 |
<span class="grey">Johansson & Sarker Experimental [Page 34]</span></pre> |
| 1901 |
<hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-35" ></span> |
| 1902 |
<span class="grey"><a href="./rfc8298">RFC 8298</a> SCReAM December 2017</span> |
| 1903 |
|
| 1904 |
|
| 1905 |
[<a id="ref-RFC7661">RFC7661</a>] Fairhurst, G., Sathiaseelan, A., and R. Secchi, "Updating |
| 1906 |
TCP to Support Rate-Limited Traffic", <a href="./rfc7661">RFC 7661</a>, |
| 1907 |
DOI 10.17487/RFC7661, October 2015, |
| 1908 |
<<a href="https://www.rfc-editor.org/info/rfc7661">https://www.rfc-editor.org/info/rfc7661</a>>. |
| 1909 |
|
| 1910 |
[<a id="ref-SCReAM-CPP-implementation">SCReAM-CPP-implementation</a>] |
| 1911 |
Ericsson Research, "SCReAM - Mobile optimised congestion |
| 1912 |
control algorithm", |
| 1913 |
<<a href="https://github.com/EricssonResearch/scream">https://github.com/EricssonResearch/scream</a>>. |
| 1914 |
|
| 1915 |
[<a id="ref-SCReAM-implementation">SCReAM-implementation</a>] |
| 1916 |
Ericsson Research, "OpenWebRTC specific GStreamer |
| 1917 |
plugins", <<a href="https://github.com/EricssonResearch/openwebrtc-gst-plugins">https://github.com/EricssonResearch/</a> |
| 1918 |
<a href="https://github.com/EricssonResearch/openwebrtc-gst-plugins">openwebrtc-gst-plugins</a>>. |
| 1919 |
|
| 1920 |
[<a id="ref-SCReAM-implementation-experience">SCReAM-implementation-experience</a>] |
| 1921 |
Sarker, Z. and I. Johansson, "Updates on SCReAM: An |
| 1922 |
implementation experience", November 2015, |
| 1923 |
<<a href="https://www.ietf.org/proceedings/94/slides/slides-94-rmcat-8.pdf">https://www.ietf.org/proceedings/94/slides/</a> |
| 1924 |
<a href="https://www.ietf.org/proceedings/94/slides/slides-94-rmcat-8.pdf">slides-94-rmcat-8.pdf</a>>. |
| 1925 |
|
| 1926 |
[<a id="ref-TFWC">TFWC</a>] Choi, S. and M. Handley, "Fairer TCP-Friendly Congestion |
| 1927 |
Control Protocol for Multimedia Streaming Applications", |
| 1928 |
DOI 10.1145/1364654.1364717, December 2007, |
| 1929 |
<<a href="http://www-dept.cs.ucl.ac.uk/staff/M.Handley/papers/tfwc-conext.pdf">http://www-dept.cs.ucl.ac.uk/staff/M.Handley/papers/</a> |
| 1930 |
<a href="http://www-dept.cs.ucl.ac.uk/staff/M.Handley/papers/tfwc-conext.pdf">tfwc-conext.pdf</a>>. |
| 1931 |
|
| 1932 |
[<a id="ref-WIRELESS-TESTS">WIRELESS-TESTS</a>] |
| 1933 |
Sarker, Z., Johansson, I., Zhu, X., Fu, J., Tan, W., and |
| 1934 |
M. Ramalho, "Evaluation Test Cases for Interactive Real- |
| 1935 |
Time Media over Wireless Networks", Work in Progress, |
| 1936 |
<a href="./draft-ietf-rmcat-wireless-tests-04">draft-ietf-rmcat-wireless-tests-04</a>, May 2017. |
| 1937 |
|
| 1938 |
|
| 1939 |
|
| 1940 |
|
| 1941 |
|
| 1942 |
|
| 1943 |
|
| 1944 |
|
| 1945 |
|
| 1946 |
|
| 1947 |
|
| 1948 |
|
| 1949 |
|
| 1950 |
|
| 1951 |
|
| 1952 |
|
| 1953 |
|
| 1954 |
|
| 1955 |
|
| 1956 |
<span class="grey">Johansson & Sarker Experimental [Page 35]</span></pre> |
| 1957 |
<hr class='noprint'/><!--NewPage--><pre class='newpage'><span id="page-36" ></span> |
| 1958 |
<span class="grey"><a href="./rfc8298">RFC 8298</a> SCReAM December 2017</span> |
| 1959 |
|
| 1960 |
|
| 1961 |
Acknowledgements |
| 1962 |
|
| 1963 |
We would like to thank the following people for their comments, |
| 1964 |
questions, and support during the work that led to this memo: Markus |
| 1965 |
Andersson, Bo Burman, Tomas Frankkila, Frederic Gabin, Laurits Hamm, |
| 1966 |
Hans Hannu, Nikolas Hermanns, Stefan Haakansson, Erlendur Karlsson, |
| 1967 |
Daniel Lindstroem, Mats Nordberg, Jonathan Samuelsson, Rickard |
| 1968 |
Sjoeberg, Robert Swain, Magnus Westerlund, and Stefan Aalund. Many |
| 1969 |
additional thanks to RMCAT chairs Karen E. E. Nielsen and Mirja |
| 1970 |
Kuehlewind for patiently reading, suggesting improvements and also |
| 1971 |
for asking all the difficult but necessary questions. Thanks to |
| 1972 |
Stefan Holmer, Xiaoqing Zhu, Safiqul Islam, and David Hayes for the |
| 1973 |
additional review of this document. Thanks to Ralf Globisch for |
| 1974 |
taking time to try out SCReAM in his challenging low-bitrate use |
| 1975 |
cases, Robert Hedman for finding a few additional flaws in the |
| 1976 |
running code, and Gustavo Garcia and 'miseri' for code contributions. |
| 1977 |
|
| 1978 |
Authors' Addresses |
| 1979 |
|
| 1980 |
Ingemar Johansson |
| 1981 |
Ericsson AB |
| 1982 |
Laboratoriegraend 11 |
| 1983 |
Luleaa 977 53 |
| 1984 |
Sweden |
| 1985 |
|
| 1986 |
Phone: +46 730783289 |
| 1987 |
Email: [email protected] |
| 1988 |
|
| 1989 |
|
| 1990 |
Zaheduzzaman Sarker |
| 1991 |
Ericsson AB |
| 1992 |
Laboratoriegraend 11 |
| 1993 |
Luleaa 977 53 |
| 1994 |
Sweden |
| 1995 |
|
| 1996 |
Phone: +46 761153743 |
| 1997 |
Email: [email protected] |
| 1998 |
|
| 1999 |
|
| 2000 |
|
| 2001 |
|
| 2002 |
|
| 2003 |
|
| 2004 |
|
| 2005 |
|
| 2006 |
|
| 2007 |
|
| 2008 |
|
| 2009 |
|
| 2010 |
|
| 2011 |
|
| 2012 |
Johansson & Sarker Experimental [Page 36] |
| 2013 |
</pre> |