One document matched: draft-alvestrand-rtcweb-congestion-00.xml
<?xml version="1.0" encoding="US-ASCII"?>
<!DOCTYPE rfc SYSTEM "rfc2629.dtd">
<?rfc toc="yes"?>
<?rfc tocompact="yes"?>
<?rfc tocdepth="3"?>
<?rfc tocindent="yes"?>
<?rfc symrefs="yes"?>
<?rfc sortrefs="yes"?>
<?rfc comments="yes"?>
<?rfc inline="yes"?>
<?rfc compact="yes"?>
<?rfc subcompact="no"?>
<rfc category="info" docName="draft-alvestrand-rtcweb-congestion-00"
ipr="trust200902">
<front>
<title abbrev="Congestion Control for RTCWEB">A Google Congestion Control
for Real-Time Communication on the World Wide Web</title>
<author fullname="Henrik Lundin" initials="H." surname="Lundin">
<organization>Google</organization>
<address>
<postal>
<street>Kungsbron 2</street>
<city>Stockholm</city>
<code>11122</code>
<country>Sweden</country>
</postal>
</address>
</author>
<author fullname="Stefan Holmer" initials="S." surname="Holmer">
<organization>Google</organization>
<address>
<postal>
<street>Kungsbron 2</street>
<city>Stockholm</city>
<code>11122</code>
<country>Sweden</country>
</postal>
<email>holmer@google.com</email>
</address>
</author>
<author fullname="Harald Alvestrand" initials="H. T." role="editor"
surname="Alvestrand">
<organization>Google</organization>
<address>
<postal>
<street>Kungsbron 2</street>
<city>Stockholm</city>
<code>11122</code>
<country>Sweden</country>
</postal>
<email>harald@alvestrand.no</email>
</address>
</author>
<date day="5" month="September" year="2011" />
<abstract>
<t>This document describes two methods of congestion control when using
real-time communications on the World Wide Web (RTCWEB); one
sender-based and one receiver-based.</t>
<t>It is published to aid the discussion on mandatory-to-implement flow
control for RTCWEB applications; initial discussion is expected in the
RTCWEB WG's mailing list.</t>
</abstract>
<note title="Requirements Language">
<t>The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT",
"SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this
document are to be interpreted as described in <xref
target="RFC2119">RFC 2119</xref>.</t>
</note>
</front>
<middle>
<section title="Introduction">
<t>Congestion control is a requirement for all applications that wish to
share the Internet <xref target="RFC2914"></xref>.</t>
<t>The problem of doing congestion control for real-time media is made
difficult for a number of reasons:</t>
<t><list style="symbols">
<t>The media is usually encoded in forms that cannot be quickly
changed to accomodate varying bandwidth, and bandwidth requirements
can often be changed only in discrete, rather large steps</t>
<t>The participants may have certain specific wishes on how to
respond - which may not be reducing the bandwidth required by the
flow on which congestion is discovered</t>
<t>The encodings are usually sensitive to packet loss, while the
real time requirement precludes the repair of packet loss by
retransmission</t>
</list>This memo describes two congestion control algorithms that
together are seen to give reasonable performance and reasonable (not
perfect) bandwidth sharing with other conferences and with TCP-using
applications that share the same links.</t>
<t>The signalling used consists of standard RTP timestamps <xref
target="RFC3550"></xref>, standard RTCP feedback reports and Temporary
Maximum Media Stream Bit Rate Requests (TMMBR) as defined in <xref
target="RFC5104"></xref> section 3.5.4.</t>
<t></t>
<section title="Mathemathical notation conventions">
<t>The mathematics of this document have been transcribed from a more
formula-friendly format.</t>
<t>The following notational conventions are used:</t>
<t><list style="hanging">
<t hangText="X_bar">The variable X, where X is a vector -
conventionally marked by a bar on top of the variable name.</t>
<t hangText="X_hat">An estimate of the true value of variable X -
conventionally marked by a circumflex accent on top of the
variable name.</t>
<t hangText="X(i)">The "i"th value of X - conventionally marked by
a subscript i.</t>
<t hangText="[x y z]">A row vector consisting of elements x, y and
z.</t>
<t hangText="X_bar^T">The transpose of vector X_bar.</t>
</list></t>
</section>
</section>
<section title="System model">
<t>The following elements are in the system:</t>
<t><list style="symbols">
<t>Incoming media stream</t>
<t>Media codec - has a bandwidth control, and encodes the incoming
media stream into an RTP stream.</t>
<t>RTP sender - sends the RTP stream over the network to the RTP
receiver. Generates the RTP timestamp.</t>
<t>RTP receiver - receives the RTP stream, notes the time of
arrival. Regenerates the media stream for the recipient.</t>
<t>RTCP sender at RTP sender - sends sender reports.</t>
<t>RTCP sender at RTP receiver - sends receiver reports and TMMBR
messages.</t>
<t>RTCP receiver at RTP sender - receives receiver reports and TMMBR
messages, reports these to sender side control.</t>
<t>RTCP receiver at RTP receiver.</t>
<t>Sender side control - takes loss rate info, round trip time info,
and TMMBR messages and computes a sending bitrate.</t>
<t>Receiver side control - takes the packet arrival info at the RTP
receiver and decides when to send TMMBR messages.</t>
</list>Together, sender side control and receiver side control
implement the congestion control algorithm.</t>
<t></t>
</section>
<section anchor="receiverside" title="Receiver side control">
<t>The receive-side algorithm can be further decomposed into three
parts: an arrival-time filter, an over-use detector, and a remote
rate-control.</t>
<t></t>
<section title="Arrival-time filter">
<t>This section describes an adaptive filter that continuously updates
estimates of network parameters based on the timing of the received
frames.</t>
<t>At the receiving side we are observing groups of incoming video
packets, where each group of packets corresponding to the same frame
having timestamp T(i).</t>
<t>Each frame is assigned a receive time t(i), which corresponds to
the time at which the whole frame has been received (ignoring any
packet losses). A frame is delayed relative to its predecessor if
t(i)-t(i-1)>T(i)-T(i-1), i.e., if the arrival time difference is
larger than the timestamp difference.</t>
<t>We define the (relative) inter-arrival time, d(i) as</t>
<figure>
<artwork><![CDATA[
d(i) = t(i)-t(i-1)-(T(i)-T(i-1))
]]></artwork>
</figure>
<t>Since the time ts to send a frame of size L over a path with a
capacity of C is</t>
<figure>
<artwork><![CDATA[
ts = L/C
]]></artwork>
</figure>
<t>we can model the inter-arrival time as</t>
<figure>
<artwork><![CDATA[
L(i)-L(i-1)
d(i) = -------------- + w(i) =~ dL(i)/C+w(i)
C
]]></artwork>
</figure>
<t>Here, w(i) is a sample from a stochastic process W, which is a
function of the capacity C, the current cross traffic X(i), and the
current send bit rate R(i). We model W as a white Gaussian process. If
we are over-using the channel we expect w(i) to increase, and if a
queue on the network path is being emptied, w(i) will decrease;
otherwise the mean of w(i) will be zero.</t>
<t>Breaking out the mean of w(i) to make it zero mean, we get</t>
<figure>
<preamble>Equation 5</preamble>
<artwork><![CDATA[
d(i) = dL(i)/C + m(i) + v(i)
]]></artwork>
</figure>
<t>This is our fundamental model, where we take into account that a
large frame needs more time to traverse the link than a small frame,
thus arriving with higher relative delay. The noise term represents
network jitter and other delay effects not captured by the model.</t>
<t>When graphing the values for d(i) versus dL(i) on a scatterplot, we
find that most samples cluster around the center, and the outliers are
clustered along a line with average slope 1/C and zero offset.</t>
<t>When using a regular video codec, most frames are roughly the same
size after encoding (the central “cloud”); the exceptions
are I-frames (or key frames) which are typically much larger than the
average causing positive outliers (the I-frame itself) and negative
outliers (the frame after an I-frame).</t>
<t>The parameters d(i) and dL(i) are readily available for each frame
i, and we want to estimate C and m(i) and use those estimates to
detect whether or not we are over-using the bandwidth currently
available. These parameters are easily estimated by any adaptive
filter – we are using the Kalman filter.</t>
<t>Let</t>
<figure>
<artwork><![CDATA[
theta_bar(i) = [1/C(i) m(i)]^T
]]></artwork>
</figure>
<t>and call it the state of time i. We model the state evolution from
time i to time i+1 as</t>
<t></t>
<figure>
<artwork><![CDATA[
theta_bar(i+1) = theta_bar(i) + u_bar(i)
]]></artwork>
</figure>
<t>where u_bar(i) is the zero mean white Gaussian process noise with
covariance</t>
<t></t>
<figure>
<preamble>Equation 7</preamble>
<artwork><![CDATA[
Q(i) = E{u_bar(i) u_bar(i)^T}
]]></artwork>
</figure>
<t>Given equation 5 we get</t>
<figure>
<preamble>Equation 8</preamble>
<artwork><![CDATA[
d(i) = h_bar(i)^T theta_bar(i) + v(i)
h_bar(i) = [ dL(i) 1 ]^T
]]></artwork>
</figure>
<t>where v(i) is zero mean white Gaussian measurement noise with
variance var_v = sigma(v,i)^2</t>
<t>The Kalman filter recursively updates our estimate</t>
<figure>
<artwork><![CDATA[
theta_hat(i) = [1/C_hat(i) m_hat(i)]^T]]></artwork>
</figure>
<t>as</t>
<figure>
<artwork><![CDATA[
z(i) = d(i) - h_bar(i)^T * theta_hat(i-1)
theta_hat(i) = theta_hat(i-1) + z(i) * k_bar(i)
E(i-1) * h_bar(i)
k_bar(i) = ------------------------------
var_v_hat + h_bar(i)^T E(i-1)h_bar(i)
E(i)=(I - K_bar(i) h_bar(i)^T) * E(i-1) + Q(i)
]]></artwork>
</figure>
<t>I is the 2-by-2 identity matrix.</t>
<t>The variance var_v = sigma(v,i)^2 is estimated using an exponential
averaging filter, modified for variable sampling rate</t>
<figure>
<artwork><![CDATA[
var_v_hat = beta*sigma(v,i-1)^2 + (1-beta)*z(i)^2
beta = (1-alpha)30/(1000 * f_max)
]]></artwork>
</figure>
<t>where f_max = max {1/(T(j) - T(j-1))} for j in i-K+1...i is the
highest rate at which frames have been captured by the camera the last
K frames and alpha is a filter coefficient typically chosen as a
number in the interval [0.1, 0.001]. Since our assumption that v(i)
should be zero mean WGN is less accurate in some cases, we have
introduced an additional outlier filter around the updates of
var_v_hat. If z(i) > 3 var_v_hat the filter is updated with 3
sqrt(var_v_hat) rather than z(i). In a similar way, Q(i) is chosen as
a diagonal matrix with main diagonal elements given by</t>
<figure>
<artwork><![CDATA[
diag(Q(i)) = 30/(1000 * f_max)[10^-10 10^-2]^T
]]></artwork>
</figure>
<t>It is necessary to scale these filter parameters with the frame
rate to make the detector respond as quickly at low frame rates as at
high frame rates.</t>
</section>
<section title="Over-use detector">
<t>The offset estimate m(i) is compared with a threshold gamma_1. An
estimate above the threshold is considered as an indication of
over-use. Such an indication is not enough for the detector to signal
over-use to the rate control subsystem. Not until over-use has been
detected for at least gamma_2 milliseconds and at least gamma_3
frames, a definitive over-use will be signaled. However, if the offset
estimate m(i) was decreased in the last update, over-use will not be
signaled even if all the above conditions are met. Similarly, the
opposite state, under-use, is detected when m(i) < -gamma_1. If
neither over-use nor under-use is detected, the detector will be in
the normal state.</t>
</section>
<section title="Rate control">
<t>The rate control at the receiving side is designed to increase the
available bandwidth estimate A_hat as long as the detected state is
normal. Doing that assures that we, sooner or later, will reach the
available bandwidth of the channel and detect an over-use.</t>
<t>As soon as over-use has been detected the available bandwidth
estimate is decreased. In this way we get a recursive and adaptive
estimate of the available bandwidth.</t>
<t>In this design description we make the assumption that the rate
control subsystem is executed periodically and that this period is
constant.</t>
<t>The rate control subsystem has 3 states: Increase, Decrease and
Hold. "Increase" is the state when no congestion is detected;
"Decrease" is the state where congestion is detected, and "Hold" is a
state that waits until built-up queues have drained before going to
"increase" state.</t>
<t>The state transitions (with blank fields meaning "remain in state")
are:</t>
<t></t>
<figure>
<artwork><![CDATA[State ----> | Hold |Increase |Decrease
Signal-----------------------------------------
v | | |
Over-use | Decrease |Decrease |
-----------------------------------------------
Normal | Increase | |Hold
-----------------------------------------------
Under-use | |Hold |Hold
-----------------------------------------------
]]></artwork>
</figure>
<t>The subsystem starts in the increase state, where it will stay
until over-use or under-use has been detected by the detector
subsystem. On every update the available bandwidth is increased with a
factor which is a function of the global system response time and the
estimated measurement noise variance var_v_hat. The global system
response time is the time from an increase that causes over-use until
that over-use can be detected by the over-use detector. The variance
var_v_hat affects how responsive the Kalman filter is, and is thus
used as an indicator of the delay inflicted by the Kalman filter.</t>
<figure>
<artwork><![CDATA[
A(i) = eta*A(i-1)
1.001+B
eta(RTT, var_v_hat) = ------------------------------------------
1+e^(b(d*RTT - (c1 * var_v_hat + c2)))]]></artwork>
</figure>
<t>Here, B, b, d, c1 and c2 are design parameters.</t>
<t>Since the system depends on over-using the channel to verify the
current available bandwidth estimate, we must make sure that our
estimate doesn’t diverge from the rate at which the sender is
actually sending. Thus, if the sender is unable to produce a bit
stream with the bit rate the receiver is asking for, the available
bandwidth estimate must stay within a given bound. Therefore we
introduce a threshold</t>
<figure>
<artwork><![CDATA[
A_hat(i) < 1.5 * R_hat(i)
]]></artwork>
</figure>
<t>where R_hat(i) is the incoming bit rate measured over a T seconds
window:</t>
<figure>
<artwork><![CDATA[
R_hat(i) = 1/T * sum(L(j)) for j from 1 to N(i)]]></artwork>
</figure>
<t>N(i) is the number of frames received the past T seconds and L(j)
is the payload size of frame j.</t>
<t>When an over-use is detected the system transitions to the decrease
state, where the available bandwidth estimate is decreased to a factor
times the currently incoming bit rate.</t>
<figure>
<artwork><![CDATA[
A_hat(i) = alpha*R_hat(i)]]></artwork>
</figure>
<t>alpha is typically chosen to be in the interval [0.8, 0.95].</t>
<t>When the detector signals under-use to the rate control subsystem,
we know that queues in the network path are being emptied, indicating
that our available bandwidth estimate is lower than the actual
available bandwidth. Upon that signal the rate control subsystem will
enter the hold state, where the available bandwidth estimate will be
held constant while waiting for the queues to stabilize at a lower
level – a way of keeping the delay as low as possible. This
decrease of delay is wanted, and expected, immediately after the
estimate has been reduced due to over-use, but can also happen if the
cross traffic over some links is reduced. In either case we want to
measure the highest incoming rate during the under-use interval:</t>
<figure>
<artwork><![CDATA[
R_max = max{R_hat(i)} for i in 1..K
]]></artwork>
</figure>
<t>where K is the number of frames of under-use before returning to
the normal state. R_max is a measure of the actual bandwidth available
and is a good guess of what bit rate we should be able to transmit at.
Therefore the available bandwidth will be set to Rmax when we
transition from the hold state to the increase state.</t>
</section>
</section>
<section anchor="senderside" title="Sender side control">
<t>An additional congestion controller resides at the sending side. It
bases its decisions on the round-trip time, packet loss and available
bandwidth estimates transmitted from the receiving side.</t>
<t>The available bandwidth estimates produced by the receiving side are
only reliable when the size of the queues along the channel are large
enough. If the queues are very short, over-use will only be visible
through packet losses, which aren't used by the receiving side
algorithm.</t>
<t>This algorithm is run every time a receive report arrives at the
sender, which will happen [[how often do we expect? and why?]]. If no
receive report is recieved within [[what timeout?]], the algorithm will
take action as if all packets in the interval have been lost. [[does
that make sense?]]</t>
<t><list style="symbols">
<t>If 2-10% of the packets have been lost since the previous report
from the receiver, the sender available bandwidth estimate As(i) (As
denotes ‘sender available bandwidth’) will be kept
unchanged.</t>
<t>If more than 10% of the packets have been lost a new estimate is
calculated as As(i)=As(i-1)(1-0.5p), where p is the loss ratio.</t>
<t>As long as less than 2% of the packets have been lost As(i) will
be increased as As(i)=1.05(As(i-1)+1000)</t>
</list></t>
<t>The new send-side estimate is limited by the TCP Friendly Rate
Control formula <xref target="RFC3448"></xref> and the receive-side
estimate of the available bandwidth A(i):</t>
<figure>
<artwork><![CDATA[ 8 s
As(i) >= ----------------------------------------------------------
R*sqrt(2*b*p/3) + (t_RTO*(3*sqrt(3*b*p/8) * p * (1+32*p^2)))
As(i) <= A(i)
]]></artwork>
</figure>
<t>where b is the number of packets acknowledged by a single TCP
acknowledgement (set to 1 per TFRC recommendations), t_RTO is the TCP
retransmission timeout value in seconds (set to 4*R) and s is the
average packet size in bytes.</t>
<t>(The multiplication by 8 comes because TFRC is computing bandwidth in
bytes, while this document computes bandwidth in bits.)</t>
<t>In words: The sender-side estimate will never be larger than the
receiver-side estimate, and will never be lower than the estimate from
the TFRC formula.</t>
<t>We motivate the packet loss thresholds by noting that if we have
small amount of packet losses due to over-use, that amount will soon
increase if we don’t adjust our bit rate. Therefore we will soon
enough reach above the 10 % threshold and adjust As(i). However if the
packet loss rate does not increase, the losses are probably not related
to self-induced channel over-use and therefore we should not react on
them.</t>
</section>
<section title="Interoperability Considerations">
<t>There are three scenarios of interest, and one included for
reference</t>
<t><list style="symbols">
<t>Both parties implement the algorithms described here</t>
<t>Sender implements the algorithm described in section <xref
target="senderside"></xref>, recipient does not implement <xref
target="receiverside"></xref></t>
<t>Recipient implements the algorithm in section <xref
target="receiverside"></xref>, sender does not implement <xref
target="senderside"></xref>.</t>
</list>In the case where both parties implement the algorithms, we
expect to see most of the congestion control response to slowly varying
conditions happen by TMMBR messages from recipient to sender. At most
times, the sender will send less than the congestion-inducing bandwidth
limit C, and when he sends more, congestion will be detected before
packets are lost.</t>
<t>If sudden changes happen, packets will be lost, and the sender side
control will trigger, limiting traffic until the congestion becomes low
enough that the system switches back to the receiver-controlled
state.</t>
<t>In the case where sender only implements, we expect to see somewhat
higher loss rates and delays, but the system will still be overall TCP
friendly and self-adjusting; the governing term in the calculation will
be the TFRC formula.</t>
<t>In the case where recipient implements this algorithm and sender does
not, congestion will be avoided for slow changes as long as the sender
understands and obeys TMMBR; there will be no backoff for
packet-loss-inducing changes in capacity. Given that some kind of
congestion control is mandatory for the sender according to the TMMBR
spec, this case has to be reevaluated against the specific congestion
control implemented by the sender.</t>
</section>
<section title="Implementation Experience">
<t>This algorithm has been implemented in the open-source WebRTC
project.</t>
</section>
<section title="Further Work">
<t>This draft is offered as input to the congestion control
discussion.</t>
<t>Work that can be done on this basis includes:</t>
<t><list style="symbols">
<t>Consideration of timing info: It may be sensible to use the
proposed TFRC RTP header extensions <xref
target="I-D.gharai-avtcore-rtp-tfrc"></xref>to carry per-packet
timing information, which would both give more data points and a
timestamp applied closer to the network interface.</t>
<t>Considerations of cross-channel calculation: If all packets in
multiple streams follow the same path over the network, congestion
or queueing information should be considered across all packets
between two parties, not just per media stream.</t>
<t>Considerations of cross-channel balancing: The decision to slow
down sending in a situation with multiple media streams should be
taken across all media streams, not per stream.</t>
<t>Considerations of additional input: How and where packet loss
detected at the recipient can be added to the algorithm.</t>
<t>Considerations of locus of control: Whether the sender or the
recipient is in the best position to figure out which media streams
it makes sense to slow down, and therefore whether one should use
TMMBR to slow down one channel, signal an overall bandwidth change
and let the sender make the decision, or signal the (possibly
processed) delay info and let the sender run the algorithm.</t>
</list>These are matters for further work; since some of them involve
extensions that have not yet been standardized, this could take some
time, and it's important to consider when this work can be
completed.</t>
</section>
<section anchor="IANA" title="IANA Considerations">
<t>This document makes no request of IANA.</t>
<t>Note to RFC Editor: this section may be removed on publication as an
RFC.</t>
</section>
<section anchor="Security" title="Security Considerations">
<t>An attacker with the ability to insert or remove messages on the
connection will, of course, have the ability to mess up rate control,
causing people to send either too fast or too slow, and causing
congestion.</t>
<t>In this case, the control information is carried inside RTP, and can
be protected against modification or message insertion using SRTP, just
as for the media. Given that timestamps are carried in the RTP header,
which is not encrypted, this is not protected against disclosure, but it
seems hard to mount an attack based on timing information only.</t>
</section>
<section anchor="Acknowledgements" title="Acknowledgements">
<t></t>
</section>
</middle>
<back>
<references title="Normative References">
<?rfc include="reference.RFC.2119"?>
<?rfc include='reference.RFC.3448'?>
<?rfc include='reference.RFC.3550'?>
<?rfc include='reference.RFC.5104'?>
</references>
<references title="Informative References">
<?rfc include='reference.I-D.gharai-avtcore-rtp-tfrc'?>
<?rfc include='reference.RFC.2914'?>
</references>
</back>
</rfc>
| PAFTECH AB 2003-2026 | 2026-04-24 02:59:20 |