Rfc | 3448 |
Title | TCP Friendly Rate Control (TFRC): Protocol Specification |
Author | M.
Handley, S. Floyd, J. Padhye, J. Widmer |
Date | January 2003 |
Format: | TXT,
HTML |
Obsoleted by | RFC5348 |
Status: | PROPOSED STANDARD |
|
Network Working Group M. Handley
Request for Comments: 3448 S. Floyd
Category: Standards Track ICIR
J. Padhye
Microsoft
J. Widmer
University of Mannheim
January 2003
TCP Friendly Rate Control (TFRC):
Protocol Specification
Status of this Memo
This document specifies an Internet standards track protocol for the
Internet community, and requests discussion and suggestions for
improvements. Please refer to the current edition of the "Internet
Official Protocol Standards" (STD 1) for the standardization state
and status of this protocol. Distribution of this memo is unlimited.
Copyright Notice
Copyright (C) The Internet Society (2003). All Rights Reserved.
Abstract
This document specifies TCP-Friendly Rate Control (TFRC). TFRC is a
congestion control mechanism for unicast flows operating in a best-
effort Internet environment. It is reasonably fair when competing
for bandwidth with TCP flows, but has a much lower variation of
throughput over time compared with TCP, making it more suitable for
applications such as telephony or streaming media where a relatively
smooth sending rate is of importance.
Table of Contents
1. Introduction. . . . . . . . . . . . . . . . . . . . . . 2
2. Terminology . . . . . . . . . . . . . . . . . . . . . . 3
3. Protocol Mechanism. . . . . . . . . . . . . . . . . . . 3
3.1. TCP Throughput Equation. . . . . . . . . . . . . . 4
3.2. Packet Contents. . . . . . . . . . . . . . . . . . 6
3.2.1. Data Packets. . . . . . . . . . . . . . . . 6
3.2.2. Feedback Packets. . . . . . . . . . . . . . 7
4. Data Sender Protocol. . . . . . . . . . . . . . . . . . 7
4.1. Measuring the Packet Size. . . . . . . . . . . . . 8
4.2. Sender Initialization. . . . . . . . . . . . . . . 8
4.3. Sender behavior when a feedback packet is
received. . . . . . . . . . . . . .. . . . . . . . 8
4.4. Expiration of nofeedback timer . . . . . . . . . . 9
4.5. Preventing Oscillations. . . . . . . . . . . . . . 10
4.6. Scheduling of Packet Transmissions . . . . . . . . 11
5. Calculation of the Loss Event Rate (p). . . . . . . . . 12
5.1. Detection of Lost or Marked Packets. . . . . . . . 12
5.2. Translation from Loss History to Loss Events . . . 13
5.3. Inter-loss Event Interval. . . . . . . . . . . . . 14
5.4. Average Loss Interval. . . . . . . . . . . . . . . 14
5.5. History Discounting. . . . . . . . . . . . . . . . 15
6. Data Receiver Protocol. . . . . . . . . . . . . . . . . 17
6.1. Receiver behavior when a data packet is
received . . . . . . . . . . . . . . . . . . . . . 18
6.2. Expiration of feedback timer . . . . . . . . . . . 18
6.3. Receiver initialization. . . . . . . . . . . . . . 19
6.3.1. Initializing the Loss History after the
First Loss Event . . . . . . . . . . . . . 19
7. Sender-based Variants . . . . . . . . . . . . . . . . . 20
8. Implementation Issues . . . . . . . . . . . . . . . . . 20
9. Security Considerations . . . . . . . . . . . . . . . . 21
10. IANA Considerations . . . . . . . . . . . . . . . . . . 22
11. Acknowledgments . . . . . . . . . . . . . . . . . . . . 22
12. Non-Normative References. . . . . . . . . . . . . . . . 22
13. Authors' Addresses. . . . . . . . . . . . . . . . . . . 23
14. Full Copyright Statement. . . . . . . . . . . . . . . . 24
1. Introduction
This document specifies TCP-Friendly Rate Control (TFRC). TFRC is a
congestion control mechanism designed for unicast flows operating in
an Internet environment and competing with TCP traffic [2]. Instead
of specifying a complete protocol, this document simply specifies a
congestion control mechanism that could be used in a transport
protocol such as RTP [7], in an application incorporating end-to-end
congestion control at the application level, or in the context of
endpoint congestion management [1]. This document does not discuss
packet formats or reliability. Implementation-related issues are
discussed only briefly, in Section 8.
TFRC is designed to be reasonably fair when competing for bandwidth
with TCP flows, where a flow is "reasonably fair" if its sending rate
is generally within a factor of two of the sending rate of a TCP flow
under the same conditions. However, TFRC has a much lower variation
of throughput over time compared with TCP, which makes it more
suitable for applications such as telephony or streaming media where
a relatively smooth sending rate is of importance.
The penalty of having smoother throughput than TCP while competing
fairly for bandwidth is that TFRC responds slower than TCP to changes
in available bandwidth. Thus TFRC should only be used when the
application has a requirement for smooth throughput, in particular,
avoiding TCP's halving of the sending rate in response to a single
packet drop. For applications that simply need to transfer as much
data as possible in as short a time as possible we recommend using
TCP, or if reliability is not required, using an Additive-Increase,
Multiplicative-Decrease (AIMD) congestion control scheme with similar
parameters to those used by TCP.
TFRC is designed for applications that use a fixed packet size, and
vary their sending rate in packets per second in response to
congestion. Some audio applications require a fixed interval of time
between packets and vary their packet size instead of their packet
rate in response to congestion. The congestion control mechanism in
this document cannot be used by those applications; TFRC-PS (for
TFRC-PacketSize) is a variant of TFRC for applications that have a
fixed sending rate but vary their packet size in response to
congestion. TFRC-PS will be specified in a later document.
TFRC is a receiver-based mechanism, with the calculation of the
congestion control information (i.e., the loss event rate) in the
data receiver rather in the data sender. This is well-suited to an
application where the sender is a large server handling many
concurrent connections, and the receiver has more memory and CPU
cycles available for computation. In addition, a receiver-based
mechanism is more suitable as a building block for multicast
congestion control.
2. Terminology
In this document, the key words "MUST", "MUST NOT", "REQUIRED",
"SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY",
and "OPTIONAL" are to be interpreted as described in BCP 14, RFC 2119
and indicate requirement levels for compliant TFRC implementations.
3. Protocol Mechanism
For its congestion control mechanism, TFRC directly uses a throughput
equation for the allowed sending rate as a function of the loss event
rate and round-trip time. In order to compete fairly with TCP, TFRC
uses the TCP throughput equation, which roughly describes TCP's
sending rate as a function of the loss event rate, round-trip time,
and packet size. We define a loss event as one or more lost or
marked packets from a window of data, where a marked packet refers to
a congestion indication from Explicit Congestion Notification (ECN)
[6].
Generally speaking, TFRC's congestion control mechanism works as
follows:
o The receiver measures the loss event rate and feeds this
information back to the sender.
o The sender also uses these feedback messages to measure the
round-trip time (RTT).
o The loss event rate and RTT are then fed into TFRC's throughput
equation, giving the acceptable transmit rate.
o The sender then adjusts its transmit rate to match the calculated
rate.
The dynamics of TFRC are sensitive to how the measurements are
performed and applied. We recommend specific mechanisms below to
perform and apply these measurements. Other mechanisms are possible,
but it is important to understand how the interactions between
mechanisms affect the dynamics of TFRC.
3.1. TCP Throughput Equation
Any realistic equation giving TCP throughput as a function of loss
event rate and RTT should be suitable for use in TFRC. However, we
note that the TCP throughput equation used must reflect TCP's
retransmit timeout behavior, as this dominates TCP throughput at
higher loss rates. We also note that the assumptions implicit in the
throughput equation about the loss event rate parameter have to be a
reasonable match to how the loss rate or loss event rate is actually
measured. While this match is not perfect for the throughput
equation and loss rate measurement mechanisms given below, in
practice the assumptions turn out to be close enough.
The throughput equation we currently recommend for TFRC is a slightly
simplified version of the throughput equation for Reno TCP from [4].
Ideally we'd prefer a throughput equation based on SACK TCP, but no
one has yet derived the throughput equation for SACK TCP, and from
both simulations and experiments, the differences between the two
equations are relatively minor.
The throughput equation is:
s
X = ----------------------------------------------------------
R*sqrt(2*b*p/3) + (t_RTO * (3*sqrt(3*b*p/8) * p * (1+32*p^2)))
Where:
X is the transmit rate in bytes/second.
s is the packet size in bytes.
R is the round trip time in seconds.
p is the loss event rate, between 0 and 1.0, of the number of loss
events as a fraction of the number of packets transmitted.
t_RTO is the TCP retransmission timeout value in seconds.
b is the number of packets acknowledged by a single TCP
acknowledgement.
We further simplify this by setting t_RTO = 4*R. A more accurate
calculation of t_RTO is possible, but experiments with the current
setting have resulted in reasonable fairness with existing TCP
implementations [9]. Another possibility would be to set t_RTO =
max(4R, one second), to match the recommended minimum of one second
on the RTO [5].
Many current TCP connections use delayed acknowledgements, sending an
acknowledgement for every two data packets received, and thus have a
sending rate modeled by b = 2. However, TCP is also allowed to send
an acknowledgement for every data packet, and this would be modeled
by b = 1. Because many TCP implementations do not use delayed
acknowledgements, we recommend b = 1.
In future, different TCP equations may be substituted for this
equation. The requirement is that the throughput equation be a
reasonable approximation of the sending rate of TCP for conformant
TCP congestion control.
The parameters s (packet size), p (loss event rate) and R (RTT) need
to be measured or calculated by a TFRC implementation. The
measurement of s is specified in Section 4.1, measurement of R is
specified in Section 4.3, and measurement of p is specified in
Section 5. In the rest of this document all data rates are measured
in bytes/second.
3.2. Packet Contents
Before specifying the sender and receiver functionality, we describe
the contents of the data packets sent by the sender and feedback
packets sent by the receiver. As TFRC will be used along with a
transport protocol, we do not specify packet formats, as these depend
on the details of the transport protocol used.
3.2.1. Data Packets
Each data packet sent by the data sender contains the following
information:
o A sequence number. This number is incremented by one for each
data packet transmitted. The field must be sufficiently large
that it does not wrap causing two different packets with the same
sequence number to be in the receiver's recent packet history at
the same time.
o A timestamp indicating when the packet is sent. We denote by ts_i
the timestamp of the packet with sequence number i. The
resolution of the timestamp should typically be measured in
milliseconds. This timestamp is used by the receiver to determine
which losses belong to the same loss event. The timestamp is also
echoed by the receiver to enable the sender to estimate the
round-trip time, for senders that do not save timestamps of
transmitted data packets. We note that as an alternative to a
timestamp incremented in milliseconds, a "timestamp" that
increments every quarter of a round-trip time would be sufficient
for determining when losses belong to the same loss event, in the
context of a protocol where this is understood by both sender and
receiver, and where the sender saves the timestamps of transmitted
data packets.
o The sender's current estimate of the round trip time. The
estimate reported in packet i is denoted by R_i. The round-trip
time estimate is used by the receiver, along with the timestamp,
to determine when multiple losses belong to the same loss event.
If the sender sends a coarse-grained "timestamp" that increments
every quarter of a round-trip time, as discussed above, then the
sender does not need to send its current estimate of the round
trip time.
3.2.2. Feedback Packets
Each feedback packet sent by the data receiver contains the following
information:
o The timestamp of the last data packet received. We denote this by
t_recvdata. If the last packet received at the receiver has
sequence number i, then t_recvdata = ts_i. This timestamp is used
by the sender to estimate the round-trip time, and is only needed
if the sender does not save timestamps of transmitted data
packets.
o The amount of time elapsed between the receipt of the last data
packet at the receiver, and the generation of this feedback
report. We denote this by t_delay.
o The rate at which the receiver estimates that data was received
since the last feedback report was sent. We denote this by
X_recv.
o The receiver's current estimate of the loss event rate, p.
4. Data Sender Protocol
The data sender sends a stream of data packets to the data receiver
at a controlled rate. When a feedback packet is received from the
data receiver, the data sender changes its sending rate, based on the
information contained in the feedback report. If the sender does not
receive a feedback report for two round trip times, it cuts its
sending rate in half. This is achieved by means of a timer called
the nofeedback timer.
We specify the sender-side protocol in the following steps:
o Measurement of the mean packet size being sent.
o The sender behavior when a feedback packet is received.
o The sender behavior when the nofeedback timer expires.
o Oscillation prevention (optional)
o Scheduling of transmission on non-realtime operating systems.
4.1. Measuring the Packet Size
The parameter s (packet size) is normally known to an application.
This may not be so in two cases:
o The packet size naturally varies depending on the data. In this
case, although the packet size varies, that variation is not
coupled to the transmit rate. It should normally be safe to use
an estimate of the mean packet size for s.
o The application needs to change the packet size rather than the
number of packets per second to perform congestion control. This
would normally be the case with packet audio applications where a
fixed interval of time needs to be represented by each packet.
Such applications need to have a completely different way of
measuring parameters.
The second class of applications are discussed separately in a
separate document on TFRC-PS. For the remainder of this section we
assume the sender can estimate the packet size, and that congestion
control is performed by adjusting the number of packets sent per
second.
4.2. Sender Initialization
To initialize the sender, the value of X is set to 1 packet/second
and the nofeedback timer is set to expire after 2 seconds. The
initial values for R (RTT) and t_RTO are undefined until they are set
as described below. The initial value of tld, for the Time Last
Doubled during slow-start, is set to -1.
4.3. Sender behavior when a feedback packet is received
The sender knows its current sending rate, X, and maintains an
estimate of the current round trip time, R, and an estimate of the
timeout interval, t_RTO.
When a feedback packet is received by the sender at time t_now, the
following actions should be performed:
1) Calculate a new round trip sample.
R_sample = (t_now - t_recvdata) - t_delay.
2) Update the round trip time estimate:
If no feedback has been received before
R = R_sample;
Else
R = q*R + (1-q)*R_sample;
TFRC is not sensitive to the precise value for the filter constant q,
but we recommend a default value of 0.9.
3) Update the timeout interval:
t_RTO = 4*R.
4) Update the sending rate as follows:
If (p > 0)
Calculate X_calc using the TCP throughput equation.
X = max(min(X_calc, 2*X_recv), s/t_mbi);
Else
If (t_now - tld >= R)
X = max(min(2*X, 2*X_recv), s/R);
tld = t_now;
Note that if p == 0, then the sender is in slow-start phase, where
it approximately doubles the sending rate each round-trip time
until a loss occurs. The s/R term gives a minimum sending rate
during slow-start of one packet per RTT. The parameter t_mbi is
64 seconds, and represents the maximum inter-packet backoff
interval in the persistent absence of feedback. Thus, when p > 0
the sender sends at least one packet every 64 seconds.
5) Reset the nofeedback timer to expire after max(4*R, 2*s/X)
seconds.
4.4. Expiration of nofeedback timer
If the nofeedback timer expires, the sender should perform the
following actions:
1) Cut the sending rate in half. If the sender has received feedback
from the receiver, this is done by modifying the sender's cached
copy of X_recv (the receive rate). Because the sending rate is
limited to at most twice X_recv, modifying X_recv limits the
current sending rate, but allows the sender to slow-start,
doubling its sending rate each RTT, if feedback messages resume
reporting no losses.
If (X_calc > 2*X_recv)
X_recv = max(X_recv/2, s/(2*t_mbi));
Else
X_recv = X_calc/4;
The term s/(2*t_mbi) limits the backoff to one packet every 64
seconds in the case of persistent absence of feedback.
2) The value of X must then be recalculated as described under point
(4) above.
If the nofeedback timer expires when the sender does not yet have
an RTT sample, and has not yet received any feedback from the
receiver, then step (1) can be skipped, and the sending rate cut
in half directly:
X = max(X/2, s/t_mbi)
3) Restart the nofeedback timer to expire after max(4*R, 2*s/X)
seconds.
Note that when the sender stops sending, the receiver will stop
sending feedback. This will cause the nofeedback timer to start to
expire and decrease X_recv. If the sender subsequently starts to
send again, X_recv will limit the transmit rate, and a normal
slowstart phase will occur until the transmit rate reaches X_calc.
If the sender has been idle since this nofeedback timer was set and
X_recv is less than four packets per round-trip time, then X_recv
should not be halved in response to the timer expiration. This
ensures that the allowed sending rate is never reduced to less than
two packets per round-trip time as a result of an idle period.
4.5. Preventing Oscillations
To prevent oscillatory behavior in environments with a low degree of
statistical multiplexing it is useful to modify sender's transmit
rate to provide congestion avoidance behavior by reducing the
transmit rate as the queuing delay (and hence RTT) increases. To do
this the sender maintains an estimate of the long-term RTT and
modifies its sending rate depending on how the most recent sample of
the RTT differs from this value. The long-term sample is R_sqmean,
and is set as follows:
If no feedback has been received before
R_sqmean = sqrt(R_sample);
Else
R_sqmean = q2*R_sqmean + (1-q2)*sqrt(R_sample);
Thus R_sqmean gives the exponentially weighted moving average of the
square root of the RTT samples. The constant q2 should be set
similarly to q, and we recommend a value of 0.9 as the default.
The sender obtains the base transmit rate, X, from the throughput
function. It then calculates a modified instantaneous transmit rate
X_inst, as follows:
X_inst = X * R_sqmean / sqrt(R_sample);
When sqrt(R_sample) is greater than R_sqmean then the queue is
typically increasing and so the transmit rate needs to be decreased
for stable operation.
Note: This modification is not always strictly required, especially
if the degree of statistical multiplexing in the network is high.
However, we recommend that it is done because it does make TFRC
behave better in environments with a low level of statistical
multiplexing. If it is not done, we recommend using a very low value
of q, such that q is close to or exactly zero.
4.6. Scheduling of Packet Transmissions
As TFRC is rate-based, and as operating systems typically cannot
schedule events precisely, it is necessary to be opportunistic about
sending data packets so that the correct average rate is maintained
despite the course-grain or irregular scheduling of the operating
system. Thus a typical sending loop will calculate the correct
inter-packet interval, t_ipi, as follows:
t_ipi = s/X_inst;
When a sender first starts sending at time t_0, it calculates t_ipi,
and calculates a nominal send time t_1 = t_0 + t_ipi for packet 1.
When the application becomes idle, it checks the current time, t_now,
and then requests re-scheduling after (t_ipi - (t_now - t_0))
seconds. When the application is re-scheduled, it checks the current
time, t_now, again. If (t_now > t_1 - delta) then packet 1 is sent.
Now a new t_ipi may be calculated, and used to calculate a nominal
send time t_2 for packet 2: t2 = t_1 + t_ipi. The process then
repeats, with each successive packet's send time being calculated
from the nominal send time of the previous packet.
In some cases, when the nominal send time, t_i, of the next packet is
calculated, it may already be the case that t_now > t_i - delta. In
such a case the packet should be sent immediately. Thus if the
operating system has coarse timer granularity and the transmit rate
is high, then TFRC may send short bursts of several packets separated
by intervals of the OS timer granularity.
The parameter delta is to allow a degree of flexibility in the send
time of a packet. If the operating system has a scheduling timer
granularity of t_gran seconds, then delta would typically be set to:
delta = min(t_ipi/2, t_gran/2);
t_gran is 10ms on many Unix systems. If t_gran is not known, a value
of 10ms can be safely assumed.
5. Calculation of the Loss Event Rate (p)
Obtaining an accurate and stable measurement of the loss event rate
is of primary importance for TFRC. Loss rate measurement is
performed at the receiver, based on the detection of lost or marked
packets from the sequence numbers of arriving packets. We describe
this process before describing the rest of the receiver protocol.
5.1. Detection of Lost or Marked Packets
TFRC assumes that all packets contain a sequence number that is
incremented by one for each packet that is sent. For the purposes of
this specification, we require that if a lost packet is
retransmitted, the retransmission is given a new sequence number that
is the latest in the transmission sequence, and not the same sequence
number as the packet that was lost. If a transport protocol has the
requirement that it must retransmit with the original sequence
number, then the transport protocol designer must figure out how to
distinguish delayed from retransmitted packets and how to detect lost
retransmissions.
The receiver maintains a data structure that keeps track of which
packets have arrived and which are missing. For the purposes of
specification, we assume that the data structure consists of a list
of packets that have arrived along with the receiver timestamp when
each packet was received. In practice this data structure will
normally be stored in a more compact representation, but this is
implementation-specific.
The loss of a packet is detected by the arrival of at least three
packets with a higher sequence number than the lost packet. The
requirement for three subsequent packets is the same as with TCP, and
is to make TFRC more robust in the presence of reordering. In
contrast to TCP, if a packet arrives late (after 3 subsequent packets
arrived) in TFRC, the late packet can fill the hole in TFRC's
reception record, and the receiver can recalculate the loss event
rate. Future versions of TFRC might make the requirement for three
subsequent packets adaptive based on experienced packet reordering,
but we do not specify such a mechanism here.
For an ECN-capable connection, a marked packet is detected as a
congestion event as soon as it arrives, without having to wait for
the arrival of subsequent packets.
5.2. Translation from Loss History to Loss Events
TFRC requires that the loss fraction be robust to several consecutive
packets lost where those packets are part of the same loss event.
This is similar to TCP, which (typically) only performs one halving
of the congestion window during any single RTT. Thus the receiver
needs to map the packet loss history into a loss event record, where
a loss event is one or more packets lost in an RTT. To perform this
mapping, the receiver needs to know the RTT to use, and this is
supplied periodically by the sender, typically as control information
piggy-backed onto a data packet. TFRC is not sensitive to how the
RTT measurement sent to the receiver is made, but we recommend using
the sender's calculated RTT, R, (see Section 4.3) for this purpose.
To determine whether a lost or marked packet should start a new loss
event, or be counted as part of an existing loss event, we need to
compare the sequence numbers and timestamps of the packets that
arrived at the receiver. For a marked packet S_new, its reception
time T_new can be noted directly. For a lost packet, we can
interpolate to infer the nominal "arrival time". Assume:
S_loss is the sequence number of a lost packet.
S_before is the sequence number of the last packet to arrive with
sequence number before S_loss.
S_after is the sequence number of the first packet to arrive with
sequence number after S_loss.
T_before is the reception time of S_before.
T_after is the reception time of S_after.
Note that T_before can either be before or after T_after due to
reordering.
For a lost packet S_loss, we can interpolate its nominal "arrival
time" at the receiver from the arrival times of S_before and S_after.
Thus:
T_loss = T_before + ( (T_after - T_before)
* (S_loss - S_before)/(S_after - S_before) );
Note that if the sequence space wrapped between S_before and S_after,
then the sequence numbers must be modified to take this into account
before performing this calculation. If the largest possible sequence
number is S_max, and S_before > S_after, then modifying each sequence
number S by S' = (S + (S_max + 1)/2) mod (S_max + 1) would normally
be sufficient.
If the lost packet S_old was determined to have started the previous
loss event, and we have just determined that S_new has been lost,
then we interpolate the nominal arrival times of S_old and S_new,
called T_old and T_new respectively.
If T_old + R >= T_new, then S_new is part of the existing loss event.
Otherwise S_new is the first packet in a new loss event.
5.3. Inter-loss Event Interval
If a loss interval, A, is determined to have started with packet
sequence number S_A and the next loss interval, B, started with
packet sequence number S_B, then the number of packets in loss
interval A is given by (S_B - S_A).
5.4. Average Loss Interval
To calculate the loss event rate p, we first calculate the average
loss interval. This is done using a filter that weights the n most
recent loss event intervals in such a way that the measured loss
event rate changes smoothly.
Weights w_0 to w_(n-1) are calculated as:
If (i < n/2)
w_i = 1;
Else
w_i = 1 - (i - (n/2 - 1))/(n/2 + 1);
Thus if n=8, the values of w_0 to w_7 are:
1.0, 1.0, 1.0, 1.0, 0.8, 0.6, 0.4, 0.2
The value n for the number of loss intervals used in calculating the
loss event rate determines TFRC's speed in responding to changes in
the level of congestion. As currently specified, TFRC should not be
used for values of n significantly greater than 8, for traffic that
might compete in the global Internet with TCP. At the very least,
safe operation with values of n greater than 8 would require a slight
change to TFRC's mechanisms to include a more severe response to two
or more round-trip times with heavy packet loss.
When calculating the average loss interval we need to decide whether
to include the interval since the most recent packet loss event. We
only do this if it is sufficiently large to increase the average loss
interval.
Thus if the most recent loss intervals are I_0 to I_n, with I_0 being
the interval since the most recent loss event, then we calculate the
average loss interval I_mean as:
I_tot0 = 0;
I_tot1 = 0;
W_tot = 0;
for (i = 0 to n-1) {
I_tot0 = I_tot0 + (I_i * w_i);
W_tot = W_tot + w_i;
}
for (i = 1 to n) {
I_tot1 = I_tot1 + (I_i * w_(i-1));
}
I_tot = max(I_tot0, I_tot1);
I_mean = I_tot/W_tot;
The loss event rate, p is simply:
p = 1 / I_mean;
5.5. History Discounting
As described in Section 5.4, the most recent loss interval is only
assigned 1/(0.75*n) of the total weight in calculating the average
loss interval, regardless of the size of the most recent loss
interval. This section describes an optional history discounting
mechanism, discussed further in [3] and [9], that allows the TFRC
receiver to adjust the weights, concentrating more of the relative
weight on the most recent loss interval, when the most recent loss
interval is more than twice as large as the computed average loss
interval.
To carry out history discounting, we associate a discount factor DF_i
with each loss interval L_i, for i > 0, where each discount factor is
a floating point number. The discount array maintains the cumulative
history of discounting for each loss interval. At the beginning, the
values of DF_i in the discount array are initialized to 1:
for (i = 1 to n) {
DF_i = 1;
}
History discounting also uses a general discount factor DF, also a
floating point number, that is also initialized to 1. First we show
how the discount factors are used in calculating the average loss
interval, and then we describe later in this section how the discount
factors are modified over time.
As described in Section 5.4 the average loss interval is calculated
using the n previous loss intervals I_1, ..., I_n, and the interval
I_0 that represents the number of packets received since the last
loss event. The computation of the average loss interval using the
discount factors is a simple modification of the procedure in Section
5.4, as follows:
I_tot0 = I_0 * w_0
I_tot1 = 0;
W_tot0 = w_0
W_tot1 = 0;
for (i = 1 to n-1) {
I_tot0 = I_tot0 + (I_i * w_i * DF_i * DF);
W_tot0 = W_tot0 + w_i * DF_i * DF;
}
for (i = 1 to n) {
I_tot1 = I_tot1 + (I_i * w_(i-1) * DF_i);
W_tot1 = W_tot1 + w_(i-1) * DF_i;
}
p = min(W_tot0/I_tot0, W_tot1/I_tot1);
The general discounting factor, DF is updated on every packet arrival
as follows. First, the receiver computes the weighted average I_mean
of the loss intervals I_1, ..., I_n:
I_tot = 0;
W_tot = 0;
for (i = 1 to n) {
W_tot = W_tot + w_(i-1) * DF_i;
I_tot = I_tot + (I_i * w_(i-1) * DF_i);
}
I_mean = I_tot / W_tot;
This weighted average I_mean is compared to I_0, the number of
packets received since the last loss event. If I_0 is greater than
twice I_mean, then the new loss interval is considerably larger than
the old ones, and the general discount factor DF is updated to
decrease the relative weight on the older intervals, as follows:
if (I_0 > 2 * I_mean) {
DF = 2 * I_mean/I_0;
if (DF < THRESHOLD)
DF = THRESHOLD;
} else
DF = 1;
A nonzero value for THRESHOLD ensures that older loss intervals from
an earlier time of high congestion are not discounted entirely. We
recommend a THRESHOLD of 0.5. Note that with each new packet
arrival, I_0 will increase further, and the discount factor DF will
be updated.
When a new loss event occurs, the current interval shifts from I_0 to
I_1, loss interval I_i shifts to interval I_(i+1), and the loss
interval I_n is forgotten. The previous discount factor DF has to be
incorporated into the discount array. Because DF_i carries the
discount factor associated with loss interval I_i, the DF_i array has
to be shifted as well. This is done as follows:
for (i = 1 to n) {
DF_i = DF * DF_i;
}
for (i = n-1 to 0 step -1) {
DF_(i+1) = DF_i;
}
I_0 = 1;
DF_0 = 1;
DF = 1;
This completes the description of the optional history discounting
mechanism. We emphasize that this is an optional mechanism whose
sole purpose is to allow TFRC to response somewhat more quickly to
the sudden absence of congestion, as represented by a long current
loss interval.
6. Data Receiver Protocol
The receiver periodically sends feedback messages to the sender.
Feedback packets should normally be sent at least once per RTT,
unless the sender is sending at a rate of less than one packet per
RTT, in which case a feedback packet should be send for every data
packet received. A feedback packet should also be sent whenever a
new loss event is detected without waiting for the end of an RTT, and
whenever an out-of-order data packet is received that removes a loss
event from the history.
If the sender is transmitting at a high rate (many packets per RTT)
there may be some advantages to sending periodic feedback messages
more than once per RTT as this allows faster response to changing RTT
measurements, and more resilience to feedback packet loss. However,
there is little gain from sending a large number of feedback messages
per RTT.
6.1. Receiver behavior when a data packet is received
When a data packet is received, the receiver performs the following
steps:
1) Add the packet to the packet history.
2) Let the previous value of p be p_prev. Calculate the new value of
p as described in Section 5.
3) If p > p_prev, cause the feedback timer to expire, and perform the
actions described in Section 6.2
If p <= p_prev no action need be performed.
However an optimization might check to see if the arrival of the
packet caused a hole in the packet history to be filled and
consequently two loss intervals were merged into one. If this is
the case, the receiver might also send feedback immediately. The
effects of such an optimization are normally expected to be small.
6.2. Expiration of feedback timer
When the feedback timer at the receiver expires, the action to be
taken depends on whether data packets have been received since the
last feedback was sent.
Let the maximum sequence number of a packet at the receiver so far be
S_m, and the value of the RTT measurement included in packet S_m be
R_m. If data packets have been received since the previous feedback
was sent, the receiver performs the following steps:
1) Calculate the average loss event rate using the algorithm
described above.
2) Calculate the measured receive rate, X_recv, based on the packets
received within the previous R_m seconds.
3) Prepare and send a feedback packet containing the information
described in Section 3.2.2
4) Restart the feedback timer to expire after R_m seconds.
If no data packets have been received since the last feedback was
sent, no feedback packet is sent, and the feedback timer is restarted
to expire after R_m seconds.
6.3. Receiver initialization
The receiver is initialized by the first packet that arrives at the
receiver. Let the sequence number of this packet be i.
When the first packet is received:
o Set p=0
o Set X_recv = 0.
o Prepare and send a feedback packet.
o Set the feedback timer to expire after R_i seconds.
6.3.1. Initializing the Loss History after the First Loss Event
The number of packets until the first loss can not be used to compute
the sending rate directly, as the sending rate changes rapidly during
this time. TFRC assumes that the correct data rate after the first
loss is half of the sending rate when the loss occurred. TFRC
approximates this target rate by X_recv, the receive rate over the
most recent round-trip time. After the first loss, instead of
initializing the first loss interval to the number of packets sent
until the first loss, the TFRC receiver calculates the loss interval
that would be required to produce the data rate X_recv, and uses this
synthetic loss interval to seed the loss history mechanism.
TFRC does this by finding some value p for which the throughput
equation in Section 3.1 gives a sending rate within 5% of X_recv,
given the current packet size s and round-trip time R. The first
loss interval is then set to 1/p. (The 5% tolerance is introduced
simply because the throughput equation is difficult to invert, and we
want to reduce the costs of calculating p numerically.)
7. Sender-based Variants
It would be possible to implement a sender-based variant of TFRC,
where the receiver uses reliable delivery to send information about
packet losses to the sender, and the sender computes the packet loss
rate and the acceptable transmit rate. However, we do not specify
the details of a sender-based variant in this document.
The main advantages of a sender-based variant of TFRC would be that
the sender would not have to trust the receiver's calculation of the
packet loss rate. However, with the requirement of reliable delivery
of loss information from the receiver to the sender, a sender-based
TFRC would have much tighter constraints on the transport protocol in
which it is embedded.
In contrast, the receiver-based variant of TFRC specified in this
document is robust to the loss of feedback packets, and therefore
does not require the reliable delivery of feedback packets. It is
also better suited for applications such as streaming media from web
servers, where it is typically desirable to offload work from the
server to the client as much as possible.
The sender-based and receiver-based variants also have different
properties in terms of upgrades. For example, for changes in the
procedure for calculating the packet loss rate, the sender would have
to be upgraded in the sender-based variant, and the receiver would
have to be upgraded in the receiver-based variant.
8. Implementation Issues
This document has specified the TFRC congestion control mechanism,
for use by applications and transport protocols. This section
mentions briefly some of the few implementation issues.
For t_RTO = 4*R and b = 1, the throughput equation in Section 3.1 can
be expressed as follows:
s
X = --------
R * f(p)
for
f(p) = sqrt(2*p/3) + (12*sqrt(3*p/8) * p * (1+32*p^2)).
A table lookup could be used for the function f(p).
Many of the multiplications (e.g., q and 1-q for the round-trip time
average, a factor of 4 for the timeout interval) are or could be by
powers of two, and therefore could be implemented as simple shift
operations.
We note that the optional sender mechanism for preventing
oscillations described in Section 4.5 uses a square-root computation.
The calculation of the average loss interval in Section 5.4 involves
multiplications by the weights w_0 to w_(n-1), which for n=8 are:
1.0, 1.0, 1.0, 1.0, 0.8, 0.6, 0.4, 0.2.
With a minor loss of smoothness, it would be possible to use weights
that were powers of two or sums of powers of two, e.g.,
1.0, 1.0, 1.0, 1.0, 0.75, 0.5, 0.25, 0.25.
The optional history discounting mechanism described in Section 5.5
is used in the calculation of the average loss rate. The history
discounting mechanism is invoked only when there has been an
unusually long interval with no packet losses. For a more efficient
operation, the discount factor DF_i could be restricted to be a power
of two.
9. Security Considerations
TFRC is not a transport protocol in its own right, but a congestion
control mechanism that is intended to be used in conjunction with a
transport protocol. Therefore security primarily needs to be
considered in the context of a specific transport protocol and its
authentication mechanisms.
Congestion control mechanisms can potentially be exploited to create
denial of service. This may occur through spoofed feedback. Thus
any transport protocol that uses TFRC should take care to ensure that
feedback is only accepted from the receiver of the data. The precise
mechanism to achieve this will however depend on the transport
protocol itself.
In addition, congestion control mechanisms may potentially be
manipulated by a greedy receiver that wishes to receive more than its
fair share of network bandwidth. A receiver might do this by
claiming to have received packets that in fact were lost due to
congestion. Possible defenses against such a receiver would normally
include some form of nonce that the receiver must feed back to the
sender to prove receipt. However, the details of such a nonce would
depend on the transport protocol, and in particular on whether the
transport protocol is reliable or unreliable.
We expect that protocols incorporating ECN with TFRC will also want
to incorporate feedback from the receiver to the sender using the ECN
nonce [WES02]. The ECN nonce is a modification to ECN that protects
the sender from the accidental or malicious concealment of marked
packets. Again, the details of such a nonce would depend on the
transport protocol, and are not addressed in this document.
10. IANA Considerations
There are no IANA actions required for this document.
11. Acknowledgments
We would like to acknowledge feedback and discussions on equation-
based congestion control with a wide range of people, including
members of the Reliable Multicast Research Group, the Reliable
Multicast Transport Working Group, and the End-to-End Research Group.
We would like to thank Ken Lofgren, Mike Luby, Eduardo Urzaiz,
Vladica Stanisic, Randall Stewart, Shushan Wen, and Wendy Lee
(lhh@zsu.edu.cn) for feedback on earlier versions of this document,
and to thank Mark Allman for his extensive feedback from using the
document to produce a working implementation.
12. Informational References
[1] Balakrishnan, H., Rahul, H., and S. Seshan, "An Integrated
Congestion Management Architecture for Internet Hosts," Proc. ACM
SIGCOMM, Cambridge, MA, September 1999.
[2] Floyd, S., Handley, M., Padhye, J. and J. Widmer, "Equation-Based
Congestion Control for Unicast Applications", August 2000, Proc.
ACM SIGCOMM 2000.
[3] Floyd, S., Handley, M., Padhye, J. and J. Widmer, "Equation-Based
Congestion Control for Unicast Applications: the Extended
Version", ICSI tech report TR-00-03, March 2000.
[4] Padhye, J., Firoiu, V., Towsley, D. and J. Kurose, "Modeling TCP
Throughput: A Simple Model and its Empirical Validation", Proc.
ACM SIGCOMM 1998.
[5] Paxson V. and M. Allman, "Computing TCP's Retransmission Timer",
RFC 2988, November 2000.
[6] Ramakrishnan, K., Floyd, S. and D. Black, "The Addition of
Explicit Congestion Notification (ECN) to IP", RFC 3168,
September 2001.
[7] Schulzrinne, H., Casner, S., Frederick, R. and V. Jacobson, "RTP:
A Transport Protocol for Real-Time Applications", RFC 1889,
January 1996.
[8] Wetherall, D., Ely, D., N. Spring, S. Savage, and T. Anderson,
"Robust Congestion Signaling", IEEE International Conference on
Network Protocols, November 2001.
[9] Widmer, J., "Equation-Based Congestion Control", Diploma Thesis,
University of Mannheim, February 2000. URL
"http://www.icir.org/tfrc/".
13. Authors' Addresses
Mark Handley
ICIR/ICSI
1947 Center St, Suite 600
Berkeley, CA 94708
EMail: mjh@icir.org
Sally Floyd
ICIR/ICSI
1947 Center St, Suite 600
Berkeley, CA 94708
EMail: floyd@icir.org
Jitendra Padhye
Microsoft Research
EMail: padhye@microsoft.com
Joerg Widmer
Lehrstuhl Praktische Informatik IV
Universitat Mannheim
L 15, 16 - Room 415
D-68131 Mannheim
Germany
EMail: widmer@informatik.uni-mannheim.de
14. Full Copyright Statement
Copyright (C) The Internet Society (2003). All Rights Reserved.
This document and translations of it may be copied and furnished to
others, and derivative works that comment on or otherwise explain it
or assist in its implementation may be prepared, copied, published
and distributed, in whole or in part, without restriction of any
kind, provided that the above copyright notice and this paragraph are
included on all such copies and derivative works. However, this
document itself may not be modified in any way, such as by removing
the copyright notice or references to the Internet Society or other
Internet organizations, except as needed for the purpose of
developing Internet standards in which case the procedures for
copyrights defined in the Internet Standards process must be
followed, or as required to translate it into languages other than
English.
The limited permissions granted above are perpetual and will not be
revoked by the Internet Society or its successors or assigns.
This document and the information contained herein is provided on an
"AS IS" basis and THE INTERNET SOCIETY AND THE INTERNET ENGINEERING
TASK FORCE DISCLAIMS ALL WARRANTIES, EXPRESS OR IMPLIED, INCLUDING
BUT NOT LIMITED TO ANY WARRANTY THAT THE USE OF THE INFORMATION
HEREIN WILL NOT INFRINGE ANY RIGHTS OR ANY IMPLIED WARRANTIES OF
MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.
Acknowledgement
Funding for the RFC Editor function is currently provided by the
Internet Society.