RFC 4654

Network Working Group                                          J. Widmer
Request for Comments: 4654                              DoCoMo Euro-Labs
Category: Experimental                                        M. Handley
                                                             August 2006

          TCP-Friendly Multicast Congestion Control (TFMCC):
                         Protocol Specification

Status of This Memo

   This memo defines an Experimental Protocol for the Internet
   community.  It does not specify an Internet standard of any kind.
   Discussion and suggestions for improvement are requested.
   Distribution of this memo is unlimited.

Copyright Notice

   Copyright (C) The Internet Society (2006).


   This document specifies TCP-Friendly Multicast Congestion Control
   (TFMCC).  TFMCC is a congestion control mechanism for multicast
   transmissions in a best-effort Internet environment.  It is a
   single-rate congestion control scheme, where the sending rate is
   adapted to the receiver experiencing the worst network conditions.
   TFMCC is reasonably fair when competing for bandwidth with TCP flows
   and has a relatively low variation of throughput over time, making it
   suitable for applications where a relatively smooth sending rate is
   of importance, such as streaming media.

Widmer & Handley              Experimental                      [Page 1]

RFC 4654             TFMCC: Protocol Specification           August 2006

Table of Contents

   1. Introduction ....................................................3
      1.1. Related Documents ..........................................4
      1.2. Environmental Requirements and Considerations ..............4
   2. Protocol Overview ...............................................5
      2.1. TCP Throughput Equation ....................................6
      2.2. Packet Contents ............................................7
           2.2.1. Sender Packets ......................................8
           2.2.2. Feedback Packets ....................................9
   3. Data Sender Protocol ...........................................10
      3.1. Sender Initialization .....................................10
      3.2. Determining the Maximum RTT ...............................10
      3.3. Adjusting the Sending Rate ................................11
      3.4. Controlling Receiver Feedback .............................12
      3.5. Assisting Receiver-Side RTT Measurements ..................14
      3.6. Slowstart .................................................15
      3.7. Scheduling of Packet Transmissions ........................15
   4. Data Receiver Protocol .........................................16
      4.1. Receiver Initialization ...................................17
      4.2. Receiver Leave ............................................17
      4.3. Measurement of the Network Conditions .....................17
           4.3.1. Updating the Loss Event Rate .......................17
           4.3.2. Basic Round-Trip Time Measurement ..................17
           4.3.3. One-Way Delay Adjustments ..........................18
           4.3.4. Receive Rate Measurements ..........................19
      4.4. Setting the Desired Rate ..................................19
      4.5. Feedback and Feedback Suppression .........................20
   5. Calculation of the Loss Event Rate .............................22
      5.1. Detection of Lost or Marked Packets .......................22
      5.2. Translation from Loss History to Loss Events ..............23
      5.3. Inter-Loss Event Interval .................................24
      5.4. Average Loss Interval .....................................24
      5.5. History Discounting .......................................25
      5.6. Initializing the Loss History after the First Loss Event ..27
   6. Security Considerations ........................................28
   7. Acknowledgments ................................................29
   8. References .....................................................29
      8.1. Normative References ......................................29
      8.2. Informative References ....................................29

Widmer & Handley              Experimental                      [Page 2]

RFC 4654             TFMCC: Protocol Specification           August 2006

1.  Introduction

   This document specifies TCP-Friendly Multicast Congestion Control
   (TFMCC) [3].  TFMCC is a source-based, single-rate congestion control
   scheme that builds upon the unicast TCP-Friendly Rate Control
   mechanism (TFRC) [4].  TFMCC is stable and responsive under a wide
   range of network conditions and scales to receiver sets on the order
   of several thousand receivers.  To support scalability, as much
   congestion control functionality as possible is located at the
   receivers.  Each receiver continuously determines a desired receive
   rate that is TCP-friendly for the path from the sender to this
   receiver.  Selected receivers then report the rate to the sender in
   feedback packets.

   TFMCC is a building block as defined in RFC 3048 [1].  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 [11], in an application incorporating end-to-end
   congestion control at the application level.  This document does not
   discuss packet formats, reliability, or implementation-related

   TFMCC is designed to be reasonably fair when competing for bandwidth
   with TCP flows.  A multicast flow is "reasonably fair" if its sending
   rate is generally within a factor of two of the sending rate of a TCP
   flow from the sender to the slowest receiver of the multicast group
   under the same network conditions.

   In general, TFMCC has a low variation of throughput, which makes it
   suitable for applications where a relatively smooth sending rate is
   of importance, such as streaming media.  The penalty of having smooth
   throughput while competing fairly for bandwidth is a reduced
   responsiveness to changes in available bandwidth.  Thus TFMCC should
   be used when the application has a requirement for smooth throughput,
   in particular, avoiding halving of the sending rate in response to a
   single packet drop.  For applications that simply need to multicast
   as much data as possible in as short a time as possible, PGMCC [10]
   may be more suitable.

   This memo contains part of the definitions necessary to fully specify
   a Reliable Multicast Transport protocol in accordance with RFC 2357.
   As per RFC 2357, the use of any reliable multicast protocol in the
   Internet requires an adequate congestion control scheme.  This
   document specifies an experimental congestion control scheme.  While
   waiting for initial deployment and experience to show this scheme to
   be effective and scalable, the IETF publishes this scheme in the
   "Experimental" category.

Widmer & Handley              Experimental                      [Page 3]

RFC 4654             TFMCC: Protocol Specification           August 2006

   It is the intent of the Reliable Multicast Transport (RMT) Working
   Group to re-submit the specification as an IETF Proposed Standard as
   soon as the scheme is deemed adequate.

1.1.  Related Documents

   As described in RFC 3048 [1], TFMCC is a building block that is
   intended to be used, in conjunction with other building blocks, to
   help specify a protocol instantiation.  It follows the general
   guidelines provided in RFC 3269 [2].  In particular, TFMCC is a
   suitable congestion control building block for NACK-Oriented Reliable
   Multicast (NORM) [5].

1.2.  Environmental Requirements and Considerations

   TFMCC is intended to be a congestion control scheme that can be used
   in a complete protocol instantiation that delivers objects and
   streams (both reliable content delivery and streaming of multimedia

   TFMCC is most applicable for sessions that deliver a substantial
   amount of data (i.e., in length from hundreds of kilobytes to many
   gigabytes) and whose duration is on the order of tens of seconds or

   TFMCC is intended for multicast delivery.  There are currently two
   models of multicast delivery: the Any-Source Multicast (ASM) model as
   defined in [6] and the Source-Specific Multicast (SSM) model as
   defined in [7].  TFMCC works with both multicast models, but in a
   slightly different way.  When ASM is used, feedback from the
   receivers is multicast to the sender, as well as to all other
   receivers.  Feedback can be either multicast on the same group
   address used for sending data or on a separate multicast feedback
   group address.  For SSM, the receivers must unicast the feedback
   directly to the sender.  Hence, feedback from a receiver will not be
   received by other receivers.

   TFMCC inherently works with all types of networks that allow bi-
   directional communication, including LANs, WANs, Intranets, the
   Internet, asymmetric networks, wireless networks, and satellite
   networks.  However, in some network environments varying the sending
   rate to the receivers may not be advantageous (e.g., for a satellite
   or wireless network, there may be no mechanism for receivers to
   effectively reduce their reception rate since there may be a fixed
   transmission rate allocated to the session).

Widmer & Handley              Experimental                      [Page 4]

RFC 4654             TFMCC: Protocol Specification           August 2006

   The difference in responsiveness of TFMCC and TCP may result in
   significant throughput differences in case of a very low bitrate.
   TFMCC requires an estimate of the loss event rate to calculate a fair
   sending rate.  This estimate may be inaccurate in case TFMCC receives
   only very few packets per RTT.  TFMCC should not be used together
   with TCP if the capacity of the bottleneck link is less than 30KBit/s
   (e.g., a very slow modem connection).  TFMCC may also achieve a rate
   that is very different from the average TCP rate in case buffer space
   at the bottleneck is severely underprovisioned.  In particular, TFMCC
   is less susceptible to small buffer sizes since TFMCC spaces out
   packets in time, whereas TCP sends them back to back.  Thus TCP is
   much more likely to see a packet loss if buffer space is scarce.

   TFMCC is designed for applications that use a fixed packet size and
   vary their sending rate in packets per second in response to
   congestion.  Some applications (e.g., those using audio) 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

2.  Protocol Overview

   TFMCC extends the basic mechanisms of TFRC into the multicast domain.
   In order to compete fairly with TCP, TFMCC receivers individually
   measure the prevalent network conditions and calculate a rate that is
   TCP-friendly on the path from the sender to themselves.  The rate is
   determined using an equation for TCP throughput, which roughly
   describes TCP's sending rate as a function of the loss event rate,
   round-trip time (RTT), and packet size.  We define a loss event as
   one or more lost or marked packets from the packets received during
   one RTT, where a marked packet refers to a congestion indication from
   Explicit Congestion Notification (ECN) [9].  The sending rate of the
   multicast transmission is adapted to the receiver experiencing the
   worst network conditions.

   Basically, TFMCC's congestion control mechanism works as follows:

   o Each receiver measures the loss event rate and its RTT to the

   o Each receiver then uses this information, together with an equation
     for TCP throughput, to derive a TCP-friendly sending rate.

Widmer & Handley              Experimental                      [Page 5]

RFC 4654             TFMCC: Protocol Specification           August 2006

   o Through a distributed feedback suppression mechanism, only a subset
     of the receivers are allowed to give feedback to prevent a feedback
     implosion at the sender.  The feedback mechanism ensures that
     receivers reporting a low desired transmission rate have a high
     probability of sending feedback.

   o Receivers whose feedback is not suppressed report the calculated
     transmission rate back to the sender in so-called receiver reports.
     The receiver reports serve two purposes: they inform the sender
     about the appropriate transmit rate, and they allow the receivers
     to measure their RTT.

   o The sender selects the receiver that reports the lowest rate as
     current limiting receiver (CLR).  Whenever feedback with an even
     lower rate reaches the sender, the corresponding receiver becomes
     CLR and the sending rate is reduced to match that receiver's
     calculated rate.  The sending rate increases when the CLR reports a
     calculated rate higher than the current sending rate.

   The dynamics of TFMCC are sensitive to how the measurements are
   performed and applied and to what feedback suppression mechanism is
   chosen.  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 TFMCC.

2.1.  TCP Throughput Equation

   Any realistic equation giving TCP throughput as a function of loss
   event rate and RTT should be suitable for use in TFMCC.  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 TFMCC is a
   slightly simplified version of the throughput equation for Reno TCP
   from [8]:

                                  8 s
   X =  ---------------------------------------------------------   (1)
         R * (sqrt(2*p/3) + (12*sqrt(3*p/8) * p * (1+32*p^2)))

Widmer & Handley              Experimental                      [Page 6]

RFC 4654             TFMCC: Protocol Specification           August 2006


      X is the transmit rate in bits/second.

      s is the packet size in bytes.

      R is the round-trip time in seconds.

      p is the loss event rate, between 0.0 and 1.0, of the number of
        loss events as a fraction of the number of packets transmitted.

   In the 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 TFMCC implementation.  The
   measurement of R is specified in Section 4.3.2, and the measurement
   of p is specified in Section 5.  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 different way of measuring

   Currently, TFMCC cannot be used for the second class of applications.

2.2.  Packet Contents

   Before specifying the sender and receiver functionality, we describe
   the congestion control information contained in packets sent by the
   sender and feedback packets from the receivers.  Information from the
   sender can either be sent in separate congestion control messages or
   piggybacked onto data packets.  If separate congestion control
   messages are sent at time intervals larger than the time interval
   between data packets (e.g., once per feedback round), it is necessary
   to be able to include timestamp information destined for more than
   one receiver to allow a sufficient number of receivers to measure
   their RTT.

Widmer & Handley              Experimental                      [Page 7]

RFC 4654             TFMCC: Protocol Specification           August 2006

   As TFMCC will be used along with a transport protocol, we do not
   specify packet formats, since these depend on the details of the
   transport protocol used.  The recommended representation of the
   header fields is given below.  Alternatively, if the computational
   overhead of a floating point representation is prohibitive, fixed
   point arithmetic can be used at the expense of larger packet headers.
   Sender and receivers of a specific TFMCC instance need to agree on a
   common encoding for the header fields.

2.2.1.  Sender Packets

   Each packet sent by the data sender contains the following

   o A sequence number i.  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.  In most cases, the sequence number will be supplied
     by the transport protocol used along with TFMCC.

   o A suppression rate X_supp in bits/s.  Only receivers with a
     calculated rate lower than the suppression rate are eligible to
     give feedback, unless their RTT is higher than the maximum RTT
     described below, in which case they are also eligible to give
     feedback.  The suppression rate should be represented as a 12-bit
     floating point value with 5 bits for the unsigned exponent and 7
     bits for the unsigned mantissa (to represent rates from 100 bit/s
     to 400 Gbit/s with an error of less than 1%).

   o A timestamp ts_i indicating when the packet is sent.  The
     resolution of the timestamp should typically be milliseconds, and
     the timestamp should be an unsigned integer value no less than 16
     bits wide.

   o A receiver ID r and a copy of the timestamp tr_r' = tr_r of that
     receiver's last report, which allows the receiver to measure its
     RTT.  If there is a delay ts_d between receiving the report from
     receiver r and sending the data packet, then tr_r' = tr_r + ts_d is
     included in the packet instead.  The receiver ID is described in
     the next section.  The resolution of the timestamp echo should be
     milliseconds, and the timestamp should be an unsigned integer value
     no less than 16 bits wide.  If separate congestion control messages
     are used instead of piggybacked ones, the packet needs to contain a
     list of receiver IDs with corresponding timestamps to allow a
     sufficient number of receivers to simultaneously measure their RTT.
     For the default values used for the feedback process, this
     corresponds to a list size on the order of 10 to 20 entries.

Widmer & Handley              Experimental                      [Page 8]

RFC 4654             TFMCC: Protocol Specification           August 2006

   o A flag is_CLR indicating whether the receiver with ID r is the CLR.

   o A feedback round counter fb_nr.  This counter is incremented by the
     sender at the beginning of a new feedback round to notify the
     receivers that all feedback for older rounds should be suppressed.
     The feedback round counter should be at least 4 bits wide.

   o A maximum RTT value R_max, representing the maximum of the RTTs of
     all receivers.  The RTT should be measured in milliseconds.  An
     8-bit floating point value with 4 bits for the unsigned exponent
     and 4 bits for the unsigned mantissa (to represent RTTs from 1
     millisecond to 64 seconds with an error of ca. 6%) should be used
     for the representation.

2.2.2.  Feedback Packets

     Each feedback packet sent by a data receiver contains the following

   o A unique receiver ID r.  In most cases, the receiver ID will be
     supplied by the transport protocol, but it may simply be the IP
     address of the receiver.

   o A flag have_RTT indicating whether the receiver has made at least
     one RTT measurement since it joined the session.

   o A flag have_loss indicating whether the receiver experienced at
     least one loss event since it joined the session.

   o A flag receiver_leave indicating that the receiver will leave the
     session (and should therefore not be CLR).

   o A timestamp tr_r indicating when the feedback packet is sent.  The
     representation of the timestamp should be the same as that of the
     timestamp echo in the data packets.

   o An echo ts_i' of the timestamp of the last data packet received.
     If the last packet received at the receiver has sequence number i,
     then ts_i' = ts_i is included in the feedback.  If there is a delay
     tr_d between receiving that last data packet and sending feedback,
     then ts_i' = ts_i + tr_d is included in the feedback instead.  The
     representation of the timestamp echo should be the same as that of
     the timestamp in the data packets.

   o A feedback round echo fb_nr, reflecting the highest feedback round
     counter value received so far.  The representation of the feedback
     round echo should be the same as the one used for the feedback
     round counter in the data packets.

Widmer & Handley              Experimental                      [Page 9]

RFC 4654             TFMCC: Protocol Specification           August 2006

   o The desired sending rate X_r.  This is the rate calculated by the
     receiver to be TCP-friendly on the path from the sender to this
     receiver.  The representation of the desired sending rate should be
     the same as that of the suppression rate in the data packets.

3.  Data Sender Protocol

   The data sender multicasts a stream of data packets to the data
   receivers at a controlled rate.  Whenever feedback is received, the
   sender checks if it is necessary to switch CLRs and to readjust the
   sending rate.

   The main tasks that have to be provided by a TFMCC sender are:

   o adjusting the sending rate,

   o controlling receiver feedback, and

   o assisting receiver-side RTT measurements.

3.1.  Sender Initialization

   At initialization of the sender, the maximum RTT is set to a value
   that should be larger than the highest RTT to any of the receivers.
   It should not be smaller than 500 milliseconds for operation in the
   public Internet.  The sending rate X is initialized to 1 packet per
   maximum RTT.

3.2.  Determining the Maximum RTT

   For each feedback packet that arrives at the sender, the sender
   computes the instantaneous RTT to the receiver as

      R_r = ts_now - ts_i'

   where ts_now is the time the feedback packet arrived.  Receivers will
   have adjusted ts_i' for the time interval between receiving the last
   data packet and sending the corresponding report so that this
   interval will not be included in R_r.  If the actual RTT is smaller
   than the resolution of the timestamps and ts_now equals ts_i', then
   R_r is set to the smallest positive RTT value larger than 0 (i.e., 1
   millisecond in our case).  If the instantaneous RTT is larger than
   the current maximum RTT, the maximum RTT is increased to that value:

      R_max = R_r

Widmer & Handley              Experimental                     [Page 10]

RFC 4654             TFMCC: Protocol Specification           August 2006

   Otherwise, if no feedback with a higher instantaneous RTT than the
   maximum RTT is received during a feedback round (see Section 3.4),
   the maximum RTT is reduced to

      R_max = MAX(R_max * 0.9, R_peak)

   where R_peak is the peak receiver RTT measured during the feedback

   The maximum RTT is mainly used for feedback suppression among
   receivers with heterogeneous RTTs.  Feedback suppression is closely
   coupled to the sending of data packets, and for this reason, the
   maximum RTT must not decrease below the maximum time interval between
   consecutive data packets:

      R_max = max(R_max, 8s/X + ts_gran)

   where ts_gran is the granularity of the sender's system clock (see
   Section 3.7).

3.3.  Adjusting the Sending Rate

   When a feedback packet from receiver r arrives at the sender, the
   sender has to check whether it is necessary to adjust the
   transmission rate and to switch to a new CLR.

   How the rate is adjusted depends on the desired rate X_r of the
   receiver report.  We distinguish four cases:

   1.  If no CLR is present, receiver r becomes the current limiting
       receiver.  The sending rate X is directly set to X_r, so long as
       this would result in a rate increase of less than 8s/R_max bits/s
       (i.e., 1 packet per R_max).  Otherwise X is gradually increased
       to X_r at an increase rate of no more than 8s/R_max bits/s every
       R_max seconds.

   2.  If receiver r is not the CLR but a CLR is present, then receiver
       r becomes the current limiting receiver if X_r is less than the
       current sending rate X and the receiver_leave flag of that
       receiver's report is not set.  Furthermore, the sending rate is
       reduced to X_r.

   3.  If receiver r is not the CLR but a CLR is present and the
       receiver_leave flag of the CLR's last report was set, then
       receiver r becomes the current limiting receiver.  However, if
       X_r > X, the sending rate is not increased to X_r for the
       duration of a feedback round to allow other (lower rate)
       receivers to give feedback and be selected as CLR.

Widmer & Handley              Experimental                     [Page 11]

RFC 4654             TFMCC: Protocol Specification           August 2006

   4.  If receiver r is the CLR, the sending rate is set to the minimum
       of X_r and X + 8s/R_max bits/s.

   If the receiver has not yet measured its RTT but already experienced
   packet loss (indicated by the corresponding flags in the receiver
   report), the receiver report will include a desired rate that is
   based on the maximum RTT rather than the actual RTT to that receiver.
   In this case, the sender adjusts the desired rate using its
   measurement of the instantaneous RTT R_r to that receiver:

      X_r' = X_r * R_max / R_r

   X_r' is then used instead of X_r to detect whether to switch to a new

   If the TFMCC sender receives no reports from the CLR for 4 RTTs, the
   sending rate is cut in half unless the CLR was selected less than 10
   RTTs ago.  In addition, if the sender receives no reports from the
   CLR for at least 10 RTTs, it assumes that the CLR crashed or left the
   group.  A new CLR is selected from the feedback that subsequently
   arrives at the sender, and we increase as in case 3, above.

   If no new CLR can be selected (i.e., in the absence of any feedback
   from any of the receivers) it is necessary to reduce the sending rate
   further.  For every 10 consecutive RTTs without feedback, the sending
   rate is cut in half.  The rate is at most reduced to one packet every
   8 seconds.

   Note that when receivers stop receiving data packets, they will stop
   sending feedback.  This eventually causes the sending rate to be
   reduced in the case of network failure.  If the network subsequently
   recovers, a linear increase to the calculated rate of the CLR will
   occur at 8s/R_max bits/s every R_max.

   An application using TFMCC may have a minimum sending rate
   requirement, where the application becomes unusable if the sending
   rate continuously falls below this minimum rate.  The application
   should exclude receivers that report such a low rate from the
   multicast group.  The specific mechanism to do this is application
   dependent and beyond the scope of this document.

3.4.  Controlling Receiver Feedback

   The receivers allowed to send a receiver report are determined in so-
   called feedback rounds.  Feedback rounds have a duration T of six
   times the maximum RTT.  In case the multicast model is ASM (i.e.,
   receiver feedback is multicast to the whole group) the duration of a
   feedback round may be reduced to four times the maximum RTT.

Widmer & Handley              Experimental                     [Page 12]

RFC 4654             TFMCC: Protocol Specification           August 2006

   Only receivers wishing to report a rate that is lower than the
   suppression rate X_supp or those with a higher RTT than R_max may
   send feedback.  At the beginning of each feedback round, X_supp is
   set to the highest possible value that can be represented.  When
   feedback arrives at the sender over the course of a feedback round,
   X_supp is decreased such that more and more feedback is suppressed
   towards the end of the round.  How receiver feedback is spread out
   over the feedback round is discussed in Section 4.5.

   Whenever non-CLR feedback for the current round arrives at the
   sender, X_supp is reduced to

      X_supp = (1-g) * X_r

   if X_supp > X_r.  Feedback that causes the corresponding receiver to
   be selected as CLR, but that was from a non-CLR receiver at the time
   of sending, also contributes to the feedback suppression.  Note that
   X_r must not be adjusted by the sender to reflect the receiver's real
   RTT in case X_r was calculated using the maximum RTT, as is done for
   setting the sending rate (Section 3.3); otherwise, a feedback
   implosion is possible.  The parameter g determines to what extent
   higher rate feedback can suppress lower rate feedback.  This
   mechanism guarantees that the lowest calculated rate reported lies
   within a factor of g of the actual lowest calculated rate of the
   receiver set (see [13]).  A value of g of 0.1 is recommended.

   To allow receivers to suppress their feedback, the sender's
   suppression rate needs to be updated whenever feedback is received.
   This suppression rate has to be communicated to the receivers in a
   timely manner, either by including it in the data packet header or,
   if separate congestion control messages are used, by sending a
   message with the suppression rate whenever the rate changes
   significantly (i.e., when it is reduced to less than (1-g) times the
   previously advertised suppression rate).

   After a time span of T, the feedback round ends if non-CLR feedback
   was received during that time.  Otherwise, the feedback round ends as
   soon as the first non-CLR feedback message arrives at the sender but
   at most after 2T.  The feedback round counter is incremented by one,
   and the suppression rate X_supp is reset to the highest representable
   value.  The feedback round counter restarts with round 0 after a

Widmer & Handley              Experimental                     [Page 13]

RFC 4654             TFMCC: Protocol Specification           August 2006

3.5.  Assisting Receiver-Side RTT Measurements

   Receivers measure their RTT by sending a timestamp with a receiver
   report, which is echoed by the sender.  If congestion control
   information is piggybacked onto data packets, usually only one
   receiver ID and timestamp can be included.  If multiple feedback
   messages from different receivers arrive at the sender during the
   time interval between two data packets, the sender has to decide
   which receiver to allow to measure the RTT.  The same applies if
   separate congestion control messages allow echoing multiple receiver
   timestamps simultaneously, but the number of receivers that gave
   feedback since the last congestion control message exceeds the list

   The sender's timestamp echoes are prioritized in the following order:

   1.  a new CLR (after a change of CLR's) or a CLR without any previous
       RTT measurements

   2.  receivers without any previous RTT measurements in the order of
       the feedback round echo of the corresponding receiver report
       (i.e., older feedback first)

   3.  non-CLR receivers with previous RTT measurements, again in
       ascending order of the feedback round echo of the report

   4.  the CLR

   Ties are broken in favor of the receiver with the lowest reported

   It is necessary to account for the time that elapses between
   receiving a report and sending the next data packet.  This time needs
   to be deducted from the RTT and thus has to be added to the
   receiver's timestamp value.

   Whenever no feedback packets arrive in the interval between two data
   packets, the CLR's last timestamp, adjusted by the appropriate
   offset, is echoed.  When the number of packets per RTT is so low that
   all packets carry a non-CLR receiver's timestamp, the CLR's timestamp
   and ID are included in a data packet at least once per feedback

Widmer & Handley              Experimental                     [Page 14]

RFC 4654             TFMCC: Protocol Specification           August 2006

3.6.  Slowstart

   TFMCC uses a slowstart mechanism to quickly approach its fair
   bandwidth share at the start of a session.  During slowstart, the
   sending rate increases exponentially.  The rate increase is limited
   to the minimum of the rates included in the receiver reports, and
   receivers report twice the rate at which they currently receive data.
   As in normal congestion control mode, the receiver with the smallest
   reported rate becomes CLR.  Since a receiver can never receive data
   at a rate higher than its link bandwidth, this effectively limits the
   overshoot to twice this bandwidth.  In case the resulting increase
   over R_max is less than 8s/R_max bits/s, the sender may choose to
   increase the rate by up to 8s/R_max bits/s every R_max.  The current
   sending rate is gradually adjusted to the target rate reported in the
   receiver reports over the course of an RTT.  Slowstart is terminated
   as soon as any one of the receivers experiences its first packet
   loss.  Since that receiver's calculated rate will be lower than the
   current sending rate, the receiver will be selected as CLR.

   During slowstart, the upper bound on the rate increase of 8s/R_max
   bits/s every RTT does not apply.  Only after the TFMCC sender
   receives the first report with the have_loss flag set is the rate
   increase limited in this way.

   Slowstart may also be used after the sender has been idle for some
   time, to quickly reach the previous sending rate.  When the sender
   stops sending data packets, it records the current sending rate X' =
   X.  Every 10 RTTs, the allowed sending rate will be halved due to
   lack of receiver feedback, as specified in Section 3.3.  This halving
   may take place multiple times.  When the sender resumes, it may
   perform a slowstart from the current allowed rate up to the recorded
   rate X'.  Slowstart ends after the first packet loss by any of the
   receivers or as soon as X' is reached.

   To this end, receivers have to clear the have_loss flag after 10 RTTs
   without data packets as specified in Section 4.3.1.  The have_loss
   flag is only used during slowstart.  Therefore, clearing the flag has
   no effect if no packets arrived due to network partitioning or packet

3.7.  Scheduling of Packet Transmissions

   As TFMCC 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 coarse-grain or irregular scheduling of the operating
   system.  Thus, a typical sending loop will calculate the correct
   inter-packet interval, ts_ipi, as follows:

Widmer & Handley              Experimental                     [Page 15]

RFC 4654             TFMCC: Protocol Specification           August 2006

      ts_ipi = 8s/X

   When a sender first starts sending at time t_0, it calculates ts_ipi
   and calculates a nominal send time, t_1 = t_0 + ts_ipi, for packet 1.
   When the application becomes idle, it checks the current time,
   ts_now, and then requests re-scheduling after (ts_ipi - (ts_now -
   t_0)) seconds.  When the application is re-scheduled, it checks the
   current time, ts_now, again.  If (ts_now > t_1 - delta) then packet 1
   is sent (see below for delta).

   Now, a new ts_ipi may be calculated and used to calculate a nominal
   send time, t_2, for packet 2: t_2 = t_1 + ts_ipi.  The process then
   repeats with each successive packet's send time being calculated from
   the nominal send time of the previous packet.  Note that the actual
   send time ts_i, and not the nominal send time, is included as
   timestamp in the packet header.

   In some cases, when the nominal send time, t_i, of the next packet is
   calculated, it may already be the case that ts_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 TFMCC 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 ts_gran seconds, then delta would typically be set to:

      delta = min(ts_ipi/2, ts_gran/2)

   ts_gran is 10 milliseconds on many Unix systems.  If ts_gran is not
   known, a value of 10 milliseconds can be safely assumed.

4.  Data Receiver Protocol

   Receivers measure the current network conditions (namely, RTT and
   loss event rate) and use this information to calculate a rate that is
   fair to competing traffic.  The rate is then communicated to the
   sender in receiver reports.  Due to the potentially large number of
   receivers, it is undesirable that all receivers send reports,
   especially not at the same time.

   In the description of the receiver functionality, we will first
   address how the receivers measure the network parameters and then
   discuss the feedback process.

Widmer & Handley              Experimental                     [Page 16]

RFC 4654             TFMCC: Protocol Specification           August 2006

4.1.  Receiver Initialization

   The receiver is initialized when it receives the first data packet.
   The RTT is set to the maximum RTT value contained in the data packet.
   This initial value is used as the receiver's RTT until the first real
   RTT measurement is made.  The loss event rate is initialized to 0.
   Also, the flags receiver_leave, have_RTT, and have_loss are cleared.

4.2.  Receiver Leave

   A receiver that sends feedback but wishes to leave the TFMCC session
   within the next feedback round may indicate the pending leave by
   setting the receiver_leave flag in its report.  If the leaving
   receiver is the CLR, the receiver_leave flag should be set for all
   the reports within the feedback round before the leave takes effect.

4.3.  Measurement of the Network Conditions

   Receivers have to update their estimate of the network parameters
   with each new data packet they receive.

4.3.1.  Updating the Loss Event Rate

   When a data packet is received, the receiver adds the packet to the
   packet history.  It then recalculates the new value of the loss event
   rate p.  The loss event rate measurement mechanism is described
   separately in Section 5.

   When a loss event is detected, the flag have_loss is set.  In case no
   data packets are received for 10 consecutive RTTs, the flag is
   cleared to allow the sender to slowstart.  It is set again when new
   data packets arrive and a loss event is detected.

4.3.2.  Basic Round-Trip Time Measurement

   When a receiver gets a data packet that carries the receiver's own ID
   in the r field, the receiver updates its RTT estimate.

   1.  The current RTT is calculated as:

       R_sample = tr_now - tr_r'

       where tr_now is the time the data packet arrives at the receiver
       and tr_r' is the receiver report timestamp echoed in the data
       packet.  If the actual RTT is smaller than the resolution of the
       timestamps and tr_now equals tr_r', then R_sample is set to the
       smallest positive RTT value larger than 0 (i.e., 1 millisecond in
       our case).

Widmer & Handley              Experimental                     [Page 17]

RFC 4654             TFMCC: Protocol Specification           August 2006

   2.  The smoothed RTT estimate R is updated:

       If no feedback has been received before
           R = R_sample

           R = q*R + (1-q)*R_sample

       A filter parameter q of 0.5 is recommended for non-CLR receivers.
       The CLR performs RTT measurements much more frequently and hence
       should use a higher filter value.  We recommend using q=0.9.
       Note that TFMCC is not sensitive to the precise value for the
       filter constant.

   Optionally, sender-based RTT measurements may be used instead of
   receiver-based ones.  The sender already determines the RTT to a
   receiver from the receiver's echo of the sender's own timestamp for
   the calculation of the maximum RTT.  For sender-based RTT
   measurements, this RTT measurement needs to be communicated to the
   receiver.  Instead of including an echo of the receiver's timestamp,
   the sender includes the receiver's RTT in the next data packet, using
   the prioritization rules described in Section 3.5.

   To simplify sender operation, smoothing of RTT samples as described
   above should still be done at the receiver.

4.3.3.  One-Way Delay Adjustments

   When an RTT measurement is performed, the receiver also determines
   the one-way delay D_r from itself to the sender:

      D_r = tr_r' - ts_i

   where ts_i and tr_r' are the timestamp and receiver report timestamp
   echo contained in the data packet.  With each new data packet j, the
   receiver can now calculate an updated RTT estimate as:

      R' = max(D_r + tr_now - ts_j, 1 millisecond)

   In between RTT measurements, the updated R' is used instead of the
   smoothed RTT R.  Like the RTT samples, R' must be strictly positive.
   When a new measurement is made, all interim one-way delay
   measurements are discarded (i.e., the smoothed RTT is updated
   according to Section 4.3.2 without taking the interim one-way delay
   adjustments into account).

Widmer & Handley              Experimental                     [Page 18]

RFC 4654             TFMCC: Protocol Specification           August 2006

   For the one-way delay measurements, the clocks of sender and
   receivers need not be synchronized.  Clock skew will cancel itself
   out when both one-way measurements are added to form an RTT estimate,
   as long as clock drift between real RTT measurements is negligible.

   The same one-way delay adjustments should be applied to the RTT
   supplied by the sender when using sender-based RTT measurements.

4.3.4.  Receive Rate Measurements

   When a receiver has not experienced any loss events, it cannot
   calculate a TCP-friendly rate to include in the receiver reports.
   Instead, the receiver measures the current receive rate and sets the
   desired rate X_r to twice the receive rate.

   The receive rate in bits/s is measured as the number of bits received
   over the last k RTTs, taking into account the IP and transport packet
   headers, but excluding the link-layer packet headers.  A value for k
   between 2 and 4 is recommended.

4.4.  Setting the Desired Rate

   When a receiver measures a non-zero loss event rate, it calculates
   the desired rate using Equation (1).  In case no RTT measurement is
   available yet, the maximum RTT is used instead of the receiver's RTT.
   The desired rate X_r is updated whenever the loss event rate or the
   RTT changes.

   A receiver may decide not to report desired rates that are below 1
   packet per 8 seconds, since a sender is very slow to recover from
   such low sending rates.  In this case, the receiver reports a desired
   rate of 1 packet per 8 seconds.  However, it must leave the multicast
   group if for more than 120 seconds, the calculated rate falls below
   the reported rate and the current sending rate is higher than the
   receiver's calculated rate.

   As mentioned above, calculation of the desired rate is not possible
   before the receiver experiences the first loss event.  In that case,
   twice the rate at which data is received is included in the receiver
   reports as X_r to allow the sender to slowstart as described in
   Section 3.6.  This is also done when the sender resumes sending data
   packets after the have_loss flag was cleared due to the sender being

Widmer & Handley              Experimental                     [Page 19]

RFC 4654             TFMCC: Protocol Specification           August 2006

4.5.  Feedback and Feedback Suppression

   Let fb_nr be the highest feedback round counter value received by a
   receiver.  When a new data packet arrives with a higher feedback
   round counter than fb_nr, a new feedback round begins and fb_nr is
   updated.  Outstanding feedback for the old round is canceled.  In
   case a feedback number with a value that is more than half the
   feedback number space lower than fb_nr is received, the receiver
   assumes that the feedback round counter wrapped and also cancels the
   feedback timer and updates fb_nr.

   The CLR sends its feedback independently from all the other receivers
   once per RTT.  Its feedback does not suppress other feedback and
   cannot be suppressed by other receiver's feedback.

   Non-CLR receivers set a feedback timer at the beginning of a feedback
   round.  Using an exponentially weighted random timer mechanism, the
   feedback timer is set to expire after

      t = max(T * (1 + log(x)/log(N)), 0)


      x is a random variable uniformly distributed in (0,1],

      T is the duration of a feedback round (i.e., 6 * R_max),

      N is an estimated upper bound on the number of receivers.

   N is a constant specific to the TFMCC protocol.  Since TFMCC scales
   to up to thousands of receivers, setting N to 10,000 for all
   receivers (and limiting the TFMCC session to at most 10,000
   receivers) is recommended.

   A feedback packet is sent when the feedback timer expires, unless the
   timer is canceled beforehand.  When the multicast model is ASM,
   feedback is multicast to the whole group; otherwise, the feedback is
   unicast to the sender.  The feedback packet includes the calculated
   rate valid at the time the feedback packet is sent (not the rate at
   the point of time when the feedback timer is set).  The copy of the
   timestamp ts_i of the last data packet received, which is included in
   the feedback packet, needs to be adjusted by the time interval
   between receiving the data packet and sending the report to allow the
   sender to correctly infer the instantaneous RTT (i.e., that time
   interval has to be added to the timestamp value).

Widmer & Handley              Experimental                     [Page 20]

RFC 4654             TFMCC: Protocol Specification           August 2006

   The timer is canceled if a data packet is received that has a lower
   suppression rate than the receiver's calculated rate and a higher or
   equal maximum RTT than the receiver's RTT.  Likewise, a data packet
   indicating the beginning of a new feedback round cancels all feedback
   for older rounds.  In case of ASM, the timer is also canceled if a
   feedback packet is received from another non-CLR receiver reporting a
   lower rate.

   The feedback suppression process is complicated by the fact that the
   calculated rates of the receivers will change during a feedback
   round.  If the calculated rates decrease rapidly for all receivers,
   feedback suppression can no longer prevent a feedback implosion,
   since earlier feedback will always report a higher rate than current
   feedback.  To make the feedback suppression mechanism robust in the
   face of changing rates, it is necessary to introduce X_fbr, the
   calculated rate of a receiver at the beginning of a feedback round.
   A receiver needs to suppress its feedback not only when the
   suppression rate is less than the receiver's current calculated rate
   but also in the case that the suppression rate falls below X_fbr.

   When the maximum RTT changes significantly during one feedback round,
   it is necessary to reschedule the feedback timer in proportion to the

      t = t * R_max / R_max'

   where R_max is the new maximum RTT and R_max' is the previous maximum
   RTT.  The same considerations hold when the last data packets were
   received more than a time interval of R_max ago.  In this case, it is
   necessary to add the difference of the inter-packet gap and the
   maximum RTT to the feedback time to prevent a feedback implosion
   (e.g., in case the sender crashed).

      t = t + max(tr_now - tr_i - R_max, 0)

   where tr_i is the time when the last data packet arrived at the

   More details on the characteristics of the feedback suppression
   mechanism can be found in [13] and [3].

Widmer & Handley              Experimental                     [Page 21]

RFC 4654             TFMCC: Protocol Specification           August 2006

5.  Calculation of the Loss Event Rate

   Obtaining an accurate and stable measurement of the loss event rate
   is of primary importance for TFMCC.  Loss rate measurement is
   performed at the receiver, based on the detection of lost or marked
   packets from the sequence numbers of arriving packets.

5.1.  Detection of Lost or Marked Packets

   TFMCC 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

   The receivers each maintain 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 timestamp when each
   packet was received.  In practice, this data structure will normally
   be stored in a more compact representation, but this is

   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
   it is to make TFMCC more robust in the presence of reordering.  In
   contrast to TCP, if a packet arrives late (after 3 subsequent packets
   arrived) at a receiver, the late packet can fill the hole in the
   reception record, and the receiver can recalculate the loss event
   rate.  Future versions of TFMCC 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.

Widmer & Handley              Experimental                     [Page 22]

RFC 4654             TFMCC: Protocol Specification           August 2006

5.2.  Translation from Loss History to Loss Events

   TFMCC requires that the loss event rate 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
   receivers need 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 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 be either before or after T_after due to

   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.

      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,
   the sequence numbers must be modified to take this into account
   before the calculation is performed.  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.

Widmer & Handley              Experimental                     [Page 23]

RFC 4654             TFMCC: Protocol Specification           August 2006

   If the lost packet S_old was determined to have started the previous
   loss event, and if 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 of 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;
           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 TFMCC's speed in responding to changes in
   the level of congestion.  As currently specified, TFMCC 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 TFMCC'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

Widmer & Handley              Experimental                     [Page 24]

RFC 4654             TFMCC: Protocol Specification           August 2006

   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 4/(3*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 that
   allows the TFMCC receivers 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, 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 = 0 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.

Widmer & Handley              Experimental                     [Page 25]

RFC 4654             TFMCC: Protocol Specification           August 2006

   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, a 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_(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;

Widmer & Handley              Experimental                     [Page 26]

RFC 4654             TFMCC: Protocol Specification           August 2006

   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 TFMCC to respond more quickly to the sudden
   absence of congestion, as represented by a long current loss

5.6.  Initializing the Loss History after the First Loss Event

   The number of packets received before the first loss event usually
   does not reflect the current loss event rate.  When the first loss
   event occurs, a TFMCC receiver assumes that the correct data rate is
   the rate at which data was received during the last RTT when the loss
   occurred.  Instead of initializing the first loss interval to the
   number of packets sent until the first loss event, the TFMCC receiver
   calculates the loss interval that would be required to produce the
   receive rate X_recv, and it uses this synthetic loss interval l_0 to
   seed the loss history mechanism.

   The initial loss interval is calculated by inverting a simplified
   version of the TCP Equation (1).

Widmer & Handley              Experimental                     [Page 27]

RFC 4654             TFMCC: Protocol Specification           August 2006

      X_recv = sqrt(3/2) * -----------------
                            R * sqrt(1/l_0)

                    X_recv * R
      ==> l_0 = (----------------)^2
                  sqrt(3/2) * 8s

   The resulting initial loss interval is too small at higher loss rates
   compared to using the more accurate Equation (1), which leads to a
   more conservative initial loss event rate.

   If a receiver still uses the initial RTT R_max instead of its real
   RTT, the initial loss interval is too large in case the initial RTT
   is higher than the actual RTT.  As a consequence, the receiver will
   calculate too high a desired rate when the first RTT measurement R is
   made and the initial loss interval is still in the loss history.  The
   receiver has to adjust l_0 as follows:

      l_0 = l_0 * (R/R_max)^2

   No action needs to be taken when the first RTT measurement is made
   after the initial loss interval left the loss history.

6.  Security Considerations

   TFMCC 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 TFMCC should take care to ensure
   that feedback is only accepted from valid receivers of the data.
   However, the precise mechanism to achieve this will depend on the
   transport protocol itself.

   Congestion control mechanisms may potentially be manipulated by a
   greedy receiver that wishes to receive more than its fair share of
   network bandwidth.  However, in TFMCC a receiver can only influence
   the sending rate if it is the CLR and thus has the lowest calculated
   rate of all receivers.  If the calculated rate is then manipulated
   such that it exceeds the calculated rate of the second to lowest
   receiver, it will cease to be CLR.  A greedy receiver can only
   significantly increase the transmission rate if it is the only
   participant in the session.  If such scenarios are of concern,

Widmer & Handley              Experimental                     [Page 28]

RFC 4654             TFMCC: Protocol Specification           August 2006

   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.

   It is possible that a receiver sends feedback claiming that it has a
   very low calculated rate.  This will reduce the rate of the multicast
   session and might render it useless but obviously cannot hurt the
   network itself.

   We expect that protocols incorporating ECN with TFMCC will also want
   to incorporate feedback from the receiver to the sender using the ECN
   nonce [12].  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.

7.  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 particularly like to thank Brian Adamson, Mark Pullen, Fei
   Zhao, and Magnus Westerlund for feedback on earlier versions of this

8.  References

8.1.  Normative References

   [1]   Whetten, B., Vicisano, L., Kermode, R., Handley, M., Floyd, S.,
         and M. Luby, "Reliable Multicast Transport Building Blocks for
         One-to-Many Bulk-Data Transfer", RFC 3048, January 2001.

   [2]   Kermode, R. and L. Vicisano, "Author Guidelines for Reliable
         Multicast Transport (RMT) Building Blocks and Protocol
         Instantiation documents", RFC 3269, April 2002.

8.2.  Informative References

   [3]   J. Widmer and M. Handley, "Extending Equation-Based Congestion
         Control to Multicast Applications", Proc ACM Sigcomm 2001, San
         Diego, August 2001.

Widmer & Handley              Experimental                     [Page 29]

RFC 4654             TFMCC: Protocol Specification           August 2006

   [4]   S. Floyd, M. Handley, J. Padhye, and J. Widmer, "Equation-Based
         Congestion Control for Unicast Applications", Proc ACM SIGCOMM
         2000, Stockholm, August 2000.

   [5]   Adamson, B., Bormann, C., Handley, M., and J. Macker,
         "Negative-Acknowledgment (NACK)-Oriented Reliable Multicast
         (NORM) Building Blocks", RFC 3941, November 2004.

   [6]   Deering, S., "Host extensions for IP multicasting", STD 5, RFC
         1112, August 1989.

   [7]   H. W. Holbrook, "A Channel Model for Multicast," Ph.D.
         Dissertation, Stanford University, Department of Computer
         Science, Stanford, California, August 2001.

   [8]   J. Padhye, V. Firoiu, D. Towsley, and J. Kurose, "Modeling TCP
         Throughput: A Simple Model and its Empirical Validation", Proc
         ACM SIGCOMM 1998.

   [9]   Ramakrishnan, K., Floyd, S., and D. Black, "The Addition of
         Explicit Congestion Notification (ECN) to IP", RFC 3168,
         September 2001.

   [10]  L. Rizzo, "pgmcc: a TCP-friendly single-rate multicast
         congestion control scheme", Proc ACM Sigcomm 2000, Stockholm,
         August 2000.

   [11]  Schulzrinne, H., Casner, S., Frederick, R., and V. Jacobson,
         "RTP: A Transport Protocol for Real-Time Applications", STD 64,
         RFC 3550, July 2003.

   [12]  Spring, N., Wetherall, D., and D. Ely, "Robust Explicit
         Congestion Notification (ECN) Signaling with Nonces", RFC 3540,
         June 2003.

   [13]  J. Widmer and T. Fuhrmann, "Extremum Feedback for Very Large
         Multicast Groups", Proc NGC 2001, London, November 2001.

Widmer & Handley              Experimental                     [Page 30]

RFC 4654             TFMCC: Protocol Specification           August 2006

Authors' Addresses

   Joerg Widmer
   DoCoMo Euro-Labs
   Landsberger Str. 312, Munich, Germany
   EMail: widmer@acm.org

   Mark Handley
   UCL (University College London)
   Gower Street, London WC1E 6BT, UK
   EMail: m.handley@cs.ucl.ac.uk

Widmer & Handley              Experimental                     [Page 31]

RFC 4654             TFMCC: Protocol Specification           August 2006

Full Copyright Statement

   Copyright (C) The Internet Society (2006).

   This document is subject to the rights, licenses and restrictions
   contained in BCP 78, and except as set forth therein, the authors
   retain all their rights.

   This document and the information contained herein are provided on an

Intellectual Property

   The IETF takes no position regarding the validity or scope of any
   Intellectual Property Rights or other rights that might be claimed to
   pertain to the implementation or use of the technology described in
   this document or the extent to which any license under such rights
   might or might not be available; nor does it represent that it has
   made any independent effort to identify any such rights.  Information
   on the procedures with respect to rights in RFC documents can be
   found in BCP 78 and BCP 79.

   Copies of IPR disclosures made to the IETF Secretariat and any
   assurances of licenses to be made available, or the result of an
   attempt made to obtain a general license or permission for the use of
   such proprietary rights by implementers or users of this
   specification can be obtained from the IETF on-line IPR repository at

   The IETF invites any interested party to bring to its attention any
   copyrights, patents or patent applications, or other proprietary
   rights that may cover technology that may be required to implement
   this standard.  Please address the information to the IETF at


   Funding for the RFC Editor function is provided by the IETF
   Administrative Support Activity (IASA).

Widmer & Handley              Experimental                     [Page 32]