Network Working Group                                         B. Adamson
Request for Comments: 3941                                           NRL
Category: Experimental                                        C. Bormann
                                                 Universitaet Bremen TZI
                                                              M. Handley
                                                                     UCL
                                                               J. Macker
                                                                     NRL
                                                           November 2004


   Negative-Acknowledgment (NACK)-Oriented Reliable Multicast (NORM)
                           Building Blocks

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 (2004).

Abstract

   This document discusses the creation of negative-acknowledgment
   (NACK)-oriented reliable multicast (NORM) protocols.  The rationale
   for NORM goals and assumptions are presented.  Technical challenges
   for NACK-oriented (and in some cases general) reliable multicast
   protocol operation are identified.  These goals and challenges are
   resolved into a set of functional "building blocks" that address
   different aspects of NORM protocol operation.  It is anticipated that
   these building blocks will be useful in generating different
   instantiations of reliable multicast protocols.















Adamson, et al.               Experimental                      [Page 1]


RFC 3941                  NORM Building Blocks             November 2004


Table of Contents

   1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . .   3
   2. Rationale. . . . . . . . . . . . . . . . . . . . . . . . . . .   4
      2.1. Delivery Service Model  . . . . . . . . . . . . . . . . .   4
      2.2. Group Membership Dynamics . . . . . . . . . . . . . . . .   5
      2.3. Sender/Receiver Relationships . . . . . . . . . . . . . .   5
      2.4. Group Size Scalability. . . . . . . . . . . . . . . . . .   6
      2.5. Data Delivery Performance . . . . . . . . . . . . . . . .   6
      2.6. Network Environments. . . . . . . . . . . . . . . . . . .   6
      2.7. Router/Intermediate System Assistance . . . . . . . . . .   7
   3. Functionality. . . . . . . . . . . . . . . . . . . . . . . . .   7
      3.1. NORM Sender Transmission. . . . . . . . . . . . . . . . .  10
      3.2. NORM Repair Process . . . . . . . . . . . . . . . . . . .  11
           3.2.1. Receiver NACK Process Initiation . . . . . . . . .  11
           3.2.2. NACK Suppression . . . . . . . . . . . . . . . . .  13
           3.2.3. NACK Content . . . . . . . . . . . . . . . . . . .  17
                  3.2.3.1. NACK and FEC Repair Strategies. . . . . .  17
                  3.2.3.2. NACK Content Format . . . . . . . . . . .  20
           3.2.4. Sender Repair Response . . . . . . . . . . . . . .  21
      3.3. NORM Receiver Join Policies and Procedures. . . . . . . .  23
      3.4. Reliable Multicast Member Identification. . . . . . . . .  24
      3.5. Data Content Identification . . . . . . . . . . . . . . .  24
      3.6. Forward Error Correction (FEC). . . . . . . . . . . . . .  26
      3.7. Round-trip Timing Collection. . . . . . . . . . . . . . .  27
           3.7.1. One-to-Many Sender GRTT Measurement. . . . . . . .  27
           3.7.2. One-to-Many Receiver RTT Measurement . . . . . . .  29
           3.7.3. Many-to-Many RTT Measurement . . . . . . . . . . .  29
           3.7.4. Sender GRTT Advertisement. . . . . . . . . . . . .  30
      3.8. Group Size Determination/Estimation . . . . . . . . . . .  31
      3.9. Congestion Control Operation. . . . . . . . . . . . . . .  31
      3.10 Router/Intermediate System Assistance . . . . . . . . . .  31
      3.11 NORM Applicability. . . . . . . . . . . . . . . . . . . .  31
   4. Security Considerations. . . . . . . . . . . . . . . . . . . .  32
   5. Acknowledgements . . . . . . . . . . . . . . . . . . . . . . .  33
   6. References . . . . . . . . . . . . . . . . . . . . . . . . . .  33
      6.1. Normative References. . . . . . . . . . . . . . . . . . .  33
      6.2. Informative References. . . . . . . . . . . . . . . . . .  33
   7. Authors' Addresses . . . . . . . . . . . . . . . . . . . . . .  35
      Full Copyright Statement . . . . . . . . . . . . . . . . . . .  36











Adamson, et al.               Experimental                      [Page 2]


RFC 3941                  NORM Building Blocks             November 2004


1.  Introduction

   Reliable multicast transport is a desirable technology for the
   efficient and reliable distribution of data to a group on the
   Internet.  The complexities of group communication paradigms
   necessitate different protocol types and instantiations to meet the
   range of performance and scalability requirements of different
   potential reliable multicast applications and users [3].  This
   document addresses the creation of negative-acknowledgment (NACK)-
   oriented reliable multicast (NORM) protocols.  While different
   protocol instantiations may be required to meet specific application
   and network architecture demands [4], there are a number of
   fundamental components that may be common to these different
   instantiations.  This document describes the framework and common
   "building block" components relevant to multicast protocols based
   primarily on NACK operation for reliable transport.  While this
   document discusses a large set of reliable multicast components and
   issues relevant to NORM protocol design, it specifically addresses in
   detail the following building blocks which are not addressed in other
   IETF documents:

      1) NORM sender transmission strategies,

      2) NACK-oriented repair process with timer-based feedback
         suppression, and

      3) Round-trip timing for adapting NORM timers.

   The potential relationships to other reliable multicast transport
   building blocks (Forward Error Correction (FEC), congestion control)
   and general issues with NORM protocols are also discussed.  This
   document is a product of the IETF RMT WG and follows the guidelines
   provided in RFC 3269 [5].  The key words "MUST", "MUST NOT",
   "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT",
   "RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be
   interpreted as described in BCP 14, RFC 2119 [1].

Statement of Intent

   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.

   While waiting for such a scheme to be available, or for an existing
   scheme to be proven adequate, the Reliable Multicast Transport
   working group (RMT) publishes this Request for Comments in the
   "Experimental" category.



Adamson, et al.               Experimental                      [Page 3]


RFC 3941                  NORM Building Blocks             November 2004


   It is the intent of RMT to re-submit this specification as an IETF
   Proposed Standard as soon as the above condition is met.

2.  Rationale

   Each potential protocol instantiation using the building blocks
   presented here (and in other applicable building block documents)
   will have specific criteria that may influence individual protocol
   design.  To support the development of applicable building blocks, it
   is useful to identify and summarize driving general protocol design
   goals and assumptions.  These are areas that each protocol
   instantiation will need to address in detail.  Each building block
   description in this document will include a discussion of the impact
   of these design criteria.  The categories of design criteria
   considered here include:

      1) Delivery Service Model,
      2) Group Membership Dynamics,
      3) Sender/receiver relationships,
      4) Group Size Scalability,
      5) Data Delivery Performance,
      6) Network Environments, and
      7) Router/Intermediate System Interactions.

   All of these areas are at least briefly discussed.  Additionally,
   other reliable multicast transport building block documents such as
   [9] have been created to address areas outside of the scope of this
   document.  NORM protocol instantiations may depend upon these other
   building blocks as well as the ones presented here.  This document
   focuses on areas that are unique to NORM but may be used in concert
   with the other building block areas.  In some cases, a building block
   may be able address a wide range of assumptions, while in other cases
   there will be trade-offs required to meet different application needs
   or operating  environments.  Where necessary, building block features
   are designed to be parametric to meet different requirements.  Of
   course, an underlying goal will be to minimize design complexity and
   to at least recommend default values for any such parameters that
   meet a general purpose "bulk data transfer" requirement in a typical
   Internet environment.

2.1.  Delivery Service Model

   The implicit goal of a reliable multicast transport protocol is the
   reliable delivery of data among a group of members communicating
   using IP multicast datagram service.  However, the specific service
   the application is attempting to provide can impact design decisions.
   A most basic service model for reliable multicast transport is that
   of "bulk transfer" which is a primary focus of this and other related



Adamson, et al.               Experimental                      [Page 4]


RFC 3941                  NORM Building Blocks             November 2004


   RMT working group documents.  However, the same principles in
   protocol design may also be applied to other services models, e.g.,
   more interactive exchanges of small messages such as with white-
   boarding or text chat.  Within these different models there are
   issues such as the sender's ability to cache transmitted data (or
   state referencing it) for retransmission or repair.  The needs for
   ordering and/or causality in the sequence of transmissions and
   receptions among members in the group may be different depending upon
   data content.  The group communication paradigm differs significantly
   from the point-to-point model in that, depending upon the data
   content type, some receivers may complete reception of a portion of
   data content and be able to act upon it before other members have
   received the content.  This may be acceptable (or even desirable) for
   some applications but not for others.  These varying requirements
   drive the need for a number of different protocol instantiation
   designs.  A significant challenge in developing generally useful
   building block mechanisms is accommodating even a limited range of
   these capabilities without defining specific application-level
   details.

2.2.  Group Membership Dynamics

   One area where group communication can differ from point-to-point
   communications is that even if the composition of the group changes,
   the "thread" of communication can still exist.  This contrasts with
   the point-to-point communication model where, if either of the two
   parties leave, the communication process (exchange of data) is
   terminated (or at least paused).  Depending upon application goals,
   senders and receivers participating in a reliable multicast transport
   "session" may be able to join late, leave, and/or potentially rejoin
   while the ongoing group communication "thread" still remains
   functional and useful.  Also note that this can impact protocol
   message content.  If "late joiners" are supported, some amount of
   additional information may be placed in message headers to
   accommodate this functionality.  Alternatively, the information may
   be sent in its own message (on demand or intermittently) if the
   impact to the overhead of typical message transmissions is deemed too
   great.  Group dynamics can also impact other protocol mechanisms such
   as NACK timing, congestion control operation, etc.

2.3.  Sender/Receiver Relationships

   The relationship of senders and receivers among group members
   requires consideration.  In some applications, there may be a single
   sender multicasting to a group of receivers.  In other cases, there
   may be more than one sender or the potential for everyone in the
   group to be a sender _and_ receiver of data may exist.




Adamson, et al.               Experimental                      [Page 5]


RFC 3941                  NORM Building Blocks             November 2004


2.4.  Group Size Scalability

   Native IP multicast [2] may scale to extremely large group sizes.  It
   may be desirable for some applications to scale along with the
   multicast infrastructure's ability to scale.  In its simplest form,
   there are limits to the group size to which a NACK-oriented protocol
   can apply without NACK implosion problems.  Research suggests that
   NORM group sizes on the order of tens of thousands of receivers may
   operate with modest feedback to the sender using probabilistic,
   timer-based suppression techniques [7].  However, the potential for
   router assistance and/or other NACK suppression heuristics may enable
   these protocols to scale to very large group sizes.  In large scale
   cases, it may be prohibitive for members to maintain state on all
   other members (in particular, other receivers) in the group.  The
   impact of group size needs to be considered in the development of
   applicable building blocks.

2.5.  Data Delivery Performance

   There is a trade-off between scalability and data delivery latency
   when designing NACK-oriented protocols.  If probabilistic, timer-
   based NACK suppression is to be used, there will be some delays built
   into the NACK process to allow suppression to occur and for the
   sender of data to identify appropriate content for efficient repair
   transmission.  For example, backoff timeouts can be used to ensure
   efficient NACK suppression and repair transmission, but this comes at
   a cost of increased delivery latency and increased buffering
   requirements for both senders and receivers.  The building blocks
   SHOULD allow applications to establish bounds for data delivery
   performance.  Note that application designers must be aware of the
   scalability trade-off that is made when such bounds are applied.

2.6.  Network Environments

   The Internet Protocol has historically assumed a role of providing
   service across heterogeneous network topologies.  It is desirable
   that a reliable multicast protocol be capable of effectively
   operating across a wide range of the networks to which general
   purpose IP service applies.  The bandwidth available on the links
   between the members of a single group today may vary between low
   numbers of kbit/s for wireless links and multiple Gbit/s for high
   speed LAN connections, with varying degrees of contention from other
   flows.  Recently, a number of asymmetric network services including
   56K/ADSL modems, CATV Internet service, satellite and other wireless
   communication services have begun to proliferate.  Many of these are
   inherently broadcast media with potentially large "fan-out" to which
   IP multicast service is highly applicable.  Additionally, policy
   and/or technical issues may result in topologies where multicast



Adamson, et al.               Experimental                      [Page 6]


RFC 3941                  NORM Building Blocks             November 2004


   connectivity is limited to a single source multicast (SSM) model from
   a specific source [8].  Receivers in the group may be restricted to
   unicast feedback for NACKs and other messages.  Consideration must be
   given, in building block development and protocol design, to the
   nature of the underlying networks.

2.7.  Router/Intermediate System Assistance

   While intermediate assistance from devices/systems with direct
   knowledge of the underlying network topology may be used to leverage
   the performance and scalability of reliable multicast protocols,
   there will continue to be a number of instances where this is not
   available or practical.  Any building block components for NACK-
   oriented reliable multicast SHALL be capable of operating without
   such assistance.  However, it is RECOMMENDED that such protocols also
   consider utilizing these features when available.

3.  Functionality

   The previous section has presented the role of protocol building
   blocks and some of the criteria that may affect NORM building block
   identification/design.  This section describes different building
   block areas applicable to NORM protocols.  Some of these areas are
   specific to NACK-oriented protocols.  Detailed descriptions of such
   areas are provided.  In other cases, the areas (e.g., node
   identifiers, forward error correction (FEC), etc.) may be applicable
   to other forms of reliable multicast.  In those cases, the discussion
   below describes requirements placed on those other general building
   block areas from the standpoint of NACK-oriented reliable multicast.
   Where applicable, other building block documents are referenced for
   possible contribution to NORM protocols.

   For each building block, a notional "interface description" is
   provided to illustrate any dependencies of one building block
   component upon another or upon other protocol parameters.  A building
   block component may require some form of "input" from another
   building block component or other source to perform its function.
   Any "inputs" required by a building block component and/or any
   resultant "output" provided will be defined and described in each
   building block component's interface description.  Note that the set
   of building blocks presented here do not fully satisfy each other's
   "input" and "output" needs.  In some cases, "inputs" for the building
   blocks here must come from other building blocks external to this
   document (e.g., congestion control or FEC).  In other cases NORM
   building block "inputs" must be satisfied by the specific protocol
   instantiation or implementation (e.g., application data and control).





Adamson, et al.               Experimental                      [Page 7]


RFC 3941                  NORM Building Blocks             November 2004


   The following building block components relevant to NORM are
   identified:

   (NORM-Specific)
        1)   NORM Sender Transmission
        2)   NORM Repair Process
        3)   NORM Receiver Join Policies
   (General Purpose)
        4)   Node (member) Identification
        5)   Data Content Identification
        6)   Forward Error Correction (FEC)
        7)   Round-trip Timing Collection
        8)   Group Size Determination/Estimation
        9)   Congestion Control Operation
        10)  Router/Intermediate System Assistance
        11)  Ancillary Protocol Mechanisms

   Figure 1 provides a pictorial overview of these building block areas
   and some of their relationships.  For example, the content of the
   data messages that a sender initially transmits depends upon the
   "Node Identification", "Data Content Identification", and "FEC"
   components while the rate of message transmission will generally
   depend upon the "Congestion Control" component.  Subsequently, the
   receivers' response to these transmissions (e.g., NACKing for repair)
   will depend upon the data message content and inputs from other
   building block components.  Finally, the sender's processing of
   receiver responses will feed back into its transmission strategy.
























Adamson, et al.               Experimental                      [Page 8]


RFC 3941                  NORM Building Blocks             November 2004


                                     Application Data and Control
                                                 |
                                                 v
    .---------------------.            .-----------------------.
    | Node Identification |----------->|  Sender Transmission  |<------.
    `---------------------'       _.-' `-----------------------'       |
    .---------------------.   _.-' .'            | .--------------.    |
    | Data Identification |--'   .''             | |  Join Policy |    |
    `---------------------'    .' '              v `--------------'    |
    .---------------------.  .'  '     .------------------------.      |
 .->| Congestion Control  |-'   '      | Receiver NACK          |      |
 |  `---------------------'   .'       | Repair Process         |      |
 |  .---------------------. .'         | .------------------.   |      |
 |  |        FEC          |'.          | | NACK Initiation  |   |      |
 |  `---------------------'` `._       | `------------------'   |      |
 |  .---------------------. ``. `-._   | .------------------.   |      |
 `--|    RTT Collection   |._` `    `->| | NACK Content     |   |      |
    `---------------------' .`- `      | `------------------'   |      |
    .---------------------.  \ `-`._   | .------------------.   |      |
    |    Group Size Est.  |---.-`---`->| | NACK Suppression |   |      |
    `---------------------'`.  ` `     | `------------------'   |      |
    .---------------------.  `  ` `    `------------------------'      |
    |       Other         |   `  ` `             | .-----------------. |
    `---------------------'    `  ` `            | |Router Assistance| |
                                `. ` `           v `-----------------' |
                                  `.`' .-------------------------.     |
                                     `>| Sender NACK Processing  |_____/
                                       | and Repair Response     |
                                       `-------------------------'

                    ^                         ^
                    |                         |
                  .-----------------------------.
                  |         (Security)          |
                  `-----------------------------'

                Fig. 1 - NORM Building Block Framework

   The components on the left side of this figure are areas that may be
   applicable beyond NORM.  The most significant of these components are
   discussed in other building block documents such as [9].  A brief
   description of these areas and their role in the NORM protocol is
   given below.  The components on the right are seen as specific to
   NORM protocols, most notably the NACK repair process.  These areas
   are discussed in detail below.  Some other components (e.g.,
   "Security") impact many aspects of the protocol, and others such as
   "Router Assistance" may be more transparent to the core protocol
   processing.  The sections below describe the "NORM Sender



Adamson, et al.               Experimental                      [Page 9]


RFC 3941                  NORM Building Blocks             November 2004


   Transmission", "NORM Repair Process", and "RTT Collection" building
   blocks in detail.  The relationships to and among the other building
   block areas are also discussed, focusing on issues applicable to NORM
   protocol design.  Where applicable, specific technical
   recommendations are made for mechanisms that will properly satisfy
   the goals of NORM transport for the Internet.

3.1.  NORM Sender Transmission

   NORM senders will transmit data content to the multicast session.
   The data content will be application dependent.  The sender will
   transmit data content at a rate, and with message sizes, determined
   by application and/or network architecture requirements.  Any FEC
   encoding of sender transmissions SHOULD conform with the guidelines
   of [9].  When congestion control mechanisms are needed (REQUIRED for
   general Internet operation), NORM transmission SHALL be controlled by
   the congestion control mechanism.  In any case, it is RECOMMENDED
   that all data transmissions from  NORM senders be subject to rate
   limitations determined by the application or congestion control
   algorithm.  The sender's transmissions SHOULD make good utilization
   of the available capacity (which may be limited by the application
   and/or by congestion control).  As a result, it is expected there
   will be overlap and multiplexing of new data content transmission
   with repair content.  Other factors related to application operation
   may determine sender transmission formats and methods.  For example,
   some consideration needs to be given to the sender's behavior during
   intermittent idle periods when it has no data to transmit.

   In addition to data content, other sender messages or commands may be
   employed as part of protocol operation.  These messages may occur
   outside of the scope of application data transfer.  In NORM
   protocols, reliability of such protocol messages may be attempted by
   redundant transmission when positive acknowledgement is prohibitive
   due to group size scalability concerns.  Note that protocol design
   SHOULD provide mechanisms for dealing with cases where such messages
   are not received by the group.  As an example, a command message
   might be redundantly transmitted by a sender to indicate that it is
   temporarily (or permanently) halting transmission.  At this time, it
   may be appropriate for receivers to respond with NACKs for any
   outstanding repairs they require following the rules of the NORM NACK
   procedure.  For efficiency, the sender should allow sufficient time
   between the redundant transmissions to receive any NACK-oriented
   responses from the receivers to this command.

   In general, when there is any resultant NACK or other feedback
   operation, the timing of redundant transmission of control messages
   issued by a sender and other NORM protocol timeouts should be
   dependent upon the group greatest round trip timing (GRTT) estimate



Adamson, et al.               Experimental                     [Page 10]


RFC 3941                  NORM Building Blocks             November 2004


   and any expected resultant NACK or other feedback operation.  The
   NORM GRTT is an estimate of the worst-case round-trip timing from a
   sender to any receivers in the group.  It is assumed that the GRTT
   interval is a conservative estimate of the maximum span (with respect
   to delay) of the multicast group across a network topology with
   respect to given sender.  NORM instantiations SHOULD be able to
   dynamically adapt to a wide range of multicast network topologies.

   Sender Transmission Interface Description

   Inputs:

      1) Application data and control
      2) Sender node identifier
      3) Data identifiers
      4) Segmentation and FEC parameters
      5) Transmission rate
      6) Application controls
      7) Receiver feedback messages (e.g., NACKs)

   Outputs:

      1) Controlled transmission of messages with headers uniquely
         identifying data or repair content within the context of the
         NORM session.
      2) Commands indicating sender's status or other transport
         control actions to be taken.

3.2.  NORM Repair Process

   A critical component of NORM protocols is the NACK repair process.
   This includes the receiver's role in detecting and requesting repair
   needs, and the sender's response to such requests.  There are four
   primary elements of the NORM repair process:

      1) Receiver NACK process initiation,

      3) NACK suppression,

      2) NACK message content,

      4) Sender NACK processing and response.

3.2.1.  Receiver NACK Process Initiation

   The NORM NACK process (cycle) will be initiated by receivers that
   detect a need for repair transmissions from a specific sender to
   achieve reliable reception.  When FEC is applied, a receiver should



Adamson, et al.               Experimental                     [Page 11]


RFC 3941                  NORM Building Blocks             November 2004


   initiate the NACK process only when it is known its repair
   requirements exceed the amount of pending FEC transmission for a
   given coding block of data content.  This can be determined at the
   end of the current transmission block (if it is indicated) or upon
   the start of reception of a subsequent coding block or transmission
   object.  This implies the NORM data content is marked to identify its
   FEC block number and that ordinal relationship is preserved in order
   of transmission.

   Alternatively, if the sender's transmission advertises the quantity
   of repair packets it is already planning to send for a block, the
   receiver may be able to initiate the NACK processor earlier.
   Allowing receivers to initiate NACK cycles at any time they detect
   their repair needs have exceeded pending repair transmissions may
   result in slightly quicker repair cycles.  However, it may be useful
   to limit NACK process initiation to specific events such as at the
   end-of-transmission of an FEC coding block or upon detection of
   subsequent coding blocks.  This can allow receivers to aggregate NACK
   content into a smaller number of NACK messages and provide some
   implicit loose synchronization among the receiver set to help
   facilitate effective probabilistic suppression of NACK feedback.  The
   receiver MUST maintain a history of data content received from the
   sender to determine its current repair needs.  When FEC is employed,
   it is expected that the history will correspond to a record of
   pending or partially-received coding blocks.

   For probabilistic, timer-base suppression of feedback, the NACK cycle
   should begin with receivers observing backoff timeouts.  In
   conjunction with initiating this backoff timeout, it is important
   that the receivers record the current position in the sender's
   transmission sequence at which they initiate the NACK cycle.  When
   the suppression backoff timeout expires, the receivers should only
   consider their repair needs up to this recorded transmission position
   in making the decision to transmit or suppress a NACK.  Without this
   restriction, suppression is greatly reduced as additional content is
   received from the sender during the time a NACK message propagates
   across the network to the sender and other receivers.

   Receiver NACK Process Initiation Interface Description

   Inputs:

      1) Sender data content with sequencing identifiers from sender
         transmissions.
      2) History of content received from sender.






Adamson, et al.               Experimental                     [Page 12]


RFC 3941                  NORM Building Blocks             November 2004


   Outputs:

      1) NACK process initiation decision
      2) Recorded sender transmission sequence position.

3.2.2.  NACK Suppression

   An effective NORM feedback suppression mechanism is the use of random
   backoff timeouts prior to NACK transmission by receivers requiring
   repairs [10].  Upon expiration of the backoff timeout, a receiver
   will request repairs unless its pending repair needs have been
   completely superseded by NACK messages heard from other receivers
   (when receivers are multicasting NACKs) or from some indicator from
   the sender.  When receivers are unicasting NACK messages, the sender
   may facilitate NACK suppression by forwarding a representation of
   NACK content it has received to the group at large or provide some
   other indicator of the repair information it will be subsequently
   transmitting.

   For effective and scalable suppression performance, the backoff
   timeout periods used by receivers should be independently, randomly
   picked by receivers with a truncated exponential distribution [6].
   This results in the majority of the receiver set holding off
   transmission of NACK messages under the assumption that the smaller
   number of "early NACKers" will supersede the repair needs of the
   remainder of the group.  The mean of the distribution should be
   determined as a function of the current estimate of sender<->group
   GRTT and a group size estimate that is determined by other mechanisms
   within the protocol or preset by the multicast application.

   A simple algorithm can be constructed to generate random backoff
   timeouts with the appropriate distribution.  Additionally, the
   algorithm may be designed to optimize the backoff distribution given
   the number of receivers (R) potentially generating feedback.  This
   "optimization" minimizes the number of feedback messages (e.g., NACK)
   in the worst-case situation where all receivers generate a NACK.  The
   maximum backoff timeout (T_maxBackoff) can be set to control reliable
   delivery latency versus volume of feedback traffic.  A larger value
   of T_maxBackoff will result in a lower density of feedback traffic
   for a given repair cycle.  A smaller value of T_maxBackoff results in
   shorter latency which also reduces the buffering requirements of
   senders and receivers for reliable transport.

   Given the receiver group size (R), and maximum allowed backoff
   timeout (T_maxBackoff), random backoff timeouts (t') with a truncated
   exponential distribution can be picked with the following algorithm:





Adamson, et al.               Experimental                     [Page 13]


RFC 3941                  NORM Building Blocks             November 2004


   1) Establish an optimal mean (L) for the exponential backoff based on
      the group size:

                                L = ln(R) + 1

   2) Pick a random number (x) from a uniform distribution over a range
      of:

               L                           L                   L
       --------------------  to   --------------------  +  ----------
      T_maxBackoff*(exp(L)-1)    T_maxBackoff*(exp(L)-1)  T_maxBackoff


   3) Transform this random variate to generate the desired random
      backoff time (t') with the following equation:

      t' = T_maxBackoff/L * ln(x * (exp(L) - 1) * (T_maxBackoff/L))

   This C language function can be used to generate an appropriate
   random backoff time interval:

      double RandomBackoff(double maxTime, double groupSize)
      {
          double lambda = log(groupSize) + 1;
          double x = UniformRand(lambda/maxTime) +
                     lambda / (maxTime*(exp(lambda)-1));
          return ((maxTime/lambda) *
                  log(x*(exp(lambda)-1)*(maxTime/lambda)));
      }  // end RandomBackoff()

   where UniformRand(double max) returns random numbers with a uniform
   distribution from the range of 0..max.  For example, based on the
   POSIX "rand()" function, the following C code can be used:

      double UniformRand(double max)
      {
          return (max * ((double)rand()/(double)RAND_MAX));
      }

   The number of expected NACK messages generated (N) within the first
   round trip time for a single feedback event is approximately:

      N = exp(1.2 * L / (2*T_maxBackoff/GRTT))

   Thus the maximum backoff time can be adjusted to tradeoff worst-case
   NACK feedback volume versus latency.  This is derived from [6] and
   assumes  T_maxBackoff >= GRTT, and L is the mean of the distribution
   optimized for the given group size as shown in the algorithm above.



Adamson, et al.               Experimental                     [Page 14]


RFC 3941                  NORM Building Blocks             November 2004


   Note that other mechanisms within the protocol may work to reduce
   redundant NACK generation further.  It is suggested that T_maxBackoff
   be selected as an integer multiple of the sender's current advertised
   GRTT estimate such that:

      T_maxBackoff = K * GRTT ;where K >= 1

   For general Internet operation, a default value of K=4 is RECOMMENDED
   for operation with multicast (to the group at large) NACK delivery
   and a value of K=6 for unicast NACK delivery.  Alternate values may
   be used to for buffer utilization, reliable delivery latency and
   group size scalability tradeoffs.

   Given that (K*GRTT) is the maximum backoff time used by the receivers
   to initiate NACK transmission, other timeout periods related to the
   NACK repair process can be scaled accordingly.  One of those timeouts
   is the amount of time a receiver should wait after generating a NACK
   message before allowing itself to initiate another NACK
   backoff/transmission cycle (T_rcvrHoldoff).  This delay should be
   sufficient for the sender to respond to the received NACK with repair
   messages.  An appropriate value depends upon the amount of time for
   the NACK to reach the sender and the sender to provide a repair
   response.  This MUST include any amount of sender NACK aggregation
   period during which possible multiple NACKs are accumulated to
   determine an efficient repair response.  These timeouts are further
   discussed in the section below on "Sender NACK Processing and Repair
   Response".

   There are also secondary measures that can be applied to improve the
   performance of feedback suppression.  For example, the sender's data
   content transmissions can follow an ordinal sequence of transmission.
   When repairs for data content occur, the receiver can note that the
   sender has "rewound" its data content transmission position by
   observing the data object, FEC block number, and FEC symbol
   identifiers.  Receivers SHOULD limit transmission of NACKs to only
   when the sender's current transmission position exceeds the point to
   which the receiver has incomplete reception.  This reduces premature
   requests for repair of data the sender may be planning to provide in
   response to other receiver requests.  This mechanism can be very
   effective for protocol convergence in high loss conditions when
   transmissions of NACKs from other receivers (or indicators from the
   sender) are lost.  Another mechanism (particularly applicable when
   FEC is used) is for the sender to embed an indication of impending
   repair transmissions in current packets sent.  For example, the
   indication may be as simple as an advertisement of the number of FEC
   packets to be sent for the current applicable coding block.





Adamson, et al.               Experimental                     [Page 15]


RFC 3941                  NORM Building Blocks             November 2004


   Finally, some consideration might be given to using the NACKing
   history of receivers to weight their selection of NACK backoff
   timeout intervals.  For example, if a receiver has historically been
   experiencing the greatest degree of loss, it may promote itself to
   statistically NACK sooner than other receivers.  Note this requires
   there is correlation over successive intervals of time in the loss
   experienced by a receiver.  Such correlation MAY not be present in
   multicast networks.  This adjustment of backoff timeout selection may
   require the creation of an "early NACK" slot for these historical
   NACKers.  This additional slot in the NACK backoff window will result
   in a longer repair cycle process that may not be desirable for some
   applications.  The resolution of these trade-offs may be dependent
   upon the protocol's target application set or network.

   After the random backoff timeout has expired, the receiver will make
   a decision on whether to generate a NACK repair request or not (i.e.,
   it has been suppressed).  The NACK will be suppressed when any of the
   following conditions has occurred:

   1) The accumulated state of NACKs heard from other receivers (or
      forwarding of this state by the sender) is equal to or supersedes
      the repair needs of the local receiver.  Note that the local
      receiver should consider its repair needs only up to the sender
      transmission position recorded at the NACK cycle initiation (when
      the backoff timer was activated).

   2) The sender's data content transmission position "rewinds" to a
      point ordinally less than that of the lowest sequence position of
      the local receiver's repair needs.  (This detection of sender
      "rewind" indicates the sender has already responded to other
      receiver repair needs of which the local receiver may not have
      been aware).  This "rewind" event can occur any time between 1)
      when the NACK cycle was initiated with the backoff timeout
      activation and 2) the current moment when the backoff timeout has
      expired to suppress the NACK.  Another NACK cycle must be
      initiated by the receiver when the sender's transmission sequence
      position exceeds the receiver's lowest ordinal repair point.  Note
      it is possible that the local receiver may have had its repair
      needs satisfied as a result of the sender's response to the repair
      needs of other receivers and no further NACKing is required.

   If these conditions have not occurred and the receiver still has
   pending repair needs, a NACK message is generated and transmitted.
   The NACK should consist of an accumulation of repair needs from the
   receiver's lowest ordinal repair point up to the current sender
   transmission sequence position.  A single NACK message should be
   generated and the NACK message content should be truncated if it
   exceeds the payload size of single protocol message.  When such NACK



Adamson, et al.               Experimental                     [Page 16]


RFC 3941                  NORM Building Blocks             November 2004


   payload limits occur, the NACK content SHOULD contain requests for
   the ordinally lowest repair content needed from the sender.

   NACK Suppression Interface Description

   Inputs:

      1) NACK process initiation decision.
      2) Recorded sender transmission sequence position.
      3) Sender GRTT.
      4) Sender group size estimate.
      5) Application-defined bound on backoff timeout period.
      6) NACKs from other receivers.
      7) Pending repair indication from sender (may be forwarded
         NACKs).
      8) Current sender transmission sequence position.

   Outputs:

      1) Yes/no decision to generate NACK message upon backoff timer
         expiration.

3.2.3.  NACK Content

   The content of NACK messages generated by reliable multicast
   receivers will include information detailing their current repair
   needs.  The specific information depends on the use and type of FEC
   in the NORM repair process.  The identification of repair needs is
   dependent upon the data content identification (See Section 3.5
   below).  At the highest level the NACK content will identify the
   sender to which the NACK is addressed and the data transport object
   (or stream) within the sender's transmission that needs repair.  For
   the indicated transport entity, the NACK content will then identify
   the specific FEC coding blocks and/or symbols it requires to
   reconstruct the complete transmitted data.  This content may consist
   of FEC block erasure counts and/or explicit indication of missing
   blocks or symbols (segments) of data and FEC content.  It should also
   be noted that NORM can be effectively instantiated without a
   requirement for reliable NACK delivery using the techniques discussed
   here.

3.2.3.1.  NACK and FEC Repair Strategies

   Where FEC-based repair is used, the NACK message content will
   minimally need to identify the coding block(s) for which repair is
   needed and a count of erasures (missing packets) for the coding
   block.  An exact count of erasures implies the FEC algorithm is
   capable of repairing _any_ loss combination within the coding block.



Adamson, et al.               Experimental                     [Page 17]


RFC 3941                  NORM Building Blocks             November 2004


   This count may need to be adjusted for some FEC algorithms.
   Considering that multiple repair rounds may be required to
   successfully complete repair, an erasure count also implies that the
   quantity of unique FEC parity packets the server has available to
   transmit is essentially unlimited (i.e., the server will always be
   able to provide new, unique, previously unsent parity packets in
   response to any subsequent repair requests for the same coding
   block).  Alternatively, the sender may "round-robin" transmit through
   its available set of FEC symbols for a given coding block, and
   eventually affect repair.  For a most efficient repair strategy, the
   NACK content will need to also _explicitly_ identify which symbols
   (information and/or parity) the receiver requires to successfully
   reconstruct the content of the coding block.  This will be
   particularly true of small to medium size block FEC codes (e.g., Reed
   Solomon) that are capable of provided a limited number of parity
   symbols per FEC coding block.

   When FEC is not used as part of the repair process, or the protocol
   instantiation is required to provide reliability even when the sender
   has transmitted all available parity for a given coding block (or the
   sender's ability to buffer transmission history is exceeded by the
   delay*bandwidth*loss characteristics of the network topology), the
   NACK content will need to contain _explicit_ coding block and/or
   segment loss information so that the sender can provide appropriate
   repair packets and/or data retransmissions.  Explicit loss
   information in NACK content may also potentially serve other
   purposes.  For example, it may be useful for decorrelating loss
   characteristics among a group of receivers to help differentiate
   candidate congestion control bottlenecks among the receiver set.

   When FEC is used and NACK content is designed to contain explicit
   repair requests, there is a strategy where the receivers can NACK for
   specific content that will help facilitate NACK suppression and
   repair efficiency.  The assumptions for this strategy are that sender
   may potentially exhaust its supply of new, unique parity packets
   available for a given coding block and be required to explicitly
   retransmit some data or parity symbols to complete reliable transfer.
   Another assumption is that an FEC algorithm where any parity packet
   can fill any erasure within the coding block (e.g., Reed Solomon) is
   used.  The goal of this strategy is to make maximum use of the
   available parity and provide the minimal amount of data and repair
   transmissions during reliable transfer of data content to the group.

   When systematic FEC codes are used, the sender transmits the data
   content of the coding block (and optionally some quantity of parity
   packets) in its initial transmission.  Note that a systematic FEC
   coding block is considered to be logically made up of the contiguous
   set of data vectors plus parity vectors for the given FEC algorithm



Adamson, et al.               Experimental                     [Page 18]


RFC 3941                  NORM Building Blocks             November 2004


   used.  For example, a coding scheme that provides for 64 data symbols
   and 32 parity symbols per coding block would contain FEC symbol
   identifiers in the range of 0 to 95.

   Receivers then can construct NACK messages requesting sufficient
   content to satisfy their repair needs.  For example, if the receiver
   has three erasures in a given received coding block, it will request
   transmission of the three lowest ordinal parity vectors in the coding
   block.  In our example coding scheme from the previous paragraph, the
   receiver would explicitly request parity symbols 64 to 66 to fill its
   three erasures for the coding block.  Note that if the receiver's
   loss for the coding block exceeds the available parity quantity
   (i.e., greater than 32 missing symbols in our example), the receiver
   will be required to construct a NACK requesting all (32) of the
   available parity symbols plus some additional portions of its missing
   data symbols in order to reconstruct the block.  If this is done
   consistently across the receiver group, the resulting NACKs will
   comprise a minimal set of sender transmissions to satisfy their
   repair needs.

   In summary, the rule is to request the lower ordinal portion of the
   parity content for the FEC coding block to satisfy the erasure repair
   needs on the first NACK cycle.  If the available number of parity
   symbols is insufficient, the receiver will also request the subset of
   ordinally highest missing data symbols to cover what the parity
   symbols will not fill.  Note this strategy assumes FEC codes such as
   Reed-Solomon for which a single parity symbol can repair any erased
   symbol.  This strategy would need minor modification to take into
   account the possibly limited repair capability of other FEC types.
   On subsequent NACK repair cycles where the receiver may have received
   some portion of its previously requested repair content, the receiver
   will use the same strategy, but only NACK for the set of parity
   and/or data symbols it has not yet received.  Optionally, the
   receivers could also provide a count of erasures as a convenience to
   the sender or intermediate systems assisting NACK operation.

   After receipt and accumulation of NACK messages during the
   aggregation period, the sender can begin transmission of fresh
   (previously untransmitted) parity symbols for the coding block based
   on the highest receiver erasure count _if_ it has a sufficient
   quantity of parity symbols that were _not_ previously transmitted.
   Otherwise, the sender MUST resort to transmitting the explicit set of
   repair vectors requested.  With this approach, the sender needs to
   maintain very little state on requests it has received from the group
   without need for synchronization of repair requests from the group.
   Since all receivers use the same consistent algorithm to express
   their explicit repair needs, NACK suppression among receivers is
   simplified over the course of multiple repair cycles.  The receivers



Adamson, et al.               Experimental                     [Page 19]


RFC 3941                  NORM Building Blocks             November 2004


   can simply compare NACKs heard from other receivers against their own
   calculated repair needs to determine whether they should transmit or
   suppress their pending NACK messages.

3.2.3.2.  NACK Content Format

   The format of NACK content will depend on the protocol's data service
   model and the format of data content identification the protocol
   uses.  This NACK format also depends upon the type of FEC encoding
   (if any) used.  Figure 2 illustrates a logical, hierarchical
   transmission content identification scheme, denoting that the notion
   of objects (or streams) and/or FEC blocking is optional at the
   protocol instantiation's discretion.  Note that the identification of
   objects is with respect to a given sender.  It is recommended that
   transport data content identification is done within the context of a
   sender in a given session.  Since the notion of session "streams" and
   "blocks" is optional, the framework degenerates to that of typical
   transport data segmentation and reassembly in its simplest form.

   Session_
           \_
              Sender_
                     \_
                        [Object/Stream(s)]_
                                           \_
                                              [FEC Blocks]_
                                                           \_
                                                              Symbols

            Fig. 2: NORM Data Content Identification Hierarchy

   The format of NACK messages should meet the following goals:

   1) Able to identify transport data unit transmissions required to
      repair a portion of the received content, whether it is an entire
      missing object/stream (or range), entire FEC coding block(s), or
      sets of symbols,

   2) Be simple to process for NACK aggregation and suppression,

   3) Be capable of including NACKs for multiple objects, FEC coding
      blocks and/or symbols in a single message, and

   4) Have a reasonably compact format.

   If the NORM transport object/stream is identified with an <objectId>
   and the FEC symbol being transmitted is identified with an
   <fecPayloadId>, the concatenation of <objectId::fecPayloadId>



Adamson, et al.               Experimental                     [Page 20]


RFC 3941                  NORM Building Blocks             November 2004


   comprises a basic transport protocol data unit (TPDU) identifier for
   symbols from a given source.  NACK content can be composed of lists
   and/or ranges of these TPDU identifiers to build up NACK messages to
   describe the receivers repair needs.  If no hierarchical object
   delineation or FEC blocking is used, the TPDU is a simple linear
   representation of the data symbols transmitted by the sender.  When
   the TPDU represents a hierarchy for purposes of object/stream
   delineation and/or FEC blocking, the NACK content unit may require
   flags to indicate which portion of the TPDU is applicable.  For
   example, if an entire "object" (or range of objects) is missing in
   the received data, the receiver will not necessarily know the
   appropriate range of <sourceBlockNumbers> or <encodingSymbolIds> for
   which to request repair and thus requires some mechanism to request
   repair (or retransmission) of the entire unit represented by an
   <objectId>.  The same is true if entire FEC coding blocks represented
   by one or a range of <sourceBlockNumbers> have been lost.

   NACK Content Interface Description

   Inputs:

      1) Sender identification.
      2) Sender data identification.
      3) Sender FEC Object Transmission Information.
      4) Recorded sender transmission sequence position.
      5) Current sender transmission sequence position.  History of
         repair needs for this sender.

   Outputs:

      1)   NACK message with repair requests.

3.2.4.  Sender Repair Response

   Upon reception of a repair request from a receiver in the group, the
   sender will initiate a repair response procedure.  The sender may
   wish to delay transmission of repair content until it has had
   sufficient time to accumulate potentially multiple NACKs from the
   receiver set.  This allows the sender to determine the most efficient
   repair strategy for a given transport stream/object or FEC coding
   block.  Depending upon the approach used, some protocols may find it
   beneficial for the sender to provide an indicator of pending repair
   transmissions as part of its current transmitted message content.
   This can aid some NACK suppression mechanisms.  The amount of time to
   perform this NACK aggregation should be sufficient to allow for the
   maximum receiver NACK backoff window ("T_maxBackoff" from Section
   3.2.2) and propagation of NACK messages from the receivers to the
   sender.  Note the maximum transmission delay of a message from a



Adamson, et al.               Experimental                     [Page 21]


RFC 3941                  NORM Building Blocks             November 2004


   receiver to the sender may be approximately (1*GRTT) in the case of
   very asymmetric network topology with respect to transmission delay.
   Thus, if the maximum receiver NACK backoff time is T_maxBackoff =
   K*GRTT, the sender NACK aggregation period should be equal to at
   least:

           T_sndrAggregate = T_maxBackoff + 1*GRTT = (K+1)*GRTT

   Immediately after the sender NACK aggregation period, the sender will
   begin transmitting repair content determined from the aggregate NACK
   state and continue with any new transmission.  Also, at this time,
   the sender should observe a "holdoff" period where it constrains
   itself from initiating a new NACK aggregation period to allow
   propagation of the new transmission sequence position due to the
   repair response to the receiver group.  To allow for worst case
   asymmetry, this "holdoff" time should be:

                          T_sndrHoldoff = 1*GRTT

   Recall that the receivers will also employ a "holdoff" timeout after
   generating a NACK message to allow time for the sender's response.
   Given a sender <T_sndrAggregate> plus <T_sndrHoldoff> time of
   (K+1)*GRTT, the receivers should use holdoff timeouts of:

       T_rcvrHoldoff = T_sndrAggregate + T_sndrHoldoff = (K+2)*GRTT

   This allows for a worst-case propagation time of the receiver's NACK
   to the sender, the sender's aggregation time and propagation of the
   sender's response back to the receiver.  Additionally, in the case of
   unicast feedback from the receiver set, it may be useful for the
   sender to forward (via multicast) a representation of its aggregated
   NACK content to the group to allow for NACK suppression when there is
   not multicast connectivity among the receiver set.

   At the expiration of the <T_sndrAggregate> timeout, the sender will
   begin transmitting repair messages according to the accumulated
   content of NACKs received.  There are some guidelines with regards to
   FEC-based repair and the ordering of the repair response from the
   sender that can improve reliable multicast efficiency:

   1) When FEC is used, it is beneficial that the sender transmit
      previously untransmitted parity content as repair messages
      whenever possible.  This  maximizes the receiving nodes' ability
      to reconstruct the entire transmitted content from their
      individual subsets of received messages.






Adamson, et al.               Experimental                     [Page 22]


RFC 3941                  NORM Building Blocks             November 2004


   2) The transmitted object and/or stream data and repair content
      should be indexed with  monotonically increasing sequence numbers
      (within a reasonably large ordinal space).  If the sender observes
      the discipline of transmitting repair for the earliest content
      (e.g., ordinally lowest FEC blocks) first, the receivers can use a
      strategy of withholding repair requests for later content until
      the sender once again returns to that point in the object/stream
      transmission sequence.  This can increase overall message
      efficiency among the group and help work to keep repair cycles
      relatively synchronized without dependence upon strict time
      synchronization among the sender and receivers.  This also helps
      minimize the buffering requirements of receivers and senders and
      reduces redundant transmission of data to the group at large.

   Sender Repair Response Interface Description

   Inputs:

      1) Receiver NACK messages
      2) Group timing information

   Outputs

      1) Repair messages (FEC and/or Data content retransmission)
      2) Advertisement of current pending repair transmissions when
         unicast receiver feedback is detected.

3.3.  NORM Receiver Join Policies and Procedures

   Consideration should be given to the policies and procedures by which
   new receivers join a group (perhaps where reliable transmission is
   already in progress) and begin requesting repair.  If receiver joins
   are unconstrained, the dynamics of group membership may impede the
   application's ability to meet its goals for forward progression of
   data transmission.  Policies limiting the opportunities when
   receivers begin participating in the NACK process may be used to
   achieve the desired behavior.  For example, it may be beneficial for
   receivers to attempt reliable reception from a newly-heard sender
   only upon non-repair transmissions of data in the first FEC block of
   an object or logical portion of a stream.  The sender may also
   implement policies limiting the receivers from which it will accept
   NACK requests, but this may be prohibitive for scalability reasons in
   some situations.  Alternatively, it may be desirable to have a looser
   transport synchronization policy and rely upon session management
   mechanisms to limit group dynamics that can cause poor performance,
   in some types of bulk transfer applications (or for potential
   interactive reliable multicast applications).




Adamson, et al.               Experimental                     [Page 23]


RFC 3941                  NORM Building Blocks             November 2004


   Group Join Policy Interface Description

   Inputs:

      1) Current object/stream data/repair content and sequencing
         identifiers from sender transmissions.

   Outputs:

      1) Receiver yes/no decision to begin receiving and NACKing for
         reliable reception of data

3.4.  Reliable Multicast Member Identification

   In a NORM protocol (or other multicast protocols) where there is the
   potential for multiple sources of data, it is necessary to provide
   some mechanism to uniquely identify the sources (and possibly some or
   all receivers in some cases) within the group.  Identity based on
   arriving packet source addresses is insufficient for several reasons.
   These reasons include routing changes for hosts with multiple
   interfaces that result in different packet source addresses for a
   given host over time, network address translation (NAT) or firewall
   devices, or other transport/network bridging approaches.  As a
   result, some type of unique source identifier <sourceId> field should
   be present in packets transmitted by reliable multicast session
   members.

3.5.  Data Content Identification

   The data and repair content transmitted by a NORM sender requires
   some form of identification in the protocol header fields.  This
   identification is required to facilitate the reliable NACK-oriented
   repair process.  These identifiers will also be used in NACK messages
   generated.  This building block document assumes two very general
   types of data that may comprise bulk transfer session content.  One
   type is static, discrete objects of finite size and the other is
   continuous non-finite streams.  A given application  may wish to
   reliably multicast data content using either one or both of these
   paradigms.  While it may be possible for some applications to further
   generalize this model and provide mechanisms to encapsulate static
   objects as content embedded within a stream, there are advantages in
   many applications to provide distinct support for static bulk objects
   and messages with the context of a reliable multicast session.  These
   applications may include content caching servers, file transfer, or
   collaborative tools with bulk content.  Applications with
   requirements for these static object types can then take advantage of
   transport layer mechanisms (i.e., segmentation/reassembly, caching,
   integrated forward error correction coding, etc.) rather than being



Adamson, et al.               Experimental                     [Page 24]


RFC 3941                  NORM Building Blocks             November 2004


   required to provide their own mechanisms for these functions at the
   application layer.

   As noted, some applications may alternatively desire to transmit bulk
   content in the form of one or more streams of non-finite size.
   Example streams include continuous quasi-real-time message broadcasts
   (e.g., stock ticker) or some content types that are part of
   collaborative tools or other applications.  And, as indicated above,
   some applications may wish to encapsulate other bulk content (e.g.,
   files) into one or more streams within a multicast session.

   The components described within this building block document are
   envisioned to be applicable to both of these models with the
   potential for a mix of both types within a single multicast session.
   To support this requirement, the normal data content identification
   should include a field to uniquely identify the object or stream
   <objectId> within some reasonable temporal or ordinal interval.  Note
   that it is _not_ expected that this data content identification will
   be globally unique.  It is expected that the object/stream identifier
   will be unique with respect to a given sender within the reliable
   multicast session and during the time that sender is supporting a
   specific transport instance of that object or stream.

   Since the "bulk" object/stream content usually requires segmentation,
   some form of segment identification must also be  provided.  This
   segment identifier will be relative to any object or stream
   identifier that has been provided.  Thus, in some cases, NORM
   protocol instantiations may be able to receive transmissions and
   request repair for multiple streams and one or more sets of static
   objects in parallel.  For protocol instantiations employing FEC the
   segment identification portion of the data content identifier may
   consist of a logical concatenation of a coding block identifier
   <sourceBlockNumber> and an identifier for the specific data or parity
   symbol <encodingSymbolId> of the code block.  The FEC Building Block
   document [9] provides a standard message format for identifying FEC
   transmission content.  NORM protocol instantiations using FEC SHOULD
   follow that document's guidelines.

   Additionally, flags to determine the usage of the content identifier
   fields (e.g., stream vs. object) may be applicable.  Flags may also
   serve other purposes in data content identification.  It is expected
   that any flags defined will be dependent upon individual protocol
   instantiations.

   In summary, the following data content identification fields may be
   required for NORM protocol data content messages:

   1) Source node identifier (<sourceId>)



Adamson, et al.               Experimental                     [Page 25]


RFC 3941                  NORM Building Blocks             November 2004


   2) Object/Stream identifier (<objectId>), if applicable.

   3) FEC Block identifier (<sourceBlockNumber>), if applicable.

   4) FEC Symbol identifier (<encodingSymbolId>)

   5) Flags to differentiate interpretation of identifier fields or
      identifier structure that implicitly indicates usage.

   6) Additional FEC transmission content fields per FEC Building Block

   These fields have been identified because any generated NACK messages
   will use these identifiers in requesting repair or retransmission of
   data.  NORM protocols that use these data content fields should also
   be compatible with support for intermediate system assistance to
   reliable multicast transport operation when available.

3.6.  Forward Error Correction (FEC)

   Multiple forward error correction (FEC) approaches have been
   identified that can provide great performance enhancements to the
   repair process of NACK-oriented and other reliable multicast
   protocols [11], [12], [13].  NORM protocols can reap additional
   benefits since FEC-based repair does not _generally_ require explicit
   knowledge of repair content within the bounds of its coding block
   size (in symbols).  In NORM, parity repair packets generated will
   generally be transmitted only in response to NACK repair requests
   from receiving nodes.  However, there are benefits in some network
   environments for transmitting some predetermined quantity of FEC
   repair packets multiplexed with the regular data symbol transmissions
   [14].  This can reduce the amount of NACK traffic generated with
   relatively little overhead cost when group sizes are very large or
   the network connectivity has a large delay*bandwidth product with
   some nominal level of expected packet loss.  While the application of
   FEC is not unique to NORM, these sorts of requirements may dictate
   the types of algorithms and protocol approaches that are applicable.

   A specific issue related to the use of FEC with NORM is the mechanism
   used to identify the portion(s) of transmitted data content to which
   specific FEC packets are applicable.  It is expected that FEC
   algorithms will be based on generating a set of parity repair packets
   for a corresponding block of transmitted data packets.  Since data
   content packets are uniquely identified by the concatenation of
   <sourceId::objectId::sourceBlockNumber::encodingSymbolId> during
   transport, it is expected that FEC packets will be identified in a
   similar manner.  The FEC Building Block document [9] provides
   detailed recommendations concerning application of FEC and standard
   formats for related reliable multicast protocol messages.



Adamson, et al.               Experimental                     [Page 26]


RFC 3941                  NORM Building Blocks             November 2004


3.7.  Round-trip Timing Collection

   The measurement of packet propagation round-trip time (RTT) among
   members of the group is required to support timer-based NACK
   suppression algorithms, timing of sender commands or certain repair
   functions, and congestion control operation.  The nature of the
   round-trip information collected is dependent upon the type of
   interaction among the members of the group.  In the case where only
   "one-to-many" transmission is required, it may be that only the
   sender require RTT knowledge of the greatest RTT (GRTT) among the
   receiver set and/or RTT knowledge of only a portion of the group.
   Here, the GRTT information might be collected in a reasonably
   scalable manner.  For congestion control operation, it is possible
   that RTT information may be required by each receiver in the group.
   In this case, an alternative RTT collection scheme may be utilized
   where receivers collect individual RTT measurements with respect to
   the sender and advertise them to the group or sender.  Where it is
   likely that exchange of reliable multicast data will occur among the
   group on a "many-to-many" basis, there are alternative measurement
   techniques that might be employed for increased efficiency [15].  And
   in some cases, there might be absolute time synchronization among
   hosts that may simplify RTT measurement.  There are trade-offs in
   multicast congestion control design that require further
   consideration before a universal recommendation on RTT (or GRTT)
   measurement can be specified.  Regardless of how the RTT information
   is collected (and more specifically GRTT) with respect to congestion
   control or other requirements, the sender will need to advertise its
   current GRTT estimate to the group for various timeouts used by
   receivers.

3.7.1.  One-to-Many Sender GRTT Measurement

   The goal of this form of RTT measurement is for the sender to learn
   the GRTT among the receivers who are actively participating in NORM
   operation.  The set of receivers participating in this process may be
   the entire group or some subset of the group determined from another
   mechanism within the protocol instantiation.  An approach to collect
   this GRTT information follows.

   The sender periodically polls the group with a message (independent
   or "piggy-backed" with other transmissions) containing a <sendTime>
   timestamp relative to an internal clock at the sender.  Upon
   reception of this message, the receivers will record this <sendTime>
   timestamp and the time (referenced to their own clocks) at which it
   was received <recvTime>.  When the receiver provides feedback to the
   sender (either explicitly or as part of other feedback messages
   depending upon protocol instantiation specification), it will
   construct a "response" using the formula:



Adamson, et al.               Experimental                     [Page 27]


RFC 3941                  NORM Building Blocks             November 2004


            grttResponse = sendTime + (currentTime - recvTime)

   where the <sendTime> is the timestamp from the last probe message
   received from the source and the (<currentTime> - <recvTime>) is the
   amount of time differential since that request was received until the
   receiver generated the response.

   The sender processes each receiver response by calculating a current
   RTT measurement for the receiver from whom the response was received
   using the following formula:

                   RTT_rcvr = currentTime - grttResponse

   During the each periodic GRTT probing interval, the source keeps the
   peak round trip timing measurement (RTT_peak) from the set of
   responses it has received.  A conservative estimate of GRTT is kept
   to maximize the efficiency of redundant NACK suppression and repair
   aggregation.  The update to the source's ongoing estimate of GRTT is
   done observing the following rules:

   1) If a receiver's response round trip time (RTT_rcvr) is greater
      than the current GRTT estimate, the GRTT is immediately updated to
      this new peak value:

                               GRTT = RTT_rcvr

   2) At the end of the response collection period (i.e., the GRTT probe
      interval), if the recorded "peak" response RTT_peak) is less than
      the current GRTT estimate, the GRTT is updated to:

                        GRTT = MAX(0.9*GRTT, RTT_peak)

   3) If no feedback is received, the sender GRTT estimate remains
      unchanged.

   4) At the end of the response collection period, the peak tracking
      value (RTT_peak) is reset to ZERO for subsequent peak detection.

   The GRTT collection period (i.e., period of probe transmission) could
   be fixed at a value on the order of that expected for group
   membership and/or network topology dynamics.  For robustness, more
   rapid probing could be used at protocol startup before settling to a
   less frequent, steady-state interval.  Optionally, an algorithm may
   be developed to adjust the GRTT collection period dynamically in
   response to the current GRTT estimate (or variations in it) and to an
   estimation of packet loss.  The overhead of probing messages could
   then be reduced when the GRTT estimate is stable and unchanging, but
   be adjusted to track more dynamically during periods of variation



Adamson, et al.               Experimental                     [Page 28]


RFC 3941                  NORM Building Blocks             November 2004


   with correspondingly shorter GRTT collection periods.  GRTT
   collection may also be coupled with collection of other information
   for congestion control purposes.

   In summary, although NORM repair cycle timeouts are based on GRTT, it
   should be noted that convergent operation of the protocol does not
   _strictly_ depend on highly accurate GRTT estimation.  The current
   mechanism has proved sufficient in simulations and in the
   environments where NORM-like protocols have been deployed to date.
   The estimate provided by the algorithm tracks the peak envelope of
   actual GRTT (including operating system effect as well as network
   delays) even in relatively high loss connectivity.  The steady-state
   probing/update interval may potentially be varied to accommodate
   different levels of expected network dynamics in different
   environments.

3.7.2.  One-to-Many Receiver RTT Measurement

   In this approach, receivers send messages with timestamps to the
   sender.  To control the volume of these receiver-generated messages,
   a suppression mechanism similar to that described for NACK
   suppression my be used.  The "age" of receivers' RTT measurement
   should be kept by receivers and used as a metric in competing for
   feedback opportunities in the suppression scheme.  For example,
   receiver who have not made any RTT measurement or whose RTT
   measurement has aged most should have precedence over other
   receivers.  In turn the sender may have limited capacity to provide
   an "echo" of the receiver timestamps back to the group, and it could
   use this RTT "age" metric to determine which receivers get
   precedence.  The sender can determine the GRTT as described in 3.7.1
   if it provides sender timestamps to the group.  Alternatively,
   receivers who note their RTT is greater than the sender GRTT can
   compete in the feedback opportunity/suppression scheme to provide the
   sender and group with this information.

3.7.3.  Many-to-Many RTT Measurement

   For reliable multicast sessions that involve multiple senders, it may
   be useful to have RTT measurements occur on a true "many-to-many"
   basis rather than have each sender independently tracking RTT.  Some
   protocol efficiency can be gained when receivers can infer an
   approximation of their RTT with respect to a sender based on RTT
   information they have on another sender and that other sender's RTT
   with respect to the new sender of interest.  For example, for
   receiver "a" and sender's "b" and "c", it is likely that:

                  RTT(a<->b) <= RTT(a<->c)) + RTT(b<->c)




Adamson, et al.               Experimental                     [Page 29]


RFC 3941                  NORM Building Blocks             November 2004


   Further refinement of this estimate can be conducted if RTT
   information is available to a node concerning its own RTT to a small
   subset of other group members and RTT information among those other
   group members it learns during protocol operation.

3.7.4.  Sender GRTT Advertisement

   To facilitate deterministic NORM protocol operation, the sender
   should robustly advertise its current estimation of GRTT to the
   receiver set.  Common, robust knowledge of the sender's current
   operating GRTT estimate among the group will allow the protocol to
   progress in its most efficient manner.  The sender's GRTT estimate
   can be robustly advertised to the group by simply embedding the
   estimate into all pertinent messages transmitted by the sender.  The
   overhead of this can be made quite small by quantizing (compressing)
   the GRTT estimate to a single byte of information.  The following C-
   language functions allows this to be done over a wide range (RTT_MIN
   through RTT_MAX) of GRTT values while maintaining a greater range of
   precision for small GRTT values and less precision for large values.
   Values of 1.0e-06 seconds and 1000 seconds are RECOMMENDED for
   RTT_MIN and RTT_MAX respectively.  NORM applications may wish to
   place an additional, smaller upper limit on the GRTT advertised by
   senders to meet application data delivery latency constraints at the
   expense of greater feedback volume in some network environments.

      unsigned char QuantizeGrtt(double grtt)
      {
          if (grtt > RTT_MAX)
              grtt = RTT_MAX;
          else if (grtt < RTT_MIN)
              grtt = RTT_MIN;
          if (grtt < (33*RTT_MIN))
              return ((unsigned char)(grtt / RTT_MIN) - 1);
          else
              return ((unsigned char)(ceil(255.0-
                                      (13.0 * log(RTT_MAX/grtt)))));
      }

      double UnquantizeRtt(unsigned char qrtt)
      {
           return ((qrtt <= 31) ?
                     (((double)(qrtt+1))*(double)RTT_MIN) :
                    (RTT_MAX/exp(((double)(255-qrtt))/(double)13.0)));
      }







Adamson, et al.               Experimental                     [Page 30]


RFC 3941                  NORM Building Blocks             November 2004


   Note that this function is useful for quantizing GRTT times in the
   range of 1 microsecond to 1000 seconds.  Of course, NORM protocol
   implementations may wish to further constrain advertised GRTT
   estimates (e.g., limit the maximum value) for practical reasons.

3.8.  Group Size Determination/Estimation

   When NORM protocol operation includes mechanisms that excite feedback
   from the group at large (e.g., congestion control), it may be
   possible to roughly estimate the group size based on the number of
   feedback messages received with respect to the distribution of the
   probabilistic suppression mechanism used.  Note the timer-based
   suppression mechanism described in this document does not require a
   very accurate estimate of group size to perform adequately.  Thus, a
   rough estimate, particularly if conservatively managed, may suffice.
   Group size may also be determined administratively.  In absence of a
   group size determination mechanism a default group size value of
   10,000 is RECOMMENDED for reasonable management of feedback given the
   scalability of expected NORM usage.

3.9.  Congestion Control Operation

   Congestion control that fairly shares available network capacity
   with other reliable multicast and TCP instantiations is REQUIRED for
   general Internet operation.  The TCP-Friendly Multicast Congestion
   Control (TFMCC) [16] or Pragmatic General Multicast Congestion
   Control (PGMCC) techniques [17] may be applied to NORM operation to
   meet this requirement.

3.10.  Router/Intermediate System Assistance

   NACK-oriented protocols may benefit from general purpose router
   assistance.  In particular, additional NACK suppression where routers
   or intermediate systems can aggregate NACK content (or filter
   duplicate NACK content) from receivers as it is relayed toward the
   sender could enhance NORM group size scalability.  For NORM protocols
   using FEC, it is possible that intermediate systems may be able to
   filter FEC repair messages to provide an intelligent "subcast" of
   repair content to different legs of the multicast topology depending
   on the repair needs learned from previous receiver NACKs.  Both of
   these types of assist functions would require router interpretation
   of transport data unit content identifiers and flags.

3.11.  NORM Applicability

   The NORM building block applies to protocols wishing to employ
   negative acknowledgement to achieve reliable data transfer.  Properly
   designed negative-acknowledgement (NACK)-oriented reliable multicast



Adamson, et al.               Experimental                     [Page 31]


RFC 3941                  NORM Building Blocks             November 2004


   (NORM) protocols offer scalability advantages for applications and/or
   network topologies where, for various reasons, it is prohibitive to
   construct a higher order delivery infrastructure above the basic
   Layer 3 IP multicast service (e.g., unicast or hybrid
   unicast/multicast data distribution trees).  Additionally, the
   scalability property of NACK-oriented protocols [18], [19] is
   applicable where broad "fan-out" is expected for a single network hop
   (e.g., cable-TV data delivery, satellite, or other broadcast
   communication services).  Furthermore, the simplicity of a protocol
   based on "flat" group-wide multicast distribution may offer
   advantages for a broad range of distributed services or dynamic
   networks and applications.  NORM protocols can make use of reciprocal
   (among senders and receivers) multicast communication under the Any-
   Source Multicast (ASM) model defined in RFC 1112 [2], and are capable
   of scalable operation in asymmetric topologies such as Single-Source
   Multicast (SSM) [8] where there may only be unicast routing service
   from the receivers to the sender(s).

   NORM operation is compatible with transport layer forward error
   correction coding techniques as described in [13] and congestion
   control mechanisms such as those described in [16] and [17].  A
   principal limitation of NORM operation involves group size
   scalability when network capacity for receiver feedback is very
   limited.  NORM operation is also governed by implementation buffering
   constraints.  Buffering greater than that required for typical
   point-to-point reliable transport (e.g., TCP) is recommended to allow
   for disparity in the receiver group connectivity and to allow for the
   feedback delays required to attain group size scalability.

4.  Security Considerations

   NORM protocols are expected to be subject to the same sort of
   security vulnerabilities as other IP and IP multicast protocols.
   NORM is compatible with IP security (IPsec) authentication mechanisms
   [20] that are RECOMMENDED for protection against session intrusion
   and denial of service attacks.  A particular threat for NACK based
   protocols is that of NACK replay attacks that would prevent a NORM
   sender from making forward progress in transmission.  Any standard
   IPsec mechanisms that can provide protection against such replay
   attacks are RECOMMENDED for use.  Additionally, NORM protocol
   instantiations SHOULD consider providing support for their own NACK
   replay attack protection when network layer mechanisms are not
   available.  The IETF Multicast Security (msec) Working Group is also
   developing solutions which may be applicable to NORM in the future.







Adamson, et al.               Experimental                     [Page 32]


RFC 3941                  NORM Building Blocks             November 2004


5.  Acknowledgements (and these are not Negative)

   The authors would like to thank Rick Jones, and Joerg Widmer for
   their valuable comments on this document.  The authors would also
   like to thank the RMT working group chairs, Roger Kermode and Lorenzo
   Vicisano, for their support in development of this specification, and
   Sally Floyd for her early inputs into this document.

6.  References

6.1.  Normative References

   [1]  Bradner, S., "Key words for use in RFCs to Indicate Requirement
        Levels", BCP 14, RFC 2119, March 1997.

   [2]  Deering, S., "Host Extensions for IP Multicasting", STD 5, RFC
        1112, August 1989.

6.2.  Informative References

   [3]  Mankin, A., Romanow, A., Bradner, S., and V. Paxson, "IETF
        Criteria for Evaluating Reliable Multicast Transport and
        Application Protocols", RFC 2357, June 1998.

   [4]  Clark, D. and D. Tennenhouse, "Architectural Considerations for
        a New Generation of Protocols". In Proc. ACM SIGCOMM, pages
        201--208, September 1990.

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

   [6]  Nonnenmacher, J. and E. Biersack, "Optimal Multicast Feedback,"
        in IEEE Infocom, San Francisco, California, p. 964, March/April
        1998.

   [7]  Macker, J. and R. Adamson, "Quantitative Prediction of Nack
        Oriented Reliable Multicast (NORM) Feedback", Proc. IEEE MILCOM
        2002, October 2002.

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

   [9]  Luby, M., Vicisano, L., Gemmell, J., Rizzo, L., Handley, M., and
        J. Crowcroft, "Forward Error Correction (FEC) Building Block",
        RFC 3452, December 2002.




Adamson, et al.               Experimental                     [Page 33]


RFC 3941                  NORM Building Blocks             November 2004


   [10] Floyd, S., Jacobson, V., McCanne, S., Liu, C., and L. Zhang. "A
        Reliable Multicast Framework for Light-weight Sessions and
        Application Level Framing", Proc. ACM SIGCOMM, August 1995.

   [11] Metzner, J., "An Improved Broadcast Retransmission Protocol",
        IEEE Transactions on Communications, Vol. Com-32, No.6, June
        1984.

   [12] Macker, J., "Reliable Multicast Transport and Integrated
        Erasure-based Forward Error Correction", Proc. IEEE MILCOM 97,
        October 1997.

   [13] Luby, M., Vicisano, L., Gemmell, J., Rizzo, L., Handley, M., and
        J. Crowcroft, "The Use of Forward Error Correction (FEC) in
        Reliable Multicast", RFC 3453, December 2002.

   [14] Gossink, D. and J. Macker, "Reliable Multicast and Integrated
        Parity Retransmission with Channel Estimation", IEEE GLOBECOM
        98'.

   [15] Ozdemir, V., Muthukrishnan, S., and I. Rhee, "Scalable, Low-
        Overhead Network Delay Estimation", NCSU/AT&T White Paper,
        February 1999.

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

   [17] Rizzo, L., "pgmcc: A TCP-Friendly Single-Rate Multicast
        Congestion Control Scheme", Proc ACM SIGCOMM 2000, Stockholm,
        August 2000.

   [18] Pingali, S., Towsley, D., and J. Kurose, "A Comparison of
        Sender-Initiated and Receiver-Initiated Reliable Multicast
        Protocols".  In Proc. INFOCOM, San Francisco, CA, October 1993.

   [19] B.N. Levine, J.J. Garcia-Luna-Aceves, "A Comparison of Known
        Classes of Reliable Multicast Protocols", Proc. International
        Conference on Network Protocols (ICNP-96), Columbus, Ohio, Oct
        29--Nov 1, 1996.

   [20] Kent, S. and R. Atkinson, "Security Architecture for the
        Internet Protocol", RFC 2401, November 1998.








Adamson, et al.               Experimental                     [Page 34]


RFC 3941                  NORM Building Blocks             November 2004


7.  Authors' Addresses

   Brian Adamson
   Naval Research Laboratory
   Washington, DC 20375

   EMail: adamson@itd.nrl.navy.mil


   Carsten Bormann
   Universitaet Bremen TZI
   Postfach 330440
   D-28334 Bremen, Germany

   EMail: cabo@tzi.org


   Mark Handley
   Department of Computer Science
   University College London
   Gower Street
   London
   WC1E 6BT
   UK

   EMail: M.Handley@cs.ucl.ac.uk


   Joe Macker
   Naval Research Laboratory
   Washington, DC 20375

   EMail: macker@itd.nrl.navy.mil


















Adamson, et al.               Experimental                     [Page 35]


RFC 3941                  NORM Building Blocks             November 2004


Full Copyright Statement

   Copyright (C) The Internet Society (2004).

   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
   "AS IS" basis and THE CONTRIBUTOR, THE ORGANIZATION HE/SHE REPRESENTS
   OR IS SPONSORED BY (IF ANY), THE INTERNET SOCIETY AND THE INTERNET
   ENGINEERING TASK FORCE DISCLAIM 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.

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 IETF's procedures with respect to rights in IETF 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
   http://www.ietf.org/ipr.

   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 ietf-
   ipr@ietf.org.

Acknowledgement

   Funding for the RFC Editor function is currently provided by the
   Internet Society.







Adamson, et al.               Experimental                     [Page 36]