Status of This Memo

This is an Internet Standards Track document.¶

This document is a product of the Internet Engineering Task Force (IETF). It represents the consensus of the IETF community. It has received public review and has been approved for publication by the Internet Engineering Steering Group (IESG). Further information on Internet Standards is available in Section 2 of RFC 7841.¶

Information about the current status of this document, any errata, and how to provide feedback on it may be obtained at https://www.rfc-editor.org/info/rfc0000.¶

Copyright Notice

Copyright (c) 2019 IETF Trust and the persons identified as the document authors. All rights reserved.¶

This document is subject to BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents (https://trustee.ietf.org/license-info) in effect on the date of publication of this document. Please review these documents carefully, as they describe your rights and restrictions with respect to this document. Code Components extracted from this document must include Simplified BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Simplified BSD License.¶

Table of Contents

1. Introduction

Many applications need to transport a continuous stream of packetized data from a source (sender) to one or more destinations (receivers) over networks that do not provide guaranteed packet delivery. In particular, packets may be lost, which is strictly the focus of this document: we assume that transmitted packets are either lost (e.g., because of a congested router, a poor signal-to-noise ratio in a wireless network, or because the number of bit errors exceeds the correction capabilities of the physical-layer error-correcting code) or were received by the transport protocol without any corruption (i.e., the bit errors, if any, have been fixed by the physical-layer error-correcting code and therefore are hidden to the upper layers).¶

For these use cases, Forward Error Correction (FEC) applied within the transport or application layer is an efficient technique to improve packet transmission robustness in the presence of packet losses (or "erasures") without going through packet retransmissions that create a delay often incompatible with real-time constraints. The FEC Building Block defined in [RFC5052] provides a framework for the definition of Content Delivery Protocols (CDPs) that make use of separately defined FEC schemes. Any CDP defined according to the requirements of the FEC Building Block can then easily be used with any FEC Scheme that is also defined according to the requirements of the FEC Building Block.¶

Then, FECFRAME [RFC6363] provides a framework to define Content Delivery Protocols (CDPs) that provide FEC protection for arbitrary packet flows over an unreliable datagram service transport, such as UDP. It is primarily intended for real-time or streaming media applications, using broadcast, multicast, or on-demand delivery.¶

However, [RFC6363] only considers block FEC schemes defined in accordance with the FEC Building Block [RFC5052] (e.g., [RFC6681], [RFC6816], or [RFC6865]). These codes require the input flow(s) to be segmented into a sequence of blocks. Then, FEC encoding (at a sender or an encoding middlebox) and decoding (at a receiver or a decoding middlebox) are both performed on a per-block basis. For instance, if the current block encompasses the 100's to 119's source symbols (i.e., a block of size 20 symbols) of an input flow, encoding (and decoding) will be performed on this block independently of other blocks. This approach has major impacts on FEC encoding and decoding delays. The data packets of continuous media flow(s) may be passed to the transport layer immediately, without delay. But the block creation time, which depends on the number of source symbols in this block, impacts both the FEC encoding delay (since encoding requires that all source symbols be known) and, mechanically, the packet loss recovery delay at a receiver (since no repair symbol for the current block can be generated and therefore received before that time). Therefore, a good value for the block size is necessarily a balance between the maximum FEC decoding latency at the receivers (which must be in line with the most stringent real-time requirement of the protected flow(s), hence an incentive to reduce the block size) and the desired robustness against long loss bursts (which increases with the block size, hence an incentive to increase this size).¶

This document updates [RFC6363] in order to also support FEC codes based on a sliding encoding window (a.k.a., convolutional codes) [RFC8406]. This encoding window, either fixed or variable size, slides over the set of source symbols. FEC encoding is launched whenever needed from the set of source symbols present in the sliding encoding window at that time. This approach significantly reduces FEC-related latency, since repair symbols can be generated and passed to the transport layer on the fly at any time and can be regularly received by receivers to quickly recover packet losses. Using sliding window FEC codes is therefore highly beneficial to real-time flows, one of the primary targets of FECFRAME. [RFCYYY1] provides an example of such a FEC Scheme for FECFRAME, which is built upon the simple sliding window Random Linear Codes (RLC).¶

This document is fully backward compatible with [RFC6363]. Indeed:¶

• This FECFRAME update does not prevent or compromise in any way the support of block FEC codes. Both types of codes can nicely coexist, just like different block FEC schemes can coexist.¶
• Each sliding window FEC Scheme is associated with a specific FEC Encoding ID subject to IANA registration, just like block FEC Schemes.¶
• Any receiver -- for instance, a legacy receiver that only supports block FEC schemes -- can easily identify the FEC Scheme used in a FECFRAME session. Indeed, the FEC Encoding ID that identifies the FEC Scheme is carried in FEC Framework Configuration Information (see Section 5.5 of [RFC6363]). For instance, when the Session Description Protocol (SDP) is used to carry the FEC Framework Configuration Information, the FEC Encoding ID can be communicated in the "encoding-id=" parameter of a "fec-repair-flow" attribute [RFC6364]. This mechanism is the basic approach for a FECFRAME receiver to determine whether or not it supports the FEC Scheme used in a given FECFRAME session.¶

This document leverages on [RFC6363] and reuses its structure. It proposes new sections specific to sliding window FEC codes whenever required. The only exception is Section 3, which provides a quick summary of FECFRAME in order to facilitate the understanding of this document to readers not familiar with the concepts and terminology.¶

3. Summary of Architecture Overview

The architecture of Section 3 of [RFC6363] equally applies to this FECFRAME extension and is not repeated here. However, this section includes a quick summary to facilitate the understanding of this document to readers not familiar with the concepts and terminology.¶

+----------------------+
|     Application      |
+----------------------+
           |
           | (1) Application Data Units (ADUs)
           |
           v
+----------------------+                           +----------------+
|    FEC Framework     |                           |                |
|                      |-------------------------->|   FEC Scheme   |
|(2) Construct source  |(3) Source Block           |                |
|    blocks            |                           |(4) FEC Encoding|
|(6) Construct FEC     |<--------------------------|                |
|    Source and Repair |                           |                |
|    Packets           |(5) Explicit Source FEC    |                |
+----------------------+    Payload IDs            +----------------+
           |                Repair FEC Payload IDs
           |                Repair symbols
           |
           |(7) FEC Source and Repair Packets
           v
+----------------------+
|  Transport Protocol  |
+----------------------+

The FECFRAME architecture is illustrated in Figure 1 from the sender's point of view in case of a block FEC Scheme. It shows an application generating an ADU flow (other flows from other applications may coexist). These ADUs of variable size must be somehow mapped to source symbols of a fixed size (this fixed size is a requirement of all FEC Schemes, which comes from the way mathematical operations are applied to symbols' content). This is the goal of an ADU-to-symbols mapping process that is FEC Scheme specific (see below). Once the source block is built, taking into account both the FEC Scheme constraints (e.g., in terms of maximum source block size) and the application's flow constraints (e.g., in terms of real-time constraints), the associated source symbols are handed to the FEC Scheme in order to produce an appropriate number of repair symbols. FEC Source Packets (containing ADUs) and FEC Repair Packets (containing one or more repair symbols each) are then generated and sent using an appropriate transport protocol (more precisely, Section 7 of [RFC6363] requires a transport protocol providing an unreliable datagram service, such as UDP). In practice, FEC Source Packets may be passed to the transport layer as soon as available without having to wait for FEC encoding to take place. In that case, a copy of the associated source symbols needs to be kept within FECFRAME for future FEC encoding purposes.¶

At a receiver (not shown), FECFRAME processing operates in a similar way, taking as input the incoming FEC Source and Repair Packets received. In case of FEC Source Packet losses, the FEC decoding of the associated block may recover all (in case of successful decoding) or a subset potentially empty (otherwise) of the missing source symbols. After source-symbol-to-ADU mapping, when lost ADUs are recovered, they are then assigned to their respective flow (see below). ADUs are returned to the application(s), either in their initial transmission order (in which case ADUs received after an erased one will be delayed until FEC decoding has taken place) or not (in which case each ADU is returned as soon as it is received or recovered), depending on the application requirements.¶

FECFRAME features two subtle mechanisms:¶

• ADUs-to-source-symbols mapping: in order to manage variable size ADUs, FECFRAME and FEC Schemes can use small, fixed-size symbols and create a mapping between ADUs and symbols. To each ADU, this mechanism prepends a length field (plus a flow identifier; see below) and pads the result to a multiple of the symbol size. A small ADU may be mapped to a single source symbol, while a large one may be mapped to multiple symbols. The mapping details are FEC Scheme dependent and must be defined in the associated document.¶
• Assignment of decoded ADUs to flows in multi-flow configurations: when multiple flows are multiplexed over the same FECFRAME instance, a problem is to assign a decoded ADU to the right flow (UDP port numbers and IP addresses traditionally used to map incoming ADUs to flows are not recovered during FEC decoding). To make it possible, at the FECFRAME sending instance, each ADU is prepended with a flow identifier (1 byte) during the ADU-to-source-symbols mapping (see above). The flow identifiers are also shared between all FECFRAME instances as part of the FEC Framework Configuration Information. This (flow identifier + length + application payload + padding), called ADUI, is then FEC protected. Therefore, a decoded ADUI contains enough information to assign the ADU to the right flow.¶

A few aspects are not covered by FECFRAME, namely:¶

• Section 8 of [RFC6363] does not detail any congestion control mechanisms and only provides high-level normative requirements.¶
• The possibility of having feedback from receiver(s) is considered out of scope, although such a mechanism may exist within the application (e.g., through RTCP control messages).¶
• Flow adaptation at a FECFRAME sender (e.g., how to set the FEC code rate based on transmission conditions) is not detailed, but it needs to comply with the congestion control normative requirements (see above).¶

7. Transport Protocols

The discussion in Section 7 of [RFC6363] equally applies to this FECFRAME extension and is not repeated here.¶

8. Congestion Control

The discussion in Section 8 of [RFC6363] equally applies to this FECFRAME extension and is not repeated here.¶

11. IANA Considerations

This document has no IANA actions.¶

A FEC Scheme for use with this FEC Framework is identified via its FEC Encoding ID. It is subject to IANA registration in the "FEC Framework (FECFRAME) FEC Encoding IDs" registry. All the rules of Section 11 of [RFC6363] apply and are not repeated here.¶

Appendix A. About Sliding Encoding Window Management (Informational)

The FEC Framework does not specify the management of the sliding encoding window, which is the responsibility of the FEC Scheme. This annex only provides a few informational hints.¶

Source symbols are added to the sliding encoding window each time a new ADU is available at the sender, after the ADU-to-source-symbol mapping specific to the FEC Scheme.¶

Source symbols are removed from the sliding encoding window. For instance:¶

• After a certain delay, when an "old" ADU of a real-time flow times out. The source symbol retention delay in the sliding encoding window should therefore be initialized according to the real-time features of incoming flow(s) when applicable.¶
• Once the sliding encoding window has reached its maximum size (there is usually an upper limit to the sliding encoding window size). In that case, the oldest symbol is removed each time a new source symbol is added.¶

Several considerations can impact the management of this sliding encoding window:¶

• At the source flows level: real-time constraints can limit the total time during which source symbols can remain in the encoding window.¶
• At the FEC code level: theoretical or practical limitations (e.g., because of computational complexity) can limit the number of source symbols in the encoding window.¶
• At the FEC Scheme level: signaling and window management are intrinsically related. For instance, an encoding window composed of a nonsequential set of source symbols requires appropriate signaling to inform a receiver of the composition of the encoding window, and the associated transmission overhead can limit the maximum encoding window size. On the contrary, an encoding window always composed of a sequential set of source symbols simplifies signaling: providing the identity of the first source symbol plus its number is sufficient, which creates a fixed and relatively small transmission overhead.¶

Authors' Addresses