Coding and congestion control in transport
draft-irtf-nwcrg-coding-and-congestion-00

Abstract

This document discusses the interaction between congestion control and coding mechanism at the transport layer. The scope of the document is end-to-end communications. Examples of interest for the proposed solution is to better deal with tail losses or with networks with non-congestion losses. Coding for tunnels is out-of-the scope of the document.

Status of This Memo

This Internet-Draft is submitted in full conformance with the provisions of BCP 78 and BCP 79.

Internet-Drafts are working documents of the Internet Engineering Task Force (IETF). Note that other groups may also distribute working documents as Internet-Drafts. The list of current Internet-Drafts is at https://datatracker.ietf.org/drafts/current/.

Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as "work in progress."

This Internet-Draft will expire on June 7, 2020.

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
• 2. Separate channels
• 3. Base solution description
• 4. Server-side coding solutions
• 4.1. Coded packets without considering CWND progression
• 4.2. Coded packets driven by CWND progression
• 5. Server-side reaction to recovered packet signals
• 5.1. The server congestion control considers recovered packet signals as congestion-implied packet losses
• 5.2. The server adapts its window reduction to recovered packet signals
• 5.3. The server ignores recovered packet signals
• 6. Summary
• 7. Acknowledgements
• 8. IANA Considerations
• 9. Security Considerations
• 10. Informative References
• Authors' Addresses

1. Introduction

There are cases where deploying coding improves the quality of the transmission. As example, the server may hardly detect tail losses that impact may impact the application layer. Another example may be the networks where non-congestion losses are persistent and prevent the server from exploiting the link capacity. [RFC5681] defines TCP as a loss-based congestion control and coding mechanisms can hide congestion signals to the server. This memo discusses simple best practices on how coding and congestion control mechanisms could coexist.

The proposed recommendations apply for coding at the transport or application layer and coding for tunnels is out-of-the scope of the document.

2. Separate channels

Figure 1 presents the notations that will be used in this document and introduce the Congestion Control (CC) and Forward Erasure Correction (FEC) channels. Congestion Control channel carries data packets (from the server to the client) and a potential information signaling the packets that have been received (from the client to the server). Forward Erasure Correction channel carries coded packets (from the server to the client) and a potiential information signaling the packets that have been repaired (from the client to the server). It is worth pointing out that there are cases where these channels are not separated.

 Client                          Server

+------+                        +------+
|      | -- { data packet  -----|      |
|  CC  |                        |  CC  | 
|      | -- received packet } --|      |
+------+                        +------+

+------+                        +------+
|      | -- { coded packet  ----|      |
| FEC  |                        | FEC  |
|      | -- repaired packet } --|      |
+------+                        +------+
		

Figure 1: Notations and separate channels

3. Base solution description

The base solution can be described as follows:

• The client MUST indicate to the server that one or multiple packets have been recovered using a coding scheme. Such "repaired packet signal" could be based on existing signals (even if the existing signal was not designed for that purpose, such as ECN) or on new type of signals (such as a RECOVERED frame in QUIC).
• The server MUST be able to detect the "recovered packet signal". The base solution does not describe how the server reacts to such signal.

The proposed solution applies for coding in the transport layer. The proposed approach is inline with the one in [I-D.swett-nwcrg-coding-for-quic]. The proposed solution does not applies for the interaction between coding under the transport layer (i.e. not end-to-end), such as coding for tunnels.

4. Server-side coding solutions

This section presents solutions for a server to add application coding.

4.1. Coded packets without considering CWND progression

In this solution, the coded packets are sent on top of what is allowed by a congestion window. Examples of the solution could be adding a given pourcentage of the congestion window as supplementary packets or sending a given amount of coded packets at a given rate. The redundancy flow can be decorrelated from the congestion control that manages source packets : a secondary congestion control can be introduced, such as in coupled congestion control for RTP media [I-D.ietf-rmcat-coupled-cc]. An example would be to exploit a lower than best-effort congestion control [RFC6297].

The advantage of such solution is that coding would help in challenges cases where transmission losses are persistent.

The drawback of such solution is that it may result in coding solutions being unfair towards non-coding solutions. This solutions may result in adding congestion in congested networks.

4.2. Coded packets driven by CWND progression

In this solution, the coded packets are sent within what the congestion window allows, such as in [CTCP]. Examples of the solution would be sending coded packets when there is no more data to transmit or preferably send coded packets instead of the following packets in the send buffer.

The advantage of this solution is that it does not contribute in adding more congestion than the congestion window allows. Indeed, all traffic (source and redundancy) is controlled by one congestion control only and TCP metrics for fairness can be indifferently applied in this case.

The main drawback is the decrease of goodput if coded packets are sent but are not used at the client side.

5. Server-side reaction to recovered packet signals

Delay-based congestion controls ignore packets that have been repaired with coding. There is no need to define best current pratices in this case. However, more discussions are required for congestion controls that use loss as congestion signals (potentially among other congestion detection mechanism).

5.1. The server congestion control considers recovered packet signals as congestion-implied packet losses

In this solution, the server reacts to recovered packet signals as to congestion-implied packet losses. That being said, this does not necessarily means that the packets have actually been lost. The server may have other means to identify that the packet was just out-of-ordered and ignore the recovered packet signals.

The advantages of the solution are (1) that coding mechanisms do not hide congestion signals, such as packets voluntary dropped by a AQM [RFC7567] and (2) packets may be recovered faster than with traditionnal retransmission mechanisms.

The drawback of this solution is that, if there is a high non-congestion loss rate, the congestion control throughput may decrease drastically. Reporting this amount of loss to a server may reduce the application goodput when there is no actual congestion.

5.2. The server adapts its window reduction to recovered packet signals

In this solution, the server does not reduce the congestion window with the same amount when the "recovered packet signal" is received, i.e. when a packet has been lost but recovered. Example of this solution could be based on [RFC8511] or considering that recovering an isolated packet is not an actual sign of congestion.

The advantage of the solution is that in cases where there is no actual congestion, coding could help in improving the transmission without ignoring congestion signals.

The main drawback is the precised design of the solution and its interaction with AQM mechanisms [RFC7567]. Moreover there may be fairness issues since AIMD convergence may not be guaranteed.

5.3. The server ignores recovered packet signals

This is the case for delay-based congestion controls. The interaction between delay-based congestion controls and the delay induced by a coding mechanisms is an open research activity. That being said, a potential approach would be that loss-based congestion control ignores the "recovered packet signal".

The advantage of this solution is that coding would provided substantial benefits in cases where there are transmission losses.

However, this solution hides potential congestion losses, making it unfair to other congestion controls.

6. Summary

This section provides a summary on the content in previous sections. The Figure 2 sums up some recommendations. It is worth pointing out that the "coding without congestion" considers that coded packets are sent along with original data packets, in opposition with the solution where coded packets are transmitted only when there is no more original packets to transmit. Moreover, the values indicated in this Figure consider a channel that does not exhibit a high loss pattern.

+-----------------------+--------------------------------+
| Server-side reaction  | Server-side coding solutions   |
| to recovered packet   |                                |
| signals               |                                |
|                       +---------------+----------------+
|                       | Coding adding | Coding without |
|                       | congestion    | congestion     |
+-----------------------+---------------+----------------+
| React as loss         | fairness: ~   | fairness: ++   |
|                       | real-time: +  | real-time: +   |
|                       | bulk: ~       | bulk:  -       |
+-----------------------+---------------+----------------+
| Adapt window reduction| fairness: ~   | fairness: +    |
|                       | real-time: +  | real-time: +   |
|                       | bulk: +       | bulk: -        |
+-----------------------+---------------+----------------+
| Ignore signals        | fairness: -   | fairness: -    |
|                       | real-time: +  | real-time: +   |
|                       | bulk: +       | bulk: -        |
+-----------------------+---------------+----------------+
        

Figure 2: Recommendations

7. Acknowledgements

Many thanks to Spencer Dawkins, Dave Oran, Carsten Bormann, Vincent Roca and Marie-Jose Montpetit for their useful comments that helped improve the document.

8. IANA Considerations

This memo includes no request to IANA.

9. Security Considerations

There is no security considerations required for this document.

10. Informative References

Authors' Addresses