RFC 8836

Congestion Control Requirements for Interactive Real-Time Media

📅 published January 2021, from ietf

RFC 8836: Congestion Control Requirements for Interactive Real-Time Media 🎀

okay bestie, this one is SO important 💅 It's RFC 8836, titled "Congestion Control Requirements for Interactive Real-Time Media." It was written by R. Jesup of Mozilla and Z. Sarker, Ed. of Ericsson AB. It's Informational (so it's not a Standards Track spec 🧚), it was published in January 2021, and its ISSN is 2070-1721. The title block doesn't list anything this document obsoletes or updates.

Abstract ✨

okay so congestion control is needed for ALL the data moving across the Internet 🌐 It's how we keep things fair and stop the whole network from collapsing under its own traffic 😭 Interactive, point-to-point real-time media (think video calls where you need low delay and can tolerate a little loss) has different needs from bulk transfers like FTP or bursty stuff like loading web pages. And with so much RTP-based real-time traffic out there now (hi, WebRTC 👋), it's extra important that this kind of traffic is congestion controlled 🫶

This document lays out a set of requirements you can use to check whether other congestion control mechanisms are actually fit for this job 💜 It's also meant to give a set of possible requirements for a real-time media congestion avoidance technique 🦋

Status of This Memo ✨

This document is not an Internet Standards Track specification. It's published for informational purposes 🌸

It's a product of the Internet Engineering Task Force (IETF) and represents the consensus of the IETF community. It got public review and was approved for publication by the Internet Engineering Steering Group (IESG). Not every document the IESG approves is a candidate for any level of Internet Standard, so see Section 2 of RFC 7841 for that 🧐

Info about the current status of this document, any errata, and how to give feedback can be found at https://www.rfc-editor.org/info/rfc8836 🔥

Copyright (c) 2021 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) as in effect on the date of publication. Please read these carefully, because they describe your rights and restrictions for 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 they're provided without warranty as described in the Simplified BSD License 🍓

Table of Contents 🦋

1.  Introduction
  1.1.  Requirements Language
2.  Requirements
3.  Deficiencies of Existing Mechanisms
4.  IANA Considerations
5.  Security Considerations
6.  References
  6.1.  Normative References
  6.2.  Informative References
Acknowledgements
Authors' Addresses

1. Introduction 🌈

okay so most of today's TCP congestion control schemes were built with a focus on reliable bulk transfer of data that isn't time-critical, like moving huge files 📦 They've also worked really well for reliably sending smaller chunks of data as fast as possible, like when you're fetching a web page 💻

These algorithms have also been used for media streams that get watched in a non-interactive way, like "streaming" video. There, having the data ready when the viewer wants it matters, but the exact timing of delivery doesn't 🎬

Real-time interactive media is a whole different vibe 😤 You need to deliver the data continuously, within a super tight time window (we're talking no more than hundreds of milliseconds end-to-end ⏱️). The sources can adapt how much data they send within pretty wide margins, but the application can rate limit them. Sometimes they don't even have data to send 🙈 They can tolerate some packet loss, but since the data is generated in real time, sending "future" data is impossible, and since it's consumed in real time, data that shows up late is usually useless 💀

Even though real-time interactive media has different requirements from the other flow types, those other flow types are still going to be on the network 🚗 So the congestion control algorithm for real-time interactive media has to work properly when they show up as cross traffic 💪

One protocol portfolio being built for this use case is WebRTC [RFC8825]. The idea is to send multiple flows using the Real-time Transport Protocol (RTP) [RFC3550] between two peers, alongside data flows, all at the same time, without needing special arrangements with the service providers in between 🧚 RTP doesn't come with any congestion control mechanism, so a set of circuit breakers, such as the ones described in [RFC8083], are required to protect the network from excessive congestion caused by non-congestion-controlled flows 🛡️ When real-time interactive media is congestion controlled, it is recommended that the congestion control mechanism operate within the constraints defined by these circuit breakers when a circuit breaker is present, and that it should not cause congestion collapse when a circuit breaker is not implemented ✨

Since this use case is the main focus, use cases involving non-interactive media like video streaming, and those using multicast/broadcast-type technologies, are out of scope 🙅‍♀️

The terminology defined in [RFC8825] is used in this memo 📖

1.1. Requirements Language 💅

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 [RFC2119]. (Yes, these are the capitalized ones, and they're not just vibes 🫶)

2. Requirements 🎀

Here's the main event 👑 Each requirement has its own sub-points, and I've kept all the force and wording exactly as the original has it.

  1. The congestion control algorithm MUST attempt to provide as-low-as-possible-delay transit for interactive real-time traffic while still providing a useful amount of bandwidth. There may be lower limits on the amount of bandwidth that's useful, but this is largely application specific, and the application may be able to modify or remove flows so some useful flows get enough bandwidth. For example, although there might not be enough bandwidth for low-latency video+audio, there could be enough for audio only 🎧
    • a. Jitter (variation in the bitrate over short timescales) is also relevant, though moderate amounts of jitter will be absorbed by jitter buffers. Transit delay should be considered to track the short-term maximums of delay, including jitter 🌊
    • b. The algorithm should provide this as-low-as-possible-delay transit and minimize self-induced latency even when faced with intermediate bottlenecks and competing flows. Competing flows may limit what's possible to achieve 🥺
    • c. The algorithm should be resilient to the effects of events, such as routing changes, which may alter or remove bottlenecks or change the bandwidth available, especially if there is a reduction in available bandwidth or increase in observed delay. It is expected that the mechanism reacts quickly to such events to avoid delay buildup. In the context of this memo, a "quick" reaction is on the order of a few RTTs, subject to the constraints of the media codec, but is likely within a second. Reaction on the next RTT is explicitly not required, since many codecs cannot adapt their sending rate that quickly, but at the same time a response cannot be arbitrarily delayed ⏳
    • d. The algorithm should react quickly to handle both local and remote interface changes (e.g., WLAN to 3G data) that may radically change the bandwidth available or bottlenecks, especially if there is a reduction in available bandwidth or an increase in bottleneck delay. It is assumed that an interface change can generate a notification to the algorithm 📶
    • e. The real-time interactive media applications can be rate limited. This means the offered loads can be less than the available bandwidth at any given moment and may vary dramatically over time, including dropping to no load and then resuming a high load, such as in a mute/unmute operation. Hence, the algorithm must be designed to handle such behavior from a media source or application. Note that the reaction time between a change in the bandwidth available from the algorithm and a change in the offered load is variable, and it may be different when increasing versus decreasing 🎙️
    • f. The algorithm is required to avoid building up queues when competing with short-term bursts of traffic (for example, traffic generated by web browsing), which can quickly saturate a local-bottleneck router or link but clear quickly. The algorithm should also react quickly to regain its previous share of the bandwidth when the local bottleneck or link is cleared 🌷
    • g. Similarly, periodic bursty flows such as MPEG DASH [MPEG_DASH] or proprietary media streaming algorithms may compete in bursts with the algorithm and may not be adaptive within a burst. They are often layered on top of TCP but use TCP in a bursty manner that can interact poorly with competing flows during the bursts. The algorithm must not increase the already existing delay buildup during those bursts. Note that this competing traffic may be on a shared access link, or the traffic burst may cause a shift in the location of the bottleneck for the duration of the burst 💥
  2. The algorithm MUST be fair to other flows, both real-time flows (such as other instances of itself) and TCP flows, both long-lived flows and bursts such as the traffic generated by a typical web-browsing session. Note that "fair" is a rather hard-to-define term. It SHOULD be fair with itself, giving a fair share of the bandwidth to multiple flows with similar RTTs, and if possible to multiple flows with different RTTs 🤝
    • a. Existing flows at a bottleneck must also be fair to new flows to that bottleneck and must allow new flows to ramp up to a useful share of the bottleneck bandwidth as quickly as possible. A useful share will depend on the media types involved, total bandwidth available, and the user-experience requirements of a particular service. Note that relative RTTs may affect the rate at which new flows can ramp up to a reasonable share 🚀
  3. The algorithm SHOULD NOT starve competing TCP flows and SHOULD, as best as possible, avoid starvation by TCP flows 🌸
    • a. The congestion control should prioritize achieving a useful share of the bandwidth, depending on the media types and total available bandwidth, over achieving as-low-as-possible transit delay, when these two requirements are in conflict. Tbh this one is a big deal 😭🔥
  4. The algorithm SHOULD adapt as quickly as possible to initial network conditions at the start of a flow 🚀. This SHOULD happen whether the initial bandwidth is above or below the bottleneck bandwidth, no exceptions bestie.

a. The algorithm should allow different modes of adaptation ✨. For example, startup adaptation might be faster than adaptation later in a flow. It should allow both slow-start operation (adapt up) and history-based startup (start at a point expected to be at or below channel bandwidth based on historical info, which may need to adapt down quickly if the initial guess is wrong). Starting too low and/or adapting up too slowly can make a critical moment in a personal communication feel poor ("Hello!") 😭. Starting too high above the available bandwidth causes other problems for user experience, so yeah, there's a real tension here 💀. Alternative startup helpers, like probing during setup with dummy data, may be useful in some applications. In some cases there's a considerable gap in time between flow creation and the first flow of data 🎀. Again, a flow may need to change its adaptation rates due to network conditions or changes in the provided flows (like unmuting or sending data after a gap).

  1. The algorithm SHOULD be stable if the RTP streams are halted or discontinuous (for example, when using Voice Activity Detection) 🤫.

a. After stream resumption, the algorithm should attempt to rapidly regain its previous share of the bandwidth 🦋. How aggressively it does this will decay with the length of the pause.

  1. Where possible, the algorithm SHOULD merge information across multiple RTP streams sent between two endpoints when those RTP streams share a common bottleneck, whether or not those streams are multiplexed onto the same ports 🫶. This lets congestion control handle the set of streams together instead of as multiple independent streams. It also allows better overall bandwidth management, faster response to changing conditions, and fairer sharing of bandwidth with other network users 👑.

a. The algorithm should also share information and adaptation with other non-RTP flows between the same endpoints, such as a WebRTC data channel [RFC8831], when possible 🌈.

b. When there are multiple streams across the same 5-tuple coordinating their bandwidth use and congestion control, the algorithm should allow the application to control the relative split of available bandwidth 💅. The most correlated bandwidth usage is with other flows on the same 5-tuple, but there may be value in coordinating measurement and control of the local link(s). Use of information about previous flows, especially on the same 5-tuple, may be useful input to the algorithm, particularly regarding startup performance of a new flow 🧚.

  1. The algorithm SHOULD NOT require any special support from network elements to be able to convey congestion-related information 🙅‍♀️. As much as possible, it SHOULD leverage available information about the incoming flow to provide feedback to the sender 🔥. Examples include packet arrival times, acknowledgements and feedback, packet timestamps, packet losses, and Explicit Congestion Notification (ECN) [RFC3168]. All of these can tell you about the state of the path and any bottlenecks. However, how the available information gets used is algorithm dependent.

a. Extra information could be added to the packets to give more detailed info on actual send times (as opposed to sampling times), but such information should not be required.

  1. Since the assumption here is a set of RTP streams, the backchannel typically SHOULD be done via the RTP Control Protocol (RTCP) [RFC3550]. Instead, one alternative would be to include it in a reverse-RTP channel using header extensions 🦋.

a. To react sufficiently quickly when using RTCP for a backchannel, an RTP profile such as RTP/AVPF [RFC4585] or RTP/SAVPF [RFC5124] that allows sufficiently frequent feedback must be used. Note that in some cases, backchannel messages may be delayed until the RTCP channel can be allocated enough bandwidth, even under AVPF rules. This may also imply negotiating a higher maximum percentage for RTCP data, or allowing solutions to violate or modify the rules specified for AVPF.

b. Bandwidth for the feedback messages should be minimized using techniques such as those in [RFC5506], to allow RTCP without Sender/Receiver Reports 🍓.

c. Backchannel data should be minimized to avoid taking too much reverse-channel bandwidth (since this will often be used in a bidirectional set of flows). In areas of stability, backchannel data may be sent more infrequently so long as algorithm stability and fairness are maintained. When the channel is unstable or has not yet reached equilibrium after a change, backchannel feedback may be more frequent and use more reverse-channel bandwidth. This is an area with considerable flexibility of design, and different approaches to backchannel messages and frequency are expected to be evaluated 💜.

  1. Flows managed by this algorithm and flows competing against each other at a bottleneck may have different Differentiated Services Code Point (DSCP) [RFC5865] markings depending on the type of traffic, or may be subject to flow-based QoS. A particular bottleneck or section of the network path may or may not honor DSCP markings. The algorithm SHOULD attempt to leverage DSCP markings when they're available 💖.
  2. The algorithm SHOULD sense the unexpected lack of backchannel information as a possible indication of a channel-overuse problem and react accordingly to avoid burst events causing a congestion collapse 🙈.
  3. The algorithm SHOULD be stable and maintain low delay when faced with Active Queue Management (AQM) algorithms. Also note that these algorithms may apply across multiple queues in the bottleneck or to a single queue ✨.

3. Deficiencies of Existing Mechanisms 🎀

Okay so, among the existing congestion control mechanisms, TCP Friendly Rate Control (TFRC) [RFC5348] is the one that claims to be suitable for real-time interactive media 🎬. TFRC is an equation-based congestion control mechanism. It gives a reasonably fair share of bandwidth when competing with TCP flows and has way smaller throughput variations than TCP. The catch is that it gets there through a slower response to available bandwidth changes than TCP does. TFRC is designed to perform best with applications that have a fixed packet size and no fixed period between sending packets.

TFRC detects loss events and reacts to congestion-caused loss by reducing its sending rate 📉. It lets applications increase the sending rate until loss is observed in the flows. As noted in the IAB/IRTF report [RFC7295], large buffers in network elements add extra delay to the communication 😩. That's why it matters to take every possible congestion indication into account. Looking at the current Internet deployment, TFRC's biggest deficiency is that it only considers loss events as a congestion indication 💀.

A typical real-time interactive communication includes live-encoded audio and video flow(s) 🎙️. In that scenario, an audio source typically needs a fixed interval between packets, and it varies the segment size of the packets instead of the packet rate in response to congestion, so it sends smaller packets. A variant of TFRC, Small-Packet TFRC (TFRC-SP) [RFC4828], addresses the issues related to this kind of source. A video source generally varies video frame sizes. It can produce large frames that need to be further fragmented to fit into the path Maximum Transmission Unit (MTU) size, and it has an almost fixed interval between producing frames at a certain frame rate. TFRC is known to be less optimal with video sources like this 📹.

There are also some mismatches between TFRC's design assumptions and how media sources in a typical real-time interactive application actually work 🙃. TFRC is designed to maintain a smooth sending rate, but media sources can change rates in steps for both increase and decrease. TFRC can operate in two modes: i) bytes per second and ii) packets per second, whereas typical real-time interactive media sources operate on bits per second. There are also limits on how quickly media sources can adapt to specific sending rates. Modern video encoders can vary the output bitrate a lot depending on how they're configured, the scene they're encoding, and more. So the video source won't always output at an allowable bitrate. TFRC tries to increase its sending rate when transmitting at the maximum allowed rate, and it only increases to twice the current transmission rate. That can cause issues when video sources vary their bitrates (not a vibe) 😬.

Moreover, there are a number of studies on TFRC that show its limitations, including TFRC's unfairness to low statistically multiplexed links, oscillatory behavior, performance issues in highly dynamic loss-rate conditions, and more [CH09] 📚.

Looking at all of these deficiencies, it can be concluded that the requirements for a congestion control mechanism for real-time interactive media cannot be met by TFRC as defined in the standard 🥲.

4. IANA Considerations 🌸

This document has no IANA actions ✨.

5. Security Considerations 🔥

Okay bestie, this section is SERIOUS 🚨

An attacker with the ability to delete, delay, or insert messages into the flow can fake congestion signals, unless they are passed on a tamper-proof path 😱. Since some possible algorithms depend on the timing of packet arrival, even a traditional, protected channel does not fully mitigate such attacks 🫠.

An attack that reduces bandwidth is not necessarily significant, since an on-path attacker could break the connection by discarding all packets 🤷‍♀️. Attacks that increase the perceived available bandwidth are conceivable and need to be evaluated. Such attacks could result in starvation of competing flows and permit amplification attacks 🚨.

Algorithm designers should consider the possibility of malicious on-path attackers 👀.

6. References 📚

6.1. Normative References 📚✨

Okay bestie, these are the must-reads, the ones the whole doc leans on 🫶

6.2. Informative References 🧠💖

Now the extra-credit reading, the stuff that adds context and backstory. Totally worth a look if you're curious 🌈

Acknowledgements 🫶🌸

This document is the result of discussions in various fora of the WebRTC effort, in particular on the <rtp-congestion@alvestrand.no> mailing list. Many people contributed their thoughts to this. The whole crew helped make it happen 💜

Authors' Addresses 🎀

Randell Jesup Mozilla United States of America

Email: randell-ietf@jesup.org

Zaheduzzaman Sarker (editor) Ericsson AB Torshamnsgatan 23 SE-164 83 Stockholm Sweden

Phone: +46 10 717 37 43 Email: zaheduzzaman.sarker@ericsson.com

📦 where this page comes from