]> University of Oslo

PO Box 1080 Blindern 0316 Oslo Norway michawe@ifi.uio.no http://welzl.at/

University of Washington

185 Stevens Way Seattle, WA 98195 United States tomh@tomh.org https://www.tomh.org/

University of Aberdeen

Fraser Noble Building Aberdeen, AB24 3UE UK gorry@erg.abdn.ac.uk https://www.erg.abdn.ac.uk/

Transport Congestion Control Working Group This document specifies how transport protocols increase their congestion window when the sender is rate-limited, and updates RFC 5681, RFC 9002, RFC 9260, and RFC 9438. Such a limitation can be caused by the sending application not supplying data or by receiver flow control. About This Document The latest revision of this draft can be found at . Status information for this document may be found at . Discussion of this document takes place on the Congestion Control Working Group Working Group mailing list (), which is archived at . Subscribe at . Source for this draft and an issue tracker can be found at .

Introduction A sender of a congestion controlled transport protocol becomes "rate-limited" when it does not send any data even though the congestion control rules would allow it to transmit data. This could occur because the application has not provided sufficient data to fully utilise the congestion window (cwnd). It could also occur because the receiver has limited the sender using flow control (e.g., by the advertised TCP receiver window (rwnd) or by the connection or stream flow credit in QUIC). Current RFCs specifying congestion control algorithms diverge regarding the rules for increasing the cwnd when the sender is rate-limited. Congestion Window Validation (CWV) provides an experimental specification defining how to manage a cwnd that has become larger than the current flight size. In contrast, this present document concerns the increase in cwnd when a sender is rate-limited. These two topics are distinct, but are related, because both describe the management of the cwnd when the sender does not fully utilise the current cwnd. This document specifies a uniform rule that congestion control algorithms MUST apply and provides a recommendation that congestion control implementations SHOULD follow. An appendix provides an overview of the divergence in current RFCs and some current implementations regarding cwnd increase when the sender is rate-limited.

Terminology This document uses the terms defined in . Additionally, we define:

• maxFS: the largest value of FlightSize since the last time that cwnd was decreased. If cwnd has never been decreased, maxFS is the maximum value of FlightSize since the start of the data transfer.

Conventions and Definitions The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in BCP 14 when, and only when, they appear in all capitals, as shown here.

Increase rules Irrespective of the current state of a congestion control algorithm, senders using a congestion controlled transport protocol:

1. MUST include a limit to the growth of cwnd when FlightSize < cwnd.
2. SHOULD limit cwnd when FlightSize < cwnd to be no larger than limit(maxFS).
3. MAY limit maxFS as min(maxFS, pipeACK), using "pipeACK" as defined in , when FlightSize < cwnd.

In rule #2, "limit()" is a function that returns the maximum cwnd value that would result from the congestion control algorithm within one RTT, based on the "maxFS" parameter. For example, for Slow Start, as specified in , limit(maxFS)=2*maxFS, such that equation 2 in becomes: Similarly, with rule #2 applied to Congestion Avoidance, limit(maxFS)=1+maxFS, such that equation 3 in becomes: As with cwnd, without a way to reduce it when the transport sender becomes rate-limited, rule #2 allows for maxFS to stay valid for a long time, possibly not reflecting the reality of the end-to-end Internet path in use. For cwnd, this is remedied by "Congestion Window Validation" in , which also defines a "pipeACK" variable that measures the acknowledged size of the network pipe when the sender is rate-limited. Accordingly, to implement CWV, rule #3 can be used.

Discussion If the sending rate is less than permitted by cwnd for multiple RTTs, limited either by the sending application or by the receiver-advertised window, continuously increasing the cwnd would cause a mismatch between the cwnd and the capacity that the path supports (i.e., over-estimating the capacity). Such unlimited growth in the cwnd is therefore disallowed by the first rule. However, in most common congestion control algorithms, in the absence of an indication of congestion, a cwnd that has been fully utilized during an RTT is permitted to be increased during the immediately following RTT. Thus, such an increase is allowed by the second rule.

Rate-based congestion control The present document updates congestion control specifications that use a congestion window (cwnd) to limit the number of unacknowledged packets a sender is allowed to emit. Use of a congestion window variable to control sending rate is not the only mechanism available and used in practice. Congestion control algorithms can also constrain data transmission by explicitly calculating the sending rate over some time interval, by "pacing" packets (injecting pauses in between their transmission) or via combinations of the above (e.g., BBR combines these three methods ). The guiding principle behind the rules in applies to all congestion control algorithms: in the absence of a congestion indication, a sender should be allowed to increase its rate from the amount of data that it has transmitted during the previous RTT. This holds irrespective of whether the sender is rate-limited or not.

Pacing Pacing mechanisms seek to avoid the negative impacts associated with "bursts" (flights of packets transmitted back-to-back). This is usually without limiting the number of packets that are sent per RTT. The present specification introduces a limitation using "maxFS", which is measured over an RTT; thus, as long as the number of packets per RTT is unaffected by pacing, the rules in also do not constrain the use of pacing mechanisms.

Security Considerations While congestion control designs could result in unwanted competing traffic, they do not directly result in new security considerations. Transport protocols that provide authentication (including those using encryption), or are carried over protocols that provide authentication, can protect their congestion control algorithm from network attack. This is orthogonal to the congestion control rules.

IANA Considerations This document requests no IANA action.

References Normative References This document provides a mechanism to address issues that arise when TCP is used for traffic that exhibits periods where the sending rate is limited by the application rather than the congestion window. It provides an experimental update to TCP that allows a TCP sender to restart quickly following a rate-limited interval. This method is expected to benefit applications that send rate-limited traffic using TCP while also providing an appropriate response if congestion is experienced. This document also evaluates the Experimental specification of TCP Congestion Window Validation (CWV) defined in RFC 2861 and concludes that RFC 2861 sought to address important issues but failed to deliver a widely used solution. This document therefore reclassifies the status of RFC 2861 from Experimental to Historic. This document obsoletes RFC 2861. This document defines TCP's four intertwined congestion control algorithms: slow start, congestion avoidance, fast retransmit, and fast recovery. In addition, the document specifies how TCP should begin transmission after a relatively long idle period, as well as discussing various acknowledgment generation methods. This document obsoletes RFC 2581. [STANDARDS-TRACK] In many standards track documents several words are used to signify the requirements in the specification. These words are often capitalized. This document defines these words as they should be interpreted in IETF documents. This document specifies an Internet Best Current Practices for the Internet Community, and requests discussion and suggestions for improvements. RFC 2119 specifies common key words that may be used in protocol specifications. This document aims to reduce the ambiguity by clarifying that only UPPERCASE usage of the key words have the defined special meanings. CUBIC is a standard TCP congestion control algorithm that uses a cubic function instead of a linear congestion window increase function to improve scalability and stability over fast and long-distance networks. CUBIC has been adopted as the default TCP congestion control algorithm by the Linux, Windows, and Apple stacks. This document updates the specification of CUBIC to include algorithmic improvements based on these implementations and recent academic work. Based on the extensive deployment experience with CUBIC, this document also moves the specification to the Standards Track and obsoletes RFC 8312. This document also updates RFC 5681, to allow for CUBIC's occasionally more aggressive sending behavior. This document describes the Stream Control Transmission Protocol (SCTP) and obsoletes RFC 4960. It incorporates the specification of the chunk flags registry from RFC 6096 and the specification of the I bit of DATA chunks from RFC 7053. Therefore, RFCs 6096 and 7053 are also obsoleted by this document. In addition, RFCs 4460 and 8540, which describe errata for SCTP, are obsoleted by this document. SCTP was originally designed to transport Public Switched Telephone Network (PSTN) signaling messages over IP networks. It is also suited to be used for other applications, for example, WebRTC. SCTP is a reliable transport protocol operating on top of a connectionless packet network, such as IP. It offers the following services to its users: The design of SCTP includes appropriate congestion avoidance behavior and resistance to flooding and masquerade attacks. This document describes loss detection and congestion control mechanisms for QUIC. This document contains the profile for Congestion Control Identifier 2 (CCID 2), TCP-like Congestion Control, in the Datagram Congestion Control Protocol (DCCP). CCID 2 should be used by senders who would like to take advantage of the available bandwidth in an environment with rapidly changing conditions, and who are able to adapt to the abrupt changes in the congestion window typical of TCP's Additive Increase Multiplicative Decrease (AIMD) congestion control. [STANDARDS-TRACK] This document describes a simple modification to TCP's congestion control algorithms to decay the congestion window cwnd after the transition from a sufficiently-long application-limited period, while using the slow-start threshold ssthresh to save information about the previous value of the congestion window. This memo defines an Experimental Protocol for the Internet community. Informative References Google Google Google Google Google This document specifies the BBR congestion control algorithm. BBR ("Bottleneck Bandwidth and Round-trip propagation time") uses recent measurements of a transport connection's delivery rate, round-trip time, and packet loss rate to build an explicit model of the network path. BBR then uses this model to control both how fast it sends data and the maximum volume of data it allows in flight in the network at any time. Relative to loss-based congestion control algorithms such as Reno [RFC5681] or CUBIC [RFC8312], BBR offers substantially higher throughput for bottlenecks with shallow buffers or random losses, and substantially lower queueing delays for bottlenecks with deep buffers (avoiding "bufferbloat"). BBR can be implemented in any transport protocol that supports packet-delivery acknowledgment. Thus far, open source implementations are available for TCP [RFC793] and QUIC [RFC9000]. This document specifies version 2 of the BBR algorithm, also sometimes referred to as BBRv2 or bbr2.

The state of RFCs and implementations This section is provided as input for IETF discussion, and should be removed before publication.

TCP ("Reno" congestion control)

Specification does not contain a rule to limit the growth of cwnd when the sender is rate-limited. This statement (page 8) gives an impression that such cwnd growth might be expected:

• Implementation Note: An easy mistake to make is to simply use cwnd, rather than FlightSize, which in some implementations may incidentally increase well beyond rwnd.

also suggests there is no increase limitation in the standard TCP behavior (which changes), on page 4:

• Standard TCP does not impose additional restrictions on the growth of the congestion window when a TCP sender is unable to send at the maximum rate allowed by the cwnd. In this case, the rate-limited sender may grow a cwnd far beyond that corresponding to the current transmit rate, resulting in a value that does not reflect current information about the state of the network path the flow is using.

Implementation

• ns-2 allows cwnd to grow when it is rate-limited by rwnd. (Rate-limited by the sending application: not tested.)
• Until release 3.42, ns-3 allowed cwnd to grow when rate-limited, either due to an application or rwnd limit. Since release 3.42, ns-3 TCP models conform to rule #2 in , following the current Linux TCP approach in this regard (see next bullet).
• In Congestion Avoidance, Linux only allows the cwnd to grow when the sender is unconstrained. Before kernel version 3.16, this also applied to Slow Start. The check for "unconstrained" is perfomed by checking if FlightSize is greater or equal to cwnd. Since kernel version 3.16, which was published in August 2014, in Slow Start, the increase implements rule #2 in in the tcp_is_cwnd_limited function in tcp.h.

Assessment Linux implements a limit to cwnd growth in accordance with rule #1 in ; in Slow Start, this limit follows rule #2, while in Congestion Avoidance, it is more conservative than rule #2. The specification and the ns-2 and (older) ns-3 implementations are in conflict with rules #1 and #2 in .

CUBIC

Specification says:

• Cubic doesn't increase cwnd when it's limited by the sending application or rwnd.

Implementation The description of Linux described in also applies to Cubic.

Assessment Both the specification and the Linux implementation limit the cwnd growth in accordance with rule #1 in ; in Congestion Avoidance, this limit is more conservative than rule #2 in , and in Slow Start, it implements rule #2 in .

SCTP

Specification says:

• When cwnd is less than or equal to ssthresh, an SCTP endpoint MUST use the slow-start algorithm to increase cwnd only if the current congestion window is being fully utilized and the data sender is not in Fast Recovery. Only when these two conditions are met can the cwnd be increased; otherwise, the cwnd MUST NOT be increased.

Assessment The quoted statement from prescribes the same cwnd growth limitation that is also specified for Cubic and implemented for both Reno and Cubic in Linux. It is in accordance with rule #1 in , and more conservative than rule #2 in . is specifically limited to Slow Start. Congestion Avoidance is discussed in However, this section neither contains a similar rule, nor does it refer back to the rule that limits the growth of cwnd in Section 7.2.1. It is thus implicitly clear that the quoted rule only applies to Slow Start, whereas the rules in apply to both Slow Start and Congestion Avoidance.

QUIC

Specification states:

• When bytes in flight is smaller than the congestion window and sending is not pacing limited, the congestion window is underutilized. This can happen due to insufficient application data or flow control limits. When this occurs, the congestion window SHOULD NOT be increased in either slow start or congestion avoidance.

Assessment With the exception of pacing, this specification conservatively limits the growth in cwnd, similar to Cubic and SCTP. It is in accordance with rule #1 in , and more conservative than rule #2 in .

DCCP CCID2

Specification states: >There are currently no standards governing TCP's use of the congestion window during an application-limited period. In particular, it is possible for TCP's congestion window to grow quite large during a long uncongested period when the sender is application limited, sending at a low rate. essentially suggests that TCP's congestion window not be increased during application-limited periods when the congestion window is not being fully utilized.

Assessment A DCCP Congestion Control ID (CCID) specifing TCP-like behaviour ought to follow the method specified in this document. The current guidance relates only to . The text in is updated by this document to specify the management of the cwnd during an application-limited period.

Change Log

• -00 was the first individual submission for feedback by CCWG.
• -01 includes editorial improvements
  • Removes application interaction with QUIC pacing, since pacing might be within the QUIC stack.
  • Adds explicit mention of DCCP/CCID2.
  • Adds this change log.
• -02 addresses comments from IETF-119
  • Discusses rate-based controls and pacing.
  • Trims the list of possible RFCs to update.
  • Some editorial fixes: "congestion control algorithm" instead of "mechanism" for consistency with RFC5033.bis; earlier definition of maxFS; explicit mention of RFCs to update in abstract.
• -03 addresses comments from IETF-120
  • Introduces a third rule, with MAY, that avoids having an unvalidated long-lived maxFS (using pipeACK from RFC 7661).
  • Changes "inc" to "limit" and adapts the wording of rule 2 to make it clearer (thanks to Neal Cardwell).
  • Appendix: updates ns-3 in line with the recent implementation.
  • Appendix: makes the RFC 9002 text clearer and shorter.

Acknowledgments The authors would like to thank Neal Cardwell for suggesting improvements to this document.