draft-ietf-tsvwg-aqm-dualq-coupled-10.txt   draft-ietf-tsvwg-aqm-dualq-coupled-11.txt 
Transport Area working group (tsvwg) K. De Schepper Transport Area working group (tsvwg) K. De Schepper
Internet-Draft Nokia Bell Labs Internet-Draft Nokia Bell Labs
Intended status: Experimental B. Briscoe, Ed. Intended status: Experimental B. Briscoe, Ed.
Expires: January 9, 2020 G. White Expires: September 10, 2020 Independent
G. White
CableLabs CableLabs
July 8, 2019 March 9, 2020
DualQ Coupled AQMs for Low Latency, Low Loss and Scalable Throughput DualQ Coupled AQMs for Low Latency, Low Loss and Scalable Throughput
(L4S) (L4S)
draft-ietf-tsvwg-aqm-dualq-coupled-10 draft-ietf-tsvwg-aqm-dualq-coupled-11
Abstract Abstract
The Low Latency Low Loss Scalable Throughput (L4S) architecture The Low Latency Low Loss Scalable Throughput (L4S) architecture
allows data flows over the public Internet to achieve consistent allows data flows over the public Internet to achieve consistent low
ultra-low queuing latency, generally zero congestion loss and scaling queuing latency, generally zero congestion loss and scaling of per-
of per-flow throughput without the scaling problems of traditional flow throughput without the scaling problems of standard TCP Reno-
TCP. To achieve this, L4S data flows have to use one of the family friendly congestion controls. To achieve this, L4S data flows have
of 'Scalable' congestion controls (Data Centre TCP and TCP Prague are to use one of the family of 'Scalable' congestion controls (TCP
examples) and a form of Explicit Congestion Notification (ECN) with Prague and Data Center TCP are examples) and a form of Explicit
modified behaviour. However, until now, Scalable congestion controls Congestion Notification (ECN) with modified behaviour. However,
did not co-exist with existing TCP Reno/Cubic traffic --- Scalable until now, Scalable congestion controls did not co-exist with
controls are so aggressive that 'Classic' TCP algorithms drive existing Reno/Cubic traffic --- Scalable controls are so aggressive
themselves to a small capacity share. Therefore, until now, L4S that 'Classic' (e.g. Reno-friendly) algorithms sharing an ECN-
controls could only be deployed where a clean-slate environment could capable queue would drive themselves to a small capacity share.
be arranged, such as in private data centres (hence the name DCTCP). Therefore, until now, L4S controls could only be deployed where a
This specification defines `DualQ Coupled Active Queue Management clean-slate environment could be arranged, such as in private data
(AQM)', which enables these Scalable congestion controls to safely centres (hence the name DCTCP). This specification defines `DualQ
co-exist with Classic Internet traffic. Coupled Active Queue Management (AQM)', which enables Scalable
congestion controls that comply with the Prague L4S requirements to
co-exist safely with Classic Internet traffic.
Analytical study and implementation testing of the Coupled AQM have Analytical study and implementation testing of the Coupled AQM have
shown that Scalable and Classic flows competing under similar shown that Scalable and Classic flows competing under similar
conditions run at roughly the same rate. It achieves this conditions run at roughly the same rate. It achieves this
indirectly, without having to inspect transport layer flow indirectly, without having to inspect transport layer flow
identifiers. When tested in a residential broadband setting, DCTCP identifiers. When tested in a residential broadband setting, DCTCP
also achieves sub-millisecond average queuing delay and zero also achieves sub-millisecond average queuing delay and zero
congestion loss under a wide range of mixes of DCTCP and `Classic' congestion loss under a wide range of mixes of DCTCP and `Classic'
broadband Internet traffic, without compromising the performance of broadband Internet traffic, without compromising the performance of
the Classic traffic. The solution also reduces network complexity the Classic traffic. The solution has low complexity and requires no
and requires no configuration for the public Internet. configuration for the public Internet.
Status of This Memo Status of This Memo
This Internet-Draft is submitted in full conformance with the This Internet-Draft is submitted in full conformance with the
provisions of BCP 78 and BCP 79. provisions of BCP 78 and BCP 79.
Internet-Drafts are working documents of the Internet Engineering Internet-Drafts are working documents of the Internet Engineering
Task Force (IETF). Note that other groups may also distribute Task Force (IETF). Note that other groups may also distribute
working documents as Internet-Drafts. The list of current Internet- working documents as Internet-Drafts. The list of current Internet-
Drafts is at https://datatracker.ietf.org/drafts/current/. Drafts is at https://datatracker.ietf.org/drafts/current/.
Internet-Drafts are draft documents valid for a maximum of six months Internet-Drafts are draft documents valid for a maximum of six months
and may be updated, replaced, or obsoleted by other documents at any and may be updated, replaced, or obsoleted by other documents at any
time. It is inappropriate to use Internet-Drafts as reference time. It is inappropriate to use Internet-Drafts as reference
material or to cite them other than as "work in progress." material or to cite them other than as "work in progress."
This Internet-Draft will expire on January 9, 2020. This Internet-Draft will expire on September 10, 2020.
Copyright Notice Copyright Notice
Copyright (c) 2019 IETF Trust and the persons identified as the Copyright (c) 2020 IETF Trust and the persons identified as the
document authors. All rights reserved. document authors. All rights reserved.
This document is subject to BCP 78 and the IETF Trust's Legal This document is subject to BCP 78 and the IETF Trust's Legal
Provisions Relating to IETF Documents Provisions Relating to IETF Documents
(https://trustee.ietf.org/license-info) in effect on the date of (https://trustee.ietf.org/license-info) in effect on the date of
publication of this document. Please review these documents publication of this document. Please review these documents
carefully, as they describe your rights and restrictions with respect carefully, as they describe your rights and restrictions with respect
to this document. Code Components extracted from this document must to this document. Code Components extracted from this document must
include Simplified BSD License text as described in Section 4.e of include Simplified BSD License text as described in Section 4.e of
the Trust Legal Provisions and are provided without warranty as the Trust Legal Provisions and are provided without warranty as
described in the Simplified BSD License. described in the Simplified BSD License.
Table of Contents Table of Contents
1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . 3 1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . 3
1.1. Outline of the Problem . . . . . . . . . . . . . . . . . 3 1.1. Outline of the Problem . . . . . . . . . . . . . . . . . 3
1.2. Scope . . . . . . . . . . . . . . . . . . . . . . . . . . 5 1.2. Scope . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.3. Terminology . . . . . . . . . . . . . . . . . . . . . . . 7 1.3. Terminology . . . . . . . . . . . . . . . . . . . . . . . 7
1.4. Features . . . . . . . . . . . . . . . . . . . . . . . . 8 1.4. Features . . . . . . . . . . . . . . . . . . . . . . . . 9
2. DualQ Coupled AQM . . . . . . . . . . . . . . . . . . . . . . 9 2. DualQ Coupled AQM . . . . . . . . . . . . . . . . . . . . . . 10
2.1. Coupled AQM . . . . . . . . . . . . . . . . . . . . . . . 9 2.1. Coupled AQM . . . . . . . . . . . . . . . . . . . . . . . 10
2.2. Dual Queue . . . . . . . . . . . . . . . . . . . . . . . 10 2.2. Dual Queue . . . . . . . . . . . . . . . . . . . . . . . 12
2.3. Traffic Classification . . . . . . . . . . . . . . . . . 11 2.3. Traffic Classification . . . . . . . . . . . . . . . . . 12
2.4. Overall DualQ Coupled AQM Structure . . . . . . . . . . . 11 2.4. Overall DualQ Coupled AQM Structure . . . . . . . . . . . 13
2.5. Normative Requirements for a DualQ Coupled AQM . . . . . 14 2.5. Normative Requirements for a DualQ Coupled AQM . . . . . 16
2.5.1. Functional Requirements . . . . . . . . . . . . . . . 14 2.5.1. Functional Requirements . . . . . . . . . . . . . . . 16
2.5.1.1. Requirements in Unexpected Cases . . . . . . . . 15 2.5.1.1. Requirements in Unexpected Cases . . . . . . . . 17
2.5.2. Management Requirements . . . . . . . . . . . . . . . 16 2.5.2. Management Requirements . . . . . . . . . . . . . . . 18
2.5.2.1. Configuration . . . . . . . . . . . . . . . . . . 16 2.5.2.1. Configuration . . . . . . . . . . . . . . . . . . 18
2.5.2.2. Monitoring . . . . . . . . . . . . . . . . . . . 18 2.5.2.2. Monitoring . . . . . . . . . . . . . . . . . . . 19
2.5.2.3. Anomaly Detection . . . . . . . . . . . . . . . . 18 2.5.2.3. Anomaly Detection . . . . . . . . . . . . . . . . 20
2.5.2.4. Deployment, Coexistence and Scaling . . . . . . . 19 2.5.2.4. Deployment, Coexistence and Scaling . . . . . . . 20
3. IANA Considerations . . . . . . . . . . . . . . . . . . . . . 19 3. IANA Considerations . . . . . . . . . . . . . . . . . . . . . 21
4. Security Considerations . . . . . . . . . . . . . . . . . . . 19 4. Security Considerations . . . . . . . . . . . . . . . . . . . 21
4.1. Overload Handling . . . . . . . . . . . . . . . . . . . . 19 4.1. Overload Handling . . . . . . . . . . . . . . . . . . . . 21
4.1.1. Avoiding Classic Starvation: Sacrifice L4S Throughput 4.1.1. Avoiding Classic Starvation: Sacrifice L4S Throughput
or Delay? . . . . . . . . . . . . . . . . . . . . . . 20 or Delay? . . . . . . . . . . . . . . . . . . . . . . 21
4.1.2. Congestion Signal Saturation: Introduce L4S Drop or 4.1.2. Congestion Signal Saturation: Introduce L4S Drop or
Delay? . . . . . . . . . . . . . . . . . . . . . . . 21 Delay? . . . . . . . . . . . . . . . . . . . . . . . 23
4.1.3. Protecting against Unresponsive ECN-Capable Traffic . 22 4.1.3. Protecting against Unresponsive ECN-Capable Traffic . 24
5. Acknowledgements . . . . . . . . . . . . . . . . . . . . . . 22 5. Acknowledgements . . . . . . . . . . . . . . . . . . . . . . 24
6. Contributors . . . . . . . . . . . . . . . . . . . . . . . . 23 6. Contributors . . . . . . . . . . . . . . . . . . . . . . . . 24
7. References . . . . . . . . . . . . . . . . . . . . . . . . . 23 7. References . . . . . . . . . . . . . . . . . . . . . . . . . 25
7.1. Normative References . . . . . . . . . . . . . . . . . . 23 7.1. Normative References . . . . . . . . . . . . . . . . . . 25
7.2. Informative References . . . . . . . . . . . . . . . . . 24 7.2. Informative References . . . . . . . . . . . . . . . . . 26
Appendix A. Example DualQ Coupled PI2 Algorithm . . . . . . . . 27 Appendix A. Example DualQ Coupled PI2 Algorithm . . . . . . . . 30
A.1. Pass #1: Core Concepts . . . . . . . . . . . . . . . . . 28 A.1. Pass #1: Core Concepts . . . . . . . . . . . . . . . . . 31
A.2. Pass #2: Overload Details . . . . . . . . . . . . . . . . 36 A.2. Pass #2: Overload Details . . . . . . . . . . . . . . . . 39
Appendix B. Example DualQ Coupled Curvy RED Algorithm . . . . . 40 Appendix B. Example DualQ Coupled Curvy RED Algorithm . . . . . 43
B.1. Curvy RED in Pseudocode . . . . . . . . . . . . . . . . . 40 B.1. Curvy RED in Pseudocode . . . . . . . . . . . . . . . . . 43
B.2. Efficient Implementation of Curvy RED . . . . . . . . . . 46 B.2. Efficient Implementation of Curvy RED . . . . . . . . . . 49
Appendix C. Guidance on Controlling Throughput Equivalence . . . 48 Appendix C. Choice of Coupling Factor, k . . . . . . . . . . . . 51
Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . 49 C.1. RTT-Dependence . . . . . . . . . . . . . . . . . . . . . 51
C.2. Guidance on Controlling Throughput Equivalence . . . . . 52
Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . 53
1. Introduction 1. Introduction
This document specifies a framework for DualQ Coupled AQMs, which is This document specifies a framework for DualQ Coupled AQMs, which is
the network part of the L4S architecture [I-D.ietf-tsvwg-l4s-arch]. the network part of the L4S architecture [I-D.ietf-tsvwg-l4s-arch].
L4S enables both ultra-low queuing latency and high throughput at the L4S enables both ultra-low queuing latency (sub-millisecond on
same time, for ad hoc numbers of capacity-seeking applications all average) and high throughput at the same time, for ad hoc numbers of
sharing the same capacity. capacity-seeking applications all sharing the same capacity.
1.1. Outline of the Problem 1.1. Outline of the Problem
Latency is becoming the critical performance factor for many (most?) Latency is becoming the critical performance factor for many (most?)
applications on the public Internet, e.g. interactive Web, Web applications on the public Internet, e.g. interactive Web, Web
services, voice, conversational video, interactive video, interactive services, voice, conversational video, interactive video, interactive
remote presence, instant messaging, online gaming, remote desktop, remote presence, instant messaging, online gaming, remote desktop,
cloud-based applications, and video-assisted remote control of cloud-based applications, and video-assisted remote control of
machinery and industrial processes. In the developed world, further machinery and industrial processes. In the developed world, further
increases in access network bit-rate offer diminishing returns, increases in access network bit-rate offer diminishing returns,
skipping to change at page 4, line 20 skipping to change at page 4, line 28
progress. L4S involves a recognition that both approaches are progress. L4S involves a recognition that both approaches are
yielding diminishing returns: yielding diminishing returns:
o Recent state-of-the-art active queue management (AQM) in the o Recent state-of-the-art active queue management (AQM) in the
network, e.g. fq_CoDel [RFC8290], PIE [RFC8033], Adaptive network, e.g. fq_CoDel [RFC8290], PIE [RFC8033], Adaptive
RED [ARED01] ) has reduced queuing delay for all traffic, not just RED [ARED01] ) has reduced queuing delay for all traffic, not just
a select few applications. However, no matter how good the AQM, a select few applications. However, no matter how good the AQM,
the capacity-seeking (sawtoothing) rate of TCP-like congestion the capacity-seeking (sawtoothing) rate of TCP-like congestion
controls represents a lower limit that will either cause queuing controls represents a lower limit that will either cause queuing
delay to vary or cause the link to be under-utilized. These AQMs delay to vary or cause the link to be under-utilized. These AQMs
are tuned to allow a typical capacity-seeking TCP-Friendly flow to are tuned to allow a typical capacity-seeking Reno-friendly flow
induce an average queue that roughly doubles the base RTT, adding to induce an average queue that roughly doubles the base RTT,
5-15 ms of queuing on average (cf. 500 microseconds with L4S for adding 5-15 ms of queuing on average (cf. 500 microseconds with
the same mix of long-running and web traffic). However, for many L4S for the same mix of long-running and web traffic). However,
applications low delay is not useful unless it is consistently for many applications low delay is not useful unless it is
low. With these AQMs, 99th percentile queuing delay is 20-30 ms consistently low. With these AQMs, 99th percentile queuing delay
(cf. 2 ms with the same traffic over L4S). is 20-30 ms (cf. 2 ms with the same traffic over L4S).
o Similarly, recent research into using e2e congestion control o Similarly, recent research into using e2e congestion control
without needing an AQM in the network (e.g.BBRv1 [BBRv1]) seems to without needing an AQM in the network (e.g.BBRv1 [BBRv1]) seems to
have hit a similar lower limit to queuing delay of about 20ms on have hit a similar lower limit to queuing delay of about 20ms on
average (and any additional BBRv1 flow adds another 20ms of average (and any additional BBRv1 flow adds another 20ms of
queuing) but there are also regular 25ms delay spikes due to queuing) but there are also regular 25ms delay spikes due to
bandwidth probes and 60ms spikes due to flow-starts. bandwidth probes and 60ms spikes due to flow-starts.
L4S learns from the experience of Data Center TCP [RFC8257], which L4S learns from the experience of Data Center TCP [RFC8257], which
shows the power of complementary changes both in the network and on shows the power of complementary changes both in the network and on
end-systems. DCTCP teaches us that two small but radical changes to end-systems. DCTCP teaches us that two small but radical changes to
congestion control are needed to cut the two major outstanding causes congestion control are needed to cut the two major outstanding causes
of queuing delay variability: of queuing delay variability:
1. Far smaller rate variations (sawteeth) than TCP-Friendly 1. Far smaller rate variations (sawteeth) than Reno-friendly
congestion controls; congestion controls;
2. A shift of smoothing and hence smoothing delay from network to 2. A shift of smoothing and hence smoothing delay from network to
sender. sender.
Without the former, a 'Classic' flow's round trip time (RTT) varies Without the former, a 'Classic' (e.g. Reno-friendly) flow's round
between roughly 1 and 2 times the base RTT between the machines in trip time (RTT) varies between roughly 1 and 2 times the base RTT
question. Without the latter a 'Classic' flow's response to changing between the machines in question. Without the latter a 'Classic'
events is delayed by a worst-case (transcontinental) RTT, which could flow's response to changing events is delayed by a worst-case
be hundreds of times the actual smoothing delay needed for the RTT of (transcontinental) RTT, which could be hundreds of times the actual
typical traffic from localized CDNs. smoothing delay needed for the RTT of typical traffic from localized
CDNs.
These changes are the two main features of the family of so-called These changes are the two main features of the family of so-called
'Scalable' congestion controls (which includes DCTCP). Both these 'Scalable' congestion controls (which includes DCTCP). Both these
changes only reduce delay in combination with a complementary change changes only reduce delay in combination with a complementary change
in the network and they are both only feasible with ECN, not drop, in the network and they are both only feasible with ECN, not drop,
for the signalling: for the signalling:
1. The smaller sawteeth need an extremely shallow ECN packet-marking 1. The smaller sawteeth allow an extremely shallow ECN packet-
threshold in the queue. marking threshold in the queue.
2. And no smoothing in the network means that every fluctuation of 2. And no smoothing in the network means that every fluctuation of
the queue is signalled immediately. the queue is signalled immediately.
Without ECN, either of these would lead to very high loss levels. Without ECN, either of these would lead to very high loss levels.
But, with ECN, the resulting high marking levels are fine. But, with ECN, the resulting high marking levels are just signals,
not impairments.
However, until now, Scalable congestion controls (like DCTCP) did not However, until now, Scalable congestion controls (like DCTCP) did not
co-exist with existing ECN-capable TCP Reno [RFC5681] or Cubic co-exist well in a shared ECN-capable queue with existing ECN-capable
[RFC8312] traffic --- Scalable controls are so aggressive that these TCP Reno [RFC5681] or Cubic [RFC8312] congestion controls ---
'Classic' TCP algorithms drive themselves to a small capacity share. Scalable controls are so aggressive that these 'Classic' algorithms
Therefore, until now, L4S controls could only be deployed where a would drive themselves to a small capacity share. Therefore, until
clean-slate environment could be arranged, such as in private data now, L4S controls could only be deployed where a clean-slate
centres (hence the name DCTCP). environment could be arranged, such as in private data centres (hence
the name DCTCP).
This document specifies a `DualQ Coupled AQM' extension that solves This document specifies a `DualQ Coupled AQM' extension that solves
the problem of coexistence between Scalable and Classic flows, the problem of coexistence between Scalable and Classic flows,
without having to inspect flow identifiers. It is not like flow- without having to inspect flow identifiers. It is not like flow-
queuing approaches [RFC8290] that classify packets by flow identifier queuing approaches [RFC8290] that classify packets by flow identifier
into separate queues in order to isolate sparse flows from the higher into separate queues in order to isolate sparse flows from the higher
latency in the queues assigned to heavier flows. If a flow needs latency in the queues assigned to heavier flows. If a flow needs
both low delay and high throughput, having a queue to itself does not both low delay and high throughput, having a queue to itself does not
isolate it from the harm it causes to itself. In contrast, L4S isolate it from the harm it causes to itself. In contrast, L4S
addresses the root cause of the latency problem --- it is an enabler addresses the root cause of the latency problem --- it is an enabler
skipping to change at page 5, line 50 skipping to change at page 6, line 13
controls, so that every packet in every flow can enjoy very low controls, so that every packet in every flow can enjoy very low
latency, then there is no need to isolate each flow into a separate latency, then there is no need to isolate each flow into a separate
queue. queue.
1.2. Scope 1.2. Scope
L4S involves complementary changes in the network and on end-systems: L4S involves complementary changes in the network and on end-systems:
Network: A DualQ Coupled AQM (defined in the present document); Network: A DualQ Coupled AQM (defined in the present document);
End-system: A Scalable congestion control (defined in Section 2.1. End-system: A Scalable congestion control (defined in Section 2.1).
Packet identifier: The network and end-system parts of L4S can be Packet identifier: The network and end-system parts of L4S can be
deployed incrementally, because they both identify L4S packets deployed incrementally, because they both identify L4S packets
using the experimentally assigned explicit congestion notification using the experimentally assigned explicit congestion notification
(ECN) codepoints in the IP header: ECT(1) and CE [RFC8311] (ECN) codepoints in the IP header: ECT(1) and CE [RFC8311]
[I-D.ietf-tsvwg-ecn-l4s-id]. [I-D.ietf-tsvwg-ecn-l4s-id].
Data Center TCP (DCTCP [RFC8257]) is an example of a Scalable Data Center TCP (DCTCP [RFC8257]) is an example of a Scalable
congestion control that has been deployed for some time in Linux, congestion control that has been deployed for some time in Linux,
Windows and FreeBSD operating systems and Relentless TCP [Mathis09] Windows and FreeBSD operating systems and Relentless TCP [Mathis09]
is another example. During the progress of this document through the is another example. During the progress of this document through the
IETF a number of other Scalable congestion controls were implemented, IETF a number of other Scalable congestion controls were implemented,
e.g. TCP Prague [PragueLinux], QUIC Prague and the L4S variant of e.g. TCP Prague [PragueLinux], QUIC Prague and the L4S variant of
SCREAM for real-time media [RFC8298]. (Note: after the v3.19 Linux SCREAM for real-time media [RFC8298]. (Note: after the v3.19 Linux
kernel, bugs were introduced into DCTCP's scalable behaviour and not kernel, bugs were introduced into DCTCP's scalable behaviour and not
all the patches applied for L4S evaluation had been applied to the all the patches applied for L4S evaluation had been applied to the
mainline Linux kernel, which was at v5.2 at the time of writing). mainline Linux kernel, which was at v5.5 at the time of writing. TCP
Prague includes these patches and is available for all these Linux
kernels).
The focus of this specification is to get the network part of the L4S The focus of this specification is to enable deployment of the
service in place. Then, without any management intervention, network part of the L4S service. Then, without any management
applications can exploit this new network capability as their intervention, applications can exploit this new network capability as
operating systems migrate to Scalable congestion controls, which can their operating systems migrate to Scalable congestion controls,
then evolve _while_ their benefits are being enjoyed by everyone on which can then evolve _while_ their benefits are being enjoyed by
the Internet. everyone on the Internet.
The DualQ Coupled AQM framework can incorporate any AQM designed for The DualQ Coupled AQM framework can incorporate any AQM designed for
a single queue that generates a statistical or deterministic mark/ a single queue that generates a statistical or deterministic mark/
drop probability driven by the queue dynamics. Pseudocode examples drop probability driven by the queue dynamics. Pseudocode examples
of two different DualQ Coupled AQMs are given the appendices. In of two different DualQ Coupled AQMs are given in the appendices. In
many cases the framework simplifies the basic control algorithm, and many cases the framework simplifies the basic control algorithm, and
requires little extra processing. Therefore it is believed the requires little extra processing. Therefore it is believed the
Coupled AQM would be applicable and easy to deploy in all types of Coupled AQM would be applicable and easy to deploy in all types of
buffers; buffers in cost-reduced mass-market residential equipment; buffers; buffers in cost-reduced mass-market residential equipment;
buffers in end-system stacks; buffers in carrier-scale equipment buffers in end-system stacks; buffers in carrier-scale equipment
including remote access servers, routers, firewalls and Ethernet including remote access servers, routers, firewalls and Ethernet
switches; buffers in network interface cards, buffers in virtualized switches; buffers in network interface cards, buffers in virtualized
network appliances, hypervisors, and so on. network appliances, hypervisors, and so on.
For the public Internet, nearly all the benefit will typically be For the public Internet, nearly all the benefit will typically be
skipping to change at page 7, line 7 skipping to change at page 7, line 21
office, a campus or mobile user equipment. office, a campus or mobile user equipment.
Latency is not the only concern of L4S: Latency is not the only concern of L4S:
o The 'Low Loss" part of the name denotes that L4S generally o The 'Low Loss" part of the name denotes that L4S generally
achieves zero congestion loss (which would otherwise cause achieves zero congestion loss (which would otherwise cause
retransmission delays), due to its use of ECN. retransmission delays), due to its use of ECN.
o The "Scalable throughput" part of the name denotes that the per- o The "Scalable throughput" part of the name denotes that the per-
flow throughput of Scalable congestion controls should scale flow throughput of Scalable congestion controls should scale
indefinitely, avoiding the imminent scaling problems with TCP- indefinitely, avoiding the imminent scaling problems with 'TCP-
Friendly congestion control algorithms [RFC3649]. Friendly' congestion control algorithms [RFC3649].
The former is clearly in scope of this AQM document. However, the The former is clearly in scope of this AQM document. However, the
latter is an outcome of the end-system behaviour, and therefore latter is an outcome of the end-system behaviour, and therefore
outside the scope of this AQM document, even though the AQM is an outside the scope of this AQM document, even though the AQM is an
enabler. enabler.
The overall L4S architecture [I-D.ietf-tsvwg-l4s-arch] gives more The overall L4S architecture [I-D.ietf-tsvwg-l4s-arch] gives more
detail, including on wider deployment aspects such as backwards detail, including on wider deployment aspects such as backwards
compatibility of Scalable congestion controls in bottlenecks where a compatibility of Scalable congestion controls in bottlenecks where a
DualQ Coupled AQM has not been deployed. The supporting papers [PI2] DualQ Coupled AQM has not been deployed. The supporting papers [PI2]
skipping to change at page 7, line 33 skipping to change at page 7, line 47
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT",
"SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this
document are to be interpreted as described in [RFC2119] when, and document are to be interpreted as described in [RFC2119] when, and
only when, they appear in all capitals, as shown here. only when, they appear in all capitals, as shown here.
The DualQ Coupled AQM uses two queues for two services. Each of the The DualQ Coupled AQM uses two queues for two services. Each of the
following terms identifies both the service and the queue that following terms identifies both the service and the queue that
provides the service: provides the service:
Classic (denoted by subscript C): The `Classic' service is intended Classic service/queue: The Classic service is intended for all the
for all the behaviours that currently co-exist with TCP Reno (TCP congestion control behaviours that co-exist with Reno [RFC5681]
Cubic, Compound, SCTP, etc). (e.g. Reno itself, Cubic [RFC8312], TFRC [RFC5348]).
Low-Latency, Low-Loss and Scalable (L4S, denoted by subscript L): Low-Latency, Low-Loss Scalable throughput (L4S) service/queue: The
The `L4S' service is intended for a set of congestion controls 'L4S' service is intended for traffic from scalable congestion
with scalable properties, such as TCP Prague and DCTCP. For the control algorithms, such as Data Center TCP [RFC8257]. The L4S
public Internet an L4S transport has to comply with the service is for more general traffic than just DCTCP--it allows the
requirements in Section 4 of [I-D.ietf-tsvwg-ecn-l4s-id] (aka. set of congestion controls with similar scaling properties to
the 'Prague L4S requirements'). DCTCP to evolve (e.g. Relentless TCP [Mathis09], TCP Prague
[PragueLinux] and the L4S variant of SCREAM for real-time media
[RFC8298]).
Either service can cope with a proportion of unresponsive or less- Classic Congestion Control: A congestion control behaviour that can
responsive traffic as well, as long (e.g. DNS, VoIP, game sync co-exist with standard TCP Reno [RFC5681] without causing
datagrams, etc), just as a single queue AQM can if this traffic makes significantly negative impact on its flow rate [RFC5033]. With
minimal contribution to queuing. The DualQ Coupled AQM behaviour Classic congestion controls, as flow rate scales, the number of
below is defined to be similar to a single FIFO queue with respect to round trips between congestion signals (losses or ECN marks) rises
unresponsive and overload traffic. with the flow rate. So it takes longer and longer to recover
after each congestion event. Therefore control of queuing and
utilization becomes very slack, and the slightest disturbance
prevents a high rate from being attained [RFC3649].
Scalable Congestion Control: A congestion control where the average
time from one congestion signal to the next (the recovery time)
remains invariant as the flow rate scales, all other factors being
equal. This maintains the same degree of control over queueing
and utilization whatever the flow rate, as well as ensuring that
high throughput is robust to disturbances. For instance, DCTCP
averages 2 congestion signals per round-trip whatever the flow
rate. For the public Internet a Scalable transport has to comply
with the requirements in Section 4 of [I-D.ietf-tsvwg-ecn-l4s-id]
(aka. the 'Prague L4S requirements').
C: Abbreviation for Classic, e.g. when used as a subscript.
L: Abbreviation for L4S, e.g. when used as a subscript.
The terms Classic or L4S can also qualify other nouns, such as
'codepoint', 'identifier', 'classification', 'packet', 'flow'.
For example: an L4S packet means a packet with an L4S identifier
sent from an L4S congestion control.
Both Classic and L4S queues can cope with a proportion of
unresponsive or less-responsive traffic as well (e.g. DNS, VoIP,
game sync datagrams), just as a single queue AQM can if this
traffic makes minimal contribution to queuing. The DualQ Coupled
AQM behaviour is defined to be similar to a single FIFO queue with
respect to unresponsive and overload traffic.
Reno-friendly: The subset of Classic traffic that excludes
unresponsive traffic and excludes experimental congestion controls
intended to coexist with Reno but without always being strictly
friendly to it (as allowed by [RFC5033]). Reno-friendly is used
in place of 'TCP-friendly', given that the TCP protocol is used
with many different congestion control behaviours.
Classic ECN: The original Explicit Congestion Notification (ECN)
protocol [RFC3168], which requires ECN signals to be treated the
same as drops, both when generated in the network and when
responded to by the sender.
The names used for the four codepoints of the 2-bit IP-ECN field
are as defined in [RFC3168]: Not ECT, ECT(0), ECT(1) and CE, where
ECT stands for ECN-Capable Transport and CE stands for Congestion
Experienced.
1.4. Features 1.4. Features
The AQM couples marking and/or dropping from the Classic queue to the The AQM couples marking and/or dropping from the Classic queue to the
L4S queue in such a way that a flow will get roughly the same L4S queue in such a way that a flow will get roughly the same
throughput whichever it uses. Therefore both queues can feed into throughput whichever it uses. Therefore both queues can feed into
the full capacity of a link and no rates need to be configured for the full capacity of a link and no rates need to be configured for
the queues. The L4S queue enables Scalable congestion controls like the queues. The L4S queue enables Scalable congestion controls like
DCTCP or TCP Prague to give stunningly low and predictably low DCTCP or TCP Prague to give ultra-low and predictably low latency,
latency, without compromising the performance of competing 'Classic' without compromising the performance of competing 'Classic' Internet
Internet traffic. traffic.
Thousands of tests have been conducted in a typical fixed residential Thousands of tests have been conducted in a typical fixed residential
broadband setting. Experiments used a range of base round trip broadband setting. Experiments used a range of base round trip
delays up to 100ms and link rates up to 200 Mb/s between the data delays up to 100ms and link rates up to 200 Mb/s between the data
centre and home network, with varying amounts of background traffic centre and home network, with varying amounts of background traffic
in both queues. For every L4S packet, the AQM kept the average in both queues. For every L4S packet, the AQM kept the average
queuing delay below 1ms (or 2 packets where serialization delay queuing delay below 1ms (or 2 packets where serialization delay
exceeded 1ms on slower links), with 99th percentile no worse than exceeded 1ms on slower links), with 99th percentile no worse than
2ms. No losses at all were introduced by the L4S AQM. Details of 2ms. No losses at all were introduced by the L4S AQM. Details of
the extensive experiments are available [PI2] [DCttH15]. the extensive experiments are available [PI2] [DCttH15].
skipping to change at page 9, line 12 skipping to change at page 10, line 26
Classic queue that can be 'overtaken'. It gives low latency to L4S Classic queue that can be 'overtaken'. It gives low latency to L4S
traffic whether or not there is Classic traffic, and the latency of traffic whether or not there is Classic traffic, and the latency of
Classic traffic does not suffer when a proportion of the traffic is Classic traffic does not suffer when a proportion of the traffic is
L4S. L4S.
The two queues are only necessary because: The two queues are only necessary because:
o the large variations (sawteeth) of Classic flows need roughly a o the large variations (sawteeth) of Classic flows need roughly a
base RTT of queuing delay to ensure full utilization base RTT of queuing delay to ensure full utilization
o while Scalable flows do not need a queue to keep utilization high, o Scalable flows do not need a queue to keep utilization high, but
but they cannot keep latency predictably low if they are mixed they cannot keep latency predictably low if they are mixed with
with legacy TCP flows, Classic traffic,
The L4S queue has latency priority, but the coupling from the Classic The L4S queue has latency priority, but the coupling from the Classic
to the L4S AQM (explained below) ensures that it does not have to the L4S AQM (explained below) ensures that it does not have
bandwidth priority over the Classic queue. bandwidth priority over the Classic queue.
2. DualQ Coupled AQM 2. DualQ Coupled AQM
There are two main aspects to the approach: There are two main aspects to the approach:
o the Coupled AQM that addresses throughput equivalence between o the Coupled AQM that addresses throughput equivalence between
Classic (e.g. Reno, Cubic) flows and L4S flows (that satisfy the Classic (e.g. Reno, Cubic) flows and L4S flows (that satisfy the
Prague L4S requirements). Prague L4S requirements).
o the Dual Queue structure that provides latency separation for L4S o the Dual Queue structure that provides latency separation for L4S
flows to isolate them from the typically large Classic queue. flows to isolate them from the typically large Classic queue.
2.1. Coupled AQM 2.1. Coupled AQM
In the 1990s, the `TCP formula' was derived for the relationship In the 1990s, the `TCP formula' was derived for the relationship
between TCP's congestion window, cwnd, and its drop probability, p. between the steady-state congestion window, cwnd, and the drop
To a first order approximation, cwnd of TCP Reno is inversely probability, p of standard Reno congestion control [RFC5681] . To a
first order approximation, the steady-state cwnd of Reno is inversely
proportional to the square root of p. proportional to the square root of p.
The design focuses on Reno as the worst case, because if it does no The design focuses on Reno as the worst case, because if it does no
harm to Reno, it will not harm Cubic or any traffic designed to be harm to Reno, it will not harm Cubic or any traffic designed to be
friendly to Reno. TCP Cubic implements a Reno-compatibility mode, friendly to Reno. TCP Cubic implements a Reno-compatibility mode,
which is relevant for typical RTTs under 20ms as long as the which is relevant for typical RTTs under 20ms as long as the
throughput of a single flow is less than about 700Mb/s. In such throughput of a single flow is less than about 700Mb/s. In such
cases it can be assumed that Cubic traffic behaves similarly to Reno cases it can be assumed that Cubic traffic behaves similarly to Reno
(but with a slightly different constant of proportionality). The (but with a slightly different constant of proportionality). The
term 'Classic' will be used for the collection of Reno-friendly term 'Classic' will be used for the collection of Reno-friendly
traffic including Cubic in Reno mode. traffic including Cubic and potentially other experimental congestion
controls intended not to significantly impact the flow rate of Reno.
The supporting paper [PI2] includes the derivation of the equivalent A supporting paper [PI2] includes the derivation of the equivalent
rate equation for DCTCP, for which cwnd is inversely proportional to rate equation for DCTCP, for which cwnd is inversely proportional to
p (not the square root), where in this case p is the ECN marking p (not the square root), where in this case p is the ECN marking
probability. DCTCP is not the only congestion control that behaves probability. DCTCP is not the only congestion control that behaves
like this, so the term 'Scalable' will be used for all similar like this, so the term 'Scalable' will be used for all similar
congestion control behaviours (see examples in Section 1.2). The congestion control behaviours (see examples in Section 1.2). The
term 'L4S' is also used for traffic driven by a Scalable congestion term 'L4S' is also used for traffic driven by a Scalable congestion
control that also complies with the additional 'Prague L4S' control that also complies with the additional 'Prague L4S'
requirements [I-D.ietf-tsvwg-ecn-l4s-id]. requirements [I-D.ietf-tsvwg-ecn-l4s-id].
For safe co-existence, under stationary conditions, a Scalable flow For safe co-existence, under stationary conditions, a Scalable flow
has to run at roughly the same rate as a Reno TCP flow (all other has to run at roughly the same rate as a Reno TCP flow (all other
factors being equal). So the drop or marking probability for Classic factors being equal). So the drop or marking probability for Classic
traffic, p_C has to be distinct from the marking probability for L4S traffic, p_C has to be distinct from the marking probability for L4S
traffic, p_L. [RFC8311] updates the original ECN specification traffic, p_L. The original ECN specification [RFC3168] required
[RFC3168] to allow these probabilities to be distinct, because RFC these probabilities to be the same, but [RFC8311] updates RFC 3168 to
3168 required them to be the same. enable experiments in which these probabilities are different.
Also, to remain stable, Classic sources need the network to smooth Also, to remain stable, Classic sources need the network to smooth
p_C so it changes relatively slowly. It is hard for a network node p_C so it changes relatively slowly. It is hard for a network node
to know the RTTs of all the flows, so a Classic AQM adds a _worst- to know the RTTs of all the flows, so a Classic AQM adds a _worst-
case_ RTT of smoothing delay (about 100-200 ms). In contrast, L4S case_ RTT of smoothing delay (about 100-200 ms). In contrast, L4S
shifts responsibility for smoothing ECN feedback to the sender, which shifts responsibility for smoothing ECN feedback to the sender, which
only delays its response by its _own_ RTT, and allows a more only delays its response by its _own_ RTT, as well as allowing a more
immediate response if necessary. immediate response if necessary.
The Coupled AQM achieves safe coexistence by making the Classic drop The Coupled AQM achieves safe coexistence by making the Classic drop
probability p_C proportional to the square of the coupled L4S probability p_C proportional to the square of the coupled L4S
probability p_CL. p_CL is an input to the instantaneous L4S marking probability p_CL. p_CL is an input to the instantaneous L4S marking
probability p_L but it changes as slowly as p_C. This makes the Reno probability p_L but it changes as slowly as p_C. This makes the Reno
flow rate roughly equal the DCTCP flow rate, because the squaring of flow rate roughly equal the DCTCP flow rate, because the squaring of
p_CL counterbalances the square root of p_C in the Classic 'TCP p_CL counterbalances the square root of p_C in the 'TCP formula' of
formula'. Classic Reno congestion control.
Stating this as a formula, the relation between Classic drop Stating this as a formula, the relation between Classic drop
probability, p_C, and the coupled L4S probability p_CL needs to take probability, p_C, and the coupled L4S probability p_CL needs to take
the form: the form:
p_C = ( p_CL / k )^2 (1) p_C = ( p_CL / k )^2 (1)
where k is the constant of proportionality, which is termed the where k is the constant of proportionality, which is termed the
coupling factor. coupling factor.
skipping to change at page 11, line 10 skipping to change at page 12, line 25
is conditional to prevent starvation of Classic traffic. is conditional to prevent starvation of Classic traffic.
Nonetheless, coupled marking ensures that giving priority to L4S Nonetheless, coupled marking ensures that giving priority to L4S
traffic still leaves the right amount of spare scheduling time for traffic still leaves the right amount of spare scheduling time for
Classic flows to each get equivalent throughput to DCTCP flows (all Classic flows to each get equivalent throughput to DCTCP flows (all
other factors such as RTT being equal). other factors such as RTT being equal).
2.3. Traffic Classification 2.3. Traffic Classification
Both the Coupled AQM and DualQ mechanisms need an identifier to Both the Coupled AQM and DualQ mechanisms need an identifier to
distinguish L and C packets. Then the coupling algorithm can achieve distinguish L4S (L) and Classic (C) packets. Then the coupling
coexistence without having to inspect flow identifiers, because it algorithm can achieve coexistence without having to inspect flow
can apply the appropriate marking or dropping probability to all identifiers, because it can apply the appropriate marking or dropping
flows of each type. A separate probability to all flows of each type. A separate
specification [I-D.ietf-tsvwg-ecn-l4s-id] requires the sender to use specification [I-D.ietf-tsvwg-ecn-l4s-id] requires the network to
the ECT(1) and CE codepoints of the ECN field as this identifier, treat the ECT(1) and CE codepoints of the ECN field as this
having assessed various alternatives. An additional process document identifier, having assessed various alternatives. An additional
has proved necessary to make the ECT(1) codepoint available for process document has proved necessary to make the ECT(1) codepoint
experimentation [RFC8311]. available for experimentation [RFC8311].
For policy reasons, an operator might choose to steer certain packets For policy reasons, an operator might choose to steer certain packets
(e.g. from certain flows or with certain addresses) out of the L (e.g. from certain flows or with certain addresses) out of the L
queue, even though they identify themselves as L4S by their ECN queue, even though they identify themselves as L4S by their ECN
codepoints. In such cases, [I-D.ietf-tsvwg-ecn-l4s-id] says that the codepoints. In such cases, [I-D.ietf-tsvwg-ecn-l4s-id] says that the
device "MUST NOT alter the end-to-end L4S ECN identifier", so that it device "MUST NOT alter the end-to-end L4S ECN identifier", so that it
is preserved end-to-end. The aim is that each operator can choose is preserved end-to-end. The aim is that each operator can choose
how it treats L4S traffic locally, but an individual operator does how it treats L4S traffic locally, but an individual operator does
not alter the identification of L4S packets, which would prevent not alter the identification of L4S packets, which would prevent
other operators downstream from making their own choices on how to other operators downstream from making their own choices on how to
treat L4S traffic. treat L4S traffic.
In addition, an operator could use other identifiers to classify In addition, an operator could use other identifiers to classify
certain additional packet types into the L queue that it deems will certain additional packet types into the L queue that it deems will
not risk harm to the L4S service. For instance addresses of specific not risk harm to the L4S service. For instance addresses of specific
applications or hosts (see [I-D.ietf-tsvwg-ecn-l4s-id]), specific applications or hosts (see [I-D.ietf-tsvwg-ecn-l4s-id]), specific
Diffserv codepoints such as EF (Expedited Forwarding) and Voice-Admit Diffserv codepoints such as EF (Expedited Forwarding) and Voice-Admit
service classes (see [I-D.briscoe-tsvwg-l4s-diffserv]) or certain service classes (see [I-D.briscoe-tsvwg-l4s-diffserv]), the Non-
protocols (e.g. ARP, DNS). Note that the mechanism only reads these Queue-Building (NQB) per-hop behaviour [I-D.ietf-tsvwg-nqb] or
identifiers. [I-D.ietf-tsvwg-ecn-l4s-id] says it "MUST NOT alter certain protocols (e.g. ARP, DNS). Note that the mechanism only
these non-ECN identifiers". reads these identifiers. [I-D.ietf-tsvwg-ecn-l4s-id] says it "MUST
NOT alter these non-ECN identifiers". Thus, the L queue is not soley
an L4S queue, it can be consider more generally as a low latency
queue.
2.4. Overall DualQ Coupled AQM Structure 2.4. Overall DualQ Coupled AQM Structure
Figure 1 shows the overall structure that any DualQ Coupled AQM is Figure 1 shows the overall structure that any DualQ Coupled AQM is
likely to have. This schematic is intended to aid understanding of likely to have. This schematic is intended to aid understanding of
the current designs of DualQ Coupled AQMs. However, it is not the current designs of DualQ Coupled AQMs. However, it is not
intended to preclude other innovative ways of satisfying the intended to preclude other innovative ways of satisfying the
normative requirements in Section 2.5 that minimally define a DualQ normative requirements in Section 2.5 that minimally define a DualQ
Coupled AQM. Coupled AQM.
The classifier on the left separates incoming traffic between the two The classifier on the left separates incoming traffic between the two
queues (L and C). Each queue has its own AQM that determines the queues (L and C). Each queue has its own AQM that determines the
likelihood of marking or dropping (p_L and p_C). It has been likelihood of marking or dropping (p_L and p_C). It has been
proved [PI2] that it is preferable to control load with a linear proved [PI2] that it is preferable to control load with a linear
controller, then square the output before applying it as a drop controller, then square the output before applying it as a drop
probability to TCP (because TCP decreases its load proportional to probability to Reno-friendly traffic (because Reno congestion control
the square-root of the increase in drop). So, the AQM for Classic decreases its load proportional to the square-root of the increase in
traffic needs to be implemented in two stages: i) a base stage that drop). So, the AQM for Classic traffic needs to be implemented in
outputs an internal probability p' (pronounced p-prime); and ii) a two stages: i) a base stage that outputs an internal probability p'
squaring stage that outputs p_C, where (pronounced p-prime); and ii) a squaring stage that outputs p_C,
where
p_C = (p')^2. (2) p_C = (p')^2. (2)
Substituting for p_C in Eqn (1) gives: Substituting for p_C in Eqn (1) gives:
p' = p_CL / k p' = p_CL / k
So the slow-moving input to ECN marking in the L queue (the coupled So the slow-moving input to ECN marking in the L queue (the coupled
L4S probability) is: L4S probability) is:
p_CL = k*p', (3) p_CL = k*p'. (3)
where k is the constant coupling factor (see Appendix C).
It can be seen that these two transformations of p' implement the
required coupling given in equation (1) earlier.
The actual ECN marking probability p_L that is applied to the L queue The actual ECN marking probability p_L that is applied to the L queue
needs to track the immediate L queue delay under L-only congestion needs to track the immediate L queue delay under L-only congestion
conditions, as well as track p_CL under coupled congestion conditions, as well as track p_CL under coupled congestion
conditions. So the L queue uses a native AQM that calculates a conditions. So the L queue uses a native AQM that calculates a
probability p'_L as a function of the instantaneous L queue delay. probability p'_L as a function of the instantaneous L queue delay.
And, given the L queue has conditional strict priority over the C And, given the L queue has conditional priority over the C queue,
queue, whenever the L queue grows, the AQM should apply marking whenever the L queue grows, the AQM ought to apply marking
probability p'_L, but p_L should not fall below p_CL. This suggests: probability p'_L, but p_L ought not to fall below p_CL. This
suggests:
p_L = max(p'_L, p_CL), (4) p_L = max(p'_L, p_CL), (4)
which has also been found to work very well in practice. which has also been found to work very well in practice.
The two transformations of p' in equations (2) and (3) implement the
required coupling given in equation (1) earlier.
The constant of proportionality or coupling factor, k, in equation
(1) determines the ratio between the congestion probabilities (loss
or marking) experienced by L4S and Classic traffic. Thus k
indirectly determines the ratio between L4S and Classic flow rates,
because flows (assuming they are responsive) adjust their rate in
response to congestion probability. Appendix C.2 gives guidance on
the choice of k and its effect on relative flow rates.
_________ _________
| | ,------. | | ,------.
L4S queue | |===>| ECN | L4S queue | |===>| ECN |
,'| _______|_| |marker|\ ,'| _______|_| |marker|\
<' | | `------'\\ <' | | `------'\\
//`' v ^ p_L \\ //`' v ^ p_L \\
// ,-------. | \\ // ,-------. | \\
// |Native |p'_L | \\,. // |Native |p'_L | \\,.
// | L4S |--->(MAX) < | ___ // | L4S |--->(MAX) < | ___
,----------.// | AQM | ^ p_CL `\|.'Cond-`. ,----------.// | AQM | ^ p_CL `\|.'Cond-`.
skipping to change at page 13, line 33 skipping to change at page 14, line 48
< | _________ .------.// < | _________ .------.//
`\| | | | Drop |/ `\| | | | Drop |/
Classic |queue |===>|/mark | Classic |queue |===>|/mark |
__|______| `------' __|______| `------'
Legend: ===> traffic flow; ---> control dependency. Legend: ===> traffic flow; ---> control dependency.
Figure 1: DualQ Coupled AQM Schematic Figure 1: DualQ Coupled AQM Schematic
After the AQMs have applied their dropping or marking, the scheduler After the AQMs have applied their dropping or marking, the scheduler
forwards their packets to the link, giving priority to L4S traffic. forwards their packets to the link. Even though the scheduler gives
Priority has to be conditional in some way (see Section 4.1). Simple priority to the L queue, it is not as strong as the coupling from the
strict priority is inappropriate otherwise it could lead the L4S C queue. This is because, as the C queue grows, the base AQM applies
queue to starve the Classic queue. For example, consider the case more congestion signals to L traffic (as well as C). As L flows
where a continually busy L4S queue blocks a DNS request in the reduce their rate in response, they use less than the scheduling
Classic queue, arbitrarily delaying the start of a new Classic flow. share for L traffic. So, because the scheduler is work preserving,
it schedules any C traffic in the gaps.
Giving priority to the L queue has the benefit of very low L queue
delay, because the L queue is kept empty whenever L traffic is
controlled by the coupling. Also there only has to be a coupling in
one direction - from Classic to L4S. Priority has to be conditional
in some way to prevent the C queue starving under overload conditions
(see Section 4.1). With normal responsive traffic simple strict
priority would work, but it would make new Classic traffic wait until
its queue activated the coupling and L4S flows had in turn reduced
their rate enough to drain the L queue so that Classic traffic could
be scheduled. Giving a small weight or limited waiting time for C
traffic improves response times for short Classic messages, such as
DNS requests and improves Classic flow startup because immediate
capacity is available.
Example DualQ Coupled AQM algorithms called DualPI2 and Curvy RED are Example DualQ Coupled AQM algorithms called DualPI2 and Curvy RED are
given in Appendix A and Appendix B. Either example AQM can be used given in Appendix A and Appendix B. Either example AQM can be used
to couple packet marking and dropping across a dual Q. to couple packet marking and dropping across a dual Q.
DualPI2 uses a Proportional-Integral (PI) controller as the Base AQM. DualPI2 uses a Proportional-Integral (PI) controller as the Base AQM.
Indeed, this Base AQM with just the squared output and no L4S queue Indeed, this Base AQM with just the squared output and no L4S queue
can be used as a drop-in replacement for PIE [RFC8033], in which case can be used as a drop-in replacement for PIE [RFC8033], in which case
it is just called PI2 [PI2]. PI2 is a principled simplification of it is just called PI2 [PI2]. PI2 is a principled simplification of
PIE that is both more responsive and more stable in the face of PIE that is both more responsive and more stable in the face of
skipping to change at page 14, line 41 skipping to change at page 16, line 22
A Dual Queue Coupled AQM implementation MUST utilize two queues, each A Dual Queue Coupled AQM implementation MUST utilize two queues, each
with an AQM algorithm. The two queues can be part of a larger with an AQM algorithm. The two queues can be part of a larger
queuing hierarchy [I-D.briscoe-tsvwg-l4s-diffserv]. queuing hierarchy [I-D.briscoe-tsvwg-l4s-diffserv].
The AQM algorithm for the low latency (L) queue MUST be able to apply The AQM algorithm for the low latency (L) queue MUST be able to apply
ECN marking to ECN-capable packets. ECN marking to ECN-capable packets.
The scheduler draining the two queues MUST give L4S packets priority The scheduler draining the two queues MUST give L4S packets priority
over Classic, although priority MUST be bounded in order not to over Classic, although priority MUST be bounded in order not to
starve Classic traffic. starve Classic traffic. The scheduler SHOULD be work-conserving.
[I-D.ietf-tsvwg-ecn-l4s-id] defines the meaning of an ECN marking on [I-D.ietf-tsvwg-ecn-l4s-id] defines the meaning of an ECN marking on
L4S traffic, relative to drop of Classic traffic. In order to ensure L4S traffic, relative to drop of Classic traffic. In order to ensure
coexistence of Classic and Scalable L4S traffic, it says, "The coexistence of Classic and Scalable L4S traffic, it says, "The
likelihood that an AQM drops a Not-ECT Classic packet (p_C) MUST be likelihood that an AQM drops a Not-ECT Classic packet (p_C) MUST be
roughly proportional to the square of the likelihood that it would roughly proportional to the square of the likelihood that it would
have marked it if it had been an L4S packet (p_L)." The term have marked it if it had been an L4S packet (p_L)." The term
'likelihood' is used to allow for marking and dropping to be either 'likelihood' is used to allow for marking and dropping to be either
probabilistic or deterministic. probabilistic or deterministic.
skipping to change at page 15, line 19 skipping to change at page 16, line 47
The constant of proportionality, k, in Eqn (1) determines the The constant of proportionality, k, in Eqn (1) determines the
relative flow rates of Classic and L4S flows when the AQM concerned relative flow rates of Classic and L4S flows when the AQM concerned
is the bottleneck (all other factors being equal). is the bottleneck (all other factors being equal).
[I-D.ietf-tsvwg-ecn-l4s-id] says, "The constant of proportionality [I-D.ietf-tsvwg-ecn-l4s-id] says, "The constant of proportionality
(k) does not have to be standardised for interoperability, but a (k) does not have to be standardised for interoperability, but a
value of 2 is RECOMMENDED." value of 2 is RECOMMENDED."
Assuming Scalable congestion controls for the Internet will be as Assuming Scalable congestion controls for the Internet will be as
aggressive as DCTCP, this will ensure their congestion window will be aggressive as DCTCP, this will ensure their congestion window will be
roughly the same as that of a standards track TCP congestion control roughly the same as that of a standards track TCP Reno congestion
(Reno) [RFC5681] and other so-called TCP-friendly controls, such as control (Reno) [RFC5681] and other Reno-friendly controls, such as
TCP Cubic in its TCP-friendly mode. TCP Cubic in its Reno-compatibility mode.
The choice of k is a matter of operator policy, and operators MAY The choice of k is a matter of operator policy, and operators MAY
choose a different value using Table 1 and the guidelines in choose a different value using Table 1 and the guidelines in
Appendix C. Appendix C.2.
If multiple customers or users share capacity at a bottleneck (e.g. If multiple customers or users share capacity at a bottleneck (e.g.
in the Internet access link of a campus network), the operator's in the Internet access link of a campus network), the operator's
choice of k will determine capacity sharing between the flows of choice of k will determine capacity sharing between the flows of
different customers. However, on the public Internet, access network different customers. However, on the public Internet, access network
operators typically isolate customers from each other with some form operators typically isolate customers from each other with some form
of layer-2 multiplexing (OFDM(A) in DOCSIS3.1, CDMA in 3G, SC-FDMA in of layer-2 multiplexing (OFDM(A) in DOCSIS3.1, CDMA in 3G, SC-FDMA in
LTE) or L3 scheduling (WRR in DSL), rather than relying on TCP to LTE) or L3 scheduling (WRR in DSL), rather than relying on host
share capacity between customers [RFC0970]. In such cases, the congestion controls to share capacity between customers [RFC0970].
choice of k will solely affect relative flow rates within each In such cases, the choice of k will solely affect relative flow rates
customer's access capacity, not between customers. Also, k will not within each customer's access capacity, not between customers. Also,
affect relative flow rates at any times when all flows are Classic or k will not affect relative flow rates at any times when all flows are
all flows are L4S, and it will not affect the relative throughput of Classic or all flows are L4S, and it will not affect the relative
small flows. throughput of small flows.
2.5.1.1. Requirements in Unexpected Cases 2.5.1.1. Requirements in Unexpected Cases
The flexibility to allow operator-specific classifiers (Section 2.3) The flexibility to allow operator-specific classifiers (Section 2.3)
leads to the need to specify what the AQM in each queue ought to do leads to the need to specify what the AQM in each queue ought to do
with packets that do not carry the ECN field expected for that queue. with packets that do not carry the ECN field expected for that queue.
It is recommended that the AQM in each queue inspects the ECN field It is expected that the AQM in each queue will inspect the ECN field
to determine what sort of congestion notification to signal, then to determine what sort of congestion notification to signal, then it
decides whether to apply congestion notification to this particular will decide whether to apply congestion notification to this
packet, as follows: particular packet, as follows:
o If a packet that does not carry an ECT(1) or CE codepoint is o If a packet that does not carry an ECT(1) or CE codepoint is
classified into the L queue: classified into the L queue:
* if the packet is ECT(0), the L AQM SHOULD apply CE-marking * if the packet is ECT(0), the L AQM SHOULD apply CE-marking
using a probability appropriate to Classic congestion control using a probability appropriate to Classic congestion control
and appropriate to the target delay in the L queue and appropriate to the target delay in the L queue
* if the packet is Not-ECT, the appropriate action depends on * if the packet is Not-ECT, the appropriate action depends on
whether some other function is protecting the L queue from whether some other function is protecting the L queue from
misbehaving flows (e.g. per-flow queue protection or latency misbehaving flows (e.g. per-flow queue protection
policing): [I-D.briscoe-docsis-q-protection] or latency policing):
+ If separate queue protection is provided, the L AQM SHOULD + If separate queue protection is provided, the L AQM SHOULD
ignore the packet and forward it unchanged, meaning it ignore the packet and forward it unchanged, meaning it
should not calculate whether to apply congestion should not calculate whether to apply congestion
notification and it should neither drop nor CE-mark the notification and it should neither drop nor CE-mark the
packet (for instance, the operator might classify EF traffic packet (for instance, the operator might classify EF traffic
that is unresponsive to drop into the L queue, alongside that is unresponsive to drop into the L queue, alongside
responsive L4S-ECN traffic) responsive L4S-ECN traffic)
+ if separate queue protection is not provided, the L AQM + if separate queue protection is not provided, the L AQM
skipping to change at page 16, line 40 skipping to change at page 18, line 23
* the C AQM SHOULD apply CE-marking using the coupled AQM * the C AQM SHOULD apply CE-marking using the coupled AQM
probability p_CL (= k*p'). probability p_CL (= k*p').
The above requirements are worded as "SHOULDs", because operator- The above requirements are worded as "SHOULDs", because operator-
specific classifiers are for flexibility, by definition. Therefore, specific classifiers are for flexibility, by definition. Therefore,
alternative actions might be appropriate in the operator's specific alternative actions might be appropriate in the operator's specific
circumstances. An example would be where the operator knows that circumstances. An example would be where the operator knows that
certain legacy traffic marked with one codepoint actually has a certain legacy traffic marked with one codepoint actually has a
congestion response associated with another codepoint. congestion response associated with another codepoint.
If the DualQ Coupled AQM has detected overload, it MUST signal If the DualQ Coupled AQM has detected overload, it SHOULD signal
congestion solely using drop, irrespective of the ECN field. congestion solely using drop, irrespective of the ECN field.
Switching to drop if ECN marking is persistently high is required by Switching to drop if ECN marking is persistently high is required by
Section 7 of [RFC3168] and Section 4.2.1 of [RFC7567]. Section 7 of [RFC3168] and Section 4.2.1 of [RFC7567].
2.5.2. Management Requirements 2.5.2. Management Requirements
2.5.2.1. Configuration 2.5.2.1. Configuration
By default, a DualQ Coupled AQM SHOULD NOT need any configuration for By default, a DualQ Coupled AQM SHOULD NOT need any configuration for
use at a bottleneck on the public Internet [RFC7567]. The following use at a bottleneck on the public Internet [RFC7567]. The following
parameters MAY be operator-configurable, e.g. to tune for non- parameters MAY be operator-configurable, e.g. to tune for non-
Internet settings: Internet settings:
o Optional packet classifier(s) to use in addition to the ECN field o Optional packet classifier(s) to use in addition to the ECN field
(see Section 2.3); (see Section 2.3);
o Expected typical RTT, which can be used to determine the queuing o Expected typical RTT, which can be used to determine the queuing
delay of the Classic AQM at its operating point, in order to delay of the Classic AQM at its operating point, in order to
prevent typical lone TCP flows from under-utilizing capacity. For prevent typical lone flows from under-utilizing capacity. For
example: example:
* for the PI2 algorithm (Appendix A) the queuing delay target is * for the PI2 algorithm (Appendix A) the queuing delay target is
set to the typical RTT; set to the typical RTT;
* for the Curvy RED algorithm (Appendix B) the queuing delay at * for the Curvy RED algorithm (Appendix B) the queuing delay at
the desired operating point of the curvy ramp is configured to the desired operating point of the curvy ramp is configured to
encompass a typical RTT; encompass a typical RTT;
* if another Classic AQM was used, it would be likely to need an * if another Classic AQM was used, it would be likely to need an
operating point for the queue based on the typical RTT, and if operating point for the queue based on the typical RTT, and if
so it SHOULD be expressed in units of time. so it SHOULD be expressed in units of time.
An operating point that is manually calculated might be directly An operating point that is manually calculated might be directly
configurable instead, e.g. for links with large numbers of flows configurable instead, e.g. for links with large numbers of flows
where under-utilization by a single TCP flow would be unlikely. where under-utilization by a single flow would be unlikely.
o Expected maximum RTT, which can be used to set the stability o Expected maximum RTT, which can be used to set the stability
parameter(s) of the Classic AQM. For example: parameter(s) of the Classic AQM. For example:
* for the PI2 algorithm (Appendix A), the gain parameters of the * for the PI2 algorithm (Appendix A), the gain parameters of the
PI algorithm depend on the maximum RTT. PI algorithm depend on the maximum RTT.
* for the Curvy RED algorithm (Appendix B) the smoothing * for the Curvy RED algorithm (Appendix B) the smoothing
parameter is chosen to filter out transients in the queue parameter is chosen to filter out transients in the queue
within a maximum RTT. within a maximum RTT.
Stability parameter(s) that are manually calculated assuming a Stability parameter(s) that are manually calculated assuming a
maximum RTT might be directly configurable instead. maximum RTT might be directly configurable instead.
o Coupling factor, k; o Coupling factor, k (see Appendix C.2);
o A limit to the conditional priority of L4S. This is scheduler- o A limit to the conditional priority of L4S. This is scheduler-
dependent, but it SHOULD be expressed as a relation between the dependent, but it SHOULD be expressed as a relation between the
max delay of a C packet and an L packet. For example: max delay of a C packet and an L packet. For example:
* for a WRR scheduler a weight ratio between L and C of w:1 means * for a WRR scheduler a weight ratio between L and C of w:1 means
that the maximum delay to a C packet is w times that of an L that the maximum delay to a C packet is w times that of an L
packet. packet.
* for a time-shifted FIFO (TS-FIFO) scheduler (see Section 4.1.1) * for a time-shifted FIFO (TS-FIFO) scheduler (see Section 4.1.1)
skipping to change at page 19, line 18 skipping to change at page 20, line 50
Then, during that time, if the AQM enters and exits overload state Then, during that time, if the AQM enters and exits overload state
any number of times, the duration in overload state is accumulated any number of times, the duration in overload state is accumulated
but no new report is generated until the first time the AQM is out but no new report is generated until the first time the AQM is out
of overload once the timer has expired. of overload once the timer has expired.
2.5.2.4. Deployment, Coexistence and Scaling 2.5.2.4. Deployment, Coexistence and Scaling
[RFC5706] suggests that deployment, coexistence and scaling should [RFC5706] suggests that deployment, coexistence and scaling should
also be covered as management requirements. The raison d'etre of the also be covered as management requirements. The raison d'etre of the
DualQ Coupled AQM is to enable deployment and coexistence of Scalable DualQ Coupled AQM is to enable deployment and coexistence of Scalable
congestion controls - as incremental replacements for today's TCP- congestion controls - as incremental replacements for today's Reno-
friendly controls that do not scale with bandwidth-delay product. friendly controls that do not scale with bandwidth-delay product.
Therefore there is no need to repeat these motivating issues here Therefore there is no need to repeat these motivating issues here
given they are already explained in the Introduction and detailed in given they are already explained in the Introduction and detailed in
the L4S architecture [I-D.ietf-tsvwg-l4s-arch]. the L4S architecture [I-D.ietf-tsvwg-l4s-arch].
The descriptions of specific DualQ Coupled AQM algorithms in the The descriptions of specific DualQ Coupled AQM algorithms in the
appendices cover scaling of their configuration parameters, e.g. with appendices cover scaling of their configuration parameters, e.g. with
respect to RTT and sampling frequency. respect to RTT and sampling frequency.
3. IANA Considerations 3. IANA Considerations
skipping to change at page 19, line 40 skipping to change at page 21, line 25
This specification contains no IANA considerations. This specification contains no IANA considerations.
4. Security Considerations 4. Security Considerations
4.1. Overload Handling 4.1. Overload Handling
Where the interests of users or flows might conflict, it could be Where the interests of users or flows might conflict, it could be
necessary to police traffic to isolate any harm to the performance of necessary to police traffic to isolate any harm to the performance of
individual flows. However it is hard to avoid unintended side- individual flows. However it is hard to avoid unintended side-
effects with policing, and in a trusted environment policing is not effects with policing, and in a trusted environment policing is not
necessary. Therefore per-flow policing needs to be separable from a necessary. Therefore per-flow policing (e.g.
basic AQM, as an option under policy control. [I-D.briscoe-docsis-q-protection]) needs to be separable from a basic
AQM, as an option under policy control.
However, a basic DualQ AQM does at least need to handle overload. A However, a basic DualQ AQM does at least need to handle overload. A
useful objective would be for the overload behaviour of the DualQ AQM useful objective would be for the overload behaviour of the DualQ AQM
to be at least no worse than a single queue AQM. However, a trade- to be at least no worse than a single queue AQM. However, a trade-
off needs to be made between complexity and the risk of either off needs to be made between complexity and the risk of either
traffic class harming the other. In each of the following three traffic class harming the other. In each of the following three
subsections, an overload issue specific to the DualQ is described, subsections, an overload issue specific to the DualQ is described,
followed by proposed solution(s). followed by proposed solution(s).
Under overload the higher priority L4S service will have to sacrifice Under overload the higher priority L4S service will have to sacrifice
skipping to change at page 21, line 33 skipping to change at page 23, line 19
To keep the throughput of both L4S and Classic flows roughly equal To keep the throughput of both L4S and Classic flows roughly equal
over the full load range, a different control strategy needs to be over the full load range, a different control strategy needs to be
defined above the point where one AQM first saturates to a defined above the point where one AQM first saturates to a
probability of 100% leaving no room to push back the load any harder. probability of 100% leaving no room to push back the load any harder.
If k>1, L4S will saturate first, even though saturation could be If k>1, L4S will saturate first, even though saturation could be
caused by unresponsive traffic in either queue. caused by unresponsive traffic in either queue.
The term 'unresponsive' includes cases where a flow becomes The term 'unresponsive' includes cases where a flow becomes
temporarily unresponsive, for instance, a real-time flow that takes a temporarily unresponsive, for instance, a real-time flow that takes a
while to adapt its rate in response to congestion, or a TCP-like flow while to adapt its rate in response to congestion, or a standard Reno
that is normally responsive, but above a certain congestion level it flow that is normally responsive, but above a certain congestion
will not be able to reduce its congestion window below the minimum of level it will not be able to reduce its congestion window below the
2 segments [RFC5681], effectively becoming unresponsive. (Note that allowed minimum of 2 segments [RFC5681], effectively becoming
L4S traffic ought to remain responsive below a window of 2 segments unresponsive. (Note that L4S traffic ought to remain responsive
(see [I-D.ietf-tsvwg-ecn-l4s-id]). below a window of 2 segments (see [I-D.ietf-tsvwg-ecn-l4s-id]).
Saturation raises the question of whether to relieve congestion by Saturation raises the question of whether to relieve congestion by
introducing some drop into the L4S queue or by allowing delay to grow introducing some drop into the L4S queue or by allowing delay to grow
in both queues (which could eventually lead to tail drop too): in both queues (which could eventually lead to tail drop too):
Drop on Saturation: Saturation can be avoided by setting a maximum Drop on Saturation: Saturation can be avoided by setting a maximum
threshold for L4S ECN marking (assuming k>1) before saturation threshold for L4S ECN marking (assuming k>1) before saturation
starts to make the flow rates of the different traffic types starts to make the flow rates of the different traffic types
diverge. Above that the drop probability of Classic traffic is diverge. Above that the drop probability of Classic traffic is
applied to all packets of all traffic types. Then experiments applied to all packets of all traffic types. Then experiments
skipping to change at page 22, line 15 skipping to change at page 23, line 49
Delay on Saturation: When L4S marking saturates, instead of Delay on Saturation: When L4S marking saturates, instead of
switching to drop, the drop and marking probabilities could be switching to drop, the drop and marking probabilities could be
capped. Beyond that, delay will grow either solely in the queue capped. Beyond that, delay will grow either solely in the queue
with unresponsive traffic (if WRR is used), or in both queues (if with unresponsive traffic (if WRR is used), or in both queues (if
time-shifted FIFO is used). In either case, the higher delay time-shifted FIFO is used). In either case, the higher delay
ought to control temporary high congestion. If the overload is ought to control temporary high congestion. If the overload is
more persistent, eventually the combined DualQ will overflow and more persistent, eventually the combined DualQ will overflow and
tail drop will control congestion. tail drop will control congestion.
The example implementation in Appendix A solely applies the "drop on The example implementation in Appendix A solely applies the "drop on
saturation" policy. saturation" policy. The DOCSIS specification of a DualQ Coupled AQM
[DOCSIS3.1] also implements the 'drop on saturation' policy with a
very shallow L buffer. However, the addition of DOCSIS per-flow
Queue Protection [I-D.briscoe-docsis-q-protection] turns this into
'delay on saturation' by redirecting some packets of the flow(s) most
responsible for L queue overload into the C queue, which has a higher
delay target. If overload continues, this again becomes 'drop on
saturation' as the level of drop in the C queue rises to maintain the
target delay of the C queue.
4.1.3. Protecting against Unresponsive ECN-Capable Traffic 4.1.3. Protecting against Unresponsive ECN-Capable Traffic
Unresponsive traffic has a greater advantage if it is also ECN- Unresponsive traffic has a greater advantage if it is also ECN-
capable. The advantage is undetectable at normal low levels of drop/ capable. The advantage is undetectable at normal low levels of drop/
marking, but it becomes significant with the higher levels of drop/ marking, but it becomes significant with the higher levels of drop/
marking typical during overload. This is an issue whether the ECN- marking typical during overload. This is an issue whether the ECN-
capable traffic is L4S or Classic. capable traffic is L4S or Classic.
This raises the question of whether and when to switch off ECN This raises the question of whether and when to switch off ECN
skipping to change at page 22, line 39 skipping to change at page 24, line 32
Experiments with the DualPI2 AQM (Appendix A) have shown that Experiments with the DualPI2 AQM (Appendix A) have shown that
introducing 'drop on saturation' at 100% L4S marking addresses this introducing 'drop on saturation' at 100% L4S marking addresses this
problem with unresponsive ECN as well as addressing the saturation problem with unresponsive ECN as well as addressing the saturation
problem. It leaves only a small range of congestion levels where problem. It leaves only a small range of congestion levels where
unresponsive traffic gains any advantage from using the ECN unresponsive traffic gains any advantage from using the ECN
capability, and the advantage is hardly detectable [DualQ-Test]. capability, and the advantage is hardly detectable [DualQ-Test].
5. Acknowledgements 5. Acknowledgements
Thanks to Anil Agarwal, Sowmini Varadhan's, Gabi Bracha, Nicolas Thanks to Anil Agarwal, Sowmini Varadhan's, Gabi Bracha, Nicolas
Kuhn, Tom Henderson and David Pullen for detailed review comments Kuhn, Greg Skinner, Tom Henderson and David Pullen for detailed
particularly of the appendices and suggestions on how to make the review comments particularly of the appendices and suggestions on how
explanations clearer. Thanks also to Tom Henderson for insights on to make the explanations clearer. Thanks also to Tom Henderson for
the choice of schedulers and queue delay measurement techniques. insights on the choice of schedulers and queue delay measurement
techniques.
The early contributions of Koen De Schepper, Bob Briscoe, Olga The early contributions of Koen De Schepper, Bob Briscoe, Olga
Bondarenko and Inton Tsang were part-funded by the European Community Bondarenko and Inton Tsang were part-funded by the European Community
under its Seventh Framework Programme through the Reducing Internet under its Seventh Framework Programme through the Reducing Internet
Transport Latency (RITE) project (ICT-317700). Bob Briscoe's Transport Latency (RITE) project (ICT-317700). Bob Briscoe's
contribution was also part-funded by the Research Council of Norway contribution was also part-funded by the Comcast Innovation Fund and
through the TimeIn project. The views expressed here are solely the Research Council of Norway through the TimeIn project. The views
those of the authors. expressed here are solely those of the authors.
6. Contributors 6. Contributors
The following contributed implementations and evaluations that The following contributed implementations and evaluations that
validated and helped to improve this specification: validated and helped to improve this specification:
Olga Albisser <olga@albisser.org> of Simula Research Lab, Norway Olga Albisser <olga@albisser.org> of Simula Research Lab, Norway
(Olga Bondarenko during early drafts) implemented the prototype (Olga Bondarenko during early drafts) implemented the prototype
DualPI2 AQM for Linux with Koen De Schepper and conducted DualPI2 AQM for Linux with Koen De Schepper and conducted
extensive evaluations as well as implementing the live performance extensive evaluations as well as implementing the live performance
visualization GUI [L4Sdemo16]. visualization GUI [L4Sdemo16].
Olivier Tilmans <olivier.tilmans@nokia-bell-labs.com> of Nokia Olivier Tilmans <olivier.tilmans@nokia-bell-labs.com> of Nokia
Bell Labs, Belgium prepared and maintains the Linux implementation Bell Labs, Belgium prepared and maintains the Linux implementation
of DualPI2 for upstreaming. of DualPI2 for upstreaming.
Tom Henderson <tomh@tomh.org> of CableLabs, US implemented various Tom Henderson <tomh@tomh.org> of CableLabs, US implemented various
Coupled DualQ AQMs for ns3, including DualPI2 and DualPIE over DualQ Coupled AQMs for ns3, including DualPI2 and DualPIE over
point to point and DOCSIS 3.1 link models and conducted extensive point to point and DOCSIS 3.1 link models and conducted extensive
evaluations. evaluations.
Ing Jyh (Inton) Tsang of Nokia, Belgium built the End-to-End Data Ing Jyh (Inton) Tsang of Nokia, Belgium built the End-to-End Data
Centre to the Home broadband testbed on which Coupled DualQ Centre to the Home broadband testbed on which DualQ Coupled AQM
implementations were tested. implementations were tested.
7. References 7. References
7.1. Normative References 7.1. Normative References
[I-D.ietf-tsvwg-ecn-l4s-id] [I-D.ietf-tsvwg-ecn-l4s-id]
Schepper, K. and B. Briscoe, "Identifying Modified Schepper, K. and B. Briscoe, "Identifying Modified
Explicit Congestion Notification (ECN) Semantics for Explicit Congestion Notification (ECN) Semantics for
Ultra-Low Queuing Delay (L4S)", draft-ietf-tsvwg-ecn-l4s- Ultra-Low Queuing Delay (L4S)", draft-ietf-tsvwg-ecn-l4s-
id-06 (work in progress), March 2019. id-09 (work in progress), February 2020.
[RFC2119] Bradner, S., "Key words for use in RFCs to Indicate [RFC2119] Bradner, S., "Key words for use in RFCs to Indicate
Requirement Levels", BCP 14, RFC 2119, Requirement Levels", BCP 14, RFC 2119,
DOI 10.17487/RFC2119, March 1997, DOI 10.17487/RFC2119, March 1997,
<https://www.rfc-editor.org/info/rfc2119>. <https://www.rfc-editor.org/info/rfc2119>.
[RFC3168] Ramakrishnan, K., Floyd, S., and D. Black, "The Addition [RFC3168] Ramakrishnan, K., Floyd, S., and D. Black, "The Addition
of Explicit Congestion Notification (ECN) to IP", of Explicit Congestion Notification (ECN) to IP",
RFC 3168, DOI 10.17487/RFC3168, September 2001, RFC 3168, DOI 10.17487/RFC3168, September 2001,
<https://www.rfc-editor.org/info/rfc3168>. <https://www.rfc-editor.org/info/rfc3168>.
skipping to change at page 24, line 28 skipping to change at page 26, line 28
<https://www.cs.purdue.edu/homes/fahmy/papers/ldc.pdf>. <https://www.cs.purdue.edu/homes/fahmy/papers/ldc.pdf>.
[ARED01] Floyd, S., Gummadi, R., and S. Shenker, "Adaptive RED: An [ARED01] Floyd, S., Gummadi, R., and S. Shenker, "Adaptive RED: An
Algorithm for Increasing the Robustness of RED's Active Algorithm for Increasing the Robustness of RED's Active
Queue Management", ACIRI Technical Report , August 2001, Queue Management", ACIRI Technical Report , August 2001,
<http://www.icir.org/floyd/red.html>. <http://www.icir.org/floyd/red.html>.
[BBRv1] Cardwell, N., Cheng, Y., Hassas Yeganeh, S., and V. [BBRv1] Cardwell, N., Cheng, Y., Hassas Yeganeh, S., and V.
Jacobson, "BBR Congestion Control", Internet Draft draft- Jacobson, "BBR Congestion Control", Internet Draft draft-
cardwell-iccrg-bbr-congestion-control-00, July 2017, cardwell-iccrg-bbr-congestion-control-00, July 2017,
<https://tools.ietf.org/html/ <https://tools.ietf.org/html/draft-cardwell-iccrg-bbr-
draft-cardwell-iccrg-bbr-congestion-control-00>. congestion-control-00>.
[CoDel] Nichols, K. and V. Jacobson, "Controlling Queue Delay", [CoDel] Nichols, K. and V. Jacobson, "Controlling Queue Delay",
ACM Queue 10(5), May 2012, ACM Queue 10(5), May 2012,
<http://queue.acm.org/issuedetail.cfm?issue=2208917>. <http://queue.acm.org/issuedetail.cfm?issue=2208917>.
[CRED_Insights] [CRED_Insights]
Briscoe, B., "Insights from Curvy RED (Random Early Briscoe, B., "Insights from Curvy RED (Random Early
Detection)", BT Technical Report TR-TUB8-2015-003 Detection)", BT Technical Report TR-TUB8-2015-003
arXiv:1904.07339 [cs.NI], July 2015, arXiv:1904.07339 [cs.NI], July 2015,
<https://arxiv.org/abs/1904.07339>. <https://arxiv.org/abs/1904.07339>.
skipping to change at page 25, line 9 skipping to change at page 27, line 9
CableLabs, "MAC and Upper Layer Protocols Interface CableLabs, "MAC and Upper Layer Protocols Interface
(MULPI) Specification, CM-SP-MULPIv3.1", Data-Over-Cable (MULPI) Specification, CM-SP-MULPIv3.1", Data-Over-Cable
Service Interface Specifications DOCSIS(R) 3.1 Version i17 Service Interface Specifications DOCSIS(R) 3.1 Version i17
or later, January 2019, <https://specification- or later, January 2019, <https://specification-
search.cablelabs.com/CM-SP-MULPIv3.1>. search.cablelabs.com/CM-SP-MULPIv3.1>.
[DualPI2Linux] [DualPI2Linux]
Albisser, O., De Schepper, K., Briscoe, B., Tilmans, O., Albisser, O., De Schepper, K., Briscoe, B., Tilmans, O.,
and H. Steen, "DUALPI2 - Low Latency, Low Loss and and H. Steen, "DUALPI2 - Low Latency, Low Loss and
Scalable (L4S) AQM", Proc. Linux Netdev 0x13 , March 2019, Scalable (L4S) AQM", Proc. Linux Netdev 0x13 , March 2019,
<https://www.netdevconf.org/0x13/ <https://www.netdevconf.org/0x13/session.html?talk-
session.html?talk-DUALPI2-AQM>. DUALPI2-AQM>.
[DualQ-Test] [DualQ-Test]
Steen, H., "Destruction Testing: Ultra-Low Delay using Steen, H., "Destruction Testing: Ultra-Low Delay using
Dual Queue Coupled Active Queue Management", Masters Dual Queue Coupled Active Queue Management", Masters
Thesis, Dept of Informatics, Uni Oslo , May 2017. Thesis, Dept of Informatics, Uni Oslo , May 2017.
[I-D.briscoe-docsis-q-protection]
Briscoe, B. and G. White, "Queue Protection to Preserve
Low Latency", draft-briscoe-docsis-q-protection-00 (work
in progress), July 2019.
[I-D.briscoe-tsvwg-l4s-diffserv] [I-D.briscoe-tsvwg-l4s-diffserv]
Briscoe, B., "Interactions between Low Latency, Low Loss, Briscoe, B., "Interactions between Low Latency, Low Loss,
Scalable Throughput (L4S) and Differentiated Services", Scalable Throughput (L4S) and Differentiated Services",
draft-briscoe-tsvwg-l4s-diffserv-02 (work in progress), draft-briscoe-tsvwg-l4s-diffserv-02 (work in progress),
November 2018. November 2018.
[I-D.ietf-tsvwg-l4s-arch] [I-D.ietf-tsvwg-l4s-arch]
Briscoe, B., Schepper, K., and M. Bagnulo, "Low Latency, Briscoe, B., Schepper, K., Bagnulo, M., and G. White, "Low
Low Loss, Scalable Throughput (L4S) Internet Service: Latency, Low Loss, Scalable Throughput (L4S) Internet
Architecture", draft-ietf-tsvwg-l4s-arch-03 (work in Service: Architecture", draft-ietf-tsvwg-l4s-arch-05 (work
progress), October 2018. in progress), February 2020.
[I-D.ietf-tsvwg-nqb]
White, G. and T. Fossati, "A Non-Queue-Building Per-Hop
Behavior (NQB PHB) for Differentiated Services", draft-
ietf-tsvwg-nqb-00 (work in progress), November 2019.
[L4Sdemo16] [L4Sdemo16]
Bondarenko, O., De Schepper, K., Tsang, I., and B. Bondarenko, O., De Schepper, K., Tsang, I., and B.
Briscoe, "Ultra-Low Delay for All: Live Experience, Live Briscoe, "Ultra-Low Delay for All: Live Experience, Live
Analysis", Proc. MMSYS'16 pp33:1--33:4, May 2016, Analysis", Proc. MMSYS'16 pp33:1--33:4, May 2016,
<http://dl.acm.org/citation.cfm?doid=2910017.2910633 <http://dl.acm.org/citation.cfm?doid=2910017.2910633
(videos of demos: https://riteproject.eu/ (videos of demos:
dctth/#1511dispatchwg )>. https://riteproject.eu/dctth/#1511dispatchwg )>.
[LLD] White, G., Sundaresan, K., and B. Briscoe, "Low Latency [LLD] White, G., Sundaresan, K., and B. Briscoe, "Low Latency
DOCSIS: Technology Overview", CableLabs White Paper , DOCSIS: Technology Overview", CableLabs White Paper ,
February 2019, <https://cablela.bs/ February 2019, <https://cablela.bs/low-latency-docsis-
low-latency-docsis-technology-overview-february-2019>. technology-overview-february-2019>.
[Mathis09] [Mathis09]
Mathis, M., "Relentless Congestion Control", PFLDNeT'09 , Mathis, M., "Relentless Congestion Control", PFLDNeT'09 ,
May 2009, <http://www.hpcc.jp/pfldnet2009/ May 2009, <http://www.hpcc.jp/pfldnet2009/
Program_files/1569198525.pdf>. Program_files/1569198525.pdf>.
[MEDF] Menth, M., Schmid, M., Heiss, H., and T. Reim, "MEDF - a [MEDF] Menth, M., Schmid, M., Heiss, H., and T. Reim, "MEDF - a
simple scheduling algorithm for two real-time transport simple scheduling algorithm for two real-time transport
service classes with application in the UTRAN", Proc. IEEE service classes with application in the UTRAN", Proc. IEEE
Conference on Computer Communications (INFOCOM'03) Vol.2 Conference on Computer Communications (INFOCOM'03) Vol.2
skipping to change at page 26, line 41 skipping to change at page 29, line 5
[RFC3246] Davie, B., Charny, A., Bennet, J., Benson, K., Le Boudec, [RFC3246] Davie, B., Charny, A., Bennet, J., Benson, K., Le Boudec,
J., Courtney, W., Davari, S., Firoiu, V., and D. J., Courtney, W., Davari, S., Firoiu, V., and D.
Stiliadis, "An Expedited Forwarding PHB (Per-Hop Stiliadis, "An Expedited Forwarding PHB (Per-Hop
Behavior)", RFC 3246, DOI 10.17487/RFC3246, March 2002, Behavior)", RFC 3246, DOI 10.17487/RFC3246, March 2002,
<https://www.rfc-editor.org/info/rfc3246>. <https://www.rfc-editor.org/info/rfc3246>.
[RFC3649] Floyd, S., "HighSpeed TCP for Large Congestion Windows", [RFC3649] Floyd, S., "HighSpeed TCP for Large Congestion Windows",
RFC 3649, DOI 10.17487/RFC3649, December 2003, RFC 3649, DOI 10.17487/RFC3649, December 2003,
<https://www.rfc-editor.org/info/rfc3649>. <https://www.rfc-editor.org/info/rfc3649>.
[RFC5033] Floyd, S. and M. Allman, "Specifying New Congestion
Control Algorithms", BCP 133, RFC 5033,
DOI 10.17487/RFC5033, August 2007,
<https://www.rfc-editor.org/info/rfc5033>.
[RFC5348] Floyd, S., Handley, M., Padhye, J., and J. Widmer, "TCP
Friendly Rate Control (TFRC): Protocol Specification",
RFC 5348, DOI 10.17487/RFC5348, September 2008,
<https://www.rfc-editor.org/info/rfc5348>.
[RFC5681] Allman, M., Paxson, V., and E. Blanton, "TCP Congestion [RFC5681] Allman, M., Paxson, V., and E. Blanton, "TCP Congestion
Control", RFC 5681, DOI 10.17487/RFC5681, September 2009, Control", RFC 5681, DOI 10.17487/RFC5681, September 2009,
<https://www.rfc-editor.org/info/rfc5681>. <https://www.rfc-editor.org/info/rfc5681>.
[RFC5706] Harrington, D., "Guidelines for Considering Operations and [RFC5706] Harrington, D., "Guidelines for Considering Operations and
Management of New Protocols and Protocol Extensions", Management of New Protocols and Protocol Extensions",
RFC 5706, DOI 10.17487/RFC5706, November 2009, RFC 5706, DOI 10.17487/RFC5706, November 2009,
<https://www.rfc-editor.org/info/rfc5706>. <https://www.rfc-editor.org/info/rfc5706>.
[RFC7567] Baker, F., Ed. and G. Fairhurst, Ed., "IETF [RFC7567] Baker, F., Ed. and G. Fairhurst, Ed., "IETF
skipping to change at page 28, line 16 skipping to change at page 30, line 37
variant of the PIE AQM [RFC8033]. variant of the PIE AQM [RFC8033].
The pseudocode will be introduced in two passes. The first pass The pseudocode will be introduced in two passes. The first pass
explains the core concepts, deferring handling of overload to the explains the core concepts, deferring handling of overload to the
second pass. To aid comparison, line numbers are kept in step second pass. To aid comparison, line numbers are kept in step
between the two passes by using letter suffixes where the longer code between the two passes by using letter suffixes where the longer code
needs extra lines. needs extra lines.
All variables are assumed to be floating point in their basic units All variables are assumed to be floating point in their basic units
(size in bytes, time in seconds, rates in bytes/second, alpha and (size in bytes, time in seconds, rates in bytes/second, alpha and
beta in Hz, and probabilities from 0 to 1. Constants expressed in k, beta in Hz, and probabilities from 0 to 1. Constants expressed in k
M, G, u, m, %, ... are assumed to be converted to their appropriate (kilo), M (mega), G (giga), u (micro), m (milli) , %, ... are assumed
multiple or fraction. A real implementation that wants to use to be converted to their appropriate multiple or fraction to
represent the basic units. A real implementation that wants to use
integer values needs to handle appropriate scaling factors and allow integer values needs to handle appropriate scaling factors and allow
accordingly appropriate resolution of its integer types (including accordingly appropriate resolution of its integer types (including
temporary internal values during calculations). temporary internal values during calculations).
A full open source implementation for Linux is available at: A full open source implementation for Linux is available at:
https://github.com/L4STeam/sch_dualpi2_upstream and explained in https://github.com/L4STeam/sch_dualpi2_upstream and explained in
[DualPI2Linux]. The specification of the DualQ Coupled AQM for [DualPI2Linux]. The specification of the DualQ Coupled AQM for
DOCSIS cable modems and CMTSs is available in [DOCSIS3.1] and DOCSIS cable modems and CMTSs is available in [DOCSIS3.1] and
explained in [LLD]. explained in [LLD].
skipping to change at page 28, line 43 skipping to change at page 31, line 19
pseudocode consists of the following six functions: pseudocode consists of the following six functions:
o the initialization function dualpi2_params_init(...) (Figure 2) o the initialization function dualpi2_params_init(...) (Figure 2)
that sets parameter defaults (the API for setting non-default that sets parameter defaults (the API for setting non-default
values is omitted for brevity) values is omitted for brevity)
o the enqueue function dualpi2_enqueue(lq, cq, pkt) (Figure 3) o the enqueue function dualpi2_enqueue(lq, cq, pkt) (Figure 3)
o the dequeue function dualpi2_dequeue(lq, cq, pkt) (Figure 4) o the dequeue function dualpi2_dequeue(lq, cq, pkt) (Figure 4)
o recur(likelihood) for de-randomized ECN marking (shown at the end o recur(q, likelihood) for de-randomized ECN marking (shown at the
of Figure 4). end of Figure 4).
o the L4S AQM function laqm(qdelay) (Figure 5) used to calculate the o the L4S AQM function laqm(qdelay) (Figure 5) used to calculate the
ECN-marking probability for the L4S queue ECN-marking probability for the L4S queue
o the base AQM function that implements the PI algorithm o the base AQM function that implements the PI algorithm
dualpi2_update(lq, cq) (Figure 6) used to regularly update the dualpi2_update(lq, cq) (Figure 6) used to regularly update the
base probability (p'), which is squared for the Classic AQM as base probability (p'), which is squared for the Classic AQM as
well as being coupled across to the L4S queue. well as being coupled across to the L4S queue.
It also uses the following functions that are not shown in full here: It also uses the following functions that are not shown in full here:
skipping to change at page 30, line 22 skipping to change at page 32, line 22
8: RTT_max = 100 ms % Worst case RTT expected 8: RTT_max = 100 ms % Worst case RTT expected
9: RTT_typ = 15 ms % Typical RTT 9: RTT_typ = 15 ms % Typical RTT
11: % PI2 constants derived from above PI2 parameters 11: % PI2 constants derived from above PI2 parameters
10: p_Cmax = min(1/k^2, 1) % Max Classic drop/mark prob 10: p_Cmax = min(1/k^2, 1) % Max Classic drop/mark prob
12: target = RTT_typ % PI AQM Classic queue delay target 12: target = RTT_typ % PI AQM Classic queue delay target
13: Tupdate = min(RTT_typ, RTT_max/3) % PI sampling interval 13: Tupdate = min(RTT_typ, RTT_max/3) % PI sampling interval
14: alpha = 0.1 * Tupdate / RTT_max^2 % PI integral gain in Hz 14: alpha = 0.1 * Tupdate / RTT_max^2 % PI integral gain in Hz
15: beta = 0.3 / RTT_max % PI proportional gain in Hz 15: beta = 0.3 / RTT_max % PI proportional gain in Hz
16: 16:
17: % L4S ramp AQM parameters 17: % L4S ramp AQM parameters
18: minTh = 475 us % L4S min marking threshold in time units 18: minTh = 800 us % L4S min marking threshold in time units
19: range = 525 us % Range of L4S ramp in time units 19: range = 400 us % Range of L4S ramp in time units
20: Th_len = 2 * MTU % Min L4S marking threshold in bytes 20: Th_len = 2 * MTU % Min L4S marking threshold in bytes
21: % L4S constants incl. those derived from other parameters 21: % L4S constants incl. those derived from other parameters
22: p_Lmax = 1 % Max L4S marking prob 22: p_Lmax = 1 % Max L4S marking prob
23: floor = Th_len / MIN_LINK_RATE 23: floor = Th_len / MIN_LINK_RATE
24: if (minTh < floor) { 24: if (minTh < floor) {
25: % Shift ramp so minTh >= serialization time of 2 MTU 25: % Shift ramp so minTh >= serialization time of 2 MTU
26: minTh = floor 26: minTh = floor
27: } 27: }
28: maxTh = minTh+range % L4S max marking threshold in time units 28: maxTh = minTh+range % L4S max marking threshold in time units
29: } 29: }
skipping to change at page 31, line 19 skipping to change at page 33, line 19
5: % Packet classifier 5: % Packet classifier
6: if ( ecn(pkt) modulo 2 == 1 ) % ECN bits = ECT(1) or CE 6: if ( ecn(pkt) modulo 2 == 1 ) % ECN bits = ECT(1) or CE
7: lq.enqueue(pkt) 7: lq.enqueue(pkt)
8: else % ECN bits = not-ECT or ECT(0) 8: else % ECN bits = not-ECT or ECT(0)
9: cq.enqueue(pkt) 9: cq.enqueue(pkt)
10: } 10: }
Figure 3: Example Enqueue Pseudocode for DualQ Coupled PI2 AQM Figure 3: Example Enqueue Pseudocode for DualQ Coupled PI2 AQM
1: dualpi2_dequeue(lq, cq, pkt) { % Couples L4S & Classic queues 1: dualpi2_dequeue(lq, cq, pkt) { % Couples L4S & Classic queues
2: while ( lq.len() + cq.len() > 0 ) 2: while ( lq.len() + cq.len() > 0 ) {
3: if ( scheduler() == lq ) { 3: if ( scheduler() == lq ) {
4: lq.dequeue(pkt) % Scheduler chooses lq 4: lq.dequeue(pkt) % Scheduler chooses lq
5: p'_L = laqm(lq.time()) % Native L4S AQM 5: p'_L = laqm(lq.time()) % Native L4S AQM
6: p_L = max(p'_L, p_CL) % Combining function 6: p_L = max(p'_L, p_CL) % Combining function
7: if ( recur(p_L) ) % Linear marking 7: if ( recur(lq, p_L) ) % Linear marking
8: mark(pkt) 8: mark(pkt)
9: } else { 9: } else {
10: cq.dequeue(pkt) % Scheduler chooses cq 10: cq.dequeue(pkt) % Scheduler chooses cq
11: if ( p_C > rand() ) { % probability p_C = p'^2 11: if ( recur(cq, p_C) ) { % probability p_C = p'^2
12: if ( ecn(pkt) == 0 ) { % if ECN field = not-ECT 12: if ( ecn(pkt) == 0 ) { % if ECN field = not-ECT
13: drop(pkt) % squared drop 13: drop(pkt) % squared drop
14: continue % continue to the top of the while loop 14: continue % continue to the top of the while loop
15: } 15: }
16: mark(pkt) % squared mark 16: mark(pkt) % squared mark
17: } 17: }
18: } 18: }
19: return(pkt) % return the packet and stop 19: return(pkt) % return the packet and stop
20: } 20: }
21: return(NULL) % no packet to dequeue 21: return(NULL) % no packet to dequeue
22: } 22: }
23: recur(likelihood) { % Returns TRUE with a certain likelihood 23: recur(q, likelihood) { % Returns TRUE with a certain likelihood
24: count += likelihood 24: q.count += likelihood
25: if (count > 1) { 25: if (q.count > 1) {
26: count -= 1 26: q.count -= 1
27: return TRUE 27: return TRUE
28: } 28: }
29: return FALSE 29: return FALSE
30: } 30: }
Figure 4: Example Dequeue Pseudocode for DualQ Coupled PI2 AQM Figure 4: Example Dequeue Pseudocode for DualQ Coupled PI2 AQM
When packets arrive, first a common queue limit is checked as shown When packets arrive, first a common queue limit is checked as shown
in line 2 of the enqueuing pseudocode in Figure 3. This assumes a in line 2 of the enqueuing pseudocode in Figure 3. This assumes a
shared buffer for the two queues (Note b discusses the merits of shared buffer for the two queues (Note b discusses the merits of
skipping to change at page 33, line 28 skipping to change at page 35, line 28
to the RED algorithm, but simplified as follows: to the RED algorithm, but simplified as follows:
o The extent of the ramp is defined in units of queuing delay, not o The extent of the ramp is defined in units of queuing delay, not
bytes, so that configuration remains invariant as the queue bytes, so that configuration remains invariant as the queue
departure rate varies. departure rate varies.
o It uses instantaneous queueing delay, which avoids the complexity o It uses instantaneous queueing delay, which avoids the complexity
of smoothing, but also avoids embedding a worst-case RTT of of smoothing, but also avoids embedding a worst-case RTT of
smoothing delay in the network (see Section 2.1). smoothing delay in the network (see Section 2.1).
o The ramp rises linearly directly from 0 to 1, not to a an o The ramp rises linearly directly from 0 to 1, not to an
intermediate value of p'_L as RED would, because there is no need intermediate value of p'_L as RED would, because there is no need
to keep ECN marking probability low. to keep ECN marking probability low.
o Marking does not have to be randomized. Determinism is used o Marking does not have to be randomized. Determinism is used
instead of randomness; to reduce the delay necessary to smooth out instead of randomness; to reduce the delay necessary to smooth out
the noise of randomness from the signal. the noise of randomness from the signal.
The ramp function requires two configuration parameters, the minimum The ramp function requires two configuration parameters, the minimum
threshold (minTh) and the width of the ramp (range), both in units of threshold (minTh) and the width of the ramp (range), both in units of
queuing time), as shown in lines 18 & 19 of the initialization queuing time), as shown in lines 18 & 19 of the initialization
function in Figure 2. The ramp function can be configured as a step function in Figure 2. The ramp function can be configured as a step
(see Note c). (see Note c).
Although the DCTCP paper [Alizadeh-stability] recommends an ECN Although the DCTCP paper [Alizadeh-stability] recommends an ECN
marking threshold of 0.17*RTT_typ, it also shows that the threshold marking threshold of 0.17*RTT_typ, it also shows that the threshold
can be much shallower with hardly any worse under-utilization of the can be much shallower with hardly any worse under-utilization of the
link (because the amplitude of DCTCP's sawteeth is so small). Based link (because the amplitude of DCTCP's sawteeth is so small). Based
on extensive experiments, for the public Internet a default minimum on extensive experiments, for the public Internet the default minimum
ECN marking threshold of about RTT_typ/30 is recommended. ECN marking threshold in Figure 2 is considered a good compromise,
even though it is significantly smaller fraction of RTT_typ.
A minimum marking threshold parameter (Th_len) in transmission units A minimum marking threshold parameter (Th_len) in transmission units
(default 2 MTU) is also necessary to ensure that the ramp does not (default 2 MTU) is also necessary to ensure that the ramp does not
trigger excessive marking on slow links. The code in lines 24-27 of trigger excessive marking on slow links. The code in lines 24-27 of
the initialization function (Figure 2) converts 2 MTU into time units the initialization function (Figure 2) converts 2 MTU into time units
and shifts the ramp so that the min threshold is no shallower than and shifts the ramp so that the min threshold is no shallower than
this floor. this floor.
1: laqm(qdelay) { % Returns native L4S AQM probability 1: laqm(qdelay) { % Returns native L4S AQM probability
2: if (qdelay >= maxTh) 2: if (qdelay >= maxTh)
skipping to change at page 34, line 26 skipping to change at page 36, line 27
Figure 5: Example Pseudocode for the Native L4S AQM Figure 5: Example Pseudocode for the Native L4S AQM
1: dualpi2_update(lq, cq) { % Update p' every Tupdate 1: dualpi2_update(lq, cq) { % Update p' every Tupdate
2: curq = cq.time() % use queuing time of first-in Classic packet 2: curq = cq.time() % use queuing time of first-in Classic packet
3: p' = p' + alpha * (curq - target) + beta * (curq - prevq) 3: p' = p' + alpha * (curq - target) + beta * (curq - prevq)
4: p_CL = k * p' % Coupled L4S prob = base prob * coupling factor 4: p_CL = k * p' % Coupled L4S prob = base prob * coupling factor
5: p_C = p'^2 % Classic prob = (base prob)^2 5: p_C = p'^2 % Classic prob = (base prob)^2
6: prevq = curq 6: prevq = curq
7: } 7: }
(Clamping p' within the range [0,1] omitted for clarity - see text)
Figure 6: Example PI-Update Pseudocode for DualQ Coupled PI2 AQM Figure 6: Example PI-Update Pseudocode for DualQ Coupled PI2 AQM
The coupled marking probability, p_CL depends on the base probability The coupled marking probability, p_CL depends on the base probability
(p'), which is kept up to date by the core PI algorithm in Figure 6 (p'), which is kept up to date by the core PI algorithm in Figure 6
executed every Tupdate. executed every Tupdate.
Note that p' solely depends on the queuing time in the Classic queue. Note that p' solely depends on the queuing time in the Classic queue.
In line 2, the current queuing delay (curq) is evaluated from how In line 2, the current queuing delay (curq) is evaluated from how
long the head packet was in the Classic queue (cq). The function long the head packet was in the Classic queue (cq). The function
cq.time() (not shown) subtracts the time stamped at enqueue from the cq.time() (not shown) subtracts the time stamped at enqueue from the
skipping to change at page 36, line 8 skipping to change at page 38, line 12
for the L4S queue is k*alpha (with defaults alpha = 0.16 Hz and k=2, for the L4S queue is k*alpha (with defaults alpha = 0.16 Hz and k=2,
effective L4S alpha = 0.32 Hz). effective L4S alpha = 0.32 Hz).
Unlike in PIE [RFC8033], alpha and beta do not need to be tuned every Unlike in PIE [RFC8033], alpha and beta do not need to be tuned every
Tupdate dependent on p'. Instead, in PI2, alpha and beta are Tupdate dependent on p'. Instead, in PI2, alpha and beta are
independent of p' because the squaring applied to Classic traffic independent of p' because the squaring applied to Classic traffic
tunes them inherently. This is explained in [PI2], which also tunes them inherently. This is explained in [PI2], which also
explains why this more principled approach removes the need for most explains why this more principled approach removes the need for most
of the heuristics that had to be added to PIE. of the heuristics that had to be added to PIE.
Nonetheless, an implementer might wish to add selected heuristics to
either AQM. For instance the Linux reference DualPI2 implementation
includes the following:
o Prior to enqueuing an L4S packet, if the L queue contains <2
packets, the packet is flagged to suppress any native L4S AQM
marking at dequeue (which depends on sojourn time);
o Classic and coupled marking or dropping (i.e. based on p_C and
p_CL from the PI controller) is only applied to a packet if the
respective queue length in bytes is > 2 MTU (prior to enqueueing
the packet or after dequeuing it, depending on whether the AQM is
configured to be applied at enqueue or dequeue);
o In the WRR scheduler, the 'credit' indicating which queue should
transmit is only changed if there are packets in both queues (i.e.
if there is actual resource contention). This means that a
properly paced L flow might never be delayed by the WRR. The WRR
credit is reset in favour of the L queue when the link is idle.
An implementer might also wish to add other heuristics, e.g. burst
protection [RFC8033] or enhanced burst protection [RFC8034].
Notes: Notes:
a. The drain rate of the queue can vary if it is scheduled relative a. The drain rate of the queue can vary if it is scheduled relative
to other queues, or to cater for fluctuations in a wireless to other queues, or to cater for fluctuations in a wireless
medium. To auto-adjust to changes in drain rate, the queue must medium. To auto-adjust to changes in drain rate, the queue needs
be measured in time, not bytes or packets [AQMmetrics] [CoDel]. to be measured in time, not bytes or packets [AQMmetrics]
Queuing delay could be measured directly by storing a per-packet [CoDel]. Queuing delay could be measured directly by storing a
time-stamp as each packet is enqueued, and subtracting this from per-packet time-stamp as each packet is enqueued, and subtracting
the system time when the packet is dequeued. If time-stamping is this from the system time when the packet is dequeued. If time-
not easy to introduce with certain hardware, queuing delay could stamping is not easy to introduce with certain hardware, queuing
be predicted indirectly by dividing the size of the queue by the delay could be predicted indirectly by dividing the size of the
predicted departure rate, which might be known precisely for some queue by the predicted departure rate, which might be known
link technologies (see for example [RFC8034]). precisely for some link technologies (see for example [RFC8034]).
b. Line 2 of the dualpi2_enqueue() function (Figure 3) assumes an b. Line 2 of the dualpi2_enqueue() function (Figure 3) assumes an
implementation where lq and cq share common buffer memory. An implementation where lq and cq share common buffer memory. An
alternative implementation could use separate buffers for each alternative implementation could use separate buffers for each
queue, in which case the arriving packet would have to be queue, in which case the arriving packet would have to be
classified first to determine which buffer to check for available classified first to determine which buffer to check for available
space. The choice is a trade off; a shared buffer can use less space. The choice is a trade off; a shared buffer can use less
memory whereas separate buffers isolate the L4S queue from tail- memory whereas separate buffers isolate the L4S queue from tail-
drop due to large bursts of Classic traffic (e.g. a Classic TCP drop due to large bursts of Classic traffic (e.g. a Classic Reno
during slow-start over a long RTT). TCP during slow-start over a long RTT).
c. There has been some concern that using the step function of DCTCP c. There has been some concern that using the step function of DCTCP
for the Native L4S AQM requires end-systems to smooth the signal for the Native L4S AQM requires end-systems to smooth the signal
for an unnecessarily large number of round trips to ensure for an unnecessarily large number of round trips to ensure
sufficient fidelity. A ramp is no worse than a step in initial sufficient fidelity. A ramp is no worse than a step in initial
experiments with existing DCTCP. Therefore, it is recommended experiments with existing DCTCP. Therefore, it is recommended
that a ramp is configured in place of a step, which will allow that a ramp is configured in place of a step, which will allow
congestion control algorithms to investigate faster smoothing congestion control algorithms to investigate faster smoothing
algorithms. algorithms.
A ramp is more general that a step, because an operator can A ramp is more general that a step, because an operator can
effectively turn the ramp into a step function, as used by DCTCP, effectively turn the ramp into a step function, as used by DCTCP,
by setting the range to zero. There will not be a divide by zero by setting the range to zero. There will not be a divide by zero
problem at line 4 of Figure 5 because, if minTh is equal to problem at line 5 of Figure 5 because, if minTh is equal to
maxTh, the condition for this ramp calculation cannot arise. maxTh, the condition for this ramp calculation cannot arise.
A.2. Pass #2: Overload Details A.2. Pass #2: Overload Details
Figure 7 repeats the dequeue function of Figure 4, but with overload Figure 7 repeats the dequeue function of Figure 4, but with overload
details added. Similarly Figure 8 repeats the core PI algorithm of details added. Similarly Figure 8 repeats the core PI algorithm of
Figure 6 with overload details added. The initialization, enqueue, Figure 6 with overload details added. The initialization, enqueue,
L4S AQM and recur functions are unchanged. L4S AQM and recur functions are unchanged.
In line 10 of the initialization function (Figure 2), the maximum In line 10 of the initialization function (Figure 2), the maximum
skipping to change at page 37, line 24 skipping to change at page 39, line 49
In practice, 25% has been found to be a good threshold to preserve In practice, 25% has been found to be a good threshold to preserve
fairness between ECN capable and non ECN capable traffic. This fairness between ECN capable and non ECN capable traffic. This
protects the queues against both temporary overload from responsive protects the queues against both temporary overload from responsive
flows and more persistent overload from any unresponsive traffic that flows and more persistent overload from any unresponsive traffic that
falsely claims to be responsive to ECN. falsely claims to be responsive to ECN.
When the Classic ECN marking probability reaches the p_Cmax threshold When the Classic ECN marking probability reaches the p_Cmax threshold
(1/k^2), the marking probability coupled to the L4S queue, p_CL will (1/k^2), the marking probability coupled to the L4S queue, p_CL will
always be 100% for any k (by equation (1) in Section 2). So, for always be 100% for any k (by equation (1) in Section 2). So, for
readability, the constant p_Lmax is defined as 1 in line 22 of the readability, the constant p_Lmax is defined as 1 in line 22 of the
initialization function Figure 2. This is intended to ensure that initialization function (Figure 2). This is intended to ensure that
the L4S queue starts to introduce dropping once ECN-marking saturates the L4S queue starts to introduce dropping once ECN-marking saturates
at 100% and can rise no further. The 'Prague L4S' requirements at 100% and can rise no further. The 'Prague L4S' requirements
[I-D.ietf-tsvwg-ecn-l4s-id] state that, when an L4S congestion [I-D.ietf-tsvwg-ecn-l4s-id] state that, when an L4S congestion
control detects a drop, it falls back to a response that coexists control detects a drop, it falls back to a response that coexists
with 'Classic' TCP. So it is correct that, when the L4S queue drops with 'Classic' Reno congestion control. So it is correct that, when
packets, it drops them proportional to p'^2, as if they are Classic the L4S queue drops packets, it drops them proportional to p'^2, as
packets. if they are Classic packets.
Both these switch-overs are triggered by the tests for overload Both these switch-overs are triggered by the tests for overload
introduced in lines 4b and 12b of the dequeue function (Figure 7). introduced in lines 4b and 12b of the dequeue function (Figure 7).
Lines 8c to 8g drop L4S packets with probability p'^2. Lines 8h to Lines 8c to 8g drop L4S packets with probability p'^2. Lines 8h to
8i mark the remaining packets with probability p_CL. Given p_Lmax = 8i mark the remaining packets with probability p_CL. Given p_Lmax =
1, all remaining packets will be marked because, to have reached the 1, all remaining packets will be marked because, to have reached the
else block at line 8b, p_CL >= 1. else block at line 8b, p_CL >= 1.
Lines 2c to 2d in the core PI algorithm (Figure 8) deal with overload Lines 2c to 2d in the core PI algorithm (Figure 8) deal with overload
of the L4S queue when there is no Classic traffic. This is of the L4S queue when there is no Classic traffic. This is
necessary, because the core PI algorithm maintains the appropriate necessary, because the core PI algorithm maintains the appropriate
drop probability to regulate overload, but it depends on the length drop probability to regulate overload, but it depends on the length
of the Classic queue. If there is no Classic queue the naive PI of the Classic queue. If there is no Classic queue the naive PI
update function in Figure 6 would drop nothing, even if the L4S queue update function in Figure 6 would drop nothing, even if the L4S queue
were overloaded - so tail drop would have to take over (lines 2 and 3 were overloaded - so tail drop would have to take over (lines 2 and 3
of Figure 3). of Figure 3).
Instead, the test at line 2a of the full PI update function in Instead, the test at line 2a of the full PI update function in
Figure 8 keeps delay on target using drop. If the test at line 2a of Figure 8 keeps delay on target using drop. If the test at line 2a of
finds that the Classic queue is empty, line 2d measures the current Figure 8 finds that the Classic queue is empty, line 2d measures the
queue delay using the L4S queue instead. While the L4S queue is not current queue delay using the L4S queue instead. While the L4S queue
overloaded, its delay will always be tiny compared to the target is not overloaded, its delay will always be tiny compared to the
Classic queue delay. So p_CL will be driven to zero, and the L4S target Classic queue delay. So p_CL will be driven to zero, and the
queue will naturally be governed solely by p'_L from the native L4S L4S queue will naturally be governed solely by p'_L from the native
AQM (lines 5 and 6 of the dequeue algorithm in Figure 7). But, if L4S AQM (lines 5 and 6 of the dequeue algorithm in Figure 7). But,
unresponsive L4S source(s) cause overload, the DualQ transitions if unresponsive L4S source(s) cause overload, the DualQ transitions
smoothly to L4S marking based on the PI algorithm. If overload smoothly to L4S marking based on the PI algorithm. If overload
increases further, it naturally transitions from marking to dropping increases further, it naturally transitions from marking to dropping
by the switch-over mechanism already described. by the switch-over mechanism already described.
1: dualpi2_dequeue(lq, cq, pkt) { % Couples L4S & Classic queues 1: dualpi2_dequeue(lq, cq, pkt) { % Couples L4S & Classic queues
2: while ( lq.len() + cq.len() > 0 ) { 2: while ( lq.len() + cq.len() > 0 ) {
3: if ( scheduler() == lq ) { 3: if ( scheduler() == lq ) {
4a: lq.dequeue(pkt) % L4S scheduled 4a: lq.dequeue(pkt) % L4S scheduled
4b: if ( p_CL < p_Lmax ) { % Check for overload saturation 4b: if ( p_CL < p_Lmax ) { % Check for overload saturation
5: p'_L = laqm(lq.time()) % Native L4S AQM 5: p'_L = laqm(lq.time()) % Native L4S AQM
6: p_L = max(p'_L, p_CL) % Combining function 6: p_L = max(p'_L, p_CL) % Combining function
7: if ( recur(p_L) ) % Linear marking 7: if ( recur(lq, p_L) ) % Linear marking
8a: mark(pkt) 8a: mark(pkt)
8b: } else { % overload saturation 8b: } else { % overload saturation
8c: if ( p_C > rand() ) { % probability p_C = p'^2 8c: if ( recur(lq, p_C) ) { % probability p_C = p'^2
8e: drop(pkt) % revert to Classic drop due to overload 8e: drop(pkt) % revert to Classic drop due to overload
8f: continue % continue to the top of the while loop 8f: continue % continue to the top of the while loop
8g: } 8g: }
8h: if ( p_CL > rand() ) % probability p_CL = k * p' 8h: if ( recur(lq, p_CL) ) % probability p_CL = k * p'
8i: mark(pkt) % linear marking of remaining packets 8i: mark(pkt) % linear marking of remaining packets
8j: } 8j: }
9: } else { 9: } else {
10: cq.dequeue(pkt) % Classic scheduled 10: cq.dequeue(pkt) % Classic scheduled
11: if ( p_C > rand() ) { % probability p_C = p'^2 11: if ( recur(cq, p_C) ) { % probability p_C = p'^2
12a: if ( (ecn(pkt) == 0) % ECN field = not-ECT 12a: if ( (ecn(pkt) == 0) % ECN field = not-ECT
12b: OR (p_C >= p_Cmax) ) { % Overload disables ECN 12b: OR (p_C >= p_Cmax) ) { % Overload disables ECN
13: drop(pkt) % squared drop, redo loop 13: drop(pkt) % squared drop, redo loop
14: continue % continue to the top of the while loop 14: continue % continue to the top of the while loop
15: } 15: }
16: mark(pkt) % squared mark 16: mark(pkt) % squared mark
17: } 17: }
18: } 18: }
19: return(pkt) % return the packet and stop 19: return(pkt) % return the packet and stop
20: } 20: }
skipping to change at page 45, line 26 skipping to change at page 48, line 26
comparing it with the maximum out of two random numbers (assuming comparing it with the maximum out of two random numbers (assuming
U=1). Comparing it with the maximum out of two is the same as the U=1). Comparing it with the maximum out of two is the same as the
logical `AND' of two tests, which ensures drop probability rises logical `AND' of two tests, which ensures drop probability rises
with the square of queuing time. with the square of queuing time.
The AQM functions in each queue (lines 5c & 10b) are two cases of a The AQM functions in each queue (lines 5c & 10b) are two cases of a
new generalization of RED called Curvy RED, motivated as follows. new generalization of RED called Curvy RED, motivated as follows.
When the performance of this AQM was compared with fq_CoDel and PIE, When the performance of this AQM was compared with fq_CoDel and PIE,
their goal of holding queuing delay to a fixed target seemed their goal of holding queuing delay to a fixed target seemed
misguided [CRED_Insights]. As the number of flows increases, if the misguided [CRED_Insights]. As the number of flows increases, if the
AQM does not allow TCP to increase queuing delay, it has to introduce AQM does not allow host congestion controllers to increase queuing
abnormally high levels of loss. Then loss rather than queuing delay, it has to introduce abnormally high levels of loss. Then loss
becomes the dominant cause of delay for short flows, due to timeouts rather than queuing becomes the dominant cause of delay for short
and tail losses. flows, due to timeouts and tail losses.
Curvy RED constrains delay with a softened target that allows some Curvy RED constrains delay with a softened target that allows some
increase in delay as load increases. This is achieved by increasing increase in delay as load increases. This is achieved by increasing
drop probability on a convex curve relative to queue growth (the drop probability on a convex curve relative to queue growth (the
square curve in the Classic queue, if U=1). Like RED, the curve hugs square curve in the Classic queue, if U=1). Like RED, the curve hugs
the zero axis while the queue is shallow. Then, as load increases, the zero axis while the queue is shallow. Then, as load increases,
it introduces a growing barrier to higher delay. But, unlike RED, it it introduces a growing barrier to higher delay. But, unlike RED, it
requires only two parameters, not three. The disadvantage of Curvy requires only two parameters, not three. The disadvantage of Curvy
RED is that it is not adapted to a wide range of RTTs. Curvy RED can RED is that it is not adapted to a wide range of RTTs. Curvy RED can
be used as is when the RTT range to be supported is limited, be used as is when the RTT range to be supported is limited,
otherwise an adaptation mechanism is required. otherwise an adaptation mechanism is required.
From our limited experiments with Curvy RED so far, recommended From our limited experiments with Curvy RED so far, recommended
values of these parameters are: S_C = -1; g_C = 5; T = 5 * MTU at the values of these parameters are: S_C = -1; g_C = 5; T = 5 * MTU at the
link rate (about 1ms at 60Mb/s) for the range of base RTTs typical on link rate (about 1ms at 60Mb/s) for the range of base RTTs typical on
the public Internet. [CRED_Insights] explains why these parameters the public Internet. [CRED_Insights] explains why these parameters
are applicable whatever rate link this AQM implementation is deployed are applicable whatever rate link this AQM implementation is deployed
on and how the parameters would need to be adjusted for a scenario on and how the parameters would need to be adjusted for a scenario
with a different range of RTTs (e.g. a data centre). The setting of with a different range of RTTs (e.g. a data centre). The setting of
k depends on policy (see Section 2.5 and Appendix C respectively for k depends on policy (see Section 2.5 and Appendix C.2 respectively
its recommended setting and guidance on alternatives). for its recommended setting and guidance on alternatives).
There is also a cUrviness parameter, U, which is a small positive There is also a cUrviness parameter, U, which is a small positive
integer. It is likely to take the same hard-coded value for all integer. It is likely to take the same hard-coded value for all
implementations, once experiments have determined a good value. Only implementations, once experiments have determined a good value. Only
U=1 has been used in experiments so far, but results might be even U=1 has been used in experiments so far, but results might be even
better with U=2 or higher. better with U=2 or higher.
Notes: Notes:
1. The alternative of applying the AQMs at enqueue would shift some 1. The alternative of applying the AQMs at enqueue would shift some
skipping to change at page 48, line 14 skipping to change at page 51, line 14
integer nanoseconds, making the values about 2^30 times larger than integer nanoseconds, making the values about 2^30 times larger than
when the units were seconds, ii) then in lines 3 and 9 an adjustment when the units were seconds, ii) then in lines 3 and 9 an adjustment
of -2 to the right bit-shift multiplies the result by 2^2, to of -2 to the right bit-shift multiplies the result by 2^2, to
complete the scaling by 2^32. complete the scaling by 2^32.
In line 8 of the initialization function, the EWMA constant gamma is In line 8 of the initialization function, the EWMA constant gamma is
represented as an integer power of 2, g_C, so that in line 9 of the represented as an integer power of 2, g_C, so that in line 9 of the
integer code the division needed to weight the moving average can be integer code the division needed to weight the moving average can be
implemented by a right bit-shift (>> g_C). implemented by a right bit-shift (>> g_C).
Appendix C. Guidance on Controlling Throughput Equivalence Appendix C. Choice of Coupling Factor, k
C.1. RTT-Dependence
Where Classic flows compete for the same capacity, their relative
flow rates depend not only on the congestion probability, but also on
their end-to-end RTT (= base RTT + queue delay). The rates of
competing Reno [RFC5681] flows are roughly inversely proportional to
their RTTs. Cubic exhibits similar RTT-dependence when in Reno-
compatibility mode, but is less RTT-dependent otherwise.
Until the early experiments with the DualQ Coupled AQM, the
importance of the reasonably large Classic queue in mitigating RTT-
dependence had not been appreciated. Appendix A.1.5 of
[I-D.ietf-tsvwg-ecn-l4s-id] uses numerical examples to explain why
bloated buffers had concealed the RTT-dependence of Classic
congestion controls before that time. Then it explains why, the more
that queuing delays have reduced, the more that RTT-dependence has
surfaced as a potential starvation problem for long RTT flows.
Given that congestion control on end-systems is voluntary, there is
no reason why it has to be voluntarily RTT-dependent. Therefore
[I-D.ietf-tsvwg-ecn-l4s-id] requires L4S congestion controls to be
significantly less RTT-dependent than the standard Reno congestion
control [RFC5681]. Following this approach means there is no need
for network devices to address RTT-dependence, although there would
be no harm if they did, which per-flow queuing inherently does.
At the time of writing, the range of approaches to RTT-dependence in
L4S congestion controls has not settled. Therefore, the guidance on
the choice of the coupling factor in Appendix C.2 is given against
DCTCP [RFC8257], which has well-understood RTT-dependence. The
guidance is given for various RTT ratios, so that it can be adapted
to future circumstances.
C.2. Guidance on Controlling Throughput Equivalence
+---------------+------+-------+ +---------------+------+-------+
| RTT_C / RTT_L | Reno | Cubic | | RTT_C / RTT_L | Reno | Cubic |
+---------------+------+-------+ +---------------+------+-------+
| 1 | k'=1 | k'=0 | | 1 | k'=1 | k'=0 |
| 2 | k'=2 | k'=1 | | 2 | k'=2 | k'=1 |
| 3 | k'=2 | k'=2 | | 3 | k'=2 | k'=2 |
| 4 | k'=3 | k'=2 | | 4 | k'=3 | k'=2 |
| 5 | k'=3 | k'=3 | | 5 | k'=3 | k'=3 |
+---------------+------+-------+ +---------------+------+-------+
Table 1: Value of k' for which DCTCP throughput is roughly the same Table 1: Value of k' for which DCTCP throughput is roughly the same
as Reno or Cubic, for some example RTT ratios as Reno or Cubic, for some example RTT ratios
k' is related to k in Equation (1) (Section 2.1) by k=2^k'. In the above appendices that give example DualQ Coupled algorithms,
to aid efficient implementation, a coupling factor that is an integer
power of 2 is always used. k' is always used to denote the power. k'
is related to the coupling factor k in Equation (1) (Section 2.1) by
k=2^k'.
To determine the appropriate policy, the operator first has to judge To determine the appropriate coupling factor policy, the operator
whether it wants DCTCP flows to have roughly equal throughput with first has to judge whether it wants DCTCP flows to have roughly equal
Reno or with Cubic (because, even in its Reno-compatibility mode, throughput with Reno or with Cubic (because, even in its Reno-
Cubic is about 1.4 times more aggressive than Reno). Then the compatibility mode, Cubic is about 1.4 times more aggressive than
operator needs to decide at what ratio of RTTs it wants DCTCP and Reno). Then the operator needs to decide at what ratio of RTTs it
Classic flows to have roughly equal throughput. For example choosing wants DCTCP and Classic flows to have roughly equal throughput. For
k'=0 (equivalent to k=1) will make DCTCP throughput roughly the same example choosing k'=0 (equivalent to k=1) will make DCTCP throughput
as Cubic, _if their RTTs are the same_. roughly the same as Cubic, _if their RTTs are the same_.
However, even if the base RTTs are the same, the actual RTTs are However, even if the base RTTs are the same, the actual RTTs are
unlikely to be the same, because Classic (Cubic or Reno) traffic unlikely to be the same, because Classic (Cubic or Reno) traffic
needs a large queue to avoid under-utilization and excess drop, needs roughly a typical base round trip of queue to avoid under-
whereas L4S (DCTCP) does not. The operator might still choose this utilization and excess drop. Whereas L4S (DCTCP) does not. The
policy if it judges that DCTCP throughput should be rewarded for operator might still choose this policy if it judges that DCTCP
keeping its own queue short. throughput should be rewarded for keeping its own queue short.
On the other hand, the operator will choose one of the higher values On the other hand, the operator will choose one of the higher values
for k', if it wants to slow DCTCP down to roughly the same throughput for k', if it wants to slow DCTCP down to roughly the same throughput
as Classic flows, to compensate for Classic flows slowing themselves as Classic flows, to compensate for Classic flows slowing themselves
down by causing themselves extra queuing delay. down by causing themselves extra queuing delay.
The values for k' in the table are derived from the formulae, which The values for k' in the table are derived from the formulae below,
was developed in [DCttH15]: which were developed in [DCttH15]:
2^k' = 1.64 (RTT_reno / RTT_dc) (2) 2^k' = 1.64 (RTT_reno / RTT_dc) (5)
2^k' = 1.19 (RTT_cubic / RTT_dc ) (3) 2^k' = 1.19 (RTT_cubic / RTT_dc ) (6)
For localized traffic from a particular ISP's data centre, using the For localized traffic from a particular ISP's data centre, using the
measured RTTs, it was calculated that a value of k'=3 (equivalant to measured RTTs, it was calculated that a value of k'=3 (equivalant to
k=8) would achieve throughput equivalence, and experiments verified k=8) would achieve throughput equivalence, and experiments verified
the formula very closely. the formula very closely.
For a typical mix of RTTs from local data centres and across the For a typical mix of RTTs from local data centres and across the
general Internet, a value of k'=1 (equivalent to k=2) is recommended general Internet, a value of k'=1 (equivalent to k=2) is recommended
as a good workable compromise. as a good workable compromise.
skipping to change at page 49, line 31 skipping to change at page 53, line 25
Koen De Schepper Koen De Schepper
Nokia Bell Labs Nokia Bell Labs
Antwerp Antwerp
Belgium Belgium
Email: koen.de_schepper@nokia.com Email: koen.de_schepper@nokia.com
URI: https://www.bell-labs.com/usr/koen.de_schepper URI: https://www.bell-labs.com/usr/koen.de_schepper
Bob Briscoe (editor) Bob Briscoe (editor)
CableLabs Independent
UK UK
Email: ietf@bobbriscoe.net Email: ietf@bobbriscoe.net
URI: http://bobbriscoe.net/ URI: http://bobbriscoe.net/
Greg White Greg White
CableLabs CableLabs
Louisville, CO Louisville, CO
US US
 End of changes. 99 change blocks. 
280 lines changed or deleted 465 lines changed or added

This html diff was produced by rfcdiff 1.47. The latest version is available from http://tools.ietf.org/tools/rfcdiff/