draft-briscoe-tsvwg-re-ecn-tcp-06.txt   draft-briscoe-tsvwg-re-ecn-tcp-07.txt 
Transport Area Working Group B. Briscoe Transport Area Working Group B. Briscoe
Internet-Draft BT & UCL Internet-Draft BT & UCL
Intended status: Standards Track A. Jacquet Intended status: Standards Track A. Jacquet
Expires: January 15, 2009 T. Moncaster Expires: September 4, 2009 T. Moncaster
A. Smith A. Smith
BT BT
July 14, 2008 March 3, 2009
Re-ECN: Adding Accountability for Causing Congestion to TCP/IP Re-ECN: Adding Accountability for Causing Congestion to TCP/IP
draft-briscoe-tsvwg-re-ecn-tcp-06 draft-briscoe-tsvwg-re-ecn-tcp-07
Status of this Memo Status of This Memo
By submitting this Internet-Draft, each author represents that any By submitting this Internet-Draft, each author represents that any
applicable patent or other IPR claims of which he or she is aware applicable patent or other IPR claims of which he or she is aware
have been or will be disclosed, and any of which he or she becomes have been or will be disclosed, and any of which he or she becomes
aware will be disclosed, in accordance with Section 6 of BCP 79. aware will be disclosed, in accordance with Section 6 of BCP 79.
Internet-Drafts are working documents of the Internet Engineering Internet-Drafts are working documents of the Internet Engineering
Task Force (IETF), its areas, and its working groups. Note that Task Force (IETF), its areas, and its working groups. Note that
other groups may also distribute working documents as Internet- other groups may also distribute working documents as Internet-
Drafts. Drafts.
skipping to change at page 1, line 37 skipping to change at page 1, line 37
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."
The list of current Internet-Drafts can be accessed at The list of current Internet-Drafts can be accessed at
http://www.ietf.org/ietf/1id-abstracts.txt. http://www.ietf.org/ietf/1id-abstracts.txt.
The list of Internet-Draft Shadow Directories can be accessed at The list of Internet-Draft Shadow Directories can be accessed at
http://www.ietf.org/shadow.html. http://www.ietf.org/shadow.html.
This Internet-Draft will expire on January 15, 2009. This Internet-Draft will expire on September 4, 2009.
Copyright Notice Copyright Notice
Copyright (C) The IETF Trust (2008). Copyright (c) 2009 IETF Trust and the persons identified as the
document authors. All rights reserved.
This document is subject to BCP 78 and the IETF Trust's Legal
Provisions Relating to IETF Documents in effect on the date of
publication of this document (http://trustee.ietf.org/license-info).
Please review these documents carefully, as they describe your rights
and restrictions with respect to this document.
Abstract Abstract
This document introduces a new protocol for explicit congestion This document introduces a new protocol for explicit congestion
notification (ECN), termed re-ECN, which can be deployed notification (ECN), termed re-ECN, which can be deployed
incrementally around unmodified routers. It enbales the the upstream incrementally around unmodified routers. The protocol works by
party at any trust boundary in the internetwork to be held arranging an extended ECN field in each packet so that, as it crosses
responsible for the congestion they cause, or allow to be caused. any interface in an internetwork, it will carry a truthful prediction
of congestion on the remainder of its path. The purpose of this
So, networks can introduce straightforward accountability for document is to specify the re-ECN protocol at the IP layer and to
congestion and policing mechanisms for incoming traffic from end- give guidelines on any consequent changes required to transport
customers or from neighbouring network domains. The protocol works protocols. It includes the changes required to TCP both as an
by arranging an extended ECN field in each packet so that, as it example and as a specification. It briefly gives examples of
crosses any interface in an internetwork, it will carry a truthful
prediction of congestion on the remainder of its path. The purpose
of this document is to specify the re-ECN protocol at the IP layer
and to give guidelines on any consequent changes required to
transport protocols. It includes the changes required to TCP both as
an example and as a specification. It also gives examples of
mechanisms that can use the protocol to ensure data sources respond mechanisms that can use the protocol to ensure data sources respond
correctly to congestion. And it describes example mechanisms that correctly to congestion,and these are described more fully in a
ensure the dominant selfish strategy of both network domains and end- companion document [re-ecn-motive].
points will be to set the extended ECN field honestly.
Authors' Statement: Status (to be removed by the RFC Editor) Authors' Statement: Status (to be removed by the RFC Editor)
Although the re-ECN protocol is intended to make a simple but far- Although the re-ECN protocol is intended to make a simple but far-
reaching change to the Internet architecture, the most immediate reaching change to the Internet architecture, the most immediate
priority for the authors is to delay any move of the ECN nonce to priority for the authors is to delay any move of the ECN nonce to
Proposed Standard status. The argument for this position is Proposed Standard status. The argument for this position is
developed in Appendix I. developed in Appendix E.
Changes from previous drafts (to be removed by the RFC Editor) Changes from previous drafts (to be removed by the RFC Editor)
Full diffs created using the rfcdiff tool are available at Full diffs created using the rfcdiff tool are available at
<http://www.cs.ucl.ac.uk/staff/B.Briscoe/pubs.html#retcp> <http://www.cs.ucl.ac.uk/staff/B.Briscoe/pubs.html#retcp>
From -05 to -06 (current version): From -06 to -07 (current version):
Clarifications made to Section 1 and Section 3.
Minor editorial changes throughout.
From -04 to -05:
Completed justification for packet marking with FNE during slow-
start(Appendix D).
Minor editorial changes throughout.
From -03 to -04:
Clarified reasons for holding back ECN nonce (Section 3.3 &
Appendix I).
Clarified Figure 2.
Added Section 4.1.1.1 on equivalence of drops and ECN marks.
Improved precision of Section 5.6 on IP in IP tunnels.
Explained the RTT fairness is possible to enforce, but unlikely to
be required (Section 6.1.3 & Appendix F).
Explained that bulk per-user policing should be adequate but per-
flow policing is also possible if desired, though it is not likely
to be necessary (Section 6.1.5 & Appendix G).
Reinforced need for passive policing at inter-domain borders to
enable all-optical networking (Section 6.1.6).
Minor editorial changes throughout.
From -02 to -03: Major changes made following splitting this protocol document from
the related motivations document [re-ecn-motive].
Started guidelines for re-ECN support in DCCP and SCTP. Significant re-ordering of remaining text.
Added annex on limitations of nonce mechanism. New terminology introduced for clarity.
Minor editorial changes throughout. Minor editorial changes throughout.
From -01 to -02:
Explanation on informal terminology in Section 3.5 clarified.
IPv6 wire protocol encoding added (Section 5.2).
Text on (non-)issues with tunnels, encryption and link layer
congestion notification added (Section 5.6 & Section 5.7).
Section added giving evolvability arguments against encouraging
bottleneck policing (Section 6.1.2). And text on re-ECN's
evolvability by design added to Section 6.1.3
Text on inter-domain policing (Section 6.1.6) and inter-domain
fail-safes (Section 6.1.7) added.
From -00 to -01:
Encoding of re-ECN wire protocol changed for reasons given in
Appendix B and consequently draft substantially re-written.
Substantial text added in sections on applications, incremental
deployment, architectural rationale and security considerations.
Table of Contents Table of Contents
1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . 6 1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . 5
2. Requirements notation . . . . . . . . . . . . . . . . . . . . 8 2. Requirements notation . . . . . . . . . . . . . . . . . . . . 6
3. Protocol Overview . . . . . . . . . . . . . . . . . . . . . . 8 3. Terminology . . . . . . . . . . . . . . . . . . . . . . . . . 6
3.1. Background and Applicability . . . . . . . . . . . . . . . 8 4. Protocol Overview . . . . . . . . . . . . . . . . . . . . . . 7
3.2. Simplified Re-ECN Protocol . . . . . . . . . . . . . . . . 10 4.1. Simplified Re-ECN Protocol . . . . . . . . . . . . . . . . 7
3.3. Re-ECN Abstracted Network Layer Wire Protocol (IPv4 or 4.1.1. Congestion Control and Policing the Protocol . . . . . 7
v6) . . . . . . . . . . . . . . . . . . . . . . . . . . . 10 4.1.2. Background and Applicability . . . . . . . . . . . . . 8
3.4. Re-ECN Protocol Operation . . . . . . . . . . . . . . . . 12 4.2. Re-ECN Abstracted Network Layer Wire Protocol (IPv4 or
3.5. Informal Terminology . . . . . . . . . . . . . . . . . . . 14 v6) . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
4. Transport Layers . . . . . . . . . . . . . . . . . . . . . . . 15 4.3. Re-ECN Protocol Operation . . . . . . . . . . . . . . . . 10
4.1. TCP . . . . . . . . . . . . . . . . . . . . . . . . . . . 16 4.4. Positive and Negative Flows . . . . . . . . . . . . . . . 12
4.1.1. RECN mode: Full Re-ECN capable transport . . . . . . . 17 5. Network Layer . . . . . . . . . . . . . . . . . . . . . . . . 13
4.1.2. RECN-Co mode: Re-ECT Sender with a RFC3168 5.1. Re-ECN IPv4 Wire Protocol . . . . . . . . . . . . . . . . 13
compliant ECN Receiver . . . . . . . . . . . . . . . . 20 5.2. Re-ECN IPv6 Wire Protocol . . . . . . . . . . . . . . . . 15
4.1.3. Capability Negotiation . . . . . . . . . . . . . . . . 21 5.3. Router Forwarding Behaviour . . . . . . . . . . . . . . . 16
4.1.4. Extended ECN (EECN) Field Settings during Flow 5.4. Justification for Setting the First SYN to FNE . . . . . . 17
Start or after Idle Periods . . . . . . . . . . . . . 23 5.5. Control and Management . . . . . . . . . . . . . . . . . . 18
4.1.5. Pure ACKS, Retransmissions, Window Probes and 5.5.1. Negative Balance Warning . . . . . . . . . . . . . . . 18
Partial ACKs . . . . . . . . . . . . . . . . . . . . . 27 5.5.2. Rate Response Control . . . . . . . . . . . . . . . . 19
4.2. Other Transports . . . . . . . . . . . . . . . . . . . . . 27 5.6. IP in IP Tunnels . . . . . . . . . . . . . . . . . . . . . 19
4.2.1. General Guidelines for Adding Re-ECN to Other 5.7. Non-Issues . . . . . . . . . . . . . . . . . . . . . . . . 20
Transports . . . . . . . . . . . . . . . . . . . . . . 27 6. Transport Layers . . . . . . . . . . . . . . . . . . . . . . . 21
4.2.2. Guidelines for adding Re-ECN to RSVP or NSIS . . . . . 28 6.1. TCP . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
4.2.3. Guidelines for adding Re-ECN to DCCP . . . . . . . . . 28 6.1.1. RECN mode: Full Re-ECN capable transport . . . . . . . 22
4.2.4. Guidelines for adding Re-ECN to SCTP . . . . . . . . . 29 6.1.2. RECN-Co mode: Re-ECT Sender with a RFC3168
5. Network Layer . . . . . . . . . . . . . . . . . . . . . . . . 29 compliant ECN Receiver . . . . . . . . . . . . . . . . 24
5.1. Re-ECN IPv4 Wire Protocol . . . . . . . . . . . . . . . . 29 6.1.3. Capability Negotiation . . . . . . . . . . . . . . . . 26
5.2. Re-ECN IPv6 Wire Protocol . . . . . . . . . . . . . . . . 30 6.1.4. Extended ECN (EECN) Field Settings during Flow
5.3. Router Forwarding Behaviour . . . . . . . . . . . . . . . 31 Start or after Idle Periods . . . . . . . . . . . . . 27
5.4. Justification for Setting the First SYN to FNE . . . . . . 33 6.1.5. Pure ACKS, Retransmissions, Window Probes and
5.5. Control and Management . . . . . . . . . . . . . . . . . . 34 Partial ACKs . . . . . . . . . . . . . . . . . . . . . 31
5.5.1. Negative Balance Warning . . . . . . . . . . . . . . . 34 6.2. Other Transports . . . . . . . . . . . . . . . . . . . . . 32
5.5.2. Rate Response Control . . . . . . . . . . . . . . . . 35 6.2.1. General Guidelines for Adding Re-ECN to Other
5.6. IP in IP Tunnels . . . . . . . . . . . . . . . . . . . . . 35 Transports . . . . . . . . . . . . . . . . . . . . . . 32
5.7. Non-Issues . . . . . . . . . . . . . . . . . . . . . . . . 36 6.2.2. Guidelines for adding Re-ECN to RSVP or NSIS . . . . . 32
6. Applications . . . . . . . . . . . . . . . . . . . . . . . . . 37 6.2.3. Guidelines for adding Re-ECN to DCCP . . . . . . . . . 33
6.1. Policing Congestion Response . . . . . . . . . . . . . . . 37 6.2.4. Guidelines for adding Re-ECN to SCTP . . . . . . . . . 33
6.1.1. The Policing Problem . . . . . . . . . . . . . . . . . 37 7. Incremental Deployment . . . . . . . . . . . . . . . . . . . . 33
6.1.2. The Case Against Bottleneck Policing . . . . . . . . . 38 8. Related Work . . . . . . . . . . . . . . . . . . . . . . . . . 34
6.1.3. Re-ECN Incentive Framework . . . . . . . . . . . . . . 39 8.1. Congestion Notification Integrity . . . . . . . . . . . . 34
6.1.4. Egress Dropper . . . . . . . . . . . . . . . . . . . . 46 9. Security Considerations . . . . . . . . . . . . . . . . . . . 35
6.1.5. Policing . . . . . . . . . . . . . . . . . . . . . . . 47 10. IANA Considerations . . . . . . . . . . . . . . . . . . . . . 37
6.1.6. Inter-domain Policing . . . . . . . . . . . . . . . . 49 11. Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . 37
6.1.7. Inter-domain Fail-safes . . . . . . . . . . . . . . . 52 12. Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . 37
6.1.8. Simulations . . . . . . . . . . . . . . . . . . . . . 53 13. Comments Solicited . . . . . . . . . . . . . . . . . . . . . . 38
6.2. Other Applications . . . . . . . . . . . . . . . . . . . . 53 14. References . . . . . . . . . . . . . . . . . . . . . . . . . . 38
6.2.1. DDoS Mitigation . . . . . . . . . . . . . . . . . . . 53 14.1. Normative References . . . . . . . . . . . . . . . . . . . 38
6.2.2. End-to-end QoS . . . . . . . . . . . . . . . . . . . . 54 14.2. Informative References . . . . . . . . . . . . . . . . . . 39
6.2.3. Traffic Engineering . . . . . . . . . . . . . . . . . 55 Appendix A. Precise Re-ECN Protocol Operation . . . . . . . . . . 41
6.2.4. Inter-Provider Service Monitoring . . . . . . . . . . 55
6.3. Limitations . . . . . . . . . . . . . . . . . . . . . . . 55
7. Incremental Deployment . . . . . . . . . . . . . . . . . . . . 56
7.1. Incremental Deployment Features . . . . . . . . . . . . . 56
7.2. Incremental Deployment Incentives . . . . . . . . . . . . 57
8. Architectural Rationale . . . . . . . . . . . . . . . . . . . 62
9. Related Work . . . . . . . . . . . . . . . . . . . . . . . . . 65
9.1. Policing Rate Response to Congestion . . . . . . . . . . . 65
9.2. Congestion Notification Integrity . . . . . . . . . . . . 66
9.3. Identifying Upstream and Downstream Congestion . . . . . . 67
10. Security Considerations . . . . . . . . . . . . . . . . . . . 67
11. IANA Considerations . . . . . . . . . . . . . . . . . . . . . 68
12. Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . 69
13. Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . 69
14. Comments Solicited . . . . . . . . . . . . . . . . . . . . . . 69
15. References . . . . . . . . . . . . . . . . . . . . . . . . . . 70
15.1. Normative References . . . . . . . . . . . . . . . . . . . 70
15.2. Informative References . . . . . . . . . . . . . . . . . . 70
Appendix A. Precise Re-ECN Protocol Operation . . . . . . . . . . 74
Appendix B. Justification for Two Codepoints Signifying Zero Appendix B. Justification for Two Codepoints Signifying Zero
Worth Packets . . . . . . . . . . . . . . . . . . . . 75 Worth Packets . . . . . . . . . . . . . . . . . . . . 43
Appendix C. ECN Compatibility . . . . . . . . . . . . . . . . . . 76 Appendix C. ECN Compatibility . . . . . . . . . . . . . . . . . . 44
Appendix D. Packet Marking with FNE During Flow Start . . . . . . 78 Appendix D. Packet Marking with FNE During Flow Start . . . . . . 45
Appendix E. Example Egress Dropper Algorithm . . . . . . . . . . 80 Appendix E. Argument for holding back the ECN nonce . . . . . . . 47
Appendix F. Re-TTL . . . . . . . . . . . . . . . . . . . . . . . 80 Appendix F. Alternative Terminology Used in Other Documents . . . 49
Appendix G. Policer Designs to ensure Congestion
Responsiveness . . . . . . . . . . . . . . . . . . . 80
G.1. Per-user Policing . . . . . . . . . . . . . . . . . . . . 80
G.2. Per-flow Rate Policing . . . . . . . . . . . . . . . . . . 82
Appendix H. Downstream Congestion Metering Algorithms . . . . . . 84
H.1. Bulk Downstream Congestion Metering Algorithm . . . . . . 84
H.2. Inflation Factor for Persistently Negative Flows . . . . . 85
Appendix I. Argument for holding back the ECN nonce . . . . . . . 86
Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . . 88
Intellectual Property and Copyright Statements . . . . . . . . . . 90
1. Introduction 1. Introduction
This document aims: This document aims to provide a complete specification of the
addition of the re-ECN protocol to IP and guidelines on how to add it
o To provide a complete specification of the addition of the re-ECN to transport layer protocols, including a complete specification of
protocol to IP and guidelines on how to add it to transport layer re-ECN in TCP as an example. The motivation behind this proposal is
protocols, including a complete specification of re-ECN in TCP as given in [re-ecn-motive], but we include a brief summary here.
an example;
o To show how a number of hard problems become much easier to solve
once re-ECN is available in IP.
In ECN [RFC3168] congested queues probabilistically mark packets as Re-ECN is intended to allow senders to inform the network of the
they approach a congested state. The receiver informs the sender level of congestion they expect their flows to see. This information
that they have seen one or more marks. In re-ECN the sender must is currently only visible at the transport layer. ECN [RFC3168]
predict the level of congestion on the path by re-inserting feedback reveals the upstream congestion state of any path by monitoring the
according to the marking scheme described later in this draft. This rate of CE marks. The receiver then informs the sender when they
results in packets that carry a prediction of downstream congestion. have seen a marked packet. Re-ECN builds on ECN by providing new
codepoints that allow the sender to declare the level of congestion
they expect on the forward path. It is closely related to ECN and
indeed we define a compatability mode to allow a re-ECN sender to
communicate with an ECN receiver [xref].
If a sender understates expected congestion compared to actual If a sender understates expected congestion compared to actual
congestion then the network could discard packets or enact some other congestion then the network could discard packets or enact some other
sanction. A policer can also be introduced at the ingress of sanction. A policer can also be introduced at the ingress of
networks that can limit the congestion caused (or base penalties on networks that can limit the level of congestion being caused.
it).
It is important to add a few key points.
o It can be seen that it takes one round trip before any feedback is
received. For this reason a sender must make a conservative
prediction by transmitting IP packets with a special Feedback Not
Established (FNE) marking.
o It should be noted that the prediction is carried in-band in
normal data packets and for many transports feedback can be
carried in the normal acknowledgements or control packets.
o The re-ECN protocol is independent of the transport. In TCP,
acknowledgments are used to convey the feedback from receiver to
sender. This memo concentrates on TCP as an example transport
protocol, however the re-ECN protocol is compatible with any
transport where feedback can be sent from receiver to sender.
A general statement of the problem solved by re-ECN is to provide A general statement of the problem solved by re-ECN is to provide
sufficient information in each IP datagram to be able to hold senders sufficient information in each IP datagram to be able to hold senders
and whole networks accountable for the congestion they cause and whole networks accountable for the congestion they cause
downstream, before they cause it. But the every-day problems that downstream, before they cause it. But the every-day problems that
re-ECN can solve are much more recognisable than this rather generic re-ECN can solve are much more recognisable than this rather generic
statement: mitigating distributed denial of service (DDoS); statement: mitigating distributed denial of service (DDoS);
simplifying differentiation of quality of service (QoS); policing simplifying differentiation of quality of service (QoS); policing
compliance to congestion control; and so on. compliance to congestion control; and so on.
Uniquely, re-ECN manages to enable solutions to these problems It is important to add a few key points.
without unduly stifling innovative new ways to use the Internet.
This was a hard balance to strike, given it could be argued that DDoS
is an innovative way to use the Internet. The most valuable insight
was to allow each network to choose the level of constraint it wishes
to impose. Also re-ECN has been carefully designed so that networks
that choose to use it conservatively can protect themselves against
the congestion caused in their network by users on other networks
with more liberal policies.
For instance, some network owners want to block applications like
voice and video unless their network is compensated for the extra
share of bottleneck bandwidth taken. These real-time applications
tend to be unresponsive when congestion arises. Whereas elastic TCP-
based applications back away quickly, ending up taking a much smaller
share of congested capacity for themselves. Other network owners
want to invest in large amounts of capacity and make their gains from
simplicity of operation and economies of scale.
re-ECN allows the more conservative networks to police out flows that
have not asked to be unresponsive to congestion---not because they
are voice or video---just because they don't respond to congestion.
But it also allows other networks to choose not to police.
Crucially, when flows from liberal networks cross into a conservative
network, re-ECN enables the conservative network to apply penalties
to its neighbouring networks for the congestion they allow to be
caused. And these penalties can be applied to bulk data, without
regard to flows.
Then, if unresponsive applications become so dominant that some of o In any stnadard network it always takes one round trip before any
the more liberal networks experience congestion collapse [RFC3714], feedback is received. For this reason a sender must make a
they can change their minds and use re-ECN to apply tighter controls conservative prediction by transmitting IP packets with a special
in order to bring congestion back under control. Cautious marking.
re-ECN works by arranging that each packet arrives at each network o It should be noted that the prediction is carried in-band in
element carrying a view of expected congestion on its own downstream normal data packets and for many transports feedback can be
path, albeit averaged over multiple packets. Most usefully, carried in the normal acknowledgements or control packets.
congestion on the remainder of the path becomes visible in the IP
header at the first ingress. Many of the applications of re-ECN
involve a policer at this ingress using the view of downstream
congestion arriving in packets to police or control the packet rate.
Importantly, the scheme is recursive: a whole network harbouring o The re-ECN protocol is independent of the transport. In TCP,
users causing congestion in downstream networks can be held acknowledgments are used to convey the feedback from receiver to
responsible or policed by its downstream neighbour. sender. This memo concentrates on TCP as an example transport
protocol, however the re-ECN protocol is compatible with any
transport where feedback can be sent from receiver to sender.
This document is structured as follows. First an overview of the re- This document is structured as follows. First an overview of the re-
ECN protocol is given (Section 3), outlining its attributes and ECN protocol is given (Section 4), outlining its attributes and
explaining conceptually how it works as a whole. The two main parts explaining conceptually how it works as a whole. The two main parts
of the document follow. That is, the protocol specification divided of the document follow. That is, the protocol specification divided
into transport (Section 4) and network (Section 5) layers which into network (Section 5) and transport (Section 6) layers.
contain most of the standards compliance terminology, then the
applications re-ECN can be put to, such as policing DDoS, QoS and
congestion control (Section 6). Although these applications do not
require standardisation themselves, they are described in a fair
degree of detail in order to explain how re-ECN can be used. Given
re-ECN proposes to use the last undefined bit in the IPv4 header, we
felt it necessary to outline the potential that re-ECN could release
in return for being given that bit.
Deployment issues discussed throughout the document are brought Deployment issues discussed throughout the document are brought
together in Section 7, which is followed by a brief section together in Section 7. Related work is discussed in (Section 8).
explaining the somewhat subtle rationale for the design from an
architectural perspective (Section 8). We end by describing related
work (Section 9), listing security considerations (Section 10) and
finally drawing conclusions (Section 12).
2. Requirements notation 2. Requirements notation
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]. document are to be interpreted as described in [RFC2119].
This document first specifies a protocol, then describes a framework 3. Terminology
that creates the right incentives to ensure compliance to the
protocol. This could cause confusion because the second part of the
document considers many cases where malicious nodes may not comply
with the protocol. When such contingencies are described, if any of
the above keywords are not capitalised, that is deliberate. So, for
instance, the following two apparently contradictory sentences would
be perfectly consistent: i) x MUST do this; ii) x may not do this.
3. Protocol Overview The following terminology is used throughout this memo. Some of this
terminology is new and, to avoid confusion, Appendix F sets out all
the alternative terminology that has been used in other re-ECN
related documents.
3.1. Background and Applicability o Neutral packet - a packet that is able to be congestion marked by
an ECN or re-ECN queue.
o Negative packet - a Neutral packet that has been congestion marked
by an ECN or re-ECN queue.
o Positive packet - a packet that has been marked by the sender to
indicate the expected level of congestion along its path. In
general Positive packets should only be sent in response to
feedback received from the receiver.*
o Cancelled packet - a Positive Packet that has been congestion
marked by an ECN or re-ECN queue.
o Cautious packet - a packet that has been marked by the sender to
indeiate the expected level of congestion along its path. In
general Cautious packets should be used when there is insufficient
feedback to be confident about the congestion state of the
network.*
o * the difference between positive and cautious packets is
explained in detail later in the document along with guidelines on
the use of Cautious packets.
All the above terms have related IP codepoints as defined in
(Section 5).
4. Protocol Overview
4.1. Simplified Re-ECN Protocol
We describe here the simplified re-ECN protocol. To simplify the
description we assume packets and segments are synonymous.
Packets are sent from a sender to a receiver. In Figure 1 the queues
(Q1 and Q2) are ECN enabled as per RFC 3168 [RFC3168]. If congestion
occurs then packets are marked with the congestion experienced (CE)
flag exactly as in the ECN protocol [RFC3168]; the routers do not
need to be modified and do not need to know the re-ECN protocol. The
receiver constantly informs the sender of the current count of
Positive packets it has seen. The sender uses this information
determine how many Positive packets it must send into the network.
The receiver's aim is to balance the number of bytes that have been
congestion marked with the number of Positive bytes it has sent.
+--------- Feedback----------+
| |
v |
+---+ +----+ +----+ +---+
| | | | | | | |
| S |--->| Q1 |--->| Q2 |--->| R |
| | | | | | | |
+---+ +----+ +----+ +---+
Figure 1: Simple Re-ECN
4.1.1. Congestion Control and Policing the Protocol
The arrangement of the protocol ensures that packets carry a
declaration of the amount of congestion that will be experienced on
the path. The re-ECN protocol is orthogonal to to any congestion
control algorithms, but can be used to ensure that congestion control
is being applied by the sender.
In general we assume that there will be a policer at the network
ingress which can rate limit traffic based on the amount of
congestion declared.
At the network egress there is a droper which can impose sanctions on
flows that incorrectly declare congestion.
Policers and droppers are explained in more detail in
[re-ecn-motive].
4.1.2. Background and Applicability
The re-ECN protocol makes no changes and has no effect on the TCP The re-ECN protocol makes no changes and has no effect on the TCP
congestion control algorithm or on other rate responses to congestion control algorithm or on other rate responses to
congestion. re-ECN is not a new congestion control protocol, rather congestion. re-ECN is not a new congestion control protocol, rather
it is orthogonal to congestion control itself. Re-ECN is concerned it is orthogonal to congestion control itself. Re-ECN is concerned
with revealing information about congestion so that users and with revealing information about congestion so that users and
networks can be held accountable for the congestion they cause, or networks can be held accountable for the congestion they cause, or
allow to be caused. allow to be caused.
Re-ECN builds on ECN so we briefly recap the essentials of the ECN Re-ECN builds on ECN so we briefly recap the essentials of the ECN
protocol [RFC3168]. Two bits in the IP protocol (v4 or v6) are protocol [RFC3168]. Two bits in the IP protocol (v4 or v6) are
assigned to the ECN field. The sender clears the field to "00" (Not- assigned to the ECN field. The sender clears the field to "00" (Not-
ECT) if either end-point transport is not ECN-capable. Otherwise it ECT) if either end-point transport is not ECN-capable. Otherwise it
indicates an ECN-capable transport (ECT) using either of the two indicates an ECN-capable transport (ECT) using either of the two
code-points "10" or "01" (ECT(0) and ECT(1) resp.). code-points "10" or "01" (ECT(0) and ECT(1) resp.).
ECN-capable queues probabilistically set "11" if congestion is ECN-capable queues probabilistically set this field to "11" if
experienced (CE), the marking probability increasing with the length congestion is experienced (CE). In general this marking probability
of the queue at its egress link (typically using the RED will increase with the length of the queue at its egress link
algorithm [RFC2309]). However, they still drop rather than mark Not- (typically using the RED algorithm [RFC2309]). However, they still
ECT packets. With multiple ECN-capable queues on a path, a flow of drop rather than mark Not-ECT packets. With multiple ECN-capable
packets accumulates the fraction of CE marking that each queue adds. queues on a path, a flow of packets accumulates the fraction of CE
The combined effect of the packet marking of all the queues along the marking that each queue adds. The combined effect of the packet
path signals congestion of the whole path to the receiver. So, for marking of all the queues along the path signals congestion of the
example, if one queue early in a path is marking 1% of packets and whole path to the receiver. So, for example, if one queue early in a
another later in a path is marking 2%, flows that pass through both path is marking 1% of packets and another later in a path is marking
queues will experience approximately 3% marking (see Appendix A for a 2%, flows that pass through both queues will experience approximately
precise treatment). 3% marking (see Appendix A for a precise treatment).
The choice of two ECT code-points in the ECN field [RFC3168] The choice of two ECT code-points in the ECN field [RFC3168]
permitted future flexibility, optionally allowing the sender to permitted future flexibility, optionally allowing the sender to
encode the experimental ECN nonce [RFC3540] in the packet stream. encode the experimental ECN nonce [RFC3540] in the packet stream.
The nonce is designed to allow a sender to check the integrity of The nonce is designed to allow a sender to check the integrity of
congestion feedback. But Section 9.2 explains that it still gives no congestion feedback. But Section 8.1 explains that it still gives no
control over how fast the sender transmits as a result of the control over how fast the sender transmits as a result of the
feedback. On the other hand, re-ECN is designed both to ensure that feedback. On the other hand, re-ECN is designed both to ensure that
congestion is declared honestly and that the sender's rate responds congestion is declared honestly and that the sender's rate responds
appropriately. appropriately.
Re-ECN is based on a feedback arrangement called `re- Re-ECN is based on a feedback arrangement called `re-
feedback' [Re-fb]. The word is short for either receiver-aligned, feedback' [Re-fb]. The word is short for either receiver-aligned,
re-inserted or re-echoed feedback. But it actually works even when re-inserted or re-echoed feedback. But it actually works even when
no feedback is available. In fact it has been carefully designed to no feedback is available. In fact it has been carefully designed to
work for single datagram flows. It also encourages aggregation of work for single datagram flows. It also encourages aggregation of
single packet flows by congestion control proxies. Then, even if the single packet flows by congestion control proxies. Then, even if the
traffic mix of the Internet were to become dominated by short traffic mix of the Internet were to become dominated by short
messages, it would still be possible to control congestion messages, it would still be possible to control congestion
effectively and efficiently. effectively and efficiently.
Changing the Internet's feedback architecture seems to imply Changing the Internet's feedback architecture seems to imply
considerable upheaval. But re-ECN can be deployed incrementally at considerable upheaval. But re-ECN can be deployed incrementally at
the transport layer around unmodified queues using existing fields in the transport layer around unmodified queues using existing fields in
IP (v4 or v6). However it does also require the last undefined bit IP (v4 or v6). However it does also require the last undefined bit
in the IPv4 header, which it uses in combination with the 2-bit ECN in the IPv4 header, which it uses in combination with the 2-bit ECN
field to create four new codepoints. Nonetheless, we RECOMMENDED field to create four new codepoints. Nonetheless, we RECOMMEND
adding optional preferentail drop to IP queues based on the re-ECN adding optional preferentail drop to IP queues based on the re-ECN
fields in order to improve resilience against DoS attacks. fields in order to improve resilience against DoS attacks.
Similarly, re-ECN works best if both the sender and receiver Similarly, re-ECN works best if both the sender and receiver
transports are re-ECN-capable, but it can work with just sender transports are re-ECN-capable, but it can work with just sender
support. Section 7.1 summarises the incremental deployment strategy. support(Section 6.1.2).
Before re-ECN can be considered worthy of using up the last bit in
the IP header, we must be sure that all our claims are robust. We
have gradually been reducing the list of outstanding issues, but the
few that still remain are listed in Section 6.3. We expect new
attacks may still be found, but we offer the re-ECN protocol on the
basis that it is built on fairly solid theoretical foundations and,
so far, it has proved possible to keep it relatively robust.
3.2. Simplified Re-ECN Protocol
We describe here the simplified re-ECN protocol. In this first
description we assume packets and segments are synonymous.
Packets are sent from a sender to a receiver. In Figure 1 the queues
(Q1 and Q2) are ECN enabled as per RFC 3168 [ref]. If congestion
occurs then packets are marked with the congestion experienced (CE)
flag exactly as in the ECN protocol [RFC3168]; the routers do not
need to be modified and do not need to know the re-ECN protocol. On
reception of marked packets the receiver notifies the sender of the
current count of marked packets. Note that this is the number of
packets marked rather than the setting of the ECE flag in ECN. The
sender uses this information to re-echo mark packets in exact
correspondence to the number of CE marked bytes observed at the
receiver.
+--------- Feedback----------+
| |
v |
+---+ +----+ +----+ +---+
| | RE | | | | | |
| S |--->| Q1 |--->| Q2 |--->| R |
| | | | | | | |
+---+ +----+ +----+ +---+
Figure 1: Simple Re-ECN
3.3. Re-ECN Abstracted Network Layer Wire Protocol (IPv4 or v6) 4.2. Re-ECN Abstracted Network Layer Wire Protocol (IPv4 or v6)
The re-ECN wire protocol uses the two bit ECN field broadly as in The re-ECN wire protocol uses the two bit ECN field broadly as in
RFC3168 [RFC3168] as described above, but with five differences of RFC3168 [RFC3168] as described above, but with five differences of
detail (brought together in a list in Section 7.1). This detail (brought together in a list in Section 7). This specification
specification defines a new re-ECN extension (RE) flag. We will defines a new re-ECN extension (RE) flag. We will defer the
defer the definition of the actual position of the RE flag in the definition of the actual position of the RE flag in the IPv4 & v6
IPv4 & v6 headers until Section 5. When we don't need to choose headers until Section 5. When we don't need to choose between IPv4
between IPv4 and v6 wire protocols it will suffice call it the RE and v6 wire protocols it will suffice call it the RE flag.
flag.
Unlike the ECN field, the RE flag is intended to be set by the sender Unlike the ECN field, the RE flag is intended to be set by the sender
and remain unchanged along the path, although it can be read by and SHOULD remain unchanged along the path, although it can be read
network elements that understand the re-ECN protocol. It is feasible by network elements that understand the re-ECN protocol. It is
that a network element MAY change the setting of the RE flag, perhaps feasible that a network element MAY change the setting of the RE
acting as a proxy for an end-point, but such a protocol would have to flag, perhaps acting as a proxy for an end-point, but such a protocol
be defined in another specification (e.g. [Re-PCN]). would have to be defined in another specification (e.g. [Re-PCN]).
Although the RE flag is a separate, single bit field, it can be read Although the RE flag is a separate, single bit field, it can be read
as an extension to the two-bit ECN field; the three concatenated bits as an extension to the two-bit ECN field; the three concatenated bits
in what we will call the extended ECN field (EECN) giving eight in what we will call the extended ECN field (EECN) giving eight
codepoints. We will use the RFC3168 names of the ECN codepoints to codepoints. We will use the RFC3168 names of the ECN codepoints to
describe settings of the ECN field when the RE flag setting is "don't describe settings of the ECN field when the RE flag setting is "don't
care", but we also define the following six extended ECN codepoint care", but we also define the following six extended ECN codepoint
names for when we need to be more specific. names for when we need to be more specific.
One of re-ECN's codepoints is an alternative use of the codepoint set One of re-ECN's codepoints is an alternative use of the codepoint set
aside in RFC3168 for the ECN nonce (ECT(1)). Transports using re-ECN aside in RFC3168 for the ECN nonce (ECT(1)). Transports using re-ECN
do not need to use the ECN nonce as long as the sender is also do not need to use the ECN nonce as long as the sender is also
checking for transport protocol compliance checking for transport protocol compliance
[I-D.moncaster-tcpm-rcv-cheat]. The case for doing this is given in [I-D.moncaster-tcpm-rcv-cheat]. The case for doing this is given in
Appendix I. Two re-ECN codepoints are given compatible uses to those Appendix E. Two re-ECN codepoints are given compatible uses to those
defined in RFC3168 (Not-ECT and CE). The other codepoint used by defined in RFC3168 (Not-ECT and CE). The other codepoint used by
RFC3168 (ECT(0)) isn't used for re-ECN. Altogether this leave one RFC3168 (ECT(0)) isn't used for re-ECN. Altogether this leave one
codepoint of the eight unused by ECN or re-ECN and available for codepoint of the eight unused by ECN or re-ECN and available for
future use. future use.
+-------+------------+------+--------------+------------------------+ +--------+-------------+-------+-----------+------------------------+
| ECN | RFC3168 | RE | Extended ECN | Re-ECN meaning | | ECN | RFC3168 | RE | EECN | re-ECN meaning |
| field | codepoint | flag | codepoint | | | field | codepoint | flag | codepoint | |
+-------+------------+------+--------------+------------------------+ +--------+-------------+-------+-----------+------------------------+
| 00 | Not-ECT | 0 | Not-ECT | Not re-ECN-capable | | 00 | Not-ECT | 0 | Not-ECT | Not re-ECN-capable |
| | | | | transport | | | | | | transport (Legacy) |
| 00 | --- | 1 | FNE | Feedback not | | 00 | --- | 1 | FNE | Feedback not |
| | | | | established | | | | | | established (Cautious) |
| 01 | ECT(1) | 0 | Re-Echo | Re-echoed congestion | | 01 | ECT(1) | 0 | Re-Echo | Re-echoed congestion |
| | | | | and RECT | | | | | | and RECT (Positive) |
| 01 | --- | 1 | RECT | Re-ECN capable | | 01 | --- | 1 | RECT | Re-ECN capable |
| | | | | transport | | | | | | transport (Neutral) |
| 10 | ECT(0) | 0 | ECT(0) | RFC3168 ECN use only | | 10 | ECT(0) | 0 | ECT(0) | RFC3168 ECN use only |
| | | | | | | | | | | |
| 10 | --- | 1 | --CU-- | Currently unused | | 10 | --- | 1 | --CU-- | Currently unused |
| | | | | | | | | | | |
| 11 | CE | 0 | CE(0) | Re-Echo canceled by | | 11 | CE | 0 | CE(0) | Re-Echo cancelled by |
| | | | | congestion experienced | | | | | | CE (Cancelled) |
| 11 | --- | 1 | CE(-1) | Congestion experienced | | 11 | --- | 1 | CE(-1) | Congestion Experienced |
+-------+------------+------+--------------+------------------------+ | | | | | (Negative) |
+--------+-------------+-------+-----------+------------------------+
Table 1: Extended ECN Codepoints Table 1: Extended ECN Codepoints
3.4. Re-ECN Protocol Operation 4.3. Re-ECN Protocol Operation
In this section we will give an overview of the operation of the re- In this section we will give an overview of the operation of the re-
ECN protocol for TCP/IP, leaving a detailed specification to the ECN protocol for TCP/IP, leaving a detailed specification to the
following sections. Other transports will be discussed later. following sections. Other transports will be discussed later.
In summary, the protocol adds a third `re-echo' stage to the existing In summary, the protocol adds a third `re-echo' stage to the existing
TCP/IP ECN protocol. Whenever the network adds CE congestion TCP/IP ECN protocol. Whenever the network adds CE congestion
signalling to the IP header on the forward data path, the receiver signalling to the IP header on the forward data path, the receiver
feeds it back to the ingress using TCP, then the sender re-echoes it feeds it back to the ingress using TCP, then the sender re-echoes it
into the forward data path using the RE flag in the next packet. into the forward data path using the RE flag in the next packet.
Prior to receiving any feedback a sender will not know which setting Prior to receiving any feedback a sender will not know which setting
of the RE flag to use, so it sets the feedback not established (FNE) of the RE flag to use, so it sends Cautious packets by setting the
codepoint. The network reads the FNE codepoint conservatively as FNE codepoint. The network reads the FNE codepoint conservatively as
equivalent to re-echoed congestion. equivalent to re-echoed congestion.
Specifically, once feedback from a flow is established, a re-ECN Specifically, once feedback from an ECN or re-ECN capable flow is
sender always initialises the ECN field to ECT(1). And it usually established, a re-ECN sender always initialises the ECN field to
sets the RE flag to "1". Whenever a queue marks a packet to CE, the ECT(1). And it usually sets the RE flag to "1" indicating a Neutral
receiver feeds back this event to the sender. On receiving this packet. Whenever a queue marks a packet to CE, the receiver feeds
feedback, the re-ECN sender will clear the RE flag to "0" in the next back this event to the sender. On receiving this feedback, the re-
packet it sends. ECN sender will clear the RE flag to "0" in the next packet it sends
(indicating a Positive packet).
We chose to set and clear the RE flag this way round to ease We chose to set and clear the RE flag this way round to ease
incremental deployment (see Section 7.1). To avoid confusion we will incremental deployment (see Section 7). To avoid confusion we will
use the term `blanking' (rather than marking) when the RE flag is use the term `blanking' (rather than marking) when the RE flag is
cleared to "0". So, over a stream of packets, we will talk of the cleared to "0". So, over a stream of packets, we will talk of the
`RE blanking fraction' as the fraction of octets in packets with the `RE blanking fraction' as the fraction of octets in packets with the
RE flag cleared to "0". RE flag cleared to "0".
+---+ +----+ +----+ +---+ +---+ +----+ +----+ +---+
| S |--| Q1 |----------------| Q2 |--| R | | S |--| Q1 |----------------| Q2 |--| R |
+---+ +----+ +----+ +---+ +---+ +----+ +----+ +---+
. . . . . . . .
^ . . . . ^ . . . .
skipping to change at page 14, line 5 skipping to change at page 12, line 5
horizontal line at 3% in the figure. The CE marked fraction is shown horizontal line at 3% in the figure. The CE marked fraction is shown
by the stepped line which rises to meet the RE blanking fraction line by the stepped line which rises to meet the RE blanking fraction line
with steps at at each queue where packets are marked. Two queues are with steps at at each queue where packets are marked. Two queues are
shown (Q1 and Q2) that are currently congested. Each time packets shown (Q1 and Q2) that are currently congested. Each time packets
pass through a fraction are marked; 1% at Q1 and 2% at Q2). The pass through a fraction are marked; 1% at Q1 and 2% at Q2). The
approximate downstream congestion can be measured at the observation approximate downstream congestion can be measured at the observation
points shown along the path by subtracting the CE marking fraction points shown along the path by subtracting the CE marking fraction
from the RE blanking fraction, as shown in the table below from the RE blanking fraction, as shown in the table below
(Appendix A derives these approximations from a precise analysis). (Appendix A derives these approximations from a precise analysis).
+-------------------+------------------------------+ NB due to the unary nature of ECN marking and the equivalent unary
| Observation point | Approx downstream congestion | nature of re-ECN blanking, the precise fraction of marked bytes must
+-------------------+------------------------------+ be calculated by maintaining a moving average of the number of
| L | 3% - 0% = 3% | packets that have been marked as a proportion of the total number of
| M | 3% - 1% = 2% | packets.
| N | 3% - 3% = 0% |
+-------------------+------------------------------+
Table 2: Downstream Congestion Measured at Example Observation Points
All along the path, whole-path congestion remains unchanged so it can Along the path the fraction of packets that had their RE field
be used as a reference against which to compare upstream congestion. cleared remains unchanged so it can be used as a reference against
The difference predicts downstream congestion for the rest of the which to compare upstream congestion. The difference predicts
path. Therefore, measuring the fractions of each codepoint at any downstream congestion for the rest of the path. Therefore, measuring
point in the Internet will reveal upstream, downstream and whole path the fractions of each codepoint at any point in the Internet will
congestion. reveal upstream, downstream and whole path congestion.
Note that we have introduced discussion of marking and blanking Note that we have introduced discussion of marking and blanking
fractions solely for illustration. To be absolutely clear, for TCP fractions solely for illustration. We are not saying any protocol
these fractions are averages that would result from the behaviour of handler will work with these average fractions directly. In fact the
the protocol handler mechanically blanking outgoing packets in direct protocol actually requires the number of marked and blanked bytes to
response to incoming feedback---we are not saying any protocol balance by the time the packet reaches the receiver.
handler has to work with these average fractions directly.
3.5. Informal Terminology
In the rest of this memo we will loosely talk of positive or negative 4.4. Positive and Negative Flows
flows, meaning flows where the moving average of the downstream
congestion metric is persistently positive or negative. A negative
flow is one where more CE marked packets than re-ECN blanked packets
arrive. Likewise in positive flows more re-ECN blanked packets
arrive than CE marked packets. The notion of a negative metric
arises because it is derived by subtracting one metric from another.
Of course actual downstream congestion cannot be negative, only the
metric can (whether due to time lags or deliberate malice).
Just as we will loosely talk of positive and negative flows, we will In Section 3 we introduced the terms Positive, Neutral, Negative,
also talk of positive or negative packets, meaning packets that Cautious and Cancelled. This terminology is based on the requirement
contribute positively or negatively to the downstream congestion to balance the proportion of bytes marked as CE with the proportion
metric. of bytes that are re-echo marked. In the rest of this memo we will
loosely talk of positive or negative flows, meaning flows where the
moving average of the downstream congestion metric is persistently
positive or negative. A negative flow is one where more CE marked
packets than re-ECN blanked packets arrive. Likewise in positive
flows more re-ECN blanked packets arrive than CE marked packets. The
notion of a negative metric arises because it is derived by
subtracting one metric from another. Of course actual downstream
congestion cannot be negative, only the metric can (whether due to
time lags or deliberate malice).
Therefore we will talk of packets having `worth' of +1, 0 or -1, Therefore we will talk of packets having `worth' of +1, 0 or -1,
which, when multiplied by their size, indicates their contribution to which, when multiplied by their size, indicates their contribution to
the downstream congestion metric. the downstream congestion metric. The worth of each type of packet
is given below in Table 2. The idea is that most flows start with
The idea is that most packets start with zero worth. Every time the zero worth. Every time the network decrements the worth of a packet,
network decrements the worth of a packet, the sender increments the the sender increments the worth of a later packet. Then, over time,
worth of a later packet. Then, over time, as many positive octets as many positive octets should arrive at the receiver as negative.
should arrive at the receiver as negative. Note we have said octets Note we have said octets not packets, so if packets are of different
not packets, so if packets are of different sizes, the worth should sizes, the worth should be incremented on enough octets to balance
be incremented on enough octets to balance the octets in negative the octets in negative packets arriving at the receiver. It is this
packets arriving at the receiver. It is this balance that will allow balance that will allow the network to hold the sender accountable
the network to hold the sender accountable for the congestion it for the congestion it causes.
causes.
If a packet carrying re-echoed congestion happens to also be If a packet carrying re-echoed congestion happens to also be
congestion marked, the +1 worth added by the sender will be cancelled congestion marked, the +1 worth added by the sender will be cancelled
out by the -1 network congestion marking. Although the two worth out by the -1 network congestion marking. Although the two worth
values correctly cancel out, neither the congestion marking nor the values correctly cancel out, neither the congestion marking nor the
re-echoed congestion are lost, because the RE bit and the ECN field re-echoed congestion are lost, because the RE bit and the ECN field
are orthogonal. So, whenever this happens, the receiver will are orthogonal. So, whenever this happens, the receiver will
correctly detect and re-echo the new congestion event as well. correctly detect and re-echo the new congestion event as well.
The table below specifies unambiguously the worth of each extended The table below specifies unambiguously the worth of each extended
ECN codepoint. Note the order is different from the previous table ECN codepoint. Note the order is different from the previous table
to better show how the worth increments and decrements. The FNE to better show how the worth increments and decrements.
codepoint is used in the flow bootstrap process (explained later) and
has the same positive (+1) worth as a packet with the Re-Echo
codepoint.
+--------+------+----------------+-------+--------------------------+ +---------+-------+---------------+-------+-------------------------+
| ECN | RE | Extended ECN | Worth | Re-ECN meaning | | ECN | RE | Extended ECN | Worth | Re-ECN Term |
| field | bit | codepoint | | | | field | bit | codepoint | | |
+--------+------+----------------+-------+--------------------------+ +---------+-------+---------------+-------+-------------------------+
| 00 | 0 | Not-RECT | ... | Not re-ECN-capable | | 00 | 0 | Not-RECT | ... | --- |
| | | | | transport | | 00 | 1 | FNE | +1 | Cautious |
| 00 | 1 | FNE | +1 | Feedback not established | | 01 | 0 | Re-Echo | +1 | Positive |
| 01 | 0 | Re-Echo | +1 | Re-echoed congestion and | | 10 | 0 | Legacy | ... | RFC3168 ECN use only |
| | | | | RECT | | | | | | |
| 10 | 0 | --- | ... | RFC3168 ECN use only | | 11 | 0 | CE(0) | 0 | Negative |
| 11 | 0 | CE(0) | 0 | Re-Echo canceled by | | 01 | 1 | RECT | 0 | Neutral |
| | | | | congestion experienced |
| 01 | 1 | RECT | 0 | Re-ECN capable transport |
| 10 | 1 | --CU-- | ... | Currently unused | | 10 | 1 | --CU-- | ... | Currently unused |
| | | | | | | | | | | |
| 11 | 1 | CE(-1) | -1 | Congestion experienced | | 11 | 1 | CE(-1) | -1 | Negative |
+--------+------+----------------+-------+--------------------------+ +---------+-------+---------------+-------+-------------------------+
Table 3: 'Worth' of Extended ECN Codepoints Table 2: 'Worth' of Extended ECN Codepoints
4. Transport Layers 5. Network Layer
4.1. TCP 5.1. Re-ECN IPv4 Wire Protocol
The wire protocol of the ECN field in the IP header remains largely
unchanged from [RFC3168]. However, an extension to the ECN field we
call the RE (Re-ECN extension) flag (Section 4.2) is defined in this
document. It doubles the extended ECN codepoint space, giving 8
potential codepoints. The semantics of the extra codepoints are
backward compatible with the semantics of the 4 original codepoints
[RFC3168] (Section 7 collects together and summarises all the changes
defined in this document).
For IPv4, this document proposes that the new RE control flag will be
positioned where the `reserved' control flag was at bit 48 of the
IPv4 header (counting from 0). Alternatively, some would call this
bit 0 (counting from 0) of byte 7 (counting from 1) of the IPv4
header (Figure 3).
0 1 2
+---+---+---+
| R | D | M |
| E | F | F |
+---+---+---+
Figure 3: New Definition of the Re-ECN Extension (RE) Control Flag at
the Start of Byte 7 of the IPv4 Header
The semantics of the RE flag are described in outline in Section 4
and specified fully in Section 6. The RE flag is always considered
in conjunction with the 2-bit ECN field, as if they were concatenated
together to form a 3-bit extended ECN field. If the ECN field is set
to either the ECT(1) or CE codepoint, when the RE flag is blanked
(cleared to "0") it represents a re-echo of congestion experienced by
an early packet. If the ECN field is set to the Not-ECT codepoint,
when the RE flag is set to "1" it represents the feedback not
established (FNE) codepoint, which signals that the packet was sent
without the benefit of congestion feedback.
It is believed that the FNE codepoint can simultaneously serve other
purposes, particularly where the start of a flow needs distinguishing
from packets later in the flow. For instance it would have been
useful to identify new flows for tag switching and might enable
similar developments in the future if it were adopted. It is similar
to the state set-up bit idea designed to protect against memory
exhaustion attacks. This idea was proposed informally by David Clark
and documented by Handley and Greenhalgh [Steps_DoS]. The FNE
codepoint can be thought of as a `soft-state set-up flag', because it
is idempotent (i.e. one occurrence of the flag is sufficient but
further occurrences achieve the same effect if previous ones were
lost).
We are sure there will probably be other claims pending on the use of
bit 48. We know of at least two [ARI05], [RFC3514] but neither have
been pursued in the IETF, so far, although the present proposal would
meet the needs of the latter.
The security flag proposal (commonly known as the evil bit) was
published on 1 April 2003 as Informational RFC 3514, but it was not
adopted due to confusion over whether evil-doers might set it
inappropriately. The present proposal is backward compatible with
RFC3514 because if re-ECN compliant senders were benign they would
correctly clear the evil bit to honestly declare that they had just
received congestion feedback. Whereas evil-doers would hide
congestion feedback by setting the evil bit continuously, or at least
more often than they should. So, evil senders can be identified,
because they declare that they are good less often than they should.
5.2. Re-ECN IPv6 Wire Protocol
For IPv6, this document proposes that the new RE control flag will be
positioned as the first bit of the option field of a new Congestion
hop by hop option header (Figure 4).
0 1 2 3
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Next Header | Hdr ext Len | Option Type | Opt Length =4 |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|R| Reserved for future use |
|E| |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Figure 4: Definition of a New IPv6 Congestion Hop by Hop Option
Header containing the re-ECN Extension (RE) Control Flag
0 1 2 3 4 5 6 7 8
+-+-+-+-+-+-+-+-+-
|AIU|C|Option ID|
+-+-+-+-+-+-+-+-+-
Figure 5: Congestion Hop by Hop Option Type Encoding
The Hop-by-Hop Options header enables packets to carry information to
be examined and processed by routers or nodes along the packet's
delivery path, including the source and destination nodes. For re-
ECN, the two bits of the Action If Unrecognized (AIU) flag of the
Congestion extension header MUST be set to "00" meaning if
unrecognized `skip over option and continue processing the header'.
Then, any routers or a receiver not upgraded with the optional re-ECN
features described in this memo will simply ignore this header. But
routers with these optional re-ECN features or a re-ECN policing
function, will process this Congestion extension header.
The `C' flag MUST be set to "1" to specify that the Option Data
(currently only the RE control flag) can change en-route to the
packet's final destination. This ensures that, when an
Authentication header (AH [RFC4302]) is present in the packet, for
any option whose data may change en-route, its entire Option Data
field will be treated as zero-valued octets when computing or
verifying the packet's authenticating value.
Although the RE control flag should not be changed along the path, we
expect that the rest of this option field that is currently `Reserved
for future use' could be used for a multi-bit congestion notification
field which we would expect to change en route. As the RE flag does
not need end-to-end authentication, we set the C flag to '1'.
{ToDo: A Congestion Hop by Hop Option ID will need to be registered
with IANA.}
5.3. Router Forwarding Behaviour
Re-ECN works well without modifying the forwarding behaviour of any
routers. However, below, two OPTIONAL changes to forwarding
behaviour are defined which respectively enhance performance and
improve a router's discrimination against flooding attacks. They are
both OPTIONAL additions that we propose MAY apply by default to all
Diffserv per-hop scheduling behaviours (PHBs) [RFC2475] and ECN
marking behaviours [RFC3168]. Specifications for PHBs MAY define
different forwarding behaviours from this default, but this is not
required. [Re-PCN] is one example.
FNE indicates ECT:
The FNE codepoint tells a router to assume that the packet was
sent by an ECN-capable transport (see Section 5.4). Therefore an
FNE packet MAY be marked rather than dropped. Note that the FNE
codepoint has been intentionally chosen so that, to RFC3168
compliant routers (which do not inspect the RE flag) an FNE packet
appears to be Not-ECT so it will be dropped by legacy AQM
algorithms.
A network operator MUST NOT configure a queue to ECN mark rather
than drop FNE packets unless it can guarantee that FNE packets
will be rate limited, either locally or upstream. The ingress
policers discussed in [re-ecn-motive] would count as rate limiters
for this purpose.
Preferential Drop: If a re-ECN capable router queue experiences very
high load so that it has to drop arriving packets (e.g. a DoS
attack), it MAY preferentially drop packets within the same
Diffserv PHB using the preference order for extended ECN
codepoints given in Table 3. Preferential dropping can be
difficult to implement on some hardware, but if feasible it would
discriminate against attack traffic if done as part of the overall
policing framework of [re-ecn-motive]. If nowhere else, routers
at the egress of a network SHOULD implement preferential drop
(stronger than the MAY above). For simplicity, preferences 4 & 5
MAY be merged into one preference level.
+-------+-----+------------+-------+------------+-------------------+
| ECN | RE | Extended | Worth | Drop Pref | Re-ECN meaning |
| field | bit | ECN | | (1 = drop | |
| | | codepoint | | 1st) | |
+-------+-----+------------+-------+------------+-------------------+
| 01 | 0 | Re-Echo | +1 | 5/4 | Re-echoed |
| | | | | | congestion and |
| | | | | | RECT |
| 00 | 1 | FNE | +1 | 4 | Feedback not |
| | | | | | established |
| 11 | 0 | CE(0) | 0 | 3 | Re-Echo canceled |
| | | | | | by congestion |
| | | | | | experienced |
| 01 | 1 | RECT | 0 | 3 | Re-ECN capable |
| | | | | | transport |
| 11 | 1 | CE(-1) | -1 | 3 | Congestion |
| | | | | | experienced |
| 10 | 1 | --CU-- | n/a | 2 | Currently Unused |
| 10 | 0 | --- | n/a | 2 | RFC3168 ECN use |
| | | | | | only |
| 00 | 0 | Not-RECT | n/a | 1 | Not |
| | | | | | Re-ECN-capable |
| | | | | | transport |
+-------+-----+------------+-------+------------+-------------------+
Table 3: Drop Preference of EECN Codepoints (Sorted by `Worth')
The above drop preferences are arranged to preserve packets with
more positive worth (Section 4.4), given senders of positive
packets must have honestly declared downstream congestion. A full
treatment of this is provided in the companion document desribing
the motivation and architecture for re-ECN [re-ecn-motive]
particularly when the application of re-ECN to protect against
DDoS attacks is described.
5.4. Justification for Setting the First SYN to FNE
the initial SYN MUST be set to FNE by Re-ECT client A (Section 6.1.4)
and (Section 5.3) says a queue MAY optionally treat an FNE packet as
ECN capable, so an initial SYN may be marked CE(-1) rather than
dropped. This seems dangerous, because the sender has not yet
established whether the receiver is a RFC3168 one that does not
understand congestion marking. It also seems to allow malicious
senders to take advantage of ECN marking to avoid so much drop when
launching SYN flooding attacks. Below we explain the features of the
protocol design that remove both these dangers.
ECN-capable initial SYN with a Not-ECT server: If the TCP server B
is re-ECN capable, provision is made for it to feedback a possible
congestion marked SYN in the SYN ACK (Section 6.1.4). But if the
TCP client A finds out from the SYN ACK that the server was not
ECN-capable, the TCP client MUST conservatively consider the first
SYN as congestion marked before setting itself into Not-ECT mode.
Section 6.1.4 mandates that such a TCP client MUST also set its
initial window to 1 segment. In this way we remove the need to
cautiously avoid setting the first SYN to Not-RECT. This will
give worse performance while deployment is patchy, but better
performance once deployment is widespread.
SYN flooding attacks can't exploit ECN-capability: Malicious hosts
may think they can use the advantage that ECN-marking gives over
drop in launching classic SYN-flood attacks. But Section 5.3
mandates that a router MUST only be configured to treat packets
with the FNE codepoint as ECN-capable if FNE packets are rate
limited somewhere. Introduction of the FNE codepoint was a
deliberate move to enable transport-neutral handling of flow-start
and flow state set-up in the IP layer where it belongs. It then
becomes possible to protect against flooding attacks of all forms
(not just SYN flooding) without transport-specific inspection for
things like the SYN flag in TCP headers. Then, for instance, SYN
flooding attacks using IPSec ESP encryption can also be rate
limited at the IP layer.
It might seem pedantic going to all this trouble to enable ECN on the
initial packet of a flow, but it is motivated by a much wider concern
to ensure safe congestion control will still be possible even if the
application mix evolves to the point where the majority of flows
consist of a single window or even a single packet. It also allows
denial of service attacks to be more easily isolated and prevented.
5.5. Control and Management
5.5.1. Negative Balance Warning
A new ICMP message type is being considered so that a dropper can
warn the apparent sender of a flow that it has started to sanction
the flow. The message would have similar semantics to the `Time
exceeded' ICMP message type. To ensure the sender has to invest some
work before the network will generate such a message, a dropper
SHOULD only send such a message for flows that have demonstrated that
they have started correctly by establishing a positive record, but
have later gone negative. The threshold is up to the implementation.
The purpose of the message is to deconfuse the cause of drops from
other causes, such as congestion or transmission losses. The dropper
would send the message to the sender of the flow, not the receiver.
If we did define this message type, it would be REQUIRED for all re-
ECT senders to parse and understand it. Note that a sender MUST only
use this message to explain why losses are occurring. A sender MUST
NOT take this message to mean that losses have occurred that it was
not aware of. Otherwise, spoof messages could be sent by malicious
sources to slow down a sender (c.f. ICMP source quench).
However, the need for this message type is not yet confirmed, as we
are considering how to prevent it being used by malicious senders to
scan for droppers and to test their threshold settings. {ToDo:
Complete this section.}
5.5.2. Rate Response Control
As discussed in [re-ecn-motive] the sender's access operator will be
expected to use bulk per-user policing, but they might choose to
introduce a per-flow policer. In cases where operators do introduce
per-flow policing, there may be a need for a sender to send a request
to the ingress policer asking for permission to apply a non-default
response to congestion (where TCP-friendly is assumed to be the
default). This would require the sender to know what message
format(s) to use and to be able to discover how to address the
policer. The required control protocol(s) are outside the scope of
this document, but will require definition elsewhere.
The policer is likely to be local to the sender and inline, probably
at the ingress interface to the internetwork. So, discovery should
not be hard. A variety of control protocols already exist for some
widely used rate-responses to congestion. For instance DCCP
congestion control identifiers (CCIDs [RFC4340]) fulfil this role and
so does QoS signalling (e.g. and RSVP request for controlled load
service is equivalent to a request for no rate response to
congestion, but with admission control).
5.6. IP in IP Tunnels
For re-ECN to work correctly through IP in IP tunnels, it needs
slightly different tunnel handling to regular ECN [RFC3168].
Currently there is some incosistency between how the handling of IP
in IP tunnels is defined in [RFC3168] and how it is defined in
[RFC4301], but re-ECN would work fine with the IPsec behaviour. This
inconsistency is addressed in a new Internet Draft [ECN-tunnel] that
proposes to update RFC3168 tunnel behaviour to bring it into line
with IPsec. Ideally, for re-ECN to work through a tunnel, the tunnel
entry should copy both the RE flag and the ECN field from the inner
to the outer IP header. Then at the tunnel exit, any congestion
marking of the outer ECN field should overwrite the inner ECN field
(unless the inner field is Not-ECT in which case an alarm should be
raised). The RE flag shouldn't change along a path, so the outer RE
flag should be the same as the inner. If it isn't a management alarm
should be raised. This behaviour is the same as the full-
functionality variant of [RFC3168] at tunnel exit, but different at
tunnel entry.
If tunnels are left as they are specified in [RFC3168], whether the
limited or full-functionality variants are used, a problem arises
with re-ECN if a tunnel crosses an inter-domain boundary, because the
difference between positive and negative markings will not be
correctly accounted for. In a limited functionality ECN tunnel, the
flow will appear to be RFC3168 compliant traffic, and therefore may
be wrongly rate limited. In a full-functionality ECN tunnel, the
result will depend whether the tunnel entry copies the inner RE flag
to the outer header or the RE flag in the outer header is always
cleared. If the former, the flow will tend to be too positive when
accounted for at borders. If the latter, it will be too negative.
If the rules set out in [ECN-tunnel] are followed then this will not
be an issue.
5.7. Non-Issues
The following issues might seem to cause unfavourable interactions
with re-ECN, but we will explain why they don't:
o Various link layers support explicit congestion notification, such
as Frame Relay and ATM. Explicit congestion notification is
proposed to be added to other link layers, such as Ethernet
(802.3ar Ethernet congestion management) and MPLS [RFC5129];
o Encryption and IPSec.
In the case of congestion notification at the link layer, each
particular link layer scheme either manages congestion on the link
with its own link-level feedback (the usual arrangement in the cases
of ATM and Frame Relay), or congestion notification from the link
layer is merged into congestion notification at the IP level when the
frame headers are decapsulated at the end of the link (the
recommended arrangement in the Ethernet and MPLS cases). Given the
RE flag is not intended to change along the path, this means that
downstream congestion will still be measureable at any point where IP
is processed on the path by subtracting positive from negative
markings.
In the case of encryption, as long as the tunnel issues described in
Section 5.6 are dealt with, payload encryption itself will not be a
problem. The design goal of re-ECN is to include downstream
congestion in the IP header so that it is not necessary to bury into
inner headers. Obfuscation of flow identifiers is not a problem for
re-ECN policing elements. Re-ECN doesn't ever require flow
identifiers to be valid, it only requires them to be unique. So if
an IPSec encapsulating security payload (ESP [RFC4305]) or an
authentication header (AH [RFC4302]) is used, the security parameters
index (SPI) will be a sufficient flow identifier, as it is intended
to be unique to a flow without revealing actual port numbers.
In general, even if endpoints use some locally agreed scheme to hide
port numbers, re-ECN policing elements can just consider the pair of
source and destination IP addresses as the flow identifier. Re-ECN
encourages endpoints to at least tell the network layer that a
sequence of packets are all part of the same flow, if indeed they
are. The alternative would be for the sender to make each packet
appear to be a new flow, which would require them all to be marked
FNE in order to avoid being treated with the bulk of malicious flows
at the egress dropper. Given the FNE marking is worth +1 and
networks are likely to rate limit FNE packets, endpoints are given an
incentive not to set FNE on each packet. But if the sender really
does want to hide the flow relationship between packets it can choose
to pay the cost of multiple FNE packets, which in the long run will
compensate for the extra memory required on network policing elements
to process each flow.
6. Transport Layers
6.1. TCP
Re-ECN capability at the sender is essential. At the receiver it is Re-ECN capability at the sender is essential. At the receiver it is
optional, as long as the receiver has a basic RFC3168-compliant ECN- optional, as long as the receiver has a basic RFC3168-compliant ECN-
capable transport (ECT) [RFC3168]. Given re-ECN is not the first capable transport (ECT) [RFC3168]. Given re-ECN is not the first
attempt to define the semantics of the ECN field, we give a table attempt to define the semantics of the ECN field, we give a table
below summarising what happens for various combinations of below summarising what happens for various combinations of
capabilities of the sender S and receiver R, as indicated in the capabilities of the sender S and receiver R, as indicated in the
first four columns below. The last column gives the mode a half- first four columns below. The last column gives the mode a half-
connection should be in after the first two of the three TCP connection should be in after the first two of the three TCP
handshakes. handshakes.
skipping to change at page 17, line 5 skipping to change at page 22, line 40
at least one of the transports does not understand even basic ECN at least one of the transports does not understand even basic ECN
marking. marking.
Note that we use the term Re-ECT for a host transport that is re-ECN- Note that we use the term Re-ECT for a host transport that is re-ECN-
capable but RECN for the modes of the half connections between hosts capable but RECN for the modes of the half connections between hosts
when they are both Re-ECT. If a host transport is Re-ECT, this fact when they are both Re-ECT. If a host transport is Re-ECT, this fact
alone does NOT imply either of its half connections will necessarily alone does NOT imply either of its half connections will necessarily
be in RECN mode, at least not until it has confirmed that the other be in RECN mode, at least not until it has confirmed that the other
host is Re-ECT. host is Re-ECT.
4.1.1. RECN mode: Full Re-ECN capable transport 6.1.1. RECN mode: Full Re-ECN capable transport
In full RECN mode, for each half connection, both the sender and the In full RECN mode, for each half connection, both the sender and the
receiver each maintain an unsigned integer counter we will call ECC receiver each maintain an unsigned integer counter we will call ECC
(echo congestion counter). The receiver maintains a count of how (echo congestion counter). The receiver maintains a count of how
many times a CE marked packet has arrived during the half-connection. many times a CE marked packet has arrived during the half-connection.
Once a RECN connection is established, the three TCP option flags Once a RECN connection is established, the three TCP option flags
(ECE, CWR & NS) used for ECN-related functions in other versions of (ECE, CWR & NS) used for ECN-related functions in other versions of
ECN are used as a 3-bit field for the receiver to repeatedly tell the ECN are used as a 3-bit field for the receiver to repeatedly tell the
sender the current value of ECC, modulo 8, whenever it sends a TCP sender the current value of ECC, modulo 8, whenever it sends a TCP
ACK. We will call this the echo congestion increment (ECI) field. ACK. We will call this the echo congestion increment (ECI) field.
This overloaded use of these 3 option flags as one 3-bit ECI field is This overloaded use of these 3 option flags as one 3-bit ECI field is
shown in Figure 4. The actual definition of the TCP header, shown in Figure 7. The actual definition of the TCP header,
including the addition of support for the ECN nonce, is shown for including the addition of support for the ECN nonce, is shown for
comparison in Figure 3. This specification does not redefine the comparison in Figure 6. This specification does not redefine the
names of these three TCP option flags, it merely overloads them with names of these three TCP option flags, it merely overloads them with
another definition once a flow is established. another definition once a flow is established.
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
+---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+ +---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+
| | | N | C | E | U | A | P | R | S | F | | | | N | C | E | U | A | P | R | S | F |
| Header Length | Reserved | S | W | C | R | C | S | S | Y | I | | Header Length | Reserved | S | W | C | R | C | S | S | Y | I |
| | | | R | E | G | K | H | T | N | N | | | | | R | E | G | K | H | T | N | N |
+---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+ +---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+
Figure 3: The (post-ECN Nonce) definition of bytes 13 and 14 of the Figure 6: The (post-ECN Nonce) definition of bytes 13 and 14 of the
TCP Header TCP Header
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
+---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+ +---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+
| | | | U | A | P | R | S | F | | | | | U | A | P | R | S | F |
| Header Length | Reserved | ECI | R | C | S | S | Y | I | | Header Length | Reserved | ECI | R | C | S | S | Y | I |
| | | | G | K | H | T | N | N | | | | | G | K | H | T | N | N |
+---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+ +---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+
Figure 4: Definition of the ECI field within bytes 13 and 14 of the Figure 7: Definition of the ECI field within bytes 13 and 14 of the
TCP Header, overloading the current definitions above for established TCP Header, overloading the current definitions above for established
RECN flows. RECN flows.
Receiver Action in RECN Mode Receiver Action in RECN Mode
Every time a CE marked packet arrives at a receiver in RECN mode, Every time a CE marked packet arrives at a receiver in RECN mode,
the receiver transport increments its local value of ECC and MUST the receiver transport increments its local value of ECC and MUST
echo its value, modulo 8, to the sender in the ECI field of the echo its value, modulo 8, to the sender in the ECI field of the
next ACK. It MUST repeat the same value of ECI in every next ACK. It MUST repeat the same value of ECI in every
subsequent ACK until the next CE event, when it increments ECI subsequent ACK until the next CE event, when it increments ECI
skipping to change at page 18, line 30 skipping to change at page 24, line 22
below for the sender's safety strategy). Whenever the ECI field below for the sender's safety strategy). Whenever the ECI field
increments by D (and/or d drops are detected), the sender MUST increments by D (and/or d drops are detected), the sender MUST
clear the RE flag to "0" in the IP header of the next D' data clear the RE flag to "0" in the IP header of the next D' data
packets it sends (where D' = D + d), effectively re-echoing each packets it sends (where D' = D + d), effectively re-echoing each
single increment of ECI. Otherwise the data sender MUST send all single increment of ECI. Otherwise the data sender MUST send all
data packets with RE set to "1". data packets with RE set to "1".
As a general rule, once a flow is established, as well as setting As a general rule, once a flow is established, as well as setting
or clearing the RE flag as above, a data sender in RECN mode MUST or clearing the RE flag as above, a data sender in RECN mode MUST
always set the ECN field to ECT(1). However, the settings of the always set the ECN field to ECT(1). However, the settings of the
extended ECN field during flow start are defined in Section 4.1.4. extended ECN field during flow start are defined in Section 6.1.4.
As we have already emphasised, the re-ECN protocol makes no As we have already emphasised, the re-ECN protocol makes no
changes and has no effect on the TCP congestion control algorithm. changes and has no effect on the TCP congestion control algorithm.
So, the first increment of ECI (or detection of a drop) in a RTT So, the first increment of ECI (or detection of a drop) in a RTT
triggers the standard TCP congestion response, no more than one triggers the standard TCP congestion response, no more than one
congestion response per round trip, as usual. However, the sender congestion response per round trip, as usual. However, the sender
re-echoes every increment of ECI irrespective of RTTs. re-echoes every increment of ECI irrespective of RTTs.
A TCP sender also acts as the receiver for the other half- A TCP sender also acts as the receiver for the other half-
connection. The host will maintain two ECC values S.ECC and R.ECC connection. The host will maintain two ECC values S.ECC and R.ECC
as sender and receiver respectively. Every TCP header sent by a as sender and receiver respectively. Every TCP header sent by a
host in RECN mode will also repeat the prevailing value of R.ECC host in RECN mode will also repeat the prevailing value of R.ECC
in its ECI field. If a sender in RECN mode has to retransmit a in its ECI field. If a sender in RECN mode has to retransmit a
packet due to a suspected loss, the re-transmitted packet MUST packet due to a suspected loss, the re-transmitted packet MUST
carry the latest prevailing value of R.ECC when it is re- carry the latest prevailing value of R.ECC when it is re-
transmitted, which will not necessarily be the one it carried transmitted, which will not necessarily be the one it carried
originally. originally.
4.1.1.1. Drops and Marks 6.1.2. RECN-Co mode: Re-ECT Sender with a RFC3168 compliant ECN
Re-ECN is based on the ECN protocol [RFC3168] . In turn the
congestion markings ECN uses are typically based on the RED
algorithm [RFC2309]. This algorithm marks packets as CE with a
probability that increases as the size of the router queue increases.
However, if the queue becomes too full then it will revert to
dropping packets. Because of this it is important that a re-ECN
sender treats each packet drop it detects as if it were actually a CE
mark. This ensures that it can continue to correctly echo congestion
even through a highly congested path.
In order to ensure that drops are correctly echoed the sender needs
to add the number of drops detected per RTT to the difference in ECI
value waiting to be echoed. Drop detection is defined as set out in
[RFC2581] -- if the connection is in slow start then a single
duplicate aknowledgement will be treated as an indication of a drop.
When the system is in the congestion avoidance stage then 3 duplicate
acknowledgements will be treated as a sign of a drop. In all cases,
if a re-transmission time-out occurs then that will be treatd as a
drop.
4.1.1.2. Safety against Long Pure ACK Loss Sequences
The ECI method was chosen for echoing congestion marking because a
re-ECN sender needs to know about every CE mark arriving at the
receiver, not just whether at least one arrives within a round trip
time (which is all the ECE/CWR mechanism supported). And, as pure
ACKs are not protected by TCP reliable delivery, we repeat the same
ECI value in every ACK until it changes. Even if many ACKs in a row
are lost, as soon as one gets through, the ECI field it repeats from
previous ACKs that didn't get through will update the sender on how
many CE marks arrived since the last ACK got through.
The sender will only lose a record of the arrival of a CE mark if all
the ACKS are lost (and all of them were pure ACKs) for a stream of
data long enough to contain 8 or more CE marks. So, if the marking
fraction was p, at least 8/p pure ACKs would have to be lost. For
example, if p was 5%, a sequence of 160 pure ACKs would all have to
be lost. To protect against such extremely unlikely events, if a re-
ECN sender detects a sequence of pure ACKs has been lost it SHOULD
assume the ECI field wrapped as many times as possible within the
sequence.
Specifically, if a re-ECN sender receives an ACK with an
acknowledgement number that acknowledges L segments since the
previous ACK but with a sequence number unchanged from the previously
received ACK, it SHOULD conservatively assume that the ECI field
incremented by D' = L - ((L-D) mod 8), where D is the apparent
increase in the ECI field. For example if the ACK arriving after 9
pure ACK losses apparently increased ECI by 2, the assumed increment
of ECI would still be 2. But if ECI apparently increased by 2 after
11 pure ACK losses, ECI should be assumed to have increased by 10.
A re-ECN sender MAY implement a heuristic algorithm to predict beyond
reasonable doubt that the ECI field probably did not wrap within a
sequence of lost pure ACKs. But such an algorithm is OPTIONAL. Such
an algorithm MUST NOT be used unless it is proven to work even in the
presence of correlation between high ACK loss rate on the back
channel and high CE marking rate on the forward channel.
Whatever assumption a re-ECN sender makes about potentially lost CE
marks, both its congestion control and its re-echoing behaviour
SHOULD be consistent with the assumption it makes.
4.1.2. RECN-Co mode: Re-ECT Sender with a RFC3168 compliant ECN
Receiver Receiver
If the half-connection is in RECN-Co mode, ECN feedback proceeds no If the half-connection is in RECN-Co mode, ECN feedback proceeds no
differently to that of RFC3168 compliant ECN. In other words, the differently to that of RFC3168 compliant ECN. In other words, the
receiver sets the ECE flag repeatedly in the TCP header and the receiver sets the ECE flag repeatedly in the TCP header and the
sender responds by setting the CWR flag. Although RECN-Co mode is sender responds by setting the CWR flag. Although RECN-Co mode is
used when the receiver has not implemented the re-ECN protocol, the used when the receiver has not implemented the re-ECN protocol, the
sender can infer enough from its RFC3168 compliant ECN feedback to sender can infer enough from its RFC3168 compliant ECN feedback to
set or clear the RE flag reasonably well. Specifically, every time set or clear the RE flag reasonably well. Specifically, every time
the receiver toggles the ECE field from "0" to "1" (or a loss is the receiver toggles the ECE field from "0" to "1" (or a loss is
skipping to change at page 20, line 45 skipping to change at page 25, line 19
packets with RE set to "1". Once a flow is established, a re-ECN packets with RE set to "1". Once a flow is established, a re-ECN
data sender in RECN-Co mode MUST always set the ECN field to ECT(1). data sender in RECN-Co mode MUST always set the ECN field to ECT(1).
If a CE marked packet arrives at the receiver within a round trip If a CE marked packet arrives at the receiver within a round trip
time of a previous mark, the receiver will still be echoing ECE for time of a previous mark, the receiver will still be echoing ECE for
the last CE mark. Therefore, such a mark will be missed by the the last CE mark. Therefore, such a mark will be missed by the
sender. Of course, this isn't of concern for congestion control, but sender. Of course, this isn't of concern for congestion control, but
it does mean that very occasionally the RE blanking fraction will be it does mean that very occasionally the RE blanking fraction will be
understated. Therefore flows in RECN-Co mode may occasionally be understated. Therefore flows in RECN-Co mode may occasionally be
mistaken for very lightly cheating flows and consequently might mistaken for very lightly cheating flows and consequently might
suffer a small number of packet drops through an egress dropper suffer a small number of packet drops through an egress dropper. We
(Section 6.1.4). We expect re-ECN would be deployed for some time expect re-ECN would be deployed for some time before policers and
before policers and droppers start to enforce it. So, given there is droppers start to enforce it. So, given there is not much ECN
not much ECN deployment yet anyway, this minor problem may affect deployment yet anyway, this minor problem may affect only a very
only a very small proportion of flows, reducing to nothing over the small proportion of flows, reducing to nothing over the years as
years as RFC3168 compliant ECN hosts upgrade. The use of RECN-Co RFC3168 compliant ECN hosts upgrade. The use of RECN-Co mode would
mode would need to be reviewed in the light of experience at the time need to be reviewed in the light of experience at the time of re-ECN
of re-ECN deployment. deployment.
RECN-Co mode is OPTIONAL. Re-ECN implementers who want to keep their RECN-Co mode is OPTIONAL. Re-ECN implementers who want to keep their
code simple, MAY choose not to implement this mode. If they do not, code simple, MAY choose not to implement this mode. If they do not,
a re-ECN sender SHOULD fall back to RFC3168 compliant ECT mode in the a re-ECN sender SHOULD fall back to RFC3168 compliant ECT mode in the
presence of an ECN-capable receiver. It MAY choose to fall back to presence of an ECN-capable receiver. It MAY choose to fall back to
the ECT-Nonce mode, but if re-ECN implementers don't want to be the ECT-Nonce mode, but if re-ECN implementers don't want to be
bothered with RECN-Co mode, they probably won't want to add an ECT- bothered with RECN-Co mode, they probably won't want to add an ECT-
Nonce mode either. Nonce mode either.
4.1.2.1. Re-ECN support for the ECN Nonce 6.1.2.1. Re-ECN support for the ECN Nonce
A TCP half-connection in RECN-Co mode MUST NOT support the ECN A TCP half-connection in RECN-Co mode MUST NOT support the ECN
Nonce [RFC3540]. This means that the sending code of a re-ECN Nonce [RFC3540]. This means that the sending code of a re-ECN
implementation will never need to include ECN Nonce support. Re-ECN implementation will never need to include ECN Nonce support. Re-ECN
is intended to provide wider protection than the ECN nonce against is intended to provide wider protection than the ECN nonce against
congestion control misbehaviour, and re-ECN only requires support congestion control misbehaviour, and re-ECN only requires support
from the sender, therefore it is preferable to specifically rule out from the sender, therefore it is preferable to specifically rule out
the need for dual sender implementations. As a consequence, a re-ECN the need for dual sender implementations. As a consequence, a re-ECN
capable sender will never set ECT(0), so it will be easier for capable sender will never set ECT(0), so it will be easier for
network elements to discriminate re-ECN traffic flows from other ECN network elements to discriminate re-ECN traffic flows from other ECN
skipping to change at page 21, line 41 skipping to change at page 26, line 15
RFC3540 allows an ECN nonce sender to choose whether to sanction a RFC3540 allows an ECN nonce sender to choose whether to sanction a
receiver that does not ever set the nonce sum. Given re-ECN is receiver that does not ever set the nonce sum. Given re-ECN is
intended to provide wider protection than the ECN nonce against intended to provide wider protection than the ECN nonce against
congestion control misbehaviour, implementers of re-ECN receivers MAY congestion control misbehaviour, implementers of re-ECN receivers MAY
choose not to implement backwards compatibility with the ECN nonce choose not to implement backwards compatibility with the ECN nonce
capability. This may be because they deem that the risk of sanctions capability. This may be because they deem that the risk of sanctions
is low, perhaps because significant deployment of the ECN nonce seems is low, perhaps because significant deployment of the ECN nonce seems
unlikely at implementation time. unlikely at implementation time.
4.1.3. Capability Negotiation 6.1.3. Capability Negotiation
During the TCP hand-shake at the start of a connection, an originator During the TCP hand-shake at the start of a connection, an originator
of the connection (host A) with a re-ECN-capable transport MUST of the connection (host A) with a re-ECN-capable transport MUST
indicate it is Re-ECT by setting the TCP flags NS=1, CWR=1 and ECE=1 indicate it is Re-ECT by setting the TCP flags NS=1, CWR=1 and ECE=1
in the initial SYN. in the initial SYN.
A responding Re-ECT host (host B) MUST return a SYN ACK with flags A responding Re-ECT host (host B) MUST return a SYN ACK with flags
CWR=1 and ECE=0. The responding host MUST NOT set this combination CWR=1 and ECE=0. The responding host MUST NOT set this combination
of flags unless the preceding SYN has already indicated Re-ECT of flags unless the preceding SYN has already indicated Re-ECT
support as above. Normally a Re-ECT server (B) will reply to a Re- support as above. Normally a Re-ECT server (B) will reply to a Re-
skipping to change at page 23, line 19 skipping to change at page 27, line 42
preceding SYN (because there is a broken RFC3168 compliant preceding SYN (because there is a broken RFC3168 compliant
implementation that behaves this way), RFC3168 specifies that the implementation that behaves this way), RFC3168 specifies that the
whole connection MUST revert to Not-ECT. whole connection MUST revert to Not-ECT.
Also note that, whenever the SYN flag of a TCP segment is set Also note that, whenever the SYN flag of a TCP segment is set
(including when the ACK flag is also set), the NS, CWR and ECE flags (including when the ACK flag is also set), the NS, CWR and ECE flags
( i.e the ECI field of the SYNACK) MUST NOT be interpreted as the ( i.e the ECI field of the SYNACK) MUST NOT be interpreted as the
3-bit ECI value, which is only set as a copy of the local ECC value 3-bit ECI value, which is only set as a copy of the local ECC value
in non-SYN packets. in non-SYN packets.
4.1.4. Extended ECN (EECN) Field Settings during Flow Start or after 6.1.4. Extended ECN (EECN) Field Settings during Flow Start or after
Idle Periods Idle Periods
If the originator (A) of a TCP connection supports re-ECN it MUST set If the originator (A) of a TCP connection supports re-ECN it MUST set
the extended ECN (EECN) field in the IP header of the initial SYN the extended ECN (EECN) field in the IP header of the initial SYN
packet to the feedback not established (FNE) codepoint. packet to the feedback not established (FNE) codepoint.
FNE is a new extended ECN codepoint defined by this specification FNE is a new extended ECN codepoint defined by this specification
(Section 3.3). The feedback not established (FNE) codepoint is used (Section 4.2). The feedback not established (FNE) codepoint is used
when the transport does not have the benefit of ECN feedback so it when the transport does not have the benefit of ECN feedback so it
cannot decide whether to set or clear the RE flag. cannot decide whether to set or clear the RE flag.
If after receiving a SYN the server B has set its sending half- If after receiving a SYN the server B has set its sending half-
connection into RECN mode or RECN-Co mode, it MUST set the extended connection into RECN mode or RECN-Co mode, it MUST set the extended
ECN field in the IP header of its SYN ACK to the feedback not ECN field in the IP header of its SYN ACK to the feedback not
established (FNE) codepoint. Note the careful wording here, which established (FNE) codepoint. Note the careful wording here, which
means that Re-ECT server B MUST set FNE on a SYN ACK whether it is means that Re-ECT server B MUST set FNE on a SYN ACK whether it is
responding to a SYN from a Re-ECT client or from a client that is responding to a SYN from a Re-ECT client or from a client that is
merely ECN-capable. This is because FNE indicates the transport is merely ECN-capable. This is because FNE indicates the transport is
skipping to change at page 27, line 5 skipping to change at page 31, line 9
trip time. We use the lower bound of the retransmission timeout trip time. We use the lower bound of the retransmission timeout
(RTO) [RFC2988], which is commonly used as the idle period before TCP (RTO) [RFC2988], which is commonly used as the idle period before TCP
must reduce to the restart window [RFC2581]. Note our specification must reduce to the restart window [RFC2581]. Note our specification
of re-ECN's idle period is NOT intended to change the idle period for of re-ECN's idle period is NOT intended to change the idle period for
TCP's restart, nor indeed for any other purposes. TCP's restart, nor indeed for any other purposes.
{ToDo: Describe how the sender falls back to RFC3168 modes if packets {ToDo: Describe how the sender falls back to RFC3168 modes if packets
don't appear to be getting through (to work round firewalls don't appear to be getting through (to work round firewalls
discarding packets they consider unusual).} discarding packets they consider unusual).}
4.1.5. Pure ACKS, Retransmissions, Window Probes and Partial ACKs 6.1.5. Pure ACKS, Retransmissions, Window Probes and Partial ACKs
A re-ECN sender MUST clear the RE flag to "0" and set the ECN field A re-ECN sender MUST clear the RE flag to "0" and set the ECN field
to Not-ECT in pure ACKs, retransmissions and window probes, as to Not-ECT in pure ACKs, retransmissions and window probes, as
specified in [RFC3168]. Our eventual goal is for all packets to be specified in [RFC3168]. Our eventual goal is for all packets to be
sent with re-ECN enabled, and we believe the semantics of the ECI sent with re-ECN enabled, and we believe the semantics of the ECI
field go a long way towards being able to achieve this. However, we field go a long way towards being able to achieve this. However, we
have not completed a full security analysis for these cases, have not completed a full security analysis for these cases,
therefore, currently we merely re-state current practice. therefore, currently we merely re-state current practice.
We must also reconcile the facts that congestion marking is applied We must also reconcile the facts that congestion marking is applied
skipping to change at page 27, line 47 skipping to change at page 32, line 5
through the variable R. through the variable R.
This does not ensure precisely the same number of octets have RE This does not ensure precisely the same number of octets have RE
blanked as were CE marked. But we believe positive errors will blanked as were CE marked. But we believe positive errors will
cancel negative over a long enough period. {ToDo: However, more cancel negative over a long enough period. {ToDo: However, more
research is needed to prove whether this is so. If it is not, it may research is needed to prove whether this is so. If it is not, it may
be necessary to increment and decrement R in octets rather than be necessary to increment and decrement R in octets rather than
packets, by incrementing R as the product of D and the size in octets packets, by incrementing R as the product of D and the size in octets
of packets being sent (typically the MSS).} of packets being sent (typically the MSS).}
4.2. Other Transports 6.2. Other Transports
4.2.1. General Guidelines for Adding Re-ECN to Other Transports 6.2.1. General Guidelines for Adding Re-ECN to Other Transports
As a general rule, Re-ECT sender transports that have established the As a general rule, Re-ECT sender transports that have established the
receiver transport is at least ECN-capable (not necessarily re-ECN receiver transport is at least ECN-capable (not necessarily re-ECN
capable) MUST blank the RE codepoint for at least as many octets as capable) MUST blank the RE codepoint for at least as many octets as
arrive at receiver with the CE codepoint set. Re-ECN-capable sender arrive at receiver with the CE codepoint set. Re-ECN-capable sender
transports should always initialise the ECN field to the ECT(1) transports should always initialise the ECN field to the ECT(1)
codepoint once a flow is established. codepoint once a flow is established.
If the sender transport does not have sufficient feedback to even If the sender transport does not have sufficient feedback to even
estimate the path's CE rate, it SHOULD set FNE continuously. If the estimate the path's CE rate, it SHOULD set FNE continuously. If the
skipping to change at page 28, line 32 skipping to change at page 32, line 39
following: following:
o UDP fire and forget (e.g. DNS) o UDP fire and forget (e.g. DNS)
o UDP streaming with no feedback o UDP streaming with no feedback
o UDP streaming with feedback o UDP streaming with feedback
} }
4.2.2. Guidelines for adding Re-ECN to RSVP or NSIS 6.2.2. Guidelines for adding Re-ECN to RSVP or NSIS
A separate I-D has been submitted [Re-PCN] describing how re-ECN can A separate I-D has been submitted [Re-PCN] describing how re-ECN can
be used in an edge-to-edge rather than end-to-end scenario. It can be used in an edge-to-edge rather than end-to-end scenario. It can
then be used by downstream networks to police whether upstream then be used by downstream networks to police whether upstream
networks are blocking new flow reservations when downstream networks are blocking new flow reservations when downstream
congestion is too high, even though the congestion is in other congestion is too high, even though the congestion is in other
operators' downstream networks. This relates to current IETF work on operators' downstream networks. This relates to current IETF work on
Admission Control over Diffserv using Pre-Congestion Notification Admission Control over Diffserv using Pre-Congestion Notification
(PCN) [PCN-arch]. (PCN) [PCN-arch].
4.2.3. Guidelines for adding Re-ECN to DCCP 6.2.3. Guidelines for adding Re-ECN to DCCP
Beside adjusting the initial features negotiation sequence, operating Beside adjusting the initial features negotiation sequence, operating
re-ECN in DCCP [RFC4340] could be achieved by defining a new option re-ECN in DCCP [RFC4340] could be achieved by defining a new option
to be added to acknowledgments, that would include a multibit field to be added to acknowledgments, that would include a multibit field
where the destination could copy its ECC. where the destination could copy its ECC.
4.2.4. Guidelines for adding Re-ECN to SCTP 6.2.4. Guidelines for adding Re-ECN to SCTP
Appendix A in [RFC4960] gives the specifications for SCTP to support Appendix A in [RFC4960] gives the specifications for SCTP to support
ECN. Similar steps should be taken to support re-ECN. Beside ECN. Similar steps should be taken to support re-ECN. Beside
adjusting the initial features negotiation sequence, operating re-ECN adjusting the initial features negotiation sequence, operating re-ECN
in SCTP could be achieved by defining a new control chunk, that would in SCTP could be achieved by defining a new control chunk, that would
include a multibit field where the destination could copy its ECC include a multibit field where the destination could copy its ECC
5. Network Layer
5.1. Re-ECN IPv4 Wire Protocol
The wire protocol of the ECN field in the IP header remains largely
unchanged from [RFC3168]. However, an extension to the ECN field we
call the RE (Re-ECN extension) flag (Section 3.3) is defined in this
document. It doubles the extended ECN codepoint space, giving 8
potential codepoints. The semantics of the extra codepoints are
backward compatible with the semantics of the 4 original codepoints
[RFC3168] (Section 7.1 collects together and summarises all the
changes defined in this document).
For IPv4, this document proposes that the new RE control flag will be
positioned where the `reserved' control flag was at bit 48 of the
IPv4 header (counting from 0). Alternatively, some would call this
bit 0 (counting from 0) of byte 7 (counting from 1) of the IPv4
header (Figure 5).
0 1 2
+---+---+---+
| R | D | M |
| E | F | F |
+---+---+---+
Figure 5: New Definition of the Re-ECN Extension (RE) Control Flag at
the Start of Byte 7 of the IPv4 Header
The semantics of the RE flag are described in outline in Section 3
and specified fully in Section 4. The RE flag is always considered
in conjunction with the 2-bit ECN field, as if they were concatenated
together to form a 3-bit extended ECN field. If the ECN field is set
to either the ECT(1) or CE codepoint, when the RE flag is blanked
(cleared to "0") it represents a re-echo of congestion experienced by
an early packet. If the ECN field is set to the Not-ECT codepoint,
when the RE flag is set to "1" it represents the feedback not
established (FNE) codepoint, which signals that the packet was sent
without the benefit of congestion feedback.
It is believed that the FNE codepoint can simultaneously serve other
purposes, particularly where the start of a flow needs distinguishing
from packets later in the flow. For instance it would have been
useful to identify new flows for tag switching and might enable
similar developments in the future if it were adopted. It is similar
to the state set-up bit idea designed to protect against memory
exhaustion attacks. This idea was proposed informally by David Clark
and documented by Handley and Greenhalgh [Steps_DoS]. The FNE
codepoint can be thought of as a `soft-state set-up flag', because it
is idempotent (i.e. one occurrence of the flag is sufficient but
further occurrences achieve the same effect if previous ones were
lost).
We are sure there will probably be other claims pending on the use of
bit 48. We know of at least two [ARI05], [RFC3514] but neither have
been pursued in the IETF, so far, although the present proposal would
meet the needs of the former.
The security flag proposal (commonly known as the evil bit) was
published on 1 April 2003 as Informational RFC 3514, but it was not
adopted due to confusion over whether evil-doers might set it
inappropriately. The present proposal is backward compatible with
RFC3514 because if re-ECN compliant senders were benign they would
correctly clear the evil bit to honestly declare that they had just
received congestion feedback. Whereas evil-doers would hide
congestion feedback by setting the evil bit continuously, or at least
more often than they should. So, evil senders can be identified,
because they declare that they are good less often than they should.
5.2. Re-ECN IPv6 Wire Protocol
For IPv6, this document proposes that the new RE control flag will be
positioned as the first bit of the option field of a new Congestion
hop by hop option header (Figure 6).
0 1 2 3
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Next Header | Hdr ext Len | Option Type | Opt Length =4 |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|R| Reserved for future use |
|E| |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Figure 6: Definition of a New IPv6 Congestion Hop by Hop Option
Header containing the re-ECN Extension (RE) Control Flag
0 1 2 3 4 5 6 7 8
+-+-+-+-+-+-+-+-+-
|AIU|C|Option ID|
+-+-+-+-+-+-+-+-+-
Figure 7: Congestion Hop by Hop Option Type Encoding
The Hop-by-Hop Options header enables packets to carry information to
be examined and processed by routers or nodes along the packet's
delivery path, including the source and destination nodes. For re-
ECN, the two bits of the Action If Unrecognized (AIU) flag of the
Congestion extension header MUST be set to "00" meaning if
unrecognized `skip over option and continue processing the header'.
Then, any routers or a receiver not upgraded with the optional re-ECN
features described in this memo will simply ignore this header. But
routers with these optional re-ECN features or a re-ECN policing
function, will process this Congestion extension header.
The `C' flag MUST be set to "1" to specify that the Option Data
(currently only the RE control flag) can change en-route to the
packet's final destination. This ensures that, when an
Authentication header (AH [RFC4302]) is present in the packet, for
any option whose data may change en-route, its entire Option Data
field will be treated as zero-valued octets when computing or
verifying the packet's authenticating value.
Although the RE control flag should not be changed along the path, we
expect that the rest of this option field that is currently `Reserved
for future use' could be used for a multi-bit congestion notification
field which we would expect to change en route. As the RE flag does
not need end-to-end authentication, we set the C flag to '1'.
{ToDo: A Congestion Hop by Hop Option ID will need to be registered
with IANA.}
5.3. Router Forwarding Behaviour
Re-ECN works well without modifying the forwarding behaviour of any
routers. However, below, two OPTIONAL changes to forwarding
behaviour are defined which respectively enhance performance and
improve a router's discrimination against flooding attacks. They are
both OPTIONAL additions that we propose MAY apply by default to all
Diffserv per-hop scheduling behaviours (PHBs) [RFC2475] and ECN
marking behaviours [RFC3168]. Specifications for PHBs MAY define
different forwarding behaviours from this default, but this is not
required. [Re-PCN] is one example.
FNE indicates ECT:
The FNE codepoint tells a router to assume that the packet was
sent by an ECN-capable transport (see Section 5.4). Therefore an
FNE packet MAY be marked rather than dropped. Note that the FNE
codepoint has been intentionally chosen so that, to RFC3168
compliant routers (which do not inspect the RE flag) an FNE packet
appears to be Not-ECT so it will be dropped by legacy AQM
algorithms.
A network operator MUST NOT configure a queue to ECN mark rather
than drop FNE packets unless it can guarantee that FNE packets
will be rate limited, either locally or upstream. The ingress
policers discussed in Section 6.1.5 would count as rate limiters
for this purpose.
Preferential Drop: If a re-ECN capable router queue experiences very
high load so that it has to drop arriving packets (e.g. a DoS
attack), it MAY preferentially drop packets within the same
Diffserv PHB using the preference order for extended ECN
codepoints given in Table 7. Preferential dropping can be
difficult to implement on some hardware, but if feasible it would
discriminate against attack traffic if done as part of the overall
policing framework of Section 6.1.3. If nowhere else, routers at
the egress of a network SHOULD implement preferential drop
(stronger than the MAY above). For simplicity, preferences 4 & 5
MAY be merged into one preference level.
+-------+-----+------------+-------+------------+-------------------+
| ECN | RE | Extended | Worth | Drop Pref | Re-ECN meaning |
| field | bit | ECN | | (1 = drop | |
| | | codepoint | | 1st) | |
+-------+-----+------------+-------+------------+-------------------+
| 01 | 0 | Re-Echo | +1 | 5/4 | Re-echoed |
| | | | | | congestion and |
| | | | | | RECT |
| 00 | 1 | FNE | +1 | 4 | Feedback not |
| | | | | | established |
| 11 | 0 | CE(0) | 0 | 3 | Re-Echo canceled |
| | | | | | by congestion |
| | | | | | experienced |
| 01 | 1 | RECT | 0 | 3 | Re-ECN capable |
| | | | | | transport |
| 11 | 1 | CE(-1) | -1 | 3 | Congestion |
| | | | | | experienced |
| 10 | 1 | --CU-- | n/a | 2 | Currently Unused |
| 10 | 0 | --- | n/a | 2 | RFC3168 ECN use |
| | | | | | only |
| 00 | 0 | Not-RECT | n/a | 1 | Not |
| | | | | | Re-ECN-capable |
| | | | | | transport |
+-------+-----+------------+-------+------------+-------------------+
Table 7: Drop Preference of EECN Codepoints (Sorted by `Worth')
The above drop preferences are arranged to preserve packets with
more positive worth (Section 3.5), given senders of positive
packets must have honestly declared downstream congestion. This
is explained fully in Section 6 on applications, particularly when
the application of re-ECN to protect against DDoS attacks is
described.
5.4. Justification for Setting the First SYN to FNE
the initial SYN MUST be set to FNE by Re-ECT client A (Section 4.1.4)
and (Section 5.3) says a queue MAY optionally treat an FNE packet as
ECN capable, so an initial SYN may be marked CE(-1) rather than
dropped. This seems dangerous, because the sender has not yet
established whether the receiver is a RFC3168 one that does not
understand congestion marking. It also seems to allow malicious
senders to take advantage of ECN marking to avoid so much drop when
launching SYN flooding attacks. Below we explain the features of the
protocol design that remove both these dangers.
ECN-capable initial SYN with a Not-ECT server: If the TCP server B
is re-ECN capable, provision is made for it to feedback a possible
congestion marked SYN in the SYN ACK (Section 4.1.4). But if the
TCP client A finds out from the SYN ACK that the server was not
ECN-capable, the TCP client MUST conservatively consider the first
SYN as congestion marked before setting itself into Not-ECT mode.
Section 4.1.4 mandates that such a TCP client MUST also set its
initial window to 1 segment. In this way we remove the need to
cautiously avoid setting the first SYN to Not-RECT. This will
give worse performance while deployment is patchy, but better
performance once deployment is widespread.
SYN flooding attacks can't exploit ECN-capability: Malicious hosts
may think they can use the advantage that ECN-marking gives over
drop in launching classic SYN-flood attacks. But Section 5.3
mandates that a router MUST only be configured to treat packets
with the FNE codepoint as ECN-capable if FNE packets are rate
limited somewhere. Introduction of the FNE codepoint was a
deliberate move to enable transport-neutral handling of flow-start
and flow state set-up in the IP layer where it belongs. It then
becomes possible to protect against flooding attacks of all forms
(not just SYN flooding) without transport-specific inspection for
things like the SYN flag in TCP headers. Then, for instance, SYN
flooding attacks using IPSec ESP encryption can also be rate
limited at the IP layer.
It might seem pedantic going to all this trouble to enable ECN on the
initial packet of a flow, but it is motivated by a much wider concern
to ensure safe congestion control will still be possible even if the
application mix evolves to the point where the majority of flows
consist of a single window or even a single packet. It also allows
denial of service attacks to be more easily isolated and prevented.
5.5. Control and Management
5.5.1. Negative Balance Warning
A new ICMP message type is being considered so that a dropper can
warn the apparent sender of a flow that it has started to sanction
the flow. The message would have similar semantics to the `Time
exceeded' ICMP message type. To ensure the sender has to invest some
work before the network will generate such a message, a dropper
SHOULD only send such a message for flows that have demonstrated that
they have started correctly by establishing a positive record, but
have later gone negative. The threshold is up to the implementation.
The purpose of the message is to deconfuse the cause of drops from
other causes, such as congestion or transmission losses. The dropper
would send the message to the sender of the flow, not the receiver.
If we did define this message type, it would be REQUIRED for all re-
ECT senders to parse and understand it. Note that a sender MUST only
use this message to explain why losses are occurring. A sender MUST
NOT take this message to mean that losses have occurred that it was
not aware of. Otherwise, spoof messages could be sent by malicious
sources to slow down a sender (c.f. ICMP source quench).
However, the need for this message type is not yet confirmed, as we
are considering how to prevent it being used by malicious senders to
scan for droppers and to test their threshold settings. {ToDo:
Complete this section.}
5.5.2. Rate Response Control
As discussed in Section 6.1.5 the sender's access operator will be
expected to use bulk per-user policing, but they might choose to
introduce a per-flow policer. In cases where operators do introduce
per-flow policing, there may be a need for a sender to send a request
to the ingress policer asking for permission to apply a non-default
response to congestion (where TCP-friendly is assumed to be the
default). This would require the sender to know what message
format(s) to use and to be able to discover how to address the
policer. The required control protocol(s) are outside the scope of
this document, but will require definition elsewhere.
The policer is likely to be local to the sender and inline, probably
at the ingress interface to the internetwork. So, discovery should
not be hard. A variety of control protocols already exist for some
widely used rate-responses to congestion. For instance DCCP
congestion control identifiers (CCIDs [RFC4340]) fulfil this role and
so does QoS signalling (e.g. and RSVP request for controlled load
service is equivalent to a request for no rate response to
congestion, but with admission control).
5.6. IP in IP Tunnels
For re-ECN to work correctly through IP in IP tunnels, it needs
slightly different tunnel handling to regular ECN [RFC3168].
Currently there is some incosistency between how the handling of IP
in IP tunnels is defined in [RFC3168] and how it is defined in
[RFC4301], but re-ECN would work fine with the IPsec behaviour. This
inconsistency is addressed in a new Internet Draft [ECN-tunnel] that
proposes to update RFC3168 tunnel behaviour to bring it into line
with IPsec. Ideally, for re-ECN to work through a tunnel, the tunnel
entry should copy both the RE flag and the ECN field from the inner
to the outer IP header. Then at the tunnel exit, any congestion
marking of the outer ECN field should overwrite the inner ECN field
(unless the inner field is Not-ECT in which case an alarm should be
raised). The RE flag shouldn't change along a path, so the outer RE
flag should be the same as the inner. If it isn't a management alarm
should be raised. This behaviour is the same as the full-
functionality variant of [RFC3168] at tunnel exit, but different at
tunnel entry.
If tunnels are left as they are specified in [RFC3168], whether the
limited or full-functionality variants are used, a problem arises
with re-ECN if a tunnel crosses an inter-domain boundary, because the
difference between positive and negative markings will not be
correctly accounted for. In a limited functionality ECN tunnel, the
flow will appear to be RFC3168 compliant traffic, and therefore may
be wrongly rate limited. In a full-functionality ECN tunnel, the
result will depend whether the tunnel entry copies the inner RE flag
to the outer header or the RE flag in the outer header is always
cleared. If the former, the flow will tend to be too positive when
accounted for at borders. If the latter, it will be too negative.
If the rules set out in [ECN-tunnel] are followed then this will not
be an issue.
5.7. Non-Issues
The following issues might seem to cause unfavourable interactions
with re-ECN, but we will explain why they don't:
o Various link layers support explicit congestion notification, such
as Frame Relay and ATM. Explicit congestion notification is
proposed to be added to other link layers, such as Ethernet
(802.3ar Ethernet congestion management) and MPLS [RFC5129];
o Encryption and IPSec.
In the case of congestion notification at the link layer, each
particular link layer scheme either manages congestion on the link
with its own link-level feedback (the usual arrangement in the cases
of ATM and Frame Relay), or congestion notification from the link
layer is merged into congestion notification at the IP level when the
frame headers are decapsulated at the end of the link (the
recommended arrangement in the Ethernet and MPLS cases). Given the
RE flag is not intended to change along the path, this means that
downstream congestion will still be measureable at any point where IP
is processed on the path by subtracting positive from negative
markings.
In the case of encryption, as long as the tunnel issues described in
Section 5.6 are dealt with, payload encryption itself will not be a
problem. The design goal of re-ECN is to include downstream
congestion in the IP header so that it is not necessary to bury into
inner headers. Obfuscation of flow identifiers is not a problem for
re-ECN policing elements. Re-ECN doesn't ever require flow
identifiers to be valid, it only requires them to be unique. So if
an IPSec encapsulating security payload (ESP [RFC4305]) or an
authentication header (AH [RFC4302]) is used, the security parameters
index (SPI) will be a sufficient flow identifier, as it is intended
to be unique to a flow without revealing actual port numbers.
In general, even if endpoints use some locally agreed scheme to hide
port numbers, re-ECN policing elements can just consider the pair of
source and destination IP addresses as the flow identifier. Re-ECN
encourages endpoints to at least tell the network layer that a
sequence of packets are all part of the same flow, if indeed they
are. The alternative would be for the sender to make each packet
appear to be a new flow, which would require them all to be marked
FNE in order to avoid being treated with the bulk of malicious flows
at the egress dropper. Given the FNE marking is worth +1 and
networks are likely to rate limit FNE packets, endpoints are given an
incentive not to set FNE on each packet. But if the sender really
does want to hide the flow relationship between packets it can choose
to pay the cost of multiple FNE packets, which in the long run will
compensate for the extra memory required on network policing elements
to process each flow.
6. Applications
6.1. Policing Congestion Response
6.1.1. The Policing Problem
The current Internet architecture trusts hosts to respond voluntarily
to congestion. Limited evidence shows that the large majority of
end-points on the Internet comply with a TCP-friendly response to
congestion. But telephony (and increasingly video) services over the
best effort Internet are attracting the interest of major commercial
operations. Most of these applications do not respond to congestion
at all. Those that can switch to lower rate codecs, still have a
lower bound below which they must become unresponsive to congestion.
Of course, the Internet is intended to support many different
application behaviours. But the problem is that this freedom can be
exercised irresponsibly. The greater problem is that we will never
be able to agree on where the boundary is between responsible and
irresponsible. Therefore re-ECN is designed to allow different
networks to set their own view of the limit to irresponsibility, and
to allow networks that choose a more conservative limit to push back
against congestion caused in more liberal networks.
As an example of the impossibility of setting a standard for
fairness, mandating TCP-friendliness would set the bar too high for
unresponsive streaming media, but still some would say the bar was
too low. Even though all known peer-to-peer filesharing applications
are TCP-compatible, they can cause a disproportionate amount of
congestion, simply by using multiple flows and by transferring data
continuously relative to other short-lived sessions. On the other
hand, if we swung the other way and set the bar low enough to allow
streaming media to be unresponsive, we would also allow denial of
service attacks, which are typically unresponsive to congestion and
consist of multiple continuous flows.
Applications that need (or choose) to be unresponsive to congestion
can effectively take (some would say steal) whatever share of
bottleneck resources they want from responsive flows. Whether or not
such free-riding is common, inability to prevent it increases the
risk of poor returns for investors in network infrastructure, leading
to under-investment. An increasing proportion of unresponsive or
free-riding demand coupled with persistent under-supply is a broken
economic cycle. Therefore, if the current, largely co-operative
consensus continues to erode, congestion collapse could become more
common in more areas of the Internet [RFC3714].
While we have designed re-ECN so that networks can choose to deploy
stringent policing, this does not imply we advocate that every
network should introduce tight controls on those that cause
congestion. Re-ECN has been specifically designed to allow different
networks to choose how conservative or liberal they wish to be with
respect to policing congestion. But those that choose to be
conservative can protect themselves from the excesses that liberal
networks allow their users.
6.1.2. The Case Against Bottleneck Policing
The state of the art in rate policing is the bottleneck policer,
which is intended to be deployed at any forwarding resource that may
become congested. Its aim is to detect flows that cause
significantly more local congestion than others. Although operators
might solve their immediate problems by deploying bottleneck
policers, we are concerned that widespread deployment would make it
extremely hard to evolve new application behaviours. We believe the
IETF should offer re-ECN as the preferred protocol on which to base
solutions to the policing problems of operators, because it would not
harm evolvability and, frankly, it would be far more effective (see
later for why).
Approaches like [XCHOKe] & [pBox] are nice approaches for rate
policing traffic without the benefit of whole path information (such
as could be provided by re-ECN). But they must be deployed at
bottlenecks in order to work. Unfortunately, a large proportion of
traffic traverses at least two bottlenecks (in two access networks),
particularly with the current traffic mix where peer-to-peer file-
sharing is prevalent. If ECN were deployed, we believe it would be
likely that these bottleneck policers would be adapted to combine ECN
congestion marking from the upstream path with local congestion
knowledge. But then the only useful placement for such policers
would be close to the egress of the internetwork.
But then, if these bottleneck policers were widely deployed (which
would require them to be more effective than they are now), the
Internet would find itself with one universal rate adaptation policy
(probably TCP-friendliness) embedded throughout the network. Given
TCP's congestion control algorithm is already known to be hitting its
scalability limits and new algorithms are being developed for high-
speed congestion control, embedding TCP policing into the Internet
would make evolution to new algorithms extremely painful. If a
source wanted to use a different algorithm, it would have to first
discover then negotiate with all the policers on its path,
particularly those in the far access network. The IETF has already
traveled that path with the Intserv architecture and found it
constrains scalability [RFC2208].
Anyway, if bottleneck policers were ever widely deployed, they would
be likely to be bypassed by determined attackers. They inherently
have to police fairness per flow or per source-destination pair.
Therefore they can easily be circumvented either by opening multiple
flows (by varying the end-point port number); or by spoofing the
source address but arranging with the receiver to hide the true
return address at a higher layer.
6.1.3. Re-ECN Incentive Framework
The aim is to create an incentive environment that ensures optimal
sharing of capacity despite everyone acting selfishly (including
lying and cheating). Of course, the mechanisms put in place for this
can lie dormant wherever co-operation is the norm.
Throughout this document we focus on path congestion. But some forms
of fairness, particularly TCP's, also depend on round trip time. If
TCP-fairness is required, we also propose to measure downstream path
delay using re-feedback. We give a simple outline of how this could
work in Appendix F. However, we do not expect this to be necessary,
as researchers tend to agree that only congestion control dynamics
need to depend on RTT, not the rate that the algorithm would converge
on after a period of stability.
Figure 8 sketches the incentive framework that we will describe piece
by piece throughout this section. We will do a first pass in
overview, then return to each piece in detail. We re-use the earlier
example of how downstream congestion is derived by subtracting
upstream congestion from path congestion (Figure 2) but depict
multiple trust boundaries to turn it into an internetwork. For
clarity, only downstream congestion is shown (the difference between
the two earlier plots). The graph displays downstream path
congestion seen in a typical flow as it traverses an example path
from sender S to receiver R, across networks N1, N2 & N3. Everyone
is shown using re-ECN correctly, but we intend to show why everyone
would /choose/ to use it correctly, and honestly.
Three main types of self-interest can be identified:
o Users want to transmit data across the network as fast as
possible, paying as little as possible for the privilege. In this
respect, there is no distinction between senders and receivers,
but we must be wary of potential malice by one on the other;
o Network operators want to maximise revenues from the resources
they invest in. They compete amongst themselves for the custom of
users.
o Attackers (whether users or networks) want to use any opportunity
to subvert the new re-ECN system for their own gain or to damage
the service of their victims, whether targeted or random.
policer dropper
| |
| |
S <-----N1----> <---N2---> <---N3--> R domain
|
3% |---------+
| |
2% | +-----------------------+
| downstream congestion |
1% | |
| |
0% +---------------------------------+======
0 i
Figure 8: Incentive Framework, showing creation of opposing pressures
to under-declare and over-declare downstream congestion, using a
policer and a dropper
Source congestion control: We want to ensure that the sender will
throttle its rate as downstream congestion increases. Whatever
the agreed congestion response (whether TCP-compatible or some
enhanced QoS), to some extent it will always be against the
sender's interest to comply.
Ingress policing: But it is in all the network operators' interests
to encourage fair congestion response, so that their investments
are employed to satisfy the most valuable demand. The re-ECN
protocol ensures packets carry the necessary information about
their own expected downstream congestion so that N1 can deploy a
policer at its ingress to check that S1 is complying with whatever
congestion control it should be using (Section 6.1.5). If N1 is
extremely conservative it could police each flow, but it is likely
to just police the bulk amount of congestion each customer causes
without regard to flows, or if it is extremely liberal it need not
police congestion control at all. Whatever, it is always
preferable to police traffic at the very first ingress into an
internetwork, before non-compliant traffic can cause any damage.
Edge egress dropper: If the policer ensures the source has less
right to a high rate the higher it declares downstream congestion,
the source has a clear incentive to understate downstream
congestion. But, if flows of packets are understated when they
enter the internetwork, they will have become negative by the time
they leave. So, we introduce a dropper at the last network
egress, which drops packets in flows that persistently declare
negative downstream congestion (see Section 6.1.4 for details).
Inter-domain traffic policing: But next we must ask, if congestion
arises downstream (say in N3), what is the ingress network's
(N1's) incentive to police its customers' response? If N1 turns a
blind eye, its own customers benefit while other networks suffer.
This is why all inter-domain QoS architectures (e.g. Intserv,
Diffserv) police traffic each time it crosses a trust boundary.
We have already shown that re-ECN gives a trustworthy measure of
the expected downstream congestion that a flow will cause by
subtracting negative volume from positive at any intermediate
point on a path. N3 (say) can use this measure to police all the
responses to congestion of all the sources beyond its upstream
neighbour (N2), but in bulk with one very simple passive
mechanism, rather than per flow, as we will now explain.
Emulating policing with inter-domain congestion penalties: Between
high-speed networks, we would rather avoid per-flow policing, and
we would rather avoid holding back traffic while it is policed.
Instead, once re-ECN has arranged headers to carry downstream
congestion honestly, N2 can contract to pay N3 penalties in
proportion to a single bulk count of the congestion metrics
crossing their mutual trust boundary (Section 6.1.6). In this
way, N3 puts pressure on N2 to suppress downstream congestion, for
every flow passing through the border interface, even though they
will all start and end in different places, and even though they
may all be allowed different responses to congestion. The figure
depicts this downward pressure on N2 by the solid downward arrow
at the egress of N2. Then N2 has an incentive either to police
the congestion response of its own ingress traffic (from N1) or to
emulate policing by applying penalties to N1 in turn on the basis
of congestion counted at their mutual boundary. In this recursive
way, the incentives for each flow to respond correctly to
congestion trace back with each flow precisely to each source,
despite the mechanism not recognising flows (see Section 6.2.2).
Inter-domain congestion charging diversity: Any two networks are
free to agree any of a range of penalty regimes between themselves
but they would only provide the right incentives if they were
within the following reasonable constraints. N2 should expect to
have to pay penalties to N3 where penalties monotonically increase
with the volume of congestion and negative penalties are not
allowed. For instance, they may agree an SLA with tiered
congestion thresholds, where higher penalties apply the higher the
threshold that is broken. But the most obvious (and useful) form
of penalty is where N3 levies a charge on N2 proportional to the
volume of downstream congestion N2 dumps into N3. In the
explanation that follows, we assume this specific variant of
volume charging between networks - charging proportionate to the
volume of congestion.
We must make clear that we are not advocating that everyone should
use this form of contract. We are well aware that the IETF tries
to avoid standardising technology that depends on a particular
business model. And we strongly share this desire to encourage
diversity. But our aim is merely to show that border policing can
at least work with this one model, then we can assume that