One document matched: draft-fairhurst-tcpm-newcwv-04.xml
<?xml version="1.0" encoding="US-ASCII"?>
<!DOCTYPE rfc SYSTEM "rfc2629.dtd">
<rfc category="std" docName="draft-fairhurst-tcpm-newcwv-04" ipr="trust200902"
obsoletes="2861" updates="5681">
<!-- Updates SDP -->
<?rfc toc="yes"?>
<!-- generate a table of contents -->
<?rfc symrefs="yes"?>
<!-- use anchors instead of numbers for references -->
<?rfc sortrefs="yes" ?>
<!-- alphabetize the references -->
<?rfc compact="yes" ?>
<!-- conserve vertical whitespace -->
<?rfc subcompact="no" ?>
<!-- but keep a blank line between list items -->
<front>
<title abbrev="new-CWV">Updating TCP to support Application-Limited
Traffic</title>
<author fullname="Godred Fairhurst" initials="G." surname="Fairhurst">
<organization>University of Aberdeen</organization>
<address>
<postal>
<street>School of Engineering</street>
<street>Fraser Noble Building</street>
<city>Aberdeen</city>
<region>Scotland</region>
<code>AB24 3UE</code>
<country>UK</country>
</postal>
<email>gorry@erg.abdn.ac.uk</email>
<uri>http://www.erg.abdn.ac.uk</uri>
</address>
</author>
<author fullname="Arjuna Sathiaseelan" initials="A."
surname="Sathiaseelan">
<organization>University of Aberdeen</organization>
<address>
<postal>
<street>School of Engineering</street>
<street>Fraser Noble Building</street>
<city>Aberdeen</city>
<region>Scotland</region>
<code>AB24 3UE</code>
<country>UK</country>
</postal>
<email>arjuna@erg.abdn.ac.uk</email>
<uri>http://www.erg.abdn.ac.uk</uri>
</address>
</author>
<date day="06" month="August" year="2012" />
<area>Transport</area>
<workgroup>TCPM Working Group</workgroup>
<keyword>CWV</keyword>
<keyword>TCP</keyword>
<abstract>
<t>This document addresses issues that arise when TCP is used to support
traffic that exhibits periods where the transmission rate is limited by
the application rather than the congestion window. It updates TCP to
allow a TCP sender to restart quickly following either an idle or
application-limited interval. The method is expected to benefit
application-limited TCP applications, while also providing an
appropriate response if congestion is experienced.</t>
<t>It also evaluates TCP Congestion Window Validation, CWV, an IETF
experimental specification defined in RFC 2861, and concludes that CWV
sought to address important issues, but failed to deliver a widely used
solution. This document therefore proposes an update to the status of
RFC 2861 by recommending it is moved from Experimental to Historic
status, and that it is replaced by the current specification.</t>
</abstract>
</front>
<middle>
<!-- text starts here -->
<section title="Introduction" toc="include">
<t>TCP is used to support a range of application behaviours. The TCP
congestion window (cwnd) controls the number of packets/bytes that a TCP
flow may have in the network at any time. A bulk application will always
have data available to transmit. The rate at which it sends is therefore
limited by the maximum permitted by the receiver and congestion windows.
In contrast, a rate-limited application will experience periods when the
sender is either idle or is unable to send at the maximum rate permitted
by the cwnd. This latter case is called application-limited. The focus
of this document is on the operation of TCP in such an idle or
application-limited case.</t>
<t>Standard TCP <xref target="RFC5681"></xref> requires the cwnd to be
reset to the restart window (RW) when an application becomes idle. <xref
target="RFC2861"></xref> noted that this TCP behaviour was not always
observed in current implementations. Recent experiments <xref
target="Bis08"></xref> confirm this to still be the case.</t>
<t>Standard TCP does not control growth of the cwnd when a TCP sender is
application-limited. An application-limited sender may therefore grow a
cwnd beyond that corresponding to the current transmit rate, resulting
in a value that does not reflect current information about the state of
the network path the flow is using. Use of such an invalid cwnd may
result in reduced application performance and/or could significantly
contribute to network congestion.</t>
<t><xref target="RFC2861"></xref> proposed a solution to these issues in
an experimental method known as Congestion Window Validation (CWV). CWV
was intended to help reduce cases where TCP accumulated an invalid cwnd.
The use and drawbacks of using CWV with an application are discussed in
Section 2.</t>
<t>Section 4 specifies an alternative to CWV that seeks to address the
same issues, but does this in a way that is expected to mitigate the
impact on an application that varies its transmission rate. The method
described applies to both an application-limited and an idle
condition.</t>
</section>
<section title="Reviewing experience with TCP-CWV">
<t>RFC 2861 described a simple modification to the TCP congestion
control algorithm that decayed the cwnd after the transition to a
“sufficiently-long” idle period. This used the slow-start
threshold (ssthresh) to save information about the previous value of the
congestion window. The approach relaxed the standard TCP behaviour <xref
target="RFC5681"></xref> for an idle session, intended to improve
application performance. CWV also modified the behaviour for an
application-limited session where a sender transmitted at a rate less
than allowed by cwnd.</t>
<t>RFC 2861 has been implemented in some mainstream operating systems as
the default behaviour <xref target="Bis08"></xref>. Analysis (e.g. <xref
target="Bis10"></xref>) has shown that a TCP sender using CWV is able to
use available capacity on a shared path after an idle period. This can
benefit some applications, especially over long delay paths, when
compared to slow-start restart specified by standard TCP. However, CWV
would only benefit an application if the idle period were less than
several Retransmission Time Out (RTO) intervals <xref
target="RFC6298"></xref>, since the behaviour would otherwise be the
same as for standard TCP, which resets the cwnd to the RW after this
period.</t>
<t>Experience with CWV suggests that although CWV benefits the network
in an application-limited scenario (reducing the probability of network
congestion), the behaviour can be too conservative for many common
rate-limited applications. This mechanism does not therefore offer the
desirable increase in application performance for rate-limited
applications and it is unclear whether applications actually use this
mechanism in the general Internet.</t>
<t>It is therefore concluded that CWV is often a poor solution for many
rate-limited applications. It has the correct motivation, but has the
wrong approach to solving this problem.</t>
</section>
<section title="Terminology" toc="include">
<t>The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT",
"SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this
document are to be interpreted as described in <xref
target="RFC2119"></xref>.</t>
<t>The document assumes familiarity with the terminology of TCP
congestion control <xref target="RFC5681"></xref>.</t>
</section>
<section title="An updated TCP response to idle and application-limited periods "
toc="include">
<t>This section proposes an update to the TCP congestion control
behaviour during an idle or application-limited period. The new method
permits a TCP sender to preserve the cwnd when an application becomes
idle for a period of time (set in this specification to 5 minutes, see
section 5). This period, where actual usage is less than allowed by
cwnd, is named the non-validated phase. The method allows an application
to resume transmission at a previous rate without incurring the delay of
slow-start. However, if the TCP sender experiences congestion using the
preserved cwnd, it is required to immediately reset the cwnd to an
appropriate value specified by the method. If a sender does not take
advantage of the preserved cwnd within five minutes, the value of cwnd
is reduced, ensuring the value then reflects the capacity that was
recently actually used.</t>
<t>The method requires that the TCP SACK option is enabled. This allows
the sender to select a cwnd following a congestion event that is based
on the measured path capacity, better reflecting the fair-share. A
similar approach was proposed by TCP Jump Start <xref
target="Liu07"></xref>, as a congestion response after more rapid
opening of a TCP connection.</t>
<t>It is expected that this update will satisfy the requirements of many
rate-limited applications and at the same time provide an appropriate
method for use in the Internet. It also reduces the incentive for an
application to send data simply to keep transport congestion state.
(This is sometimes known as "padding").</t>
<t>The new method does not differentiate between times when the sender
has become idle or application-limited. This is partly a response to
recognition that some applications wish to transmit at a rate-limited,
and that it can be hard to make a distinction between
application-limited and idle behaviour. This is expected to encourage
applications and TCP stacks to use standards-based congestion control
methods. It may also encourage the use of long-lived connections where
this offers benefit (such as persistent http).</t>
<t>The method is specified in following subsections.</t>
<section title="A method for preserving cwnd in idle and application-limited periods."
toc="include">
<t>The method described in this document updates <xref
target="RFC5681"></xref>. Use of the method REQUIRES a TCP sender and
the corresponding receiver to enable the TCP SACK option <xref
target="RFC3517"></xref>.</t>
<t><xref target="RFC5681"></xref> defines a variable FlightSize , that
indicates the amount of outstanding data in the network. This equal to
the value of Pipe calculated based on the pipe algorithm <xref
target="RFC3517"></xref>. In RFC5681 this value is used during loss
recovery, whereas in this method it is also used during normal data
transfer. A sender is not required to continuously track this value,
but SHOULD measure the volume of data in the network with a sampling
period of not less than one RTT period.</t>
</section>
<section title="The nonvalidated phase" toc="include">
<t>The updated method creates a new TCP sender phase that captures
whether the cwnd reflects a validated or non-validated value. The
phases are defined as:</t>
<t><list style="symbols">
<t>Validated phase: FlightSize >=(3/4)*cwnd. This is the normal
phase, where cwnd is expected to be an approximate indication of
the available capacity currently available along the network path,
and the standard methods are used to increase cwnd (currently
<xref target="RFC5681"></xref>).</t>
<t>Non-validated phase: FlightSize <(1/4)*cwnd. This is the
phase where the cwnd has a value based on a previous measurement
of the available capacity, and the usage of this capacity has not
been validated in the previous RTT. That is, when it is not known
whether the cwnd reflects the currently available capacity
available along the network path. The mechanisms to be used in
this phase seek to determine whether any resumed rate remains safe
for the Internet path, i.e., it quickly reduces the rate if the
flow is known to induce congestion. These mechanisms are specified
in section 4.3.</t>
</list>The values 1/4 and 3/4 were selected to reduce the effects of
variations in the measured FlightSize.</t>
</section>
<section title="TCP congestion control during the nonvalidated phase"
toc="include">
<t>A TCP sender that enters the non-validated phase MUST preserve the
cwnd (i.e., this neither grows nor reduces while the sender remains in
this phase). The phase is concluded after a fixed period of time (five
minutes, as explained in section or when the sender transmits using
the full cwnd (i.e. it is no longer application-limited).</t>
<t>The behaviour in the non-validated phase is specified as:</t>
<t><list style="symbols">
<t>If the sender consumes all the available space within the cwnd
(i.e., the remaining unused cwnd in bytes is less than one Sender
Maximum Segment Size, SMSS), then the sender MUST exit the
non-validated phase. The threshold value of cwnd required for the
sender to enter the non-validated phase is intentionally different
to that required to leave the phase. This introduces hysteresis to
avoid rapid oscillation between the phases. Note that a change
between phases does not significantly impact an
application-limited sender, but serves to determine its behaviour
if it substantially increases its transmission rate.</t>
<t>If the sender receives an indication of congestion while in the
non-validated phase (i.e. detects loss, or an Explicit Congestion
Notification, ECN, mark <xref target="RFC3168"></xref>), the
sender MUST exit the non-validated phase (reducing the cwnd as
defined in section 4.3.1).</t>
<t>If the Retransmission Time Out (RTO) expires while in the
non-validated phase, the sender MUST exit the non-validated phase.
It then resumes using the Standard TCP RTO mechanism <xref
target="RFC5681"></xref>. (The resulting reduction of cwnd
describe din section 4.3.2 is appropriate, since any accumulated
path history is considered unreliable).</t>
</list></t>
<section title="Response to congestion in the nonvalidated phase"
toc="include">
<t>Reception of congestion feedback while in the non-validated phase
is interpreted as an indication that it was inappropriate for the
sender to use the preserved cwnd. The sender is therefore required
to quickly reduce the rate to avoid further congestion. Since the
cwnd does not have a validated value, a new cwnd value must be
selected based on the utilised rate.</t>
<t>A sender that detects a packet-drop or receives an ECN marked
packet MUST calculate a safe cwnd, by setting it to the value
specified in Section 3.2 of <xref target="RFC5681"></xref>.</t>
<t>At the end of the recovery phase, the TCP sender MUST reset the
cwnd using the method below:</t>
<figure>
<artwork><![CDATA[ cwnd = ((FlightSize - R)/2).]]></artwork>
</figure>
<t></t>
<t>Where, R is the volume of data that was reported as
unacknowledged by the SACK information. This follows the method
proposed for Jump Start [<xref target="Liu07"></xref>.</t>
<t>The inclusion of the term R makes this adjustment is more
conservative than standard TCP. This is required, since the sender
may have sent more segments than Standard TCP would have done.</t>
<t>If the sender implements a method that allows it to identify the
number of ECN-marked segments within a windowthat were observed by
the receiver, the sender SHOULD use the method above, further
reducing R by the number of marked segments.</t>
</section>
<section title="Adjustment at the end of the nonvalidated phase "
toc="default">
<t>During the non-validated phase, the sender may produce bursts of
data of up to the cwnd in size. While this is no different to
standard TCP, it is desirable to control the maximum burst size,
e.g. by setting a burst size limit, using a pacing algorithm, or
some other method <xref target="Hug01"></xref>.</t>
<t>An application that remains in the non-validated phase for a
period greater than five minutes is required to adjust its
congestion control state. At the end of the non-validated phase, the
sender MUST update the ssthresh:</t>
<figure>
<artwork><![CDATA[ sthresh = max(ssthresh, 3*cwnd/4).
]]></artwork>
</figure>
<t></t>
<t>(This adjustment of ssthresh ensures that the sender records that
it has safely sustained the present rate. The change is beneficial
to application-limited flows that encounter occasional congestion,
and could otherwise suffer an unwanted additional delay in
recovering the transmission rate.)</t>
<t>The sender MUST then update cwnd:</t>
<figure>
<artwork><![CDATA[ cwnd = max(FlightSize*2, IW).
]]></artwork>
</figure>
<t>Where IW is the TCP inital window <xref
target="RFC5681"></xref>.</t>
<t>(This allows an application to continue to send at the currently
utilised rate, and not incur delay should it increase to twice the
utilised rate.)</t>
<t>After completing this adjustment, the sender MAY re-enter the
non-validated phase, if required (see section 4.2).</t>
</section>
</section>
</section>
<section title="Determining a safe period to preserve cwnd" toc="include">
<t>This section documents the rationale for selecting the maximum period
that cwnd may be preserved.</t>
<t>Preserving cwnd avoids undesirable side effects that would result if
the cwnd were to be preserved for an arbitrary long period, which was a
part of the problem that CWV originally attempted to address. The period
a sender may safely preserve the cwnd, is a function of the period that
a network path is expected to sustain the capacity reflected by cwnd.
There is no ideal choice for this time.</t>
<t>The period of five minutes was chosen as a compromise that was larger
than the idle intervals of common applications, but not sufficiently
larger than the period for which the capacity of an Internet path may
commonly be regarded as stable. The capacity of wired networks is
usually relatively stable for periods of several minutes and that load
stability increases with the capacity. This suggests that cwnd may be
preserved for at least a few minutes.</t>
<t>There are cases where the TCP throughput exhibits significant
variability over a time less than five minutes. Examples could include
wireless topologies, where TCP rate variations may fluctuate on the
order of a few seconds as a consequence of medium access protocol
instabilities. Mobility changes may also impact TCP performance over
short time scales. Senders that observe such rapid changes in the path
characteristic may also experience increased congestion with the new
method, however such variation would likely also impact TCP’s
behaviour when supporting interactive and bulk applications.</t>
<t>Routing algorithms may modify the network path, disrupting the RTT
measurement and changing the capacity available to a TCP connection,
however such changes do not often occur within a time frame of a few
minutes.</t>
<t>The value of five minutes is therefore expected to be sufficient for
most current applications. Simulation studies also suggest that for many
practical applications, the performance using this value will not be
significantly different to that observed using a non-standard method
that does not reset the cwnd after idle.</t>
<t>Finally, other TCP sender mechanisms have used a 5 minute timer, and
there could be simplifications in some implementations by reusing the
same interval. TCP defines a default user timeout of 5 minutes <xref
target="RFC0793"></xref> i.e. how long transmitted data may remain
unacknowledged before a connection is forcefully closed.</t>
</section>
<section title="Security Considerations" toc="include">
<t>General security considerations concerning TCP congestion control are
discussed in <xref target="RFC5681"></xref>. This document describes an
algorithm that updates one aspect of the congestion control procedures,
and so the considerations described in RFC 5681 also apply to this
algorithm.</t>
</section>
<section title="IANA Considerations" toc="include">
<t>There are no IANA considerations.</t>
</section>
<section title="Acknowledgments">
<t>The authors acknowledge the contributions of Dr I Biswas and Dr R
Secchi in supporting the evaluation of CWV and for their help in
developing the mechanisms proposed in this draft. We also acknowledge
comments received from the Internet Congestion Control Research Group,
in particular Yuchung Cheng, Mirja Kuehlewind, and Joe Touch.</t>
<t><!----></t>
</section>
<section title="Other related work - Author Notes">
<t>There are several issues to be discussed more widely:</t>
<t><list style="empty">
<t>• Should the method explicitly state a procedure for
limiting burstiness or pacing?</t>
<t><list style="empty">
<t>This is often regarded as good practice, but isn't a formal
part of TCP. draft-hughes-restart-00.txt provides some
discussion of this topic.</t>
</list></t>
<t>• There are potential interaction with the proposal to raise
the TCP initial Window to ten segments, do these cases need to be
elaborated?</t>
<t><list style="empty">
<t>This relates to draft-ietf-tcpm-initcwnd.</t>
<t>The two methods have different functions and different
response to loss/congestion.</t>
<t>IW=10 proposes an experimental update to TCP that would allow
faster opening of the cwnd, and also a large (same size) restart
window. This approach is based on the assumption that many
forward paths can sustain bursts of up to ten segments without
(appreciable) loss. Such a significant increase in cwnd must be
matched with an equally large reduction of cwnd if
loss/congestion is detected, and such a congestion indication is
likely to require future use of IW=10 to be disabled for this
path for some time. This guards against the unwanted behaviour
of a series of short flows continuously flooding a network path
without network congestion feedback.</t>
<t>In contrast, new-CWV proposes a standards-track update with a
rationale that relies on recent previous path history to select
an appropriate cwnd after restart.</t>
<t>The behaviour differs in three ways:</t>
<t>1) For applications that send little initially, new-cwv may
constrain more than IW=10, but would not require the connection
to reset any path information when a restart incurred loss. In
contrast, new-cwv would allow the TCP connection to preserve the
cached cwnd, any loss, would impact cwnd, but not impact other
flows.</t>
<t>2) For applications that utilise more capacity than provided
by a cwnd=10, this method would permit a larger restart window
compared to a restart using IW=10. This is justified by the
recent path history.</t>
<t>3) new-CWV is attended to also be used for
application-limited use, where the application sends, but does
not seek to fully utilise the cwnd. In this case, new-cwv
constrains the cwnd to that justified by the recent path
history. The performance trade-offs are hence different, and it
would be possible to enable new-cwv when also using IW=10, and
yield the benefits of this.</t>
</list></t>
<t>• There is potential overlap with the Laminar proposal
(draft-mathis-tcpm-tcp-laminar)</t>
<t><list style="empty">
<t>The current draft was intended as a standards-track update to
TCP, rather than a new transport variant. At least, it would be
good to understand how the two interact and whether there is a
possibility of a single method.</t>
</list></t>
</list></t>
</section>
</middle>
<back>
<references title="Normative References">
<?rfc sortrefs="yes"?>
<?rfc include="http://xml.resource.org/public/rfc/bibxml/reference.RFC.0793.xml"?>
<?rfc include="http://xml.resource.org/public/rfc/bibxml/reference.RFC.2119.xml"?>
<?rfc include="http://xml.resource.org/public/rfc/bibxml/reference.RFC.2861.xml"?>
<?rfc include="http://xml.resource.org/public/rfc/bibxml/reference.RFC.6298.xml"?>
<?rfc include="http://xml.resource.org/public/rfc/bibxml/reference.RFC.3168.xml"?>
<?rfc include="http://xml.resource.org/public/rfc/bibxml/reference.RFC.3517.xml"?>
<?rfc include="http://xml.resource.org/public/rfc/bibxml/reference.RFC.5681.xml"?>
</references>
<references title="Informative References">
<reference anchor="Bis08">
<front>
<title>A Practical Evaluation of Congestion Window Validation
Behaviour, 9th Annual Postgraduate Symposium in the Convergence of
Telecommunications, Networking and Broadcasting (PGNet), Liverpool,
UK</title>
<author fullname="I." surname="Biswas">
<organization></organization>
</author>
<author fullname="G." surname="Fairhurst">
<organization></organization>
</author>
<date day="01" month="June" year="2008" />
</front>
</reference>
<reference anchor="Liu07">
<front>
<title>Congestion Control without a Startup Phase, 5th International
Workshop on Protocols for Fast Long-Distance Networks (PFLDnet), Los
Angeles, California, USA</title>
<author fullname="D" surname="Liu">
<organization></organization>
</author>
<author fullname="M." surname="Allman">
<organization></organization>
</author>
<author fullname="S" surname="Jiny">
<organization></organization>
</author>
<author fullname="L." surname="Wang">
<organization></organization>
</author>
<date day="01" month="February" year="2007" />
</front>
</reference>
<reference anchor="Bis10">
<front>
<title>Analysing TCP for Bursty Traffic, Int'l J. of Communications,
Network and System Sciences, 7(3)</title>
<author fullname="I." surname="Biswas">
<organization></organization>
</author>
<author fullname="A." surname="Sathiaseelan">
<organization></organization>
</author>
<author fullname="R." surname="Secchi">
<organization></organization>
</author>
<author fullname="G." surname="Fairhurst">
<organization></organization>
</author>
<date day="01" month="June" year="2010" />
</front>
</reference>
<reference anchor="Hug01">
<front>
<title>Issues in TCP Slow-Start Restart After Idle
(Work-in-Progress)</title>
<author fullname="A.S." surname="Hughes">
<organization></organization>
</author>
<author fullname="J." surname="Touch">
<organization></organization>
</author>
<author fullname="J." surname="Heidemann">
<organization></organization>
</author>
<date day="01" month="December" year="2001" />
</front>
</reference>
</references>
</back>
</rfc>
| PAFTECH AB 2003-2026 | 2026-04-23 16:29:56 |