One document matched: draft-ietf-rmcat-sbd-01.xml
<?xml version="1.0" encoding="US-ASCII"?>
<!-- This template is for creating an Internet Draft using xml2rfc,
which is available here: http://xml.resource.org. -->
<!DOCTYPE rfc SYSTEM "rfc2629.dtd" [
<!-- One method to get references from the online citation libraries.
There has to be one entity for each item to be referenced.
An alternate method (rfc include) is described in the references. -->
<!ENTITY RFC2119 SYSTEM "http://xml2rfc.ietf.org/public/rfc/bibxml/reference.RFC.2119.xml">
<!ENTITY RFC3550 SYSTEM "http://xml2rfc.ietf.org/public/rfc/bibxml/reference.RFC.3550.xml">
<!ENTITY RFC4585 SYSTEM "http://xml2rfc.ietf.org/public/rfc/bibxml/reference.RFC.4585.xml">
<!ENTITY RFC5124 SYSTEM "http://xml2rfc.ietf.org/public/rfc/bibxml/reference.RFC.5124.xml">
<!ENTITY RFC5481 SYSTEM "http://xml2rfc.ietf.org/public/rfc/bibxml/reference.RFC.5481.xml">
<!ENTITY RFC6817 SYSTEM "http://xml2rfc.ietf.org/public/rfc/bibxml/reference.RFC.6817.xml">
<!ENTITY RFC1323 SYSTEM "http://xml2rfc.ietf.org/public/rfc/bibxml/reference.RFC.1323.xml">
<!ENTITY I-D.welzl-rmcat-coupled-cc SYSTEM
"http://xml2rfc.ietf.org/public/rfc/bibxml3/reference.I-D.draft-welzl-rmcat-coupled-cc-04.xml">
]>
<?xml-stylesheet type='text/xsl' href='rfc2629.xslt' ?>
<!-- used by XSLT processors -->
<!-- For a complete list and description of processing instructions (PIs),
please see http://xml.resource.org/authoring/README.html. -->
<!-- Below are generally applicable Processing Instructions (PIs) that most I-Ds might want to use.
(Here they are set differently than their defaults in xml2rfc v1.32) -->
<?rfc strict="yes" ?>
<!-- give errors regarding ID-nits and DTD validation -->
<!-- control the table of contents (ToC) -->
<?rfc toc="yes"?>
<!-- generate a ToC -->
<?rfc tocdepth="4"?>
<!-- the number of levels of subsections in ToC. default: 3 -->
<!-- control references -->
<?rfc symrefs="yes"?>
<!-- use symbolic references tags, i.e, [RFC2119] instead of [1] -->
<?rfc sortrefs="yes" ?>
<!-- sort the reference entries alphabetically -->
<!-- control vertical white space
(using these PIs as follows is recommended by the RFC Editor) -->
<?rfc compact="yes" ?>
<!-- do not start each main section on a new page -->
<?rfc subcompact="no" ?>
<!-- keep one blank line between list items -->
<!-- end of list of popular I-D processing instructions -->
<rfc category="exp" docName="draft-ietf-rmcat-sbd-01" ipr="trust200902">
<!-- category values: std, bcp, info, exp, and historic
ipr values: full3667, noModification3667, noDerivatives3667
you can add the attributes updates="NNNN" and obsoletes="NNNN"
they will automatically be output with "(if approved)" -->
<!-- ***** FRONT MATTER ***** -->
<front>
<!-- The abbreviated title is used in the page header - it is only necessary if the
full title is longer than 39 characters -->
<title abbrev="SBD for CCC with RTP Media">
Shared Bottleneck Detection for Coupled Congestion Control for
RTP Media.
</title>
<!-- add 'role="editor"' below for the editors if appropriate -->
<!-- Another author who claims to be an editor -->
<author fullname="David Hayes" initials="D.A." role="editor"
surname="Hayes">
<organization>University of Oslo</organization>
<address>
<postal>
<street>PO Box 1080 Blindern</street>
<city>Oslo</city>
<region></region>
<code>N-0316</code>
<country>Norway</country>
</postal>
<phone>+47 2284 5566</phone>
<email>davihay@ifi.uio.no</email>
</address>
</author>
<author fullname="Simone Ferlin" initials="S."
surname="Ferlin">
<organization>Simula Research Laboratory</organization>
<address>
<postal>
<street>P.O.Box 134</street>
<city>Lysaker</city>
<region></region>
<code>1325</code>
<country>Norway</country>
</postal>
<phone>+47 4072 0702</phone>
<email>ferlin@simula.no</email>
</address>
</author>
<author fullname="Michael Welzl" initials="M."
surname="Welzl">
<organization>University of Oslo</organization>
<address>
<postal>
<street>PO Box 1080 Blindern</street>
<city>Oslo</city>
<region></region>
<code>N-0316</code>
<country>Norway</country>
</postal>
<phone>+47 2285 2420</phone>
<email>michawe@ifi.uio.no</email>
</address>
</author>
<date month="July" year="2015" />
<!-- If the month and year are both specified and are the current ones, xml2rfc will fill
in the current day for you. If only the current year is specified, xml2rfc will fill
in the current day and month for you. If the year is not the current one, it is
necessary to specify at least a month (xml2rfc assumes day="1" if not specified for the
purpose of calculating the expiry date). With drafts it is normally sufficient to
specify just the year. -->
<!-- Meta-data Declarations -->
<area>General</area>
<workgroup>RTP Media Congestion Avoidance Techniques</workgroup>
<!-- WG name at the upperleft corner of the doc,
IETF is fine for individual submissions.
If this element is not present, the default is "Network Working Group",
which is used by the RFC Editor as a nod to the history of the IETF. -->
<keyword>SBD</keyword>
<!-- Keywords will be incorporated into HTML output
files in a meta tag but they have no effect on text or nroff
output. If you submit your draft to the RFC Editor, the
keywords will be used for the search engine. -->
<abstract>
<t>This document describes a mechanism to detect whether
end-to-end data flows
share a common bottleneck. It relies on summary statistics that are calculated by
a data receiver based on continuous measurements and regularly fed to a grouping algorithm that
runs wherever the knowledge is needed. This mechanism complements the coupled congestion
control mechanism in draft-welzl-rmcat-coupled-cc.</t>
</abstract>
</front>
<middle>
<section title="Introduction">
<t>In the Internet, it is not normally known if flows (e.g., TCP connections or UDP data streams)
traverse the same bottlenecks. Even flows that have the same sender and receiver may take
different paths and share a bottleneck or not. Flows that share a bottleneck link usually
compete with one another for their share of the capacity. This competition has the potential
to increase packet loss and delays. This is especially relevant for interactive applications
that communicate simultaneously with multiple peers (such as multi-party video). For RTP
media applications such as RTCWEB, <xref target="I-D.welzl-rmcat-coupled-cc"></xref> describes
a scheme that combines
the congestion controllers of flows in order to honor their priorities and avoid unnecessary
packet loss as well as delay.
This mechanism relies on some form of Shared Bottleneck Detection (SBD); here, a
measurement-based SBD approach is described.</t>
<section title="The signals">
<t>The current Internet is unable to explicitly inform
endpoints as to which flows share bottlenecks, so endpoints
need to infer this from whatever information is available to
them. The mechanism described here currently utilises packet
loss and packet delay, but is not restricted to these.</t>
<section title="Packet Loss">
<t>Packet loss is often a relatively rare
signal. Therefore, on its own it is of limited use for
SBD, however, it is a valuable supplementary measure when
it is more prevalent.</t>
</section>
<section title="Packet Delay">
<t>End-to-end delay measurements include noise from every
device along the path in addition to the delay
perturbation at the bottleneck device. The noise is
often significantly increased if the round-trip time is used. The
cleanest signal is obtained by using One-Way-Delay
(OWD).</t>
<t>Measuring absolute OWD is difficult since it requires
both the sender and receiver clocks to be
synchronised. However, since the statistics being
collected are relative to the mean OWD, a relative OWD
measurement is sufficient. Clock skew is not usually
significant over the time intervals used by this SBD
mechanism (see <xref target="RFC6817"/> A.2 for a
discussion on clock skew and OWD measurements). However,
in circumstances where it is significant, <xref
target="clockskew"/> outlines a way of adjusting the
calculations to cater for it.</t>
<t>Each packet arriving at the bottleneck buffer may
experience very different queue lengths, and therefore different
waiting times. A single OWD sample does not, therefore,
characterize the path well. However,
multiple OWD measurements do reflect the distribution of
delays experienced at the bottleneck.</t>
</section>
<section title="Path Lag">
<t>Flows that share a common bottleneck may traverse
different paths, and these paths will often have different
base delays. This makes it difficult to correlate changes
in delay or loss. This technique uses the long term shape
of the delay distribution as a base for comparison to
counter this.</t>
</section>
</section>
</section>
<section anchor="Definitions" title="Definitions">
<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">RFC 2119</xref>.</t>
<t>Acronyms used in this document:
<list hangIndent="10" style="hanging">
<t hangText=" OWD --"> One Way Delay</t>
<t hangText=" PDV --"> Packet Delay Variation</t>
<t hangText=" MAD --"> Mean Absolute Deviation</t>
<t hangText=" RTT --"> Round Trip Time</t>
<t hangText=" SBD --"> Shared Bottleneck Detection</t>
</list></t>
<t>Conventions used in this document:
<list hangIndent="18" style="hanging">
<t hangText=" T --"> the base time interval over which measurements
are made.</t>
<t hangText=" N --"> the number of base time, T, intervals
used in some calculations.</t>
<t hangText=" sum_T(...) --"> summation of all the
measurements of the variable in parentheses taken over the
interval T</t>
<t hangText=" sum(...) --"> summation of terms of the variable in parentheses</t>
<t hangText=" sum_N(...) --"> summation of N terms of the variable in parentheses</t>
<t hangText=" sum_NT(...) --"> summation of all
measurements taken over the interval N*T</t>
<t hangText=" E_T(...) --"> the expectation or mean of the
measurements of the variable in parentheses over T</t>
<t hangText=" E_N(...) --"> the expectation or mean of the last N values of
the variable in parentheses</t>
<t hangText=" E_M(...) --"> the expectation or mean of the last M values of
the variable in parentheses, where M <= N.</t>
<t hangText=" max_T(...) --"> the maximum recorded measurement
of the variable in parentheses taken over the interval T</t>
<t hangText=" min_T(...) --"> the minimum recorded measurement
of the variable in parentheses taken over the interval T</t>
<t hangText=" num_T(...) --"> the count of measurements of the
variable in parentheses taken in the interval T</t>
<t hangText=" num_VM(...) --"> the count of valid values of the
variable in parentheses given M records</t>
<t hangText=" PC --"> a boolean variable indicating the
particular flow was identified as experiencing congestion in
the previous interval T (i.e. Previously Congested)</t>
<t hangText=" skew_est --"> a measure of skewness in a OWD distribution.</t>
<t hangText=" var_est --"> a measure of variability in
OWD measurements.</t>
<t hangText=" freq_est --"> a measure of low frequency
oscillation in the OWD measurements.</t>
<t hangText=" p_l, p_f, p_pdv, p_mad, c_s, c_h, p_s, p_d, p_v --"> various
thresholds used in the mechanism</t>
<t hangText=" M and F --"> number of values related to N</t>
</list></t>
<section anchor="parameters" title="Parameters and their Effect">
<t><list hangIndent="8" style="hanging">
<t hangText="T"> T should be long enough so that there are
enough packets received during T for a useful estimate of
short term mean OWD and variation statistics. Making T too
large can limit the efficacy of PDV and freq_est. It will
also increase the response time of the mechanism. Making T
too small will make the metrics noisier.</t>
<t hangText="N & M"> N should be large enough provide a
stable estimate of oscillations in OWD and average
PDV. Usually M=N, though having M<N may be beneficial in
certain circumstances. M*T needs to be long enough provide
stable estimates of skewness and MAD (if used).</t>
<t hangText="F"> F determines the number of intervals
over which statistics are considered to be equally
weighted. When F=M recent and older measurements are
considered equal. Making F<M can increase the
responsiveness of the SBD mechanism. If F is too small,
statistics will be too noisy.</t>
<t hangText="c_s"> c_s is the threshold in skew_est used for
determining whether a flow is experiencing congestion or
not. It should be slightly negative so that a very lightly
loaded path does not give a false indication. Setting c_s
more negative makes the SBD mechanism less sensitive to
transient and light congestion episodes.</t>
<t hangText="c_s"> c_h adds hysteresis to the congestion
determination. It should be large enough to avoid constant
switching in the determination, but low enough to ensure
that grouping is not attempted when there is no congestion
and the delay and loss signals cannot be relied upon.</t>
<t hangText="p_v"> p_v determines the sensitivity of freq_est
to noise. Making it smaller will yield higher but noisier
values for freq_est. Making it too large will render it
ineffective for determining groups.</t>
<t hangText="p_*"> Flows are separated when the
skew_est|var_est|freq_est measure is greater than
p_s|p_f|p_d|(p_pdv|p_mad). Adjusting these is a compromise
between false grouping of flows that do not share a
bottleneck and false splitting of flows that do. Making them
larger can help if the measures are very noisy, but reducing
the noise in the statistical measures by adjusting T and N|M
may be a better solution.</t>
</list></t>
</section>
<section anchor="recommended-parameters" title="Recommended Parameter Values">
<t>Reference <xref target="Hayes-LCN14"/> uses T=350ms,
N=50, p_l = 0.1. The other parameters have been tightened to
reflect minor enhancements to the algorithm outlined in
<xref target="removingnoise"/>:
c_s = -0.01, p_f = p_s = p_d = 0.1, p_pdv = 0.2,
p_v = 0.2 (or p_mad=0.1, p_v=0.7). M=50, F=25, and c_h = 0.3 are additional
parameters defined in the document.
These are values that seem to work well over a wide range of practical
Internet conditions.</t>
</section>
</section>
<section anchor="Mechanism" title="Mechanism">
<t>The mechanism described in this document is based on the
observation that the distribution of delay measurements of
packets that traverse a
common bottleneck have similar shape characteristics. These
shape characteristics are described using 3 key summary
statistics:
<list style="hanging">
<t>variability (estimate var_est, see <xref target="sbd_varest"/>)</t>
<t>skewness (estimate skew_est, see <xref target="sbd_skewest"/>)</t>
<t>oscillation (estimate freq_est, see <xref target="sbd_freqest"/>)</t>
</list>
with packet loss (estimate pkt_loss, see <xref
target="sbd_pktloss"/>) used as a supplementary statistic.</t>
<t>Summary statistics help to address both the noise and the
path lag problems by describing the general shape over a
relatively long period of time. This is sufficient for their
application in coupled congestion control for RTP Media. They
can be signalled from a receiver, which measures the OWD and calculates
the summary statistics, to a sender, which is the entity that is transmitting
the media stream. An RTP Media device may
be both a sender and a receiver. SBD can be performed at either
a sender or a receiver or both.</t>
<figure align="center" anchor="sbd-topo">
<!-- <preamble>Preamble text - can be omitted or empty.</preamble> -->
<artwork align="left"><![CDATA[
+----+
| H2 |
+----+
|
| L2
|
+----+ L1 | L3 +----+
| H1 |------|------| H3 |
+----+ +----+
]]></artwork>
<postamble>A network with 3 hosts (H1, H2, H3) and 3 links (L1, L2, L3).</postamble>
</figure>
<t>In <xref target="sbd-topo" />, there are two possible cases
for shared bottleneck detection: a sender-based and a
receiver-based case.
<list style="numbers">
<t>Sender-based: consider a situation where host H1 sends media
streams to hosts H2 and H3, and L1 is a shared bottleneck.
H2 and H3 measure the OWD and calculate summary statistics,
which they send to H1 every T. H1, having this knowledge,
can determine the shared bottleneck and accordingly control
the send rates.</t>
<t>Receiver-based: consider that H2 is also sending media to
H3, and L3 is a shared bottleneck. If H3 sends summary
statistics to H1 and H2, neither H1 nor H2 alone obtain
enough knowledge to detect this shared bottleneck; H3 can
however determine it by combining the summary statistics
related to H1 and H2, respectively. This case is applicable
when send rates are controlled by the receiver; then, the
signal from H3 to the senders contains the sending rate.</t>
</list></t>
<t>A discussion of the required signalling for the receiver-based
case is beyond the scope of this document. For the sender-based
case, the messages and their data format will be defined here in
future versions of this document. We envision that an
initialization message from the sender to the receiver could
specify which key metrics are requested out of a possibly
extensible set (pkt_loss, var_est, skew_est, freq_est).
The grouping algorithm described in this
document requires all four of these metrics, and receivers MUST be
able to provide them,
but future algorithms may be able to exploit other metrics
(e.g. metrics based on explicit network signals).
Moreover, the initialization message could
specify T, N, and the necessary resolution and precision (number of bits
per field).
</t>
<section anchor="sbd-metrics" title="Key metrics and their calculation">
<t>Measurements are calculated over a base interval,
T. T should be long enough to provide enough samples
for a good estimate of skewness, but short enough so that
a measure of the oscillation can be made from N of these
estimates. Reference <xref target="Hayes-LCN14"/>
uses T = 350ms and N=M=50,
which are values that seem to work well over a wide range
of practical Internet conditions.
</t>
<section title="Mean delay">
<t>The mean delay is not a useful signal for comparisons
between flows since flows may traverse quite different paths
and clocks will not necessarily be synchronized. However, it
is a base measure for the 3 summary statistics. The mean
delay, E_T(OWD), is the average one way delay measured over
T.</t>
<t>To facilitate the other calculations, the last N
E_T(OWD) values will need to be stored in a cyclic buffer
along with the moving
average of E_T(OWD):
<list style="hanging">
<t>mean_delay = E_M(E_T(OWD)) = sum_M(E_T(OWD)) / M</t>
</list>
where M ≤ N. Generally M=N: setting M to be less than N
allows the mechanism to be more responsive to changes, but
potentially at the expense of a higher error rate (see <xref
target="improvingresponse"/> for a discussion on improving
the responsiveness of the mechanism.) </t>
</section>
<section anchor="sbd_skewest" title="Skewness Estimate">
<t>Skewness is difficult to calculate efficiently and
accurately. Ideally it should be calculated over the entire
period (M * T) from the mean OWD over that period. However this
would require storing every delay measurement over the
period. Instead, an estimate is made over M * T based on a
calculation every T using the previous T's calculation of
mean_delay.<vspace blankLines="100" /></t>
<t>The skewness is estimated using two counters, counting
the number of one way delay samples (OWD) above and below the
mean:
<list style="hanging">
<t>skew_base_T = sum_T(OWD <
mean_delay) - sum_T(OWD > mean_delay)</t>
<t>where
<list style="hanging">
<t>if (OWD < mean_delay) 1 else 0</t>
<t>if (OWD > mean_delay) 1 else 0</t>
</list></t>
<t>and mean_delay does not include the mean of the
current T interval.</t>
<t>skew_est = sum_MT(skew_base_T)/num_MT(OWD)
<list style="hanging">
<t> where skew_est is a number between -1 and 1</t>
</list></t>
</list></t>
<t>Note: Care must be taken when implementing the
comparisons to ensure that rounding does not bias
skew_est. It is important that the mean is calculated
with a higher precision than the samples.</t>
</section>
<section anchor="sbd_varest" title="Variability Estimate">
<t>Packet Delay Variation (PDV) (<xref target="RFC5481"/>
and <xref target="ITU-Y1540"/>)
is used as an estimator of
the variability of the delay signal. We define PDV
as follows:
<list style="hanging">
<t>PDV = PDV_max = max_T(OWD) - E_T(OWD)</t>
<t>var_est = E_M(PDV) = sum_M(PDV) / M</t>
</list>
This modifies PDV as outlined in <xref target="RFC5481"/>
to provide a summary statistic version that best
aids the grouping decisions of the algorithm (see <xref
target="Hayes-LCN14"/> section IVB).</t>
<t>Generally the maximum is sampled well during congestion,
though it is more sensitive to path and operating system
noise. The use of PDV = PDV_min = E_T(OWD) - min_T(OWD)
would be less sensitive to this noise, but is not well
sampled during congestion at the bottleneck and therefore
not recommended.<vspace blankLines="100" /></t>
</section>
<section anchor="sbd_freqest" title="Oscillation Estimate">
<t>An estimate of the low frequency oscillation of the delay
signal is calculated by counting and normalising the significant mean,
E_T(OWD), crossings of mean_delay:
<list style="hanging">
<t>freq_est = number_of_crossings / N
<list style="hanging">
<t> where we define a significant mean
crossing as a crossing that extends p_v * var_est from
mean_delay. In our experiments we have found that p_v =
0.2 is a good value.</t>
</list></t>
</list>
Freq_est is a number between 0 and 1. Freq_est
can be approximated incrementally as follows:
<list style="hanging">
<t> With each new calculation of E_T(OWD) a decision is
made as to whether this value of E_T(OWD) significantly
crosses the current long term mean, mean_delay, with respect to
the previous significant mean crossing.</t>
<t>A cyclic buffer, last_N_crossings, records a 1 if there is a significant
mean crossing, otherwise a 0.</t>
<t>The counter, number_of_crossings, is incremented when there
is a significant mean crossing and decremented when a
non-zero value is removed from the last_N_crossings.</t>
</list>
This approximation of freq_est was not used in <xref
target="Hayes-LCN14"/>, which calculated freq_est every T
using the current E_N(E_T(OWD)). Our tests show that
this approximation of freq_est yields results that are almost
identical to when the full calculation is performed every
T.</t>
</section>
<section anchor="sbd_pktloss" title="Packet loss">
<t>The proportion of packets lost is used as a supplementary
measure:
<list style="hanging">
<t>pkt_loss = sum_NT(lost packets) / sum_NT(total
packets)</t>
</list>
Note: When pkt_loss is small it is very variable, however,
when pkt_loss is high it becomes a stable measure for
making grouping decisions..<vspace blankLines="100" /></t>
</section>
</section>
<section title="Flow Grouping">
<section anchor="flowgrouping" title="Flow Grouping Algorithm">
<t>The following grouping algorithm is RECOMMENDED for SBD
in the RMCAT context and is sufficient and efficient for small to
moderate numbers of flows. For very large numbers of flows
(e.g. hundreds), a more complex clustering algorithm may be
substituted.</t>
<t>Since no single metric is precise enough to group flows
(due to noise), the algorithm uses multiple metrics. Each
metric offers a different "view" of the bottleneck link
characteristics, and used together they enable a more precise
grouping of flows than would otherwise be possible.</t>
<t>Flows determined to be experiencing congestion are
successively divided into groups based on freq_est, var_est, and
skew_est.</t>
<t>The first step is to determine which flows are
experiencing congestion. This is important, since if a flow
is not experiencing congestion its delay based metrics will
not describe the bottleneck, but the "noise" from the rest
of the path. Skewness, with proportion of packets loss as a
supplementary measure, is used to do this:
<?rfc needLines="8" ?>
<list counter="grouping" style="format %d.">
<t>Grouping will be performed on flows where:
<list style="hanging">
<t>skew_est < c_s
<list style="hanging">
<t>|| ( skew_est < c_h &&
PC )</t>
<t>|| pkt_loss > p_l</t>
</list></t>
</list></t>
</list></t>
<t>The parameter c_s controls how sensitive the mechanism is
in detecting congestion. C_s = 0.0 was used in <xref
target="Hayes-LCN14"/>. A value of c_s = 0.05 is a little
more sensitive, and c_s = -0.05 is a little less
sensitive. C_h controls the hysteresis on flows that were
grouped as experiencing congestion last time. </t>
<t>These flows, flows experiencing congestion, are then
progressively divided into groups based on the freq_est, PDV,
and skew_est summary statistics. The process proceeds
according to the following steps:
<list counter="grouping" style="format %d." >
<t>Group flows whose difference in sorted freq_est is less than a
threshold:
<list style="hanging">
<t> diff(freq_est) < p_f</t>
</list></t>
<t>Group flows whose difference in sorted E_N(PDV)
(highest to lowest) is less than a threshold:
<list style="hanging">
<t> diff(var_est) < (p_pdv * var_est) </t>
</list>The threshold, (p_pdv * var_est), is with respect
to the highest value in the difference.</t>
<t>Group flows whose difference in sorted skew_est or
pkt_loss is less than a threshold:
<list style="hanging">
<t> if pkt_loss < p_l
<list style="hanging">
<t>diff(skew_est) < p_s </t>
</list></t>
<t>otherwise
<list style="hanging">
<t>diff(pkt_loss) < (p_d * pkt_loss) </t>
</list>The threshold, (p_d * pkt_loss), is with respect
to the highest value in the difference.</t>
</list></t>
</list></t>
<t>This procedure involves sorting estimates from highest to
lowest. It is simple to implement, and efficient for small
numbers of flows (up to 10-20).</t>
</section>
<section title="Using the flow group signal">
<t>A grouping decisions is made every T from the second T,
though they will not attain their full design accuracy until
after the N'th T interval.</t>
<t>Network conditions, and even the congestion controllers,
can cause bottlenecks to fluctuate. A coupled congestion
controller MAY decide only to couple groups that remain
stable, say grouped together 90% of the time, depending on
its objectives. Recommendations concerning this are beyond
the scope of this draft and will be specific to the coupled
congestion controllers objectives.</t>
</section>
</section>
<section anchor="removingnoise" title="Removing Noise from the Estimates">
<t>The following describe small changes to the calculation of
the key metrics that help remove noise from them. Currently these
"tweaks" are described separately to keep the main description
succinct. In future revisions of the draft these enhancements
may replace the original key metric calculations.</t>
<section anchor="minmaxnoise" title="PDV noise">
<t>Usually during congestion the max_T(OWD) is quite well
sampled as the delay distribution is skewed toward the
maximum. However max_T(OWD) is subject to delay noise from other
queues along the path as well as the host operating
system. Min_T(OWD) is less prone to noise along the path and
from the host operating system, but is not well sampled
during congestion (i.e. when there is a
bottleneck). Flows with very different packet send rates
exacerbate the problem.</t>
<t>An alternative delay variation measure that is less
sensitive to extreme values and different send rates is Mean
Absolute Deviation (MAD). It can be implemented in an online
manner as follows:
<list style="hanging">
<t> var_base_T = sum_T(|OWD - E_T(OWD)|)
<list style="hanging"><t>where
<list style="hanging">
<t>|x| is the absolute value of x</t>
<t>E_T(OWD) is the mean OWD calculated in the previous
T</t>
</list></t>
</list></t>
<t>var_est = MAD_MT = sum_MT(var_base_T)/num_MT(OWD) </t>
</list></t>
<t>For calculation of freq_est p_v=0.7 (MAD is a smaller
number than PDV)</t>
<t>For the grouping threshold p_mad=0.1 instead of p_pdv
(MAD is less noisy so the test can be tighter)</t>
<t>Note that the method for improving responsiveness of
MAD_MT is the same as that described in <xref
target="skewrespimp"/> for skew_est.</t>
</section>
<section anchor="oscillationnoise" title="Oscillation noise">
<t>When a path has no congestion, var_est will be very small and
the recorded significant mean crossings will be the result
of path noise. Thus up to N-1 meaningless mean crossings can
be a source of error at the point a link becomes a
bottleneck and flows traversing it begin to be grouped.</t>
<t>To remove this source of noise from freq_est:
<list counter="oscn" style="format %d.">
<t>Set the current PDV to PDV = NaN (a value representing
an invalid record, i.e. Not a Number) for flows that are
deemed to not be experiencing congestion by the first
skew_est based grouping test (see <xref
target="flowgrouping"/>).</t>
<t> Then var_est = sum_M(PDV != NaN) / num_VM(PDV)</t>
<t> For freq_est, only record a significant mean crossing
if flow is experiencing congestion.</t>
</list>
These three changes will remove the non-congestion noise
from freq_est. A similar adjustment can be made for MAD
based var_est.</t>
</section>
<section anchor="clockskew" title="Clock skew">
<t>Generally sender and receiver clock skew will be too
small to cause significant errors in the
estimators. Skew_est is most sensitive to this type of
noise. In circumstances where clock skew is high, making
M < N can reduce this error.</t>
<t>A better method is to estimate the effect the clock
skew is having on the summary statistics, and then adjust
statistics accordingly. A simple online method of doing this
based on min_T(OWD) will be described here in a subsequent
version of the draft.</t>
</section>
<!--
<section title="Bias in the variability measure">
<t>Var_est can also be biased when measuring varying rate
flows. This bias can be corrected as follows.
<list style="hanging">
<t> PDV_weight = PDV * num_T(OWD)</t>
<t> var_est = sum_MT(PDV_weight)/num_MT(OWD)</t>
</list></t>
<t> This does not require additional state, however, a
cyclic buffer storing PDV_weight values will replace the one
that stored PDV values.</t>
</section>
-->
</section>
<section anchor="improvingresponse" title="Reducing lag and Improving
Responsiveness">
<t>Measurement based shared bottleneck detection makes
decisions in the present based on what has been measured in the
past. This means that there is always a lag in responding to
changing conditions. This mechanism is based on summary
statistics taken over (N*T) seconds. This mechanism can be made more
responsive to changing conditions by:
<list style="numbers">
<t>Reducing N and/or M -- but at
the expense of having less accurate metrics, and/or</t>
<t>Exploiting the fact that more recent measurements are more
valuable than older measurements and weighting them
accordingly.</t>
</list></t>
<t>Although more recent measurements are more valuable,
older measurements are still needed to gain an accurate
estimate of the distribution descriptor we are measuring.
Unfortunately, the simple exponentially weighted moving
average weights drop off too quickly for our requirements
and have an infinite tail. A simple linearly declining
weighted moving average also does not provide enough weight
to the most recent measurements. We propose a piecewise
linear distribution of weights, such that the first section
(samples 1:F)
is flat as in a simple moving average, and the second
section (samples F+1:M) is linearly declining weights to the end of the
averaging window. We choose integer weights, which allows
incremental calculation without introducing rounding
errors. </t>
<section anchor="skewrespimp" title="Improving the response of
the skewness estimate">
<t>The weighted moving average for skew_est, based on
skew_est in <xref
target="sbd_skewest"/>, can be calculated as follows:
<list style="hanging">
<t><list hangIndent="11" style="hanging">
<t hangText="skew_est =">((M-F+1)*sum(skew_base_T(1:F))
<list hangIndent="5" style="hanging">
<t>+ sum([(M-F):1].*skew_base_T(F+1:M))) </t>
</list></t>
<t>/ ((M-F+1)*sum(numsampT(1:F))
<list hangIndent="5" style="hanging">
<t>+ sum([(M-F):1].*numsampT(F+1:M)))</t>
</list></t>
</list></t>
</list></t>
<t>where numsampT is an array of the number of OWD samples
in each T (i.e. num_T(OWD)), and numsampT(1) is the most
recent; skew_base_T(1) is the most recent calculation of
skew_base_T; 1:F refers to the integer values 1 through to F, and
[(M-F):1] refers to an array of the integer values (M-F) declining through
to 1; and ".*" is the array scalar dot product operator.</t>
</section>
<section anchor="varrespimp" title="Improving the response of
the variability estimate">
<t>The weighted moving average for var_est can be
calculated as follows:
<list style="hanging">
<t><list hangIndent="10" style="hanging">
<t hangText="var_est ="> ((M-F+1)*sum(PDV(1:F)) +
sum([(M-F):1].*PDV(F+1:M)))</t>
<t>/ (F*(M-F+1) + sum([(M-F):1])</t>
</list></t>
</list></t>
<t>where 1:F refers to the integer values 1 through to F,
and [(M-F):1] refers to an array of the integer values (M-F)
declining through to 1; and ".*" is the array scalar dot product
operator. When removing oscillation noise (see <xref target="oscillationnoise"/>) this
calculation must be adjusted to allow for invalid PDV
records.<vspace blankLines="100" /></t>
</section>
</section>
</section>
<section title="Measuring OWD">
<t>This section discusses the OWD measurements required for this
algorithm to detect shared bottlenecks.
</t>
<t>The SBD mechanism described in
this draft relies on differences between OWD measurements to avoid the
practical problems with measuring absolute OWD (see <xref
target="Hayes-LCN14"/> section IIIC). Since all summary statistics are
relative to the mean OWD and sender/receiver clock offsets
should be approximately constant over the measurement periods, the
offset is subtracted out in the calculation.</t>
<section title="Time stamp resolution">
<t>The SBD mechanism requires timing information precise enough
to be able to make comparisons. As a rule of thumb, the time
resolution should be less than one hundredth of a typical path's range
of delays. In general, the lower the time resolution, the more
care that needs to be taken to ensure rounding errors do not bias the
skewness calculation.</t>
<t>Typical RTP media flows use sub-millisecond timers,
which should be adequate in most situations.</t>
</section>
<!--
<section title="System Timers">
<t>DavidH: possibly to be included as a guide in a subsequent
iteration, though probably not the TCP part.</t>
<t>
The following remarks discuss system timers, and may help in
some implementation scenarios where available timer
granularity could influence where in the system SBD is
performed.
</t>
<t>
For an implementation of SBD in kernel-space
the system's timestamp resolution is of importance: Earlier systems have the
accuracy of the timestamps given by the resolution of the clock mantained by
the kernel in jiffy, also called system's kernel tick, given by HZ or hz variables.
And the jiffy length is determined by the system's kernel tick. Newer Linux
systems have the kernel tick set by default to 250, sometimes also to 1000.
Newer FreeBSD systems have the kernel tick set 1000 by default.
Thus, yelding to jiffies of 4 or maximum of 1 ms. For the implementer of SBD
using the system's time resolution the size of one jiffy is relevant. Larger jiffy
values allow for better timer granularity and resolution, however, it comes at
the cost of more CPU cycles.
In newer systems, other timing source is the high-resolution kernel timer
introduced for sub-jiffy granularity. However, this is yet not supported in
all hardware architectures and, thus, it is recommended to the
implementer of SBD to first test its support and usability.
</t>
<t>
In particular for applications running on top of TCP, the implementer
of SBD could make use of the TCP-TS option, in similar way to LEDBAT, to get
OWD sample measurements. However, the TCP timestamp option does not
ensure higher resolution because it relies on the kernel jiffy length.
For an application sending enough traffic, the TCP-TS is updated at maximum
of 1 ms for a system's jiffy length of 1000. Also, the TCP-TS option is limited
to two four-byte fields, which also does not guarantee finer than millisecond
granularity.
Alternatively, reliable OWD samples can be also generate inside the
application itself and written into the packet's payload. The implementer of
SBD has to decide the necessary granularity given at this level by the amount
of data generated and the application's run-time performance.
</t>
<t>
In general, the implementer has to decide which granularity for SBD is necessary
depending on its application scenario. If the time granularity of SBD is limited to
a jiffy length and, thus, not higher than milliseconds, the OWD of the underlying
network path should also not be less than milliseconds. This would cause loss in
time precision and the SBD mechanism is unable to detect OWD oscillation, usually
represented by changes in the OWD's sample lowest bits.
</t>
</section>
-->
</section>
<!--
<section title="Networks and Parameter Settings">
<t>short discussion as to what parameters might be good, say for
data centers.</t>
</section>
-->
<section anchor="Acknowledgements" title="Acknowledgements">
<t>This work was part-funded by the European Community under its
Seventh Framework Programme through the Reducing Internet
Transport Latency (RITE) project (ICT-317700). The views
expressed are solely those of the authors. </t>
</section>
<!-- Possibly a 'Contributors' section ... -->
<section anchor="IANA" title="IANA Considerations">
<t>This memo includes no request to IANA.</t>
<!--
<t>All drafts are required to have an IANA considerations section (see
<xref target="I-D.narten-iana-considerations-rfc2434bis">the update of
RFC 2434</xref> for a guide). If the draft does not require IANA to do
anything, the section contains an explicit statement that this is the
case (as above). If there are no requirements for IANA, the section will
be removed during conversion into an RFC by the RFC Editor.</t>
-->
</section>
<section anchor="Security" title="Security Considerations">
<t>The security considerations of <xref target="RFC3550">RFC
3550</xref>, <xref target="RFC4585">RFC 4585</xref>, and <xref
target="RFC5124">RFC 5124</xref> are
expected to apply.</t>
<t>Non-authenticated RTCP packets carrying shared bottleneck indications and summary
statistics could allow attackers to alter the bottleneck sharing
characteristics for private gain or disruption of other parties
communication.<vspace blankLines="100" /></t>
</section>
<section anchor="ChangeHistory" title="Change history">
<t>Changes made to this document:
<list hangIndent="18" style="hanging">
<t hangText=" WG-00->WG-01 :">Moved unbiased skew section to
replace skew estimate, more robust variability estimator, the
term variance replaced with variability, clock drift term
corrected to clock skew,
revision to clock skew section with a place holder, description
of parameters.</t>
<t hangText=" 02->WG-00 :">Fixed missing 0.5 in 3.3.2 and
missing brace in 3.3.3 </t>
<t hangText=" 01->02 :">New section describing improvements
to the key metric calculations that help to remove noise,
bias, and reduce lag. Some revisions to the notation to make
it clearer. Some
tightening of the thresholds.</t>
<t hangText=" 00->01 :">Revisions to terminology for
clarity</t>
</list></t>
</section>
</middle>
<!-- *****BACK MATTER ***** -->
<back>
<!-- References split into informative and normative -->
<!-- There are 2 ways to insert reference entries from the citation libraries:
1. define an ENTITY at the top, and use "ampersand character"RFC2629; here (as shown)
2. simply use a PI "less than character"?rfc include="reference.RFC.2119.xml"?> here
(for I-Ds: include="reference.I-D.narten-iana-considerations-rfc2434bis.xml")
Both are cited textually in the same manner: by using xref elements.
If you use the PI option, xml2rfc will, by default, try to find included files in the same
directory as the including file. You can also define the XML_LIBRARY environment variable
with a value containing a set of directories to search. These can be either in the local
filing system or remote ones accessed by http (http://domain/dir/... ).-->
<references title="Normative References">
&RFC2119;
<!-- the following is the minimum to make xml2rfc happy -->
<!--
<reference anchor="min_ref">
<front>
<title>Minimal Reference</title>
<author initials="authInitials" surname="authSurName">
<organization></organization>
</author>
<date year="2006" />
</front>
</reference> -->
</references>
<references title="Informative References">
<!-- Here we use entities that we defined at the beginning. -->
&RFC3550;
&RFC4585;
&RFC5124;
&RFC5481;
&RFC6817;
&I-D.welzl-rmcat-coupled-cc;
<!-- A reference written by by an organization not a person. -->
<reference anchor="Hayes-LCN14"
target="http://heim.ifi.uio.no/davihay/hayes14__pract_passiv_shared_bottl_detec-abstract.html">
<front>
<title>Practical Passive Shared Bottleneck Detection using Shape
Summary Statistics</title>
<author initials="D. A." surname="Hayes">
<organization>University of Oslo</organization>
</author>
<author initials="S." surname="Ferlin">
<organization>Simula Research Laboratory</organization>
</author>
<author initials="M." surname="Welzl">
<organization>University of Oslo</organization>
</author>
<date year="2014" month="September"/>
</front>
<seriesInfo name="Proc. the IEEE Local Computer Networks
(LCN)" value="p150-158"/>
</reference>
<reference anchor="ITU-Y1540"
target="http://www.itu.int/rec/T-REC-Y.1540-201103-I/en">
<front>
<title>Internet Protocol Data Communication Service - IP
Packet Transfer and Availability Performance
Parameters</title>
<author>
<organization>ITU-T</organization>
</author>
<date year="2011" month="March"/>
</front>
<seriesInfo name="Series Y: Global Information
Infrastructure, Internet Protocol Aspects
and Next-Generation Networks" value=""/>
</reference>
</references>
<!-- <reference anchor="DOMINATION"
target="http://www.example.com/dominator.html"> <front>
<title>Ultimate Plan for Taking Over the World</title>
<author>
<organization>Mad Dominators, Inc.</organization>
</author>
<date year="1984" />
</front>
</reference> -->
<!-- <section anchor="app-additional" title="Additional Stuff">
<t>This becomes an Appendix.</t>
</section> -->
</back>
</rfc>
| PAFTECH AB 2003-2026 | 2026-04-23 20:36:09 |