One document matched: draft-baker-bmwg-testing-eyeball-happiness-03.xml
<?xml version="1.0" encoding="US-ASCII"?>
<!DOCTYPE rfc SYSTEM "rfc2629.dtd">
<?xml-stylesheet type='text/xsl' href='rfc2629.xslt' ?>
<!-- Some of the more generally applicable PIs that most I-Ds might want to use -->
<!-- Try to enforce the ID-nits conventions and DTD validity -->
<?rfc strict="yes" ?>
<!-- Items used when reviewing the document -->
<?rfc comments="no" ?>
<!-- Controls display of <cref> elements -->
<?rfc inline="no" ?>
<!-- When no, put comments at end in comments section, otherwise, put inline -->
<?rfc editing="no" ?>
<!-- When yes, insert editing marks: editing marks consist of a string such as <29> printed in the blank line at the beginning of each paragraph of text. -->
<!-- Create Table of Contents (ToC) and set some options for it. Note the ToC may be omitted for very short documents,but idnits insists on a ToC if the document has more than 15 pages. -->
<?rfc toc="yes"?>
<?rfc tocompact="yes"?>
<!-- If "yes" eliminates blank lines before main section entries. -->
<?rfc tocdepth="3"?>
<!-- Sets the number of levels of sections/subsections... in ToC -->
<!-- Choose the options for the references. Some like symbolic tags in the references (and citations) and others prefer numbers. The RFC Editor always uses symbolic tags. The tags used are the anchor attributes of the references. -->
<?rfc symrefs="yes"?>
<?rfc sortrefs="yes" ?>
<!-- If "yes", causes the references to be sorted in order of tags. This doesn't have any effect unless symrefs is "yes" also. -->
<!-- These two save paper: Just setting compact to "yes" makes savings by not starting each main section on a new page but does not omit the blank lines between list items. If subcompact is also "yes" the blank lines between list items are also omitted. -->
<?rfc compact="yes" ?>
<?rfc subcompact="no" ?>
<!-- end of list of popular I-D processing instructions -->
<!-- end of list of processing instructions -->
<rfc category="std" docName="draft-baker-bmwg-testing-eyeball-happiness-03"
ipr="trust200902">
<front>
<title abbrev="">Testing Eyeball Happiness</title>
<author fullname="Fred Baker" initials="F.J." surname="Baker">
<organization>Cisco Systems</organization>
<address>
<postal>
<street></street>
<city>Santa Barbara</city>
<code>93117</code>
<region>California</region>
<country>USA</country>
</postal>
<email>fred@cisco.com</email>
</address>
</author>
<date year="2011" />
<area>Operations and Management</area>
<workgroup>Benchmarking Methodology</workgroup>
<abstract>
<t>The amount of time it takes to establish a session using common
transport APIs in dual stack networks and networks with filtering such
as proposed in BCP 38 is a barrier to IPv6 deployment. This note
describes a test that can be used to determine whether an application
can reliably establish sessions quickly in a complex environment such as
dual stack (IPv4+IPv6) deployment or IPv6 deployment with multiple
prefixes and upstream ingress filtering. This test is not a test of a
specific algorithm, but of the external behavior of the system as a
black box. Any algorithm that has the intended external behavior will be
accepted by it.</t>
</abstract>
<!-- <note title="Foreword"> </note> -->
<note title="Requirements">
<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>
</note>
<!--
<?rfc needLines="10" ?>
<texttable anchor="table_example" title="A Very Simple Table">
<preamble>Tables use ttcol to define column headers and widths. Every cell then has a "c" element for its content.</preamble>
<ttcol align="center">ttcol #1</ttcol>
<ttcol align="center">ttcol #2</ttcol>
<c>c #1</c> <c>c #2</c>
<c>c #3</c> <c>c #4</c>
<c>c #5</c> <c>c #6</c>
<postamble>which is a very simple example.</postamble>
</texttable>
-->
</front>
<middle>
<!--
<t>There are multiple list styles: "symbols", "letters", "numbers", "hanging", "format", etc.</t> <t>
<list style="symbols">
<t>First bullet</t>
<t>Second bullet</t>
</list>
</t>
-->
<!--
<figure anchor="reference" title="Figure">
<artwork align="center">
<![CDATA[
ASCII artwork goes here...
]]>
</artwork>
</figure>
-->
<section title="Introduction">
<t>The <xref target="I-D.wing-v6ops-happy-eyeballs-ipv6">Happy
Eyeballs</xref> specification notes an issue in deployed multi-prefix
IPv6-only and dual stack networks, and proposes a correction. <xref
target="RFC5461"></xref> similarly looks at TCP's response to so-called
"soft errors" (ICMP host and network unreachable messages), pointing out
an issue and a set of possible solutions.</t>
<t>In a dual stack network (i.e., one that contains both <xref
target="RFC0791">IPv4</xref> and <xref target="RFC2460">IPv6</xref>
prefixes and routes), or in an IPv6-only network that uses multiple
prefixes allocated by upstream providers that implement <xref
target="RFC2827">BCP 38 Ingress Filtering</xref>, the fact that two
hosts that need to communicate have addresses using the same
architecture does not imply that the network has usable routes
connecting them, or that those addresses are useful to the applications
in question. In addition, the process of establishing a session using
the <xref target="RFC3493">Sockets API</xref> is generally described in
terms of obtaining a list of possible addresses for a peer (which will
normally include both IPv4 and IPv6 addresses) using getaddrinfo() and
trying them in sequence until one succeeds or all have failed. This
naive algorithm, if implemented as described, has the side-effect of
making the worst case delay in establishing a session far longer than
human patience normally allows.</t>
<t>This has the effect of discouraging users from enabling IPv6 in their
equipment, or content providers from offering AAAA records for their
services.</t>
<t>This note describes a test to determine how quickly an application
can reliably open sessions in a complex environment, such as dual stack
(IPv4+IPv6) deployment or IPv6 deployment with multiple prefixes and
upstream ingress filtering. This is not a test of a specific algorithm,
but a measurement of the external behavior of the application and its
host system as a black box. The "happy eyeballs" question is this: how
long does it take an application to open a session with a server or
peer, under best case and worst case conditions?</t>
<t>The methods defined here make the assumption that the initial
communication set-up of many applications can be summarized by the
measuring the DNS query/response and transport layer handshaking,
because no application-layer communication takes place without these
steps.</t>
<t>The methods and metrics defined in this note are ideally suited for
Laboratory operation, as this affords the greatest degree of control to
modify configurations quickly and produce consistent results.</t>
<t>However, if the device under test is operated as a single user with
limited query and stream generation, then there's no concern about
overloading production network devices with a single "set of eyeballs".
Therefore, these procedures and metrics MAY be applicable to production
network application, as long as the injected traffic represents a single
user's typical traffic load, and the testers adhere to the precautions
of the relevant network with respect to re-configuration of devices in
production.</t>
</section>
<section anchor="test" title="Measuring Eyeball Happiness">
<t>This measurement determines the amount of time it takes an
application to establish a session with a peer in the presence of at
least one IPv4 and multiple IPv6 prefixes and a variety of network
behaviors. ISPs are reporting that a host (MacOSX, Windows, Linux,
FreeBSD, etc) that has more than one address (an IPv4 and an IPv6
address, two global IPv6 addresses, etc) may serially try addresses,
allowing each TCP setup to expire, taking several seconds for each
attempt. There have been reports of lengthy session setup times - in
various application and OS combinations anywhere from multi-second to
half an hour - as a result. The amount of time necessary to establish
communication between two entities should be approximately the same
regardless of the type of address chosen or the viability of routing in
the specific network; users will expect this time to be consistent with
their current experience (else, happiness is at risk).</t>
<section anchor="testbed" title="Happy Eyeballs test bed configuration">
<t>The configuration of equipment and applications is as shown in
<xref target="test1"></xref>.</t>
<figure anchor="test1" title="Generic Test Environment">
<artwork align="center"><![CDATA[
+--------+ | |198.51.100.0/24
|Protocol| |192.0.2.0/24 |2001:DB8:0:2::/64
|Analyzer+-+2001:DB8:1:0::/64 |2001:DB8:1:4::/64
+--------+ |2001:DB8:0:1::/64 |2001:DB8:2:4::/64
| |
+-----+ | | +-----+
|Alice+-+ +-+ Bob |
+-----+ | +-------+ +-------+ | +-----+
+-+Router1| |Router2+-+
+-----+ | +-----+-+ +-+-----+ |
| DNS +-+ | | |
+-----+ | -+------+- |
| 203.0.113.0/24 |
| 2001:DB8:0:3::/64 |
]]></artwork>
</figure>
<t>Alice is the unit being measured, the computer running the process
that will establish a session with Bob for the application in
question. DNS is represented in the diagram as a separate system, as
is the protocol analyzer that will watch Alice's traffic. This is not
absolutely necessary; If one computer can run tcpdump and a DNS server
process - and for that matter subsume the routers - that is
acceptable. The units are separated in the test for purposes of
clarity.</t>
<t>On each test run, configuration is performed in Router 1 to permit
only one route to work. There are various ways this can be
accomplished, including but not limited to installing<list
style="symbols">
<t>a filter that drops datagrams to Bob resulting in an ICMP
"administratively prohibited",</t>
<t>a filter that silently drops datagrams to Bob,</t>
<t>a null route or removing the route to one of Bob's prefixes,
resulting in an ICMP "destination unreachable", and</t>
<t>a middleware program that responds with a TCP RST.</t>
<t>Path MTU issues</t>
</list></t>
<t>The <xref target="RFC1191">Path MTU Discovery</xref><xref
target="RFC1981"></xref> matter requires some explanation. With IPv6,
and with IPv4 when "Do Not Fragment" is set, a router with a message
too large for an interface discards it and replies with an ICMPv4
"Destination Unreachable: Datagram Too Big" or ICMPv6 "Packet Too
Big". If this packet is lost, the source doesn't know what size to
fragment to and has no indication that fragmentation is required. A
configuration for this scenario would set the MTU on 203.0.113.0/24 or
2001:DB8:0:3::/64 to the smallest allowed by the address family (576
or 1280) and disable generation of the indicated ICMP message. Note
that <xref target="RFC4821"></xref> is intended to address these
issues.</t>
<t>The tester should try different methods to determine whether
differences in this configuration make a difference in the test. For
example, one might find that the application under test responds
differently to a TCP RST than to a silent packet loss. Each of these
scenarios should be tested; if doing so is too difficult, the most
important is the silent packet loss case, as it is the worst case.</t>
</section>
<section anchor="procedure" title="Happy Eyeballs test procedure">
<t>Consider a network as described in <xref target="testbed"></xref>.
Alice and Bob each have a set of one or more IPv4 and two or more IPv6
addresses. Bob's are in DNS, where Alice can find them; Alice's and
others may be there as well, but are not relevant to the test. Routers
1 and 2 are configured to route the relevant prefixes. Different
measurement trials revise an access list or null route in Router 1
that would prevent traffic Alice->Bob using each of Bob's
addresses. If Bob has a total of N addresses, we run the measurement
at least N times, permitting exactly one of the addresses to enjoy end
to end communication each time. If the DNS service randomizes the
order of the addresses, this may not result in a test requiring
establishment of a connection to all of the addresses; in this case,
the test will have to be run repeatedly until in at least one instance
a TCP SYN or its equivalent is seen for each relevant address. The
tester should either flush the resolver cache between iterations, to
force repeated DNS resolution, or should wait for at least the DNS RR
TTL on each resource record. In the latter case, the tester should
also observe DNS re-resolving; if not, the application is not
correctly using DNS.</t>
<t>This specification assumes common LAN technology with no competing
traffic and nominal propagation delays, so that they are not a factor
in the measurement.</t>
<t>The objective is to measure the amount of time required to
establish a session. This includes the time from Alice's initial DNS
request through one or more attempts to establish a session to the
session being established, as seen in the LAN trace. The simplest way
to measure this will be to put a traffic analyzer on Alice's point of
attachment and capture the messages exchanged by Alice.</t>
<figure anchor="tcpflow" title="Message flow using TCP">
<artwork align="center"><![CDATA[
DNS Server Alice Bob
| | |
1. |<--www.example.com A------| |
2. |<--www.example.com AAAA---| |
3. |---198.51.100.1---------->| |
4. |---2001:DB8:0:2::1------->| |
5. | | |
6. | |--TCP SYN, IPv6--->X |<***********
7. | |--TCP SYN, IPv6--->X | |
8. | |--TCP SYN, IPv6--->X | TCP 3wHS
9. | | | Time
10. | |--TCP SYN, IPv4------->|(any family)
11. | |<-TCP SYN+ACK, IPv4----| |
12. | |--TCP ACK, IPv4------->|<***********
]]></artwork>
</figure>
<t>In a TCP-based application (<xref target="tcpflow"></xref>), that
would be from the DNS request on line 1 through the first completion
of a TCP three-way handshake, the transmission on line 12.</t>
<figure anchor="udpflow" title="Message flow using UDP">
<artwork align="center"><![CDATA[
DNS Server Alice Bob
| | |
1. |<--www.example.com A------| |
2. |<--www.example.com AAAA---| |
3. |---198.51.100.1---------->| |
4. |---2001:DB8:0:2::1------->| |
5. | | |
6. | |--UDP Request, IPv6-->X|<---------
7. | |--UDP Request, IPv6-->X| first
8. | |--UDP Request, IPv6-->X| request/
9. | | | response
10. | |--UDP Request, IPv4--->| success
11. | |<-UDP Response, IPv4---|<---------
]]></artwork>
</figure>
<t>In a UDP-based application (<xref target="udpflow"></xref>), that
would be from the DNS request (line 1) through one or more UDP
Requests (lines 6-10) until a UDP Response is seen (line 11).</t>
<t>When using other transports, the methodology will have to be
specified in context; it should measure the same event.</t>
</section>
<section anchor="metrics" title="Happy Eyeballs metrics">
<t>The measurements taken are the duration of the interval from the
initial DNS request until the session is seen to have been
established, as described in <xref target="procedure"></xref>. We are
interested in the shortest and longest durations (which will most
likely be those that send one SYN and succeed and those that send a
SYN to each possible address before succeeding in one of the
attempts), and the pattern of attempts sent to different addresses.
The pattern may be to simply send an attempt every <time
interval>, or may be more complex; as a result, this is in part
descriptive.</t>
<t>ALL measurement events on the sending and receiving of messages
SHALL be observed at the "Alice" attachment point and time stamps
SHOULD be applied upon reception of the last bit of the IP information
field. Use of a alternate timing reference SHALL be noted.</t>
<section anchor="session-metric"
title="Metric: Session Setup Interval">
<t><list style="hanging">
<t hangText="Name:">Session Setup Interval</t>
<t hangText="Description:">The session setup interval MUST be
the time beginning with the first DNS query sent (observed at
Alice's attachment), and ending with successful transport
connection establishment (as indicated in line 12 of <xref
target="tcpflow"></xref>, and line 11 of <xref
target="udpflow"></xref>). This interval is defined as the
session setup interval.</t>
<t>This test will be run several times, once for each possible
combination of destination address (configured on Bob) and
failure mode (configured on Router 1).</t>
<t hangText="Methodology:">In the LAN analyzer trace, note the
times of the initial DNS request and the confirmation that the
session is open as described in <xref
target="procedure"></xref>. If the session is not successfully
opened, possibly due to Alice aborting the attempt, the Session
Setup Interval is considered to be infinite.</t>
<t hangText="Units:">Session setup time is measured in
milliseconds.</t>
<t hangText="Measurement Point(s):">The measurement point is at
Alice's LAN interface, both sending and receiving, observed
using a program such as tcpdump running on Alice or an external
analyzer.</t>
<t hangText="Timing:">Continuous.</t>
</list></t>
</section>
<section anchor="Max-metric"
title="Metric: Maximum Session Setup Interval">
<t><list style="hanging">
<t hangText="Name:">Maximum Session Setup Interval</t>
<t hangText="Description:">The maximum session setup interval is
the longest period of time observed for the establishment of a
session as described in <xref
target="session-metric"></xref>.</t>
<t hangText="Methodology:">see Session Setup Interval.</t>
<t hangText="Units:">Session setup time is measured in
milliseconds.</t>
<t hangText="Measurement Point(s):">see Session Setup
Interval.</t>
<t hangText="Timing:">Continuous.</t>
</list></t>
</section>
<section anchor="Min-metric"
title="Metric: Minimum Session Setup Interval">
<t><list style="hanging">
<t hangText="Name:">Minimum Session Setup Interval</t>
<t hangText="Description:">The minimum session setup interval is
the shortest period of time observed for the establishment of a
session.</t>
<t hangText="Methodology:">see Session Setup Interval.</t>
<t hangText="Units:">Session setup time is measured in
milliseconds.</t>
<t hangText="Measurement Point(s):">see Session Setup
Interval.</t>
<t hangText="Timing:">Continuous.</t>
</list></t>
</section>
<section anchor="Variation-metric"
title="Descriptive Metric: Attempt pattern">
<t><list style="hanging">
<t hangText="Name:">Attempt pattern</t>
<t hangText="Description: ">The Attempt Pattern is a description
of the observed pattern of attempts to establish the session. In
simple cases, it may be something like "Initial TCP SYNs to a
new address were observed every <so many> milliseconds";
in more complex cases, it might be something like "Initial TCP
SYNs in IPv6 were observed every <so many> milliseconds,
and other TCP SYNs using IPv4 were observed every <so
many> milliseconds, but the two sequences were independent."
It may also comment on retransmission patterns if observed.</t>
<t hangText="Methodology:">The traffic trace is analyzed to
determine the pattern of initiation.</t>
<t hangText="Units:">milliseconds.</t>
<t hangText="Measurement Point(s):">The measurement point is at
Alice's LAN interface, observed using a program such as tcpdump
running on Alice or an external analyzer.</t>
<t hangText="Timing:">Continuous.</t>
</list></t>
</section>
</section>
</section>
<section anchor="IANA" title="IANA Considerations">
<t>This memo asks the IANA for no new parameters.</t>
<t>Note to RFC Editor: This section will have served its purpose if it
correctly tells IANA that no new assignments or registries are required,
or if those assignments or registries are created during the RFC
publication process. From the author's perspective, it may therefore be
removed upon publication as an RFC at the RFC Editor's discretion.</t>
</section>
<section anchor="Security" title="Security Considerations">
<t>This note doesn't address security-related issues.</t>
</section>
<section anchor="Acknowledgements" title="Acknowledgements">
<t>This note was discussed with Dan Wing, Andrew Yourtchenko, and
Fernando Gont. In the Benchmark Methodology Working Group, Al Morton,
David Newman, Sarah Banks, and Tore Anderson made comments.</t>
</section>
<section anchor="log" title="Change Log">
<t><list style="hanging">
<t hangText="-00:">Initial version - November, 2010</t>
<t hangText="-01:">Rewritten per suggestions by Al Morton, David
Newman, and Sarah Banks.</t>
<t hangText="-02:">Clean-up per working group comments.</t>
<t hangText="-03:">Updated per Al Morton's and Tore Anderson's
comments.</t>
</list></t>
</section>
</middle>
<back>
<!-- references split to informative and normative -->
<references title="Normative References">
<?rfc include="reference.RFC.2119" ?>
<?rfc include="reference.I-D.wing-v6ops-happy-eyeballs-ipv6" ?>
</references>
<references title="Informative References">
<?rfc include="reference.RFC.0791" ?>
<?rfc include="reference.RFC.2460" ?>
<?rfc include="reference.RFC.1191" ?>
<?rfc include="reference.RFC.1981" ?>
<?rfc include="reference.RFC.2827" ?>
<?rfc include="reference.RFC.3493" ?>
<?rfc include="reference.RFC.4821" ?>
<?rfc include="reference.RFC.5461" ?>
</references>
</back>
</rfc>
| PAFTECH AB 2003-2026 | 2026-04-24 01:36:21 |