One document matched: draft-jennings-vipr-overview-04.xml
<?xml version="1.0" encoding="US-ASCII"?>
<?rfc toc="yes" ?>
<?rfc symrefs="yes" ?>
<?rfc iprnotified="yes" ?>
<?rfc strict="no" ?>
<?rfc compact="yes" ?>
<?rfc sortrefs="no" ?>
<?rfc colonspace="yes" ?>
<?rfc rfcedstyle="no" ?>
<?rfc tocdepth="4"?>
<rfc category="std" docName="draft-jennings-vipr-overview-04"
ipr="trust200902">
<front>
<title abbrev="VIPR Overview">Verification Involving PSTN Reachability:
Requirements and Architecture Overview</title>
<author fullname="Mary Barnes" initials="M." surname="Barnes, Ed.">
<organization>Polycom</organization>
<address>
<postal>
<street></street>
<city></city>
<region>TX</region>
<country>US</country>
</postal>
<email>mary.ietf.barnes@gmail.com</email>
</address>
</author>
<author fullname="Cullen Jennings" initials="C." surname="Jennings">
<organization>Cisco</organization>
<address>
<postal>
<street>170 West Tasman Drive</street>
<street>MS: SJC-21/2</street>
<city>San Jose</city>
<region>CA</region>
<code>95134</code>
<country>USA</country>
</postal>
<phone>+1 408 421-9990</phone>
<email>fluffy@cisco.com</email>
</address>
</author>
<author fullname="Jonathan Rosenberg" initials="J.R." surname="Rosenberg">
<organization>jdrosen.net</organization>
<address>
<postal>
<street></street>
<city>Monmouth</city>
<region>NJ</region>
<country>US</country>
</postal>
<email>jdrosen@jdrosen.net</email>
<uri>http://www.jdrosen.net</uri>
</address>
</author>
<author fullname="Marc Petit-Huguenin" initials="M."
surname="Petit-Huguenin">
<organization>Unaffiliated</organization>
<address>
<email>petithug@acm.org</email>
</address>
</author>
<date month="February" year="2013" />
<area>RAI</area>
<workgroup>VIPR WG</workgroup>
<abstract>
<t>The Session Initiation Protocol (SIP) has seen widespread deployment
within individual domains, typically supporting voice and video
communications. Though it was designed from the outset to support
inter-domain federation over the public Internet, such federation has
not materialized. The primary reasons for this are the complexities of
inter-domain phone number routing and concerns over security. This
document reviews this problem space, outlines requirements, and then
describes a new model and technique for inter-domain federation with
SIP involving the Public Switched Telephone Network (PSTN),
called Verification Involving PSTN Reachability (VIPR). VIPR
addresses the problems that have prevented inter-domain federation over
the Internet. It provides fully distributed inter-domain routing for
phone numbers, authorized mappings from phone numbers to domains, a new
technique for automated SIP anti-spam, and privacy of number ownership,
all while preserving the trapezoidal model of SIP.</t>
</abstract>
</front>
<middle>
<section title="Introduction">
<t>The Session Initiation Protocol (SIP) was originally published as
<xref target="RFC2543" /> in May of 1999. This was followed
by subsequent publication of <xref target="RFC3261"/>,
which brought the protocol to sufficient maturity to enable large scale
market adoption.</t>
<t>
SIP has achieved large scale market adoption with
hundreds of implementations, spanning consumer products, enterprise
servers, and large scale carrier equipment. It carries billions and
billions of minutes of calls, and has become the standard for
interconnection between products from different vendors. If one measures
success in deployment, then clearly SIP is a success.</t>
<t> SIP was designed from the
ground up to enable communications between users in different domains,
all over the public Internet. The intention was that real-time
communications should be no different than email or the web, with the
same any-to-any connectivity that has fueled the successes of those
technologies.
However, when SIP is used between domains, it is typically
through private federation agreements. While such agreements are positive,
they have typically been limited to voice,
which has limited the use of video and the growth of advanced SIP features,
thus preventing
the innovation that SIP was expected to drive. Thus,
the any-to-any Internet
federation model envisioned by SIP has not materialized at scale.</t>
<t>This document introduces a new technology, called Verification
Involving PSTN Reachability (VIPR), that breaks down the
barriers that have prevented inter-domain voice, video and other multimedia
services. By stepping back and
changing some of the most fundamental assumptions about federation, VIPR
is able to address the key problems preventing its deployment. VIPR
focuses on incremental deployability. At
the same time, VIPR ensures that SIP's trapezoidal model of direct
federation between domains without any intermediate processing beyond IP
transport is realized. That model is required in order to allow
innovative new services to be deployed.</t>
</section>
<!-- Conventions and Terminology -->
<section anchor="sec:conventions" title="Conventions and Terminology">
<t>The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT",
"SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and
"OPTIONAL" in this
document are to be interpreted as described in <xref target="RFC2119"></xref>.
</t>
<t>
<list style="hanging">
<t hangText="Call Agent: ">
An entity in a SIP enabled domain that supports VIPR.
The Call Agent performs call processing on behalf
of one or more user agents represented by E.164 numbers within the domain.
</t>
<t hangText="Ticket: ">
A shared secret that is generated after a PSTN call to enable secure call setup
on a subsequent inter-domain IP call enabled by VIPR.
</t>
<t hangText="User Agent: ">
As defined in <xref target="RFC3261"/>, with the restriction that the user agent
must have an associated E.164 number.
</t>
</list>
</t>
</section>
<!--Terminology-->
<section title="Problem Statement">
<t>The first question that must be asked is this - why haven't we seen
widespread adoption of inter-domain SIP federation?
The reason for this is due to problems with the following - summarized
in order of importance/impact:</t>
<t>
<list style="numbers">
<t> Phone number routing </t>
<t> Open pinhole </t>
<t> Quality of service</t>
<t> Troubleshooting</t>
</list>
</t>
<t> The first two
are the most significant.</t>
<section anchor="phone-numbers" title="The Phone Number Routing Problem">
<t>
Inter-domain federation requires that the sending domain determine
the address of the receiving domain, in the form of a DNS name
(example.com) or one or more IP addresses that can be used to reach
the domain. In email and in the web, this is easy. The identifiers
used by those services - the email address and web URL respectively -
embed the address of the receiving domain. A simple DNS lookup is all
that is required to route the connection. SIP was designed to use the
same email-style identifiers.</t>
<t>
However, most SIP deployments utilize phone numbers in the form of E.164
numbers <xref target="E.164"/>, and not
email-style SIP URIs. This is due to the huge installed base of users
that continue to exist solely on the PSTN.
In order to be reached by users on the PSTN, and in order to
reach them, users in SIP deployments need to be assigned a PSTN phone
number.
Users in SIP deployments need to place that phone number
on business cards, use it in their email signatures, and in general,
give it out to their friends and colleagues, in order to be reached.
While those users could additionally have an email style SIP URI, the
phone number serves as a single, global identifier that works for
receiving calls from users on the PSTN as well as users within the
same SIP domain.
</t>
<t> There are several reasons why two identifiers are used when one
will suffice. The
universality of PSTN phone numbers is the reason why most SIP deployments
continue to use them - often exclusively.</t>
<t>Another reason is that many SIP deployments utilize hardphones or
telephony adaptors, and the user interfaces on these devices -
patterned after existing phones - only allow phone number based
dialing. Consequently, these users are only allocated PSTN phone numbers,
and not email-style SIP URI.</t>
<t>Finally, a large number of SIP deployments are in domains where the
endpoints are not IP. Rather, they are circuit based devices,
connected to a SIP network through a gateway. SIP is used within the
core of the network, providing lower cost transit, or providing add-on
services. Clearly, in these deployments, only phone numbers are
used.</t>
<t>Consequently, to make inter-domain federation incrementally
deployable and widely applicable, it needs to work with PSTN phone numbers
rather than email-style SIP URIs. Telephone numbers, unlike email
addresses, do not provide any indication of the address of the domain
which "owns" the phone number. Indeed, the notion of phone number
ownership is somewhat cloudy. Phone numbers can be ported between carriers.
They can be assigned to a user or enterprise, and then later
re-assigned to someone else. Phone numbers are granted to users and
enterprises through a complex delegation process involving the ITU,
governments, and telecommunications carriers, often involving local
regulations that vary from country to country.</t>
<t>Therefore, in order to deploy inter-domain federation, domains are
required to utilize some kind of mechanism to map phone numbers to the
address of the domain to which calls should be routed. Though several
techniques have been developed to address this issue, none have
achieved large-scale Internet deployments.</t>
</section>
<section anchor="pinhole" title="The Open Pinhole Problem">
<t>The inter-domain federation mechanism built into SIP borrows
heavily from email. Each domain runs a SIP server on an open port.
When one domain wishes to contact another, it looks up the domain name
in the DNS, and connects to that server on the open port. Here,
"open" means that the server is reachable from anywhere on the public
Internet, and is not blocked by firewalls.</t>
<t>This simple design worked well in the early days of email. However,
the email system has now become plagued with spam. This has
resulted in administrators spending a significant amount of time
maintaining spam filters. This does not always benefit the end users
as in some cases valid emails are dropped without the user being
notified. Thus, administrators of SIP domains are rightfully concerned
that if they make a SIP server available for anyone on the Internet to
contact, it will open the floodgates for SIP spam, which is far more
disruptive than email-based spam <xref target="RFC5039"></xref>.
Administrators are also concerned that an open server will
create a back-door for denial-of-service and other attacks that can
potentially disrupt their voice and video services. Administrators are often not
willing to take that risk since voice deployments
demand higher uptimes and better levels of reliability than email,
especially for enterprises.</t>
<t>Fears around spam and denial-of-service attacks, when put together,
form the "open pinhole problem" - that domains are not willing to
enable SIP on an open port facing the Internet.</t>
<t>To fix this, a new model for federation is needed - a model where
these problems are addressed as part of the fundamental design rather than
after the functionality has been deployed.</t>
</section>
<section title="Quality of Service Problem">
<t>The Internet does not provide any Quality of Service (QoS) guarantees. All traffic is
best effort. This is not an issue for data transaction services, like
web and email. It is, however, a concern when using real-time
services, such as voice and video.</t>
<t>That said, there are a large number of existing SIP deployments
that run over the Internet. Though the lack of QoS is a concern, it
has not proven a barrier to deployment. It is believed that if
if the more
fundamental issues - the phone number routing and open pinhole
problems - can be addressed, the QoS problem will be a non-issue. As
such, QoS is not discussed further in this or other VIPR specifications.</t>
</section>
<section title="Troubleshooting Problem">
<t>The final problem that is prohibing large scale inter-domain
federation is troubleshooting. When connecting calls
between domains, problems can occur. Calls can be blocked. Calls
can be misdelivered. Features sometimes don't work. There can be one-way
media or no media at all. The video may not start. Call quality can be
poor.</t>
<t>These problems are common in SIP deployments, and they are tough
to troubleshoot even within a single administrative domain. When
real-time services extend inter-domain, the problem becomes worse.</t>
<t>Fortunately, some work has been completed to improve the ability for network
administrators to diagnose SIP problems. A Common log format <xref
target="RFC6873"></xref> has been developed.
Other work underway, such as consistent session IDs <xref
target="I-D.ietf-insipid-session-id-reqts"></xref> and
<xref target="I-D.jones-insipid-session-id"></xref> can help troubleshoot
interdomain calls.</t>
<t>In addition to the above, any new technology that facilitates
inter-domain federation needs to have troubleshooting built-in, so
that it is not a barrier to deployment. Further consideration of
necessary built-in techniques for troubleshooting is required for
successful deployment of VIPR.</t>
</section>
</section>
<section title="Summary of Existing Solutions">
<t>Given the value of inter-domain SIP federation, there are
existing deployed solutions summarized below. However, each solution
approach has
fundamental limitations that have inhibited widespread deployment.</t>
<section title="Domain Routing">
<t>The first solution for SIP inter-domain
federation is built into SIP itself - domain routing. In this
technique, users utilize email-style SIP URIs as identifiers. By
utilizing the DNS lookup mechanism defined in <xref
target="RFC3263"></xref>, SIP enables calls to be routed between
domains in much the same way email is routed between domains.</t>
<t>This technique works well in theory, but it has two limitations
which have limited its deployment:</t>
<t><list style="numbers">
<t>The majority of SIP deployments utilize phone numbers, often
exclusively. In such a case, domain routing cannot be used.</t>
<t>Domain federation brings with it the possibility (and strong
likelihood) of the same levels of spam and DoS attacks that have
plagued the email system.</t>
</list></t>
<t>These issues have already been discussed in sections <xref target="phone-numbers"/> and
<xref target="pinhole"/> respectively.</t>
</section>
<section title="Public ENUM">
<t>Public ENUM, defined in <xref target="RFC6116"></xref> addresses
the phone number routing problem by placing phone
numbers into the public DNS. Clients can then perform a simple DNS
lookup on a phone number, and retrieve a SIP URI which can be used to
route to that phone number.</t>
<t>Unfortunately, public ENUM requires that the entries placed into
the DNS be populated following a chain of responsibility that mirrors
the ownership of the numbers themselves. This means that, in order for
a number to be placed into the DNS, authorization to do so must start
with the ITU, and from there, move to the country, telecom regulator,
and ultimately the end user. The number of layers of bureaucracy
required to accomplish this is non-trivial. In addition, the telecom
operators - that would be partly responsible for populating the
numbers into the DNS - have little incentive to do so. As a
consequence, public ENUM is largely empty, and is likely to remain so
for the foreseeable future.</t>
<t>Instead, ENUM has evolved into a technique for federation amongst
closed peering partners, called private ENUM or infrastructure ENUM
<xref target="RFC5067"></xref>. While there is value in this
technology, it does not enable the open federation that public ENUM
was designed to solve.</t>
</section>
<section title="Private Federations">
<t>Private federations are a cooperative formed amongst a small number
of participating domains. The cooperative agrees to use a common
technique for federation, and through it, is able to connect to each
other. There are many such federations in use today.</t>
<t>Some of these federations rely on a central database, typically run
by the federation provider, that can be queried by participating
domains. The database contains mappings from phone numbers to domains,
and is populated by each of the participating domains, often manually.
Each domain implements an agreed-upon query interface that can be used
to access the database when a number is called. Sometimes ENUM is used
for this interface (called private ENUM), other times, a SIP
redirection is used. Some federations also utilize private IP networks
in order to address QoS problems. </t>
<t>Private federations work, but they have one major limitation:
scale. As the number of participating domains grows, several problems
arise. Firstly, the size of the databases become difficult to manage. Secondly, the
correctness of the database becomes an issue, since the odds of
misconfigured numbers (either intentionally or accidentally)
increases. As the membership grows further, the odds increase that
malicious domains will be let in, introducing a source of spam and further
problems. The owner of the federation can - and often does - assume
responsibility for this, and can attempt to identify and shut down
misbehaving participants. Indeed, as the size of the federations grow,
the owner of the federation needs to spend increasing levels of
capital on maintaining it. This often results in the owners
charging for membership, which can be a barrier to entry.</t>
</section>
</section>
<section title="Key Requirements">
<t>From the discussion on the problems of inter-domain federation and
the solutions that have been attempted so far, several key requirements
emerge:</t>
<t><list style="hanging">
<t hangText="REQ-1:">The solution must allow for federation
between any number of domains.</t>
<t hangText="REQ-2:">The solution must enable users in one domain to
identify users in another domain through the use of their existing
E.164 based phone numbers.</t>
<t hangText="REQ-3:">The solution must work with deployments that
utilize any kind of endpoint, including non-IP phones connected
through gateways, IP softphones and hardphones.</t>
<t hangText="REQ-4:">The solution must not require any change in
user behavior. The devices and techniques that users have been using
previously to make inter-domain calls must continue to work, but
now result in inter-domain calls using IP.</t>
<t hangText="REQ-5:">The solution must work worldwide, for any
domain anywhere.</t>
<t hangText="REQ-6:">The solution must not require any new
services from any kind of centralized provider. A domain should be
able to deploy equipment and
connect to the federation without any interaction with or
authorization from a centralized provider.</t>
<t hangText="REQ-7:">The solution must not require any prior
arrangement between domains in order to facilitate federation
between those domains. Federation must occur opportunistically -
connections established when they can be.</t>
<t hangText="REQ-8:">The solution must work for domains of any size
- starting with a single phone up to the largest telecom operator with
tens of millions of numbers.</t>
<t hangText="REQ-9:">The solution must have built-in mechanisms for
preventing spam and DoS attacks. These mechanisms must be fully
automated.</t>
<t hangText="REQ-10:">The solution must not require any processing
whatsoever by SIP or RTP intermediaries. It must be possible for a
direct SIP connection to be established between participating
domains.</t>
<t hangText="REQ-11:"> The solution should adapt to VIPR call failures. The
solution should
allow the user to make calls using the
inter-domain calling mechanism used prior to the initial VIPR-enabled call. </t>
</list></t>
</section>
<section title="Executive Overview">
<t>Verification Involving PSTN Reachability (VIPR) is
aimed at solving the problems that have prevented large-scale
Internet-based SIP federation of voice and video. VIPR solves these
problems by creating a hybrid of three technologies - the PSTN itself, a
Peer to Peer (P2P) network, and SIP. By using these three technologies together,
VIPR enables an incrementally deployable solution to federation.</t>
<section title="Key Properties">
<t>VIPR has several important properties that enable it to solve the
federation problem:</t>
<t><list style="hanging">
<t hangText="Works With Numbers:">VIPR enables federation for
existing PSTN phone numbers. It does not require users or administrators
to know or configure email-style identifiers. It does not require
the allocation of new numbers. It does not require a change in
user behaviors. </t>
<t hangText="Works with Existing Endpoints:">VIPR does not require
any changes to endpoints. Consequently, it works with existing SIP
endpoints and with non-IP endpoints connected through
gateways.</t>
<t hangText="Verified Mappings:">VIPR
ensures that phone calls cannot be misrouted or numbers
stolen.
The biggest issue in mapping from
a phone number to a domain or IP address, is determining whether
the mapping is correct - i.e., does the domain really own the given
phone number? While solutions like ENUM have solved this problem
by relying on centralized delegations of authorization, VIPR
provides a secure mapping in a fully distributed way. </t>
<t hangText="Worldwide:">VIPR works worldwide. Any domain that is
connected to both the PSTN and the Internet can participate.
Since VIPR does not depend on availability of any
regional services beyond IP and PSTN access - both of which are
already available globally - VIPR itself is globally
available.</t>
<t hangText="Scalibility:">VIPR is scaleable. Any
number of domains can participate.</t>
<t hangText="Self-Scale:">VIPR self-scales. This means that the
amount of computation, memory, and bandwidth that a domain must
deploy scales in direct proportion to the size of their own user
base.</t>
<t hangText="Self-Learning:">VIPR is completely automated. A
domain does not require configuration of any information about another
domain. It does not require provisioning of IP addresses, domain names,
certificates, phone number prefixes or routing rules. </t>
<t hangText="Automated Anti-Spam">VIPR has a built-in
mechanism for preventing SIP spam, which is specific to SIP.
It is fundamentally different from
existing SIP anti-spam techniques which borrow from email <xref
target="RFC5039"></xref>. This new technique is fully automated,
and requires no configuration by administrators and no
participation from end users.</t>
<t hangText="Feature Velocity:">VIPR enables direct SIP
connections between two domains seeking to federate. There are no
SIP intermediaries of any sort between the two. This means that
domains have no dependencies on intermediaries for deployment of
new features.</t>
<t hangText="Secure:"> Security is
a fundamental part of VIPR and cannot be disabled.</t>
<t hangText="Reliable:">VIPR is reliable. Through its
hybridization of the PSTN and the Internet, it ensures that
calls always go through, even in cases of network failure or
limited IP connectivity.
</t>
</list></t>
<t>In order to achieve a solution with these properties, past assumptions about
how federations should work must be challenged. </t>
</section>
<section title="Challenging Past Assumptions">
<t>Two unstated assumptions of SIP federation are challenged by
VIPR.</t>
<t>The first assumption that federation solutions have made is this:
<list style="empty">
<t>The purpose of SIP federation is to eliminate the PSTN, and
consequently, we cannot assume the PSTN itself as part of the
solution.</t>
</list> Though unstated, this assumption has clearly been part of
the design of existing solutions. SIP federation based on email-style
URIs, as defined in RFC 3261, doesn't utilize nor make mention of the
PSTN. Solutions like ENUM, or private registries,
also do not utilize nor
make mention of the PSTN. However,
such approaches ignore an incremental solution - a solution which
utilizes the PSTN itself to solve the hard problems in SIP
federation.</t>
<t>There are many advantages to leveraging the PSTN. It reaches
worldwide. It provides a global numbering translation service that
maps phone numbers to circuits. It is highly reliable, and provides
QoS. It has been built up over decades to achieve these goals.
Thus, building upon rather than replacing the PSTN, can provide
the necessary functionality
once another assumption is challenged.</t>
<t>
This second assumption is: <list style="empty">
<t>A federation solution must be the same as the final target
federation architecture, and not just a step towards it.</t>
</list>
SIP's email-style federation was a pure 'target architecture'.
ENUM was the same - a worldwide global DNS database
with everyone's phone numbers providing open
connectivity.</t>
<t>Historically, technologies are more successful when they are
incrementally deployable. As such, VIPR is very much focused on incremental
deployability. It
discards the notion of perfect IP federation for a solution that
federates most, but not all calls, by relying on the PSTN to fill in
the gaps.</t>
</section>
<section title="Technical Overview">
<t>A high level view of the VIPR architecture with an example is shown in <xref
target="fig-high-arch"></xref>.
The figure shows four different
domains, example.com, example.net, example.org and example.edu, federated using VIPR
technology. Each domain is connected to both the public Internet and
to the traditional PSTN. For simplicity, the connection for the
call agents in example.org and
example.edu to the PSTN is not indicated in the diagram as that
interface is not relevant to the subsequent examples.</t>
<figure anchor="fig-high-arch" title="High Level Architecture ">
<artwork><![CDATA[
+-------+ +-------+
| Call | | Call |
example.org | Agent | | Agent | example.edu
| | | |
+-------+ +-------+
\ /
\ /
\ /
\ /
|
//--------\\
|// \\|
| Internet |
+-------+ |\\ //| +-------+
| Call |------ \\ _______//------| Call |
//\\ | Agent | | Agent | //\\
\ / | | | | \ /
\/ ---| | +-----------+ | |---- \/
User | |======| |======| | User
Agent +-------+ | PSTN | +-------+ Agent
example.com | | example.net
+-----------+
]]></artwork>
</figure>
<t>For purposes of explanation, it is easiest to think of each domain
as having a single call agent which participates in the federation
solution. The functionality
is decomposed into several
sub-components, and this is discussed in more detail below. The call
agent is connected to one or more user agents in the domain, and is
responsible for routing calls, handling features, and processing call
state. The call agent is stateful, and is aware of when calls start
and stop. Additional detail for the
functional components of this architecture are provided in
<xref target="I-D.petithuguenin-vipr-framework"/>.</t>
<t>Assume that all four domains have a 'fresh' installation of VIPR,
and that domain example.net 'owns' +1 408 555 5xxx, a block of 1000 numbers
allocated by its PSTN provider.</t>
<t>The VIPR mechanism can be broken into four basic steps: storage of
phone numbers, PSTN first call, validation and caching, and subsequent SIP
call(s).</t>
<section title="Storage of Phone Numbers">
<t>The first step is that the call agents form a single, worldwide
P2P network, using a VIPR specific usage
<xref target="I-D.petithuguenin-vipr-reload-usage"/> of
RELOAD <xref target="I-D.ietf-p2psip-base"></xref> with
a variant of the Chord algorithm. This P2P network forms a
distributed hash table
(DHT) running amongst all participating domains. A distributed hash
table is like a simple database, allowing storage of key-value
pairs, and lookup of objects by key. Unlike a normal hash table,
which resides in the memory of a single computer, a distributed hash
table is spread across all of the servers which make up the P2P
network. In this case, it is spread across all of the domains
participating in the VIPR federation.</t>
<t>The problem solved by the variant of the Chord
algorithm (and by other DHT algorithms), is
an answer to the following: given that the desired operation is to
read or write an object with key K, which node in the DHT is the box
that currently stores the object with that key? The P2P SIP variant
of the Chord algorithm provides an
algorithm which routes read and write operations through
nodes in the DHT until they eventually arrive at the right place.
With Chord, this will take no more than log2N hops, where N is the
number of nodes in the DHT. Consequently, for a DHT with 1024 nodes,
10 hops are required in the worst case. For 2048, 11 hops. And so
on. The logarithmic factor allows DHTs to achieve efficient scale
and to provide a large amount of storage summed across all of the nodes that
make up the DHT.</t>
<t>This logarithmic hopping behavior also means that each node in
the DHT does not need to establish a TCP/TLS connection to every
other node. Rather, connections are established to a smaller subset
- just log(N) of the nodes.</t>
<t>In DHTs, each participating entity is identified by a Node-ID.
The Node-ID is a 128 bit number, assigned randomly to each entity.
They have no inherent semantic meaning; they are not like domain
names or IP addresses.</t>
<t>In the case of VIPR, each call agent is identified by one or more
Node-IDs. For purposes of discussion, consider the case where the
call agent has just one Node-ID. Each participating domain, including example.net
in our example, uses the DHT to store a mapping from each phone
number that it owns, to the domain's Node-ID. In the case of example.net, it
would store 1000 entries into the DHT, each one being a mapping from
one of its phone numbers, to the domain's Node-ID. Furthermore, when the
mappings are stored, the mapping is actually from the SHA-1 hash of
the phone number, to the Node-ID of the call agent which claims
ownership of that number.</t>
<t>For example, if the Node-ID of the call agent in domain example.net is
0x1234 (a shorter 16 bit value to simplify discussion), the entries
stored into the DHT by example.net would be:</t>
<figure anchor="fig-high-arch2" title="DHT Contents">
<artwork><![CDATA[
Key | Value
----------------------------------
SHA1(+14085555000) | 0x1234
SHA1(+14085555001) | 0x1234
SHA1(+14085555002) | 0x1234
.....
SHA1(+14085555999) | 0x1234
]]></artwork>
</figure>
<t>It is important to note that the DHT does not contain phone
numbers (it contains hashes of them), nor does it contain IP
addresses or domain names. Instead, it is a mapping from the hash of
a phone number (in E.164 format) to a Node-ID.</t>
<t>example.net will store this mapping when it starts up, or when a new
number is provisioned. The information is refreshed periodically by
example.net. The actual server on which these mappings are stored depends
on the variant of the Chord algorithm. Typically, the entries will be uniformly
distributed amongst all of the call agents participating in the
network.</t>
</section>
<section title="PSTN First Call">
<t>At some point, a user agent (Alice) in example.com makes a call to +1 408 555
5432, which is her colleague Bob. Even though both sides have VIPR,
the call takes place over the plain old PSTN, per <xref target="pstncall"/>.
Alice talks to Bob for
a bit, and they hang up.</t>
<figure anchor="pstncall" title="PSTN First Call">
<artwork><![CDATA[
+-------+ +-------+
| Call | | Call |
//\\ | Agent | | Agent | //\\
\ / | | | | \ /
\/ ---| | +-----------+ | |---- \/
Alice | |<=======<========>======>| | Bob
+-------+ | PSTN | +-------+
example.com | | example.net
+-----------+
]]></artwork>
</figure>
<t>At a random point in time after the call has completed, the call
agent in example.com "wakes up" and says to itself, "that's interesting,
someone in my domain called +1 408 555 5432, and it went over the
PSTN. I wonder if that number is reachable over IP instead?". To
make this determination, it hashes the called phone number, and
looks it up in the DHT. It is important to note that this lookup is
not at the time of an actual phone call - this lookup process
happens outside of any phone call, and is a background process.</t>
<t>The query for +1 408 555 5432 will traverse the DHT, and
eventually arrive at the node that is responsible for storing the
mapping for that number. Typically, that node will not be example.net, but
rather one of the other nodes in the network (e.g., example.org).
In many cases, the called number will not find a matching mapping in
the DHT. This happens when the number that was dialed is not owned
by a domain participating in VIPR. When that happens, example.com takes no
further action. Next time there is another call to the same number,
it will repeat the process and check once more whether the dialed
number is in the DHT.</t>
<t>In this case, there is a match in the DHT, and example.com learns the
Node-ID of example.net. It then proceeds to the validation step per
<xref target="Validate-cache"/>. It is
also possible that there are multiple matches in the DHT. This can
happen if another domain - example.edu for example - also claims ownership
of that number. When there are multiple matching results, example.com
learns all of them, and performs the validation step with each.</t>
</section>
<section anchor="Validate-cache" title="Validation and Caching">
<t>Why not just store the domain in the DHT, instead of the Node-ID?
If the domain was stored in the DHT, once example.com performed the lookup,
it would immediately
learn that the number maps to example.net, and could then make a direct
SIP call next time.</t>
<t>The main reason this doesn't work is security. The information in
the DHT is completely untrusted. There is nothing so far that
enables example.com to know that example.net does, in fact, own the phone number
in question. Indeed, if multiple domains make a claim on the number,
it has no way to know which one (if any) actually owns it.</t>
<t>To address this critical problem, VIPR requires a
mechanism called phone number validation. Phone number validation is a key
concept in VIPR. There are several models for this validation as detailed
in <xref target="I-D.petithuguenin-vipr-pvp"/>.
The essential idea is that example.com will connect to
the example.net server, by asking the DHT to form a connection to example.net's
Node-ID. Once connected, example.com demands proof of ownership of the
phone number. This proof comes in the form of demonstrated knowledge
of the previous PSTN call. When a call was placed from example.com to +1
408 555 5432, the details of that call - including its caller ID,
start time, and stop time, create a shared secret referred to as a "ticket", -
information that is only known to entities that participated in the
call. Thus, to obtain proof that example.net really owns the number in
question, example.com will demand a knowledge proof - that example.net is aware
of the details of the call.
A consequence of this is that the following property is
maintained:</t>
<t><list style="empty">
<t>A domain can only call a specific number over SIP, if it had
previously called that exact same number over the PSTN.</t>
</list></t>
<t>This property is key in fighting spam and denial-of-service
attacks. Because calling numbers on the PSTN costs money -
especially international calls - VIPR creates a financial
disincentive for spammers. For a spammer to ring every phone in a
domain with a SIP call, it must have previously called every number
in the domain with a PSTN call, and had a successfully completed
call to each and every one of them.
<xref target="I-D.petithuguenin-vipr-sip-antispam"/>
provides an
overview and further details on
the security mechanisms for VIPR for mitigation of SPAM. </t>
<t>There are a great many details required for this validation
protocol to be secured.
For example, the mechanism needs to handle the fact that call start
and stop times won't exactly match on both sides. It needs to deal
with the fact that many calls start on the top of the hour. It needs
to deal with the fact that caller ID is not often delivered, and
when it is delivered, is not reliable. It needs to deal with the
fact that example.com may in fact be the attacker, trying to use the
validation protocol to extract the shared secret from example.net. All of
this is, in fact, handled by the protocol. The protocol is based on
the Secure Remote Password for TLS Authentication (SRP-TLS) <xref
target="RFC5054"></xref>, and is described more fully in <xref
target="I-D.petithuguenin-vipr-pvp"></xref>.</t>
<t>Towards the end of the validation process, domains example.com and
example.net had determined that each was, in fact in possession of the
shared secret information about the prior PSTN call. However,
neither side has any information about the domain names of the other
side. </t>
<t>At the end of the validation process, both example.com and example.net have
been able to ascertain that the other side did in fact participate
in the previous PSTN call. At that point, example.com sends its domain
name to example.net as shown in <xref target="fig-ticket-step1"/>.</t>
<figure anchor="fig-ticket-step1" title="Ticket Validation Step 1">
<artwork><![CDATA[
+-------+ +-------+
| Call | | Call |
example.org | Agent | | Agent | example.edu
| | | |
+-------+ +-------+
\ /
+----------------------+ \ /
| Hi, I am example.com.| \ /
| How do I reach you? | \ /
+--------------\-------+ //-------\\
\ // \\
+===\======>========>========>=====+
^ | Internet | |
| | | v
+-------+ |\\ //| +-------+
| Call |------ \\ _______//------| Call |
//\\ | Agent | | Agent | //\\
\ / | | | | \ /
\/ ---| | | |---- \/
Alice | | | | Bob
+-------+ +-------+
example.com example.net
]]></artwork>
</figure>
<t>Next, the example.net domain generates the ticket. The ticket has three
fundamental parts to it:</t>
<t><list style="numbers">
<t>The phone number that was just validated - in this case, +1
408 555 5432.</t>
<t>The domain name that the originating side claims it has -
example.com in this case.</t>
<t>A signature generated by example.net, using a key known to itself
only, over the other two pieces of information.</t>
</list></t>
<t>
Then, example.net
sends to example.com - all over a secured channel - a SIP URI to use for
routing calls to this number, and a ticket, as shown in
<xref target="fig-ticket-step2"/>. The ticket is a
cryptographic object, opaque to example.com, but used by example.net to allow
incoming SIP calls. It is similar in concept to kerberos tickets -
it is a grant of access. In this case, it is a grant of access for
example.com to call +1 408 555 5432, and only +1 408 555 5432.</t>
<figure anchor="fig-ticket-step2" title="Ticket Validation Step 2">
<artwork><![CDATA[
+-------+ +-------+
| Call | | Call |
example.org | Agent | | Agent | example.edu
| | | |
+-------+ +-------+
\ /
\ / +------------------------+
\ / | Here is your ticket |
\ / | & SIP URI to reach Bob |
//-------\\ +----/-------------------+
// \\ /
+==========<========<========<===/=+
| | Internet | ^
v | | |
+-------+ |\\ //| +-------+
| Call |------ \\ _______//------| Call |
//\\ | Agent | | Agent | //\\
\ / | | | | \ /
\/ ---| | | |---- \/
Alice | | | | Bob
+-------+ +-------+
example.com example.net
]]></artwork>
</figure>
<t>The example.com call agent receives the SIP URI and ticket, and stores
both of them in an internal cache. This cache builds up slowly over
time, containing the phone number, SIP URI, and ticket, for those
numbers which are called by example.com and validated using VIPR. Because
the cache entries are only built for numbers which have actually
been called by users in the enterprise, the size of the cache
self-scales. A call agent supporting only ten users will build up a
cache proportional to the volume of numbers called by ten people,
whereas a call agent supporting ten thousand users will build up a
cache which is typically a thousand times larger.</t>
<t> This cache, containing the phone number, SIP URI
and ticket will be accessed later when Alice (or another caller
from the same call agent) makes another call to
Bob, as detailed in
<xref target="sipcall"/>.</t>
</section>
<section anchor="sipcall" title="SIP Call">
<t>At some point in the future, another call is made to +1 408 555
5432. The caller could be Alice, or it could be any other user
attached to the same call agent. This time, the call agent notes
that it has a cached entry (including the SIP URI and ticket)
for the number in question.
It is possible that there are multiple entries
for a given number. For example, both an Enterprise and Service Provider
may register the same number in the RELOAD distributed database.
It may also be possible to fork a call using the multiple entries . [Editor's
note: this requires further discussion as to whether we want to allow
multiple entries.] </t>
<t>The example.com call agent attempts to contact the SIP URI by
establishing a TCP/TLS connection to the SIP URI it learned.
If a
connection cannot be made and there are no other cached entries for the
number in question,
the call agent proceeds with the call over the PSTN.
This ensures that, in the event of an Internet failure or server
failure, the call can still proceed. Assuming the connection is
established, the example.com call agent sends a SIP INVITE to
the terminating call agent, over this newly formed secure
connection. The SIP INVITE request also contains the ticket,
placed into a new SIP header field in the message.</t>
<t>When the SIP INVITE arrives at the example.net call agent,
the call agent can extract the ticket from the new SIP header field.
This ticket is an
object, opaque to example.com, that was previously generated by the example.net
call agent as
described in <xref target="Validate-cache"/>.
example.net first verifies the signature over the
ticket. Remember that the example.net agent is the one that generated the
ticket in the first place; as such, it is in possession of the key
required to validate the signature. Once validated, it performs two
checks:</t>
<t><list style="numbers">
<t>It compares the phone number in the call setup request (the
Request URI) against the phone number stored in the ticket.</t>
<t>It compares the domain name of the calling domain, learned
from the certificates in the mutual TLS exchange, against the
domain name stored in the ticket.</t>
</list></t>
<t>If both match, the example.net call agent knows that the calling party
is in fact the domain they claimed previously, and that they had in
fact gone through the validation process successfully for the number
in question. At this time, the call is now completed per normal SIP
processing. </t>
</section>
</section>
</section>
<section title="Security Considerations">
<t>This section provides an
overview of some of the key threats and how they are handled at a high level. Note that the detailed
security solutions to handle the threats are detailed in the other relevant VIPR documents
as referenced in the sections below.</t>
<section title="Attacks on the DHT">
<t>Attackers could attempt to disrupt service through a variety of
attacks on the DHT.</t>
<t>Firstly, it must be noted that the DHT is never used at call setup
time. It is accessed as a background task, solely to learn NEW numbers
and SIP URIs that are not already known. If an
attacker was able to completely destroy the P2P network, it would not result in a
single call to fail. Furthermore, it would not cause calls to revert
to the PSTN - calls to SIP URIs learned previously would still go over
the IP network. The only impact to such a devastating attack is that
a domain could not learn SIP URIs for new numbers, until the DHT is
restored to service. This service failure is hard for users and
administrators to even notice.</t>
<t>That said, VIPR prevents many of these attacks. The DHT itself is
secured using TLS - its usage is mandatory. Quota mechanisms are put
into place that prevent an attacker from storing large amounts of data
in the DHT as described in <xref target="I-D.petithuguenin-vipr-proportional-quota"/>.
Other attacks are prevented by mechanisms defined by
RELOAD <xref target="I-D.ietf-p2psip-base"/> itself, and are not VIPR specific.</t>
</section>
<section anchor="theft" title="Theft of Phone Numbers">
<t>A key security threat that VIPR is trying to address is the theft
of phone numbers. In particular, a malicious domain could store, in
the DHT, phone numbers that it does not own, in an attempt to steal
calls targeted to those numbers. This attack is prevented by the core
validation mechanism as described in <xref target="I-D.petithuguenin-vipr-pvp"/> ,
which performs a proof of knowledge check to
verify ownership of numbers.</t>
<t>An attacker could try to claim numbers it doesn't own, which are
claimed legitimately by other domains in the VIPR network. This attack
is prevented as well. Each domain storing information into the DHT can
never overwrite information stored by another domain. As a
consequence, if two domains claim the same number, two records are
stored in the DHT. An originating domain will validate against both,
and only one will validate - the real owner.</t>
<t>An attacker could actually own a phone number, use it for a while,
validate with it, and build up a cache of routes at other domains.
Then, it gives back the phone number to the PSTN provider, who
allocates it to someone else. However, the attacker still claims
ownership of the number, even though they no longer have it. This
attack is prevented by expiring the learned routes after a while.
Typically, operators do not re-assign a number for a few months, to
allow out-of-service messages to be played to people that still have
the old number. Thus, the TTL for cached routes is set to match the
duration that carriers typically hold numbers.</t>
<t>An attacker could advertise a lot of numbers, most of which are
correct, some of which are not. VIPR prevents this by requiring each
number to be validated individually.</t>
<t>An attacker could make a call so they know the call details of the
call they made and use this to forge a validation for that call. They
could then try to convince other users, which would have to be in the
same domain as the attacker, to trust this validation. This is
mitigated by not sharing validations inside of domains where the users
that can originate call from that domain are not trusted by the
domain.</t>
</section>
<section title="Spam">
<t>Another serious concern is that attackers may try to launch SIP
spam (also known as SPIT) calls into a domain. As described
in <xref target="Validate-cache"/> and as detailed in
<xref target="I-D.petithuguenin-vipr-sip-antispam"/>, VIPR prevents this by
requiring that a domain make a PSTN call to a number before it will
allow a SIP call to be accepted to that same number. This provides a
financial disincentive to spammers. The current relatively high cost
of international calling, and the presence of national do-not-call
regulations, have prevented spam on the PSTN to a large degree. VIPR
applies those same protections to SIP connections.</t>
<t>VIPR still lowers the cost of communications, but
it does so by amortizing that savings over a large number of calls.
The costs of communications remain high for infrequent calls to many
numbers, and become low for frequent calls to a smaller set of
numbers. Since the former is more interesting to spammers, VIPR gears
its cost incentives away from the spammers, and towards domains which
collaborate frequently.</t>
<t> It is important to note that VIPR does not completely address the
spam problem. A large spamming clearing house organization could
actually incur the costs of launching the PSTN calls to numbers, and
then, in turn, act as a conduit allowing other spammers to launch
their calls to those numbers for a fee. The clearinghouse would
actually need to transit the signaling traffic (or, divulge the
private keys to their domain name), which would incur some cost. As
such, while this is not an impossible situation, the barrier is set
reasonably high to start with - high enough that it is likely to
deter spammers until it becomes a highly attractive target, at which
point other mechanisms can be brought to bear. </t>
</section>
<section title="Eavesdropping">
<t>Another class of attacks involves outsiders attempting to listen in
on the calls that run over the Internet, or obtain information about
the call through observation of signaling.</t>
<t>All of these attacks are prevented by requiring the usage of SIP
over TLS and SRTP. These are mandatory to use.</t>
</section>
</section>
<section title="IANA Considerations">
<t>This specification does not require any actions from IANA.</t>
</section>
<section title="Acknowledgements">
<t>Thanks for review comments from Ken Fischer, Rob Maidhof, Michael
Procter, Eric Burger, Richard Barnes and others. Thanks to Theo Zourzouvillys for pointing out the 5th
theft of phone numbers attack as described in <xref target="theft"/> .</t>
</section>
</middle>
<back>
<references title="Normative References">
<?rfc include="reference.RFC.2119"?>
<?rfc include="reference.RFC.3261"?>
<?rfc include="reference.I-D.ietf-p2psip-base"?>
<?rfc include="reference.I-D.petithuguenin-vipr-reload-usage"?>
<?rfc include="reference.I-D.petithuguenin-vipr-framework"?>
<?rfc include="reference.I-D.petithuguenin-vipr-sip-antispam"?>
<?rfc include="reference.I-D.jennings-vipr-vap"?>
<?rfc include="reference.I-D.petithuguenin-vipr-pvp"?>
<?rfc include="reference.I-D.petithuguenin-vipr-proportional-quota"?>
</references>
<references title="Informative References">
<?rfc include="reference.RFC.2543"?>
<?rfc include="reference.RFC.3263"?>
<reference anchor="E.164" target="">
<front>
<title>
The International Public Telecommunication Number Plan
</title>
<author>
<organization>ITU-T</organization>
</author>
<date day="" month="May" year="1997" />
</front>
<seriesInfo name='Recommendation' value='E.164' />
</reference>
<?rfc include="reference.RFC.5039"?>
<?rfc include="reference.RFC.6116"?>
<?rfc include="reference.RFC.5067"?>
<?rfc include="reference.RFC.5054"?>
<!-- Using the Secure Remote Password (SRP) Protocol for TLS -->
<?rfc include="reference.RFC.6873" ?>
<?rfc include="reference.I-D.jones-insipid-session-id"?>
<?rfc include="reference.I-D.ietf-insipid-session-id-reqts"?>
</references>
<section title="Changes since last version">
<t>This section must be removed before publication as an RFC.</t>
<t> Modifications between jennings-04 and jennings-03: </t>
<t> <list style="numbers">
<t> Updating references to SIPCLF and Session ID (INSIPID) documents. </t>
</list>
</t>
<t> Modifications between jennings-03 and jennings-02: </t>
<t> <list style="numbers">
<t> Reworded REQ -11 to clarify that in the case of call failures (i.e., IP calls),
the system should fallback to inter-domain calling prior to VIPR. </t>
<t> Deleted REQ-12 (Handover) since it's really not specific functionality provided by VIPR. </t>
<t> Moved some text from the -01 version in the Technical Overview section
back into the doc (not
sure why it was removed previously). </t>
<t> Other editorial changes: </t>
</list>
</t>
<t> <list style="empty">
<t> - Added a Terminology section.</t>
<t> - Clarified the use of the term "Call Agent".</t>
<t> - Reworded discussion of email in section 2.2 (i.e., it's not useless). </t>
<t> - Either changed or removed altogether terms like "neat", "clever",
"incredible", "enormous" and
any text that read like marketing literature as much as possible.</t>
<t> - Removed some of the more subjective and superfluous language - i.e.,
condensed the text to be more concise (Section 5.2 and many others per
the previous change) </t>
<t> - Deleted explicit reference to "SIP Trunking" as the statement didn't
introduce additional information in that paragraph and the term is not
defined in this document. </t>
<t> - and other minor editorial fixes. </t>
</list>
</t>
<t> Modifications between jennings-02 and jennings-01:</t>
<t> <list style="numbers">
<t> Sections 6,7,8 moved to new VIPR framework document. </t>
<t> Editorial changes. </t>
<t> Clarifications to re-enforce that the primary objective is not PSTN bypass but
rather to enable enhanced services such as video between domains. Changed "VoIP" to
"SIP" since the focus is not specifically voice. </t>
<t> Added reference for new framework document. </t>
<t> Section 5.3: Added references to other documents as appropriate - e.g., -pvp, -spam, etc.</t>
<t> Moved validation diagrams and text (from 5.3.4) into Validation and
caching section (5.3.3). </t>
<t> Condensed discussion of spam in section 5.3.3 and updated SPAM section in
security section. </t>
</list>
</t>
<t> Modifications between jennings-01 and rosenberg-04:</t>
<t> <list style="symbols">
<t> Not specified. </t>
</list>
</t>
<t> Modifications between rosenberg-04 and rosenberg-03 </t>
<t><list style="symbols">
<t>Nits.</t>
<t>Shorter I-Ds references.</t>
<t>Changed phone numbers to follow E.123 presentation.</t>
<t>Expanded P2P initialisms.</t>
<t>Uses +1 408 555 prefix for phone numbers in examples.</t>
</list></t>
</section>
</back>
</rfc>| PAFTECH AB 2003-2026 | 2026-04-23 10:59:17 |