One document matched: draft-roach-mmusic-pof-pan-00.xml
<?xml version="1.0" encoding="US-ASCII"?>
<!-- vim: set tw=78 : -->
<!DOCTYPE rfc SYSTEM "rfc2629.dtd" [
<!ENTITY RFC2119 SYSTEM "http://xml.resource.org/public/rfc/bibxml/reference.RFC.2119.xml">
<!ENTITY RFC3261 SYSTEM "http://xml.resource.org/public/rfc/bibxml/reference.RFC.3261.xml">
<!ENTITY RFC3264 SYSTEM "http://xml.resource.org/public/rfc/bibxml/reference.RFC.3264.xml">
<!ENTITY RFC4566 SYSTEM "http://xml.resource.org/public/rfc/bibxml/reference.RFC.4566.xml">
<!ENTITY RFC5888 SYSTEM "http://xml.resource.org/public/rfc/bibxml/reference.RFC.5888.xml">
<!ENTITY RFC6086 SYSTEM "http://xml.resource.org/public/rfc/bibxml/reference.RFC.6086.xml">
<!ENTITY RFC3311 SYSTEM "http://xml.resource.org/public/rfc/bibxml/reference.RFC.3311.xml">
<!ENTITY sdpfrag SYSTEM 'http://xml.resource.org/public/rfc/bibxml3/reference.I-D.ivov-dispatch-sdpfrag.xml'>
]>
<?xml-stylesheet type="text/xsl" href="rfc2629.xslt" ?>
<?rfc toc="yes" ?>
<?rfc symrefs="yes" ?>
<?rfc strict="yes" ?>
<?rfc compact="yes"?>
<?rfc sortrefs="yes" ?>
<?rfc colonspace="no" ?>
<?rfc rfcedstyle="no" ?>
<?rfc tocdepth="4"?>
<rfc category="std" updates="3264" docName="draft-roach-mmusic-pof-pan-00" ipr="trust200902">
<front>
<title abbrev="Partial Offers and Answers">
Using Partial Offers and Partial Answers in a Multimedia Session
</title>
<author fullname="Adam Roach" initials="A. B." surname="Roach">
<organization>Mozilla</organization>
<address>
<postal>
<street></street>
<city>Dallas</city>
<region>TX</region>
<code></code>
<country>US</country>
</postal>
<phone>+1 650 903 0800 x863</phone>
<email>adam@nostrum.com</email>
</address>
</author>
<date day="14" month="October" year="2013" />
<area>RAI</area>
<abstract>
<t>
Whenever two hosts have the ability to set up and control a session on
a peer-to-peer basis, situations can arise in which both parties
attempt to change session parameters "at the same time," such that the
session control messages cross on the wire. When this happens,
implementations need to invoke extraordinary procedures to return the
shared state of the session to a common view between the endpoints.
</t>
<t>
For real-time communications, these session control messages are
typically exchanged using the session description protocol (SDP),
using an Offer/Answer model. This document expands the offer/answer
model to include the ability to exchange information relating to
discrete media streams within the session. By reducing the amount of
session data, the frequency of session state conflicts can be reduced;
and, for certain types of operations, conflicts can be eliminated
altogether.
</t>
<t>
This document updates RFC 3264.
</t>
</abstract>
</front>
<middle>
<section anchor="sec.introduction" title="Introduction">
<t>
The SDP <xref target="RFC4566"/> offer/answer model defined in <xref
target="RFC3264"/> briefly mentions "glare" as a potential issue in
the use of offer/answer exchanges, although it relegates the problem
to the "higher layer protocol" to resolve. In SIP <xref
target="RFC3261"/>, resolving state after a glare condition is
performed via a timer-based back-off mechanism. For WebRTC, detection
of glare comes in the form of an "InvalidStateError" exception.
Actual resolution of glare is currently undefined; the present
assumption is that the applications that make use of RTCWEB are
responsible for handling glare in a sensible fashion.
</t>
<t>
The penalty for glare isn't simply code complexity; it results in
delays in updating sessions state, which can end up visible to users,
leading to a less optimal user experience.
</t>
<t>
Many of the emerging uses for both SIP and RTCWEB involve sessions
with a large number of media streams, with streams being added and
removed frequently. This kind of session churn increases the incidence
of glare significantly.
</t>
<t>
To reduce the incidence of glare under these circumstances, this
document defines a procedure via which partial offer/answer exchanges
may take place. These exchanges operate on one or more media sections
at a time, rather than an entire SDP body. These operations are
defined in a way that can completely avoid glare for stream additions
and removals, and which reduces the chance of glare for changes to
active streams. This approach requires all media sections to contain
an "a=mid" <xref target="RFC5888"/> attribute.
</t>
<t>
This document focuses on the application of this technique for use in
RTCWEB and WebRTC. The author anticipates that future work will
describe its use in conjunction with SIP and SIP-derived technologies
(such as multiparty conferencing and telepresence).
</t>
</section>
<section title="Terminology">
<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"/>.
</t>
</section>
<section title="Mechanism Overview">
<t>
The core of this mechanism is the concept of "partial offers" and
"partial answers." Syntactically, these entities are SDP fragments,
consisting of exactly one o= line; one or more media sections; and any
i=, c=, b=, k=, and a= lines associated with the media sections. They
are formatted exactly as they would be if they were part of a larger
SDP document, with one key exception: unlike SDP, in an SDP fragment,
the ordering of media sections relative to each other is not
significant. Note that SDP fragments contain only that information
that pertains to media. Other than the mandatory o= line, they never
contain any session-level information. Within the o= line, only the
<sess-version> field is allowed to be changed from its previous
value. Any changes to session-level information are expected to use a
full offer/answer exchange rather than the partial offer/partial
answer mechanism defined by this document.
</t>
<t>
<list style="hanging">
<t>
OPEN ISSUE: Do we need to relax this session-level prohibition?
It makes the mechanism clean. However, it does make it difficult
to add a new stream while simultaneously associating it with a
group. On the other hand, group lines don't have unique
identifiers, so just sending a single group line over can be
ambiguous. One way around this would be requiring that partial
offers and partial answers must contain all group attributes
associated with the session, but this still gets potentially
messy if we have both sides trying to update group information
at the same time. An alternative approach might be to indicate
group information only for *new* streams, and require full
offers for any other group changes. Of course, without unique
group identifiers, we're still stuck with the challenge of
unambiguously identifying which group we're adding a new line
to. If we constrain group membership so that each group can
be uniquely identified by its type and members, then that should
be sufficient. Is such a constraint acceptable?
</t>
</list>
</t>
<t>
Using this mechanism has two key prerequisites: (1) all offer/answer
exchanges in the session prior to sending a partial offer have
contained "a=mid" attributes for each media section, and (2) both
sides are known to support the partial offer/answer technique (either
because they are part of a single domain of control, or because use of
this technique has been explicitly signaled).
</t>
<t>
The use of an SDP fragment body will be explicitly signaled, e.g.,
using a different MIME type for SIP, or using a different "type" field
for the WebRTC API.
</t>
<section title="Adding a Stream">
<t>
To add a stream glarelessly, a party creates a "partial offer"
consisting of an o= line and one or more media sections, including
all of the corresponding i=, c=, b=, k=, and a= lines. Each media
section contains an "a=mid" attribute, indicating an MID that has
not yet been used in the session.
</t>
<t>
Upon receipt of a partial offer, an implementation processes each
media section independently. For each media section, the recipient
examines the MID in it. If the MID does not match any existing MID
in the session, then it represents a new media stream. Assuming the
recipient does not have an outstanding, unanswered partial offer
that also adds a stream, this new media section is simply appended
to the end of the existing session description, the SDP sess-version
is increased, and an answering media section is created. Once all
answering media sections have been processed, they are concatenated
into a partial answer. This partial answer consists of one or more
media sections, each containing an MID matching the one from the
partial offer.
</t>
<t>
If the recipient of a partial offer that contains a new MID
has also sent a partial offer adding a new stream to the session,
then ambiguity can arise regarding the canonical ordering of media
sections within the session description. In this situation, both
partial offer/answer exchanges are allowed to complete independently
(as no fundamental data glare has occurred). However, the order in
which they are appended to the session description is synchronized
by performing a lexical comparison among each media section's MID
attribute: the media sections are appended to the session in
lexically increasing order.
</t>
</section>
<section title="Changing a Stream">
<t>
Partial offers may also be generated for modification of an existing
stream. In this case, the MID in the media section of a partial
offer will match an existing MID in the session description.
</t>
<t>
Upon receipt of a partial offer, an implementation examines the MID
in it. If the MID matches any existing MID in the session, then it
represents a modification to that media section. Assuming the
recipient does not have an outstanding, unanswered partial offer
that also modifies that exact same stream, this media section is
treated as an independent renegotiation of that stream (only). The
SDP version is increased, and a partial answer is created.
This partial answer consists of an media section and its attributes,
and has an MID matching the one from the partial offer.
</t>
<t>
<list style="hanging">
<t>
OPEN ISSUE: Since stream *changes* can result in glare, the
foregoing text assumes that only one media section will be sent
for such a change. Is this okay?
</t>
</list>
</t>
<t>
If the recipient of a partial offer that contains an existing MID
has also sent a partial offer to change that exact same stream, and
neither the received nor the sent partial offer contains an
"a=inactive" attribute, then a legitimate glare condition has
arisen. Normal glare recovery procedures -- e.g., using a
tie-breaker token or a back-off timer -- must be engaged to resolve
the conflict.
</t>
</section>
<section title="Removing a Stream">
<t>
To remove one or more a streams in a way that eliminates the chance
of glare, an implementation generates a new partial offer,
containing one or more media sections. Each media section contains
an MID matching the stream it wants to remove, and indicates a
transport port of zero, indicating that the stream is being
deactivated.
</t>
<t>
If the recipient of a partial offer that contains an existing MID
has also sent a partial offer to change that exact same stream, and
either one of the received or the sent partial offer contains a port
number of zero, then the stream is deactivated. At this point, both
partial offers are discarded, the corresponding media section in the
session is modified by changing its port to zero, and a partial
answer is generated representing this single change.
</t>
</section>
</section>
<section title="Use With Other Protocols">
<t>
Note that this document simply defines the extensions to the SDP
offer/answer model for dealing with partial offers and partial
answers. In the same way that <xref target="RFC3264"/> does not
define specific SIP, JSEP, or WebRTC handling, neither does this
document. In order for this technique to be useful,
protocol-specific mechanisms need to be defined. This additional
work is left to appropriate venues, such as the W3C WebRTC WG, the
RTCWEB WG, and the SIPCORE WG. If the higher-level protocol allows
the use of unordered message delivery, it is that protocol's
responsibility to ensure that the result of partial offer/partial
answer exchanges is a shared and identical session state between the
parties involved.
</t>
<t>
To assist in understanding the mechanism being proposed, we describe,
in a very high-level and non-normative way, how this mechanism might
be applied to a couple of specific higher-level signaling systems.
</t>
<section title="High-Level Sketch: Use With JSEP/WebRTC">
<t>
For WebRTC, we envision that such additional specification would
add a new constraint to createOffer, requesting that a partial
offer be generated (if possible). The resulting
RTCSessionDescription would contain only the media sections that
have changed since the most recent offer/answer exchange, and
would have a type of "partialOffer." When createAnswer is called
after receipt of a partialOffer, it would create a partialAnswer,
containing only the media sections referenced in the partial
offer, that can be provided to the remote party.
</t>
</section>
<section title="High-Level Sketch: Use With SIP">
<t>
For SIP, partial offers and partial answers will likely be
provided in SIP UPDATE <xref target="RFC3311"/> or INFO <xref
target="RFC6086"/> messages, containing a special
"application/sdpfrag" MIME type <xref
target="I-D.ivov-dispatch-sdpfrag"/>, and a content-disposition
that indicates that the contents are a partial offer (rather than,
say, a trickle ice candidate). Although INVITE may seem like a
natural fit for this kind of behavior, its current definition
includes strong glare resolution behaviors that makes it
unsuitable for this purpose. Naturally, any such mechanism will
be paired with a SIP feature tag that allows for negotiation of
support for partial offers and answers.
</t>
</section>
</section>
<section title="Protocol Operation">
<t>
The following sections formally defines the procedures for generating
and processing partial offers and partial answers.
</t>
<t>
At any time during an ongoing session, either agent in the session MAY
generate a new partial offer that updates the session, subject to the
restrictions described in the following sections. However, it MUST
NOT generate a new partial offer if it has received any partial or
full offer which it has not yet answered or rejected.
</t>
<t>
An agent also MUST NOT generate a partial offer if it has sent a
partial or full offer which has not yet been accepted or
rejected.
</t>
<t>
<list style="hanging">
<t>
OPEN ISSUE: It seems like we might be able to have multiple
outstanding sent partial offers at once, as long as they don't try
to act on the same media section. The reason it's disallowed in
the above paragraph is that having several partial offers
potentially outstanding in both directions makes it very, very,
very complicated to resolve the ordering of media sections if
these partial offers in opposite directions overlap temporally.
</t>
</list>
</t>
<t>
In the situations described as "glare" below, the higher layer
protocol needs to provide a means for resolving such conditions. This
will generally be the same mechanism used to resolve the glare
conditions described in <xref target="RFC3264"/>.
</t>
<section title="Common Procedures">
<t>
For all of the procedures described in the following sections,
whenever an o= line is included in a partial offer or partial
answer, its <username>, <sess-id>, <nettype>,
<addrtype>, and <unicast-address> values MUST be identical to
those sent in the most recent full offer or full answer generated by
this agent for this session. The <sess-version> value MUST be
larger than the value in all previously sent offers, partial offers,
answers, and partial offers generated by this agent for this
session.
</t>
<t>
Whenever the procedures in the following sections indicate that a
media section is to be included in a partial offer or partial
answer, that media section MUST consist of an m= line along with
all i=, c=, b=, k=, and a= lines associated with that media section.
If a line is absent from a media section in a partial offer or
partial answer, it MUST be interpreted as an explicit removal of
that value from the media section. Recipients of such messages MUST
NOT assume that a previously-established but omitted value is still
in effect.
</t>
</section>
<section title="Generating a Partial Offer">
<t>
Whenever an agent wishes to change the state of the media in an
ongoing session -- whether through addition, modification, or
removal of a stream -- it does so through either an offer or a
partial offer. In deciding which to use, the implementation first
verifies that it has received positive confirmation that the remote
implementation supports the partial offer/partial answer mechanism.
The means of negotiating such support is left to the higher-level
protocol that makes use of the offer/answer model. The
implementation then verifies that all media sessions in the current
session are associated with unique MID values. Finally, the
implementation evaluates whether the changes it needs to make can be
performed exclusively using the values present in a media section,
without any modifications necessary to session-level values (except
for the sess-version value on the session-level o= line).
</t>
<t>
If all three of the criteria described above are true, then the
implementation MAY send a partial offer to make the changes it wants
to request. If any of these criteria are not true, then the
implementation MUST use a full offer, according to the procedures
described in <xref target="RFC3264"/>.
</t>
<t>
Once the agent determines that the change it wishes to make is
eligible to use the partial offer mechanism, it forms a new SDP
fragment by following these steps:
</t>
<t>
<list style="numbers">
<t>
For each desired change to an existing media section, the agent
creates a new partial offer consisting of one o= line and a
single media section. This media section MUST contain an
"a=mid" attribute containing an MID that matches the media
section that is being modified. The media section also contains
the modifications that the agent wishes to make, as described in
section 8.3 of <xref target="RFC3264"/>. This partial offer is
sent in isolation, with no other media section changes,
additions, or removals in the same partial offer.
</t>
<t>
If the desired change involves one or more media section
additions or removals, the agent creates a new partial offer
consisting of one o=line and any media section described in the
following two steps.
</t>
<t>
For each new media section to be added, the agent creates a new
media section to be added to the aforementioned partial offer.
This media section MUST contain an "a=mid" attribute, and the
MID present in this attribute MUST contain at least 32
characters chosen randomly from full set of 79 characters
allowed in a token. The remainder of the media section contains
the various values that the agent wishes to have associated with
the corresponding media, and is created according to the
procedures described in section 5.1 or 5.2 of <xref
target="RFC3264"/>, as appropriate.
</t>
<t>
For each existing media section to be removed, the agent
creates a new media section to be added to the aforementioned
partial offer. This media section MUST contain an "a=mid"
attribute containing an MID that matches the media section that
is being removed, and MUST contain a <port> value of 0
(zero). Except for the required "a=mid" attribute, this media
section MAY omit any or all i=, c=, b=, k=, and a= lines, and
MAY list only one m= line <fmt> value.
</t>
</list>
</t>
<t>
Once the preceding steps have been followed to create one or more
partial offers, the agent makes use of the high-level signaling
protocol to convey the offers to the remote agent, one at a time.
</t>
</section>
<section title="Processing a Partial Offer">
<t>
Upon receipt of a partial offer, an agent first determines whether
it has sent any full offers for the corresponding session. If it
has, then the partial offer represents a glare condition that is
resolved via the higher-level protocol. It then verifies whether it
has received any partial or full offers to which it has not yet sent
an answer or a rejection. If so, then it rejects the partial offer
as invalid behavior.
</t>
<t>
The agent then examines the o= line in the received partial offer.
If If the <sess-version> value is less than the most recently
received full (non-partial) offer or answer, then the partial offer
is stale and MUST be rejected. The means for rejecting the partial
offer are left to the higher-level protocol.
</t>
<t>
After such validation takes place, the agent iterates through each
media section and performs the following steps:
</t>
<t>
<list style="numbers">
<t>
If the MID present in the received media section matches a media
section already present in the ongoing session and has a
non-zero port number, it represents a change to an existing
media stream.
<list style="symbols">
<t>
If the partial offer contains more than one media section,
then the recipient MUST reject the partial offer as invalid
behavior.
</t>
<!-- This isn't allowed at the moment
<t>
If the received MID matches an MID for any partial offer
that it has received but not yet answered or rejected, then
the recipient MUST reject the partial offer as invalid
behavior.
</t>
-->
<t>
If the MID matches the MID of a media section in a partial
offer that the agent has sent, AND the sent media section
contains a port number of zero, then the incoming partial
offer is rejected, as any such changes have been "overtaken
by events:" the stream will be deactivated momentarily.
</t>
<t>
The recipient verifies that the MID does not match the MID
of any media section in any partial offers that it has sent
but has not yet received a partial answer or rejection for,
unless the media section in the sent partial offer has a
port number of zero. If this verification fails, then the
received partial offer represents a glare condition that is
resolved via the higher-level protocol.
</t>
<t>
After the preceding verifications have succeeded, the agent
creates media section to include in the partial answer. To
reject the media section in the partial offer, the agent
generates a media section with a port number set to zero;
otherwise, the agent forms the media section by following
the procedures described in section 6.1 or 6.2 of <xref
target="RFC3264"/>, as appropriate.
</t>
</list>
</t>
<t>
If the MID present in the received media section matches a media
section already present in the ongoing session and has a port
number of zero, then it represents the removal of an existing
media stream. The agent creates a media section to include in
the partial answer. With the exception of the "a=mid" attribute,
this media section MAY omit any or all i=, c=, b=, k=, and a=
lines, and MAY indicate a single payload type.
<!-- Multiple POFs not currently allowed
<list style="symbols">
<t>
If the received MID matches an MID for any partial offer
that it has received but not yet answered or rejected, then
the recipient MUST reject the partial offer as invalid
behavior.
</t>
<t>
If the above validation succeeds, then the agent creates a
media section to include in the partial answer. With the
exception of the "a=mid" attribute, this media section MAY
omit any or all i=, c=, b=, k=, and a= lines, and MAY
indicate a single payload type.
</t>
</list>
-->
</t>
<t>
If the MID present in the received media section does not match
any media section already present in the ongoing session, then
it represents a new media stream.
<list style="symbols">
<!-- Multiple POFs not currently allowed
<t>
If the received MID matches an MID for any partial offer
that it has received but not yet answered or rejected, then
the recipient MUST reject the partial offer as invalid
behavior.
</t>
-->
<t>
If the received media section contains a port number of
zero, then the recipient MUST reject the partial offer as
invalid behavior: this mechanism does not support the
atomic addition and removal of the same stream.
</t>
<t>
If the above validation succeeds, the agent creates a media
section to include in the partial answer. To reject the
media section in the partial offer, the agent generates a
media section with a port number set to zero; otherwise, the
agent forms the media section by following the procedures
described in section 6.1 or 6.2 of <xref target="RFC3264"/>,
as appropriate.
</t>
</list>
</t>
</list>
</t>
<t>
All media sections that are formed in the foregoing steps MUST
contain an "a=mid" attribute matching the MID that was present in
the corresponding media section from the partial offer.
</t>
<t>
If the preceding steps have been performed for each media
section without resulting in a rejection, then the agent forms a
partial answer consisting of a single o= line, and all of the media
sections that were generated as part of the preceding steps. Note
that this processing will always yield the same number of media
sections in a partial answer as were present in the partial offer.
Unlike normal SDP processing, however, the order of the media
sections in a partial answer is not significant. This partial answer
is then sent to the remote agent using the high-level protocol.
</t>
<t>
If the above processing results in a successful partial answer, then
the agent's view of the session is updated as described in <xref
target="sess-update"/>.
</t>
</section>
<section title="Processing a Partial Answer">
<t>
When a partial answer is received, the offerer matches each media
section in the partial answer to its corresponding media section
according to its MID. The agent MUST NOT assume that the order of the
sections in the received partial answer matches the order of the
sections it sent in the partial offer. However, it can expect that
each section in the partial offer has a corresponding section in the
receive partial answer.
</t>
<t>
For each media section, the agent then updates its local view of
session state as described in <xref target="sess-update"/>, and
follows the process described in section 7 of <xref
target="RFC3264"/>.
</t>
</section>
<section anchor="sess-update"
title="Updating the Shared View of Session State">
<t>
Whenever a partial offer or partial answer is processed, the agent
performs the following steps to ensure that a common view of session
state is maintained:
</t>
<t>
<list style="numbers">
<t>
The remote session's <sess-version> value is updated
according to the value received in the o= line of the sdpfrag.
</t>
<t>
Any changed or removed media sections are modified in-place.
Their position in the overall session description remains the
same as it was before.
</t>
<t>
Any added media sections are appended to the existing session.
The order in which they are appended is determined by lexically
sorting them according to their MID values. This is not
necessarily the same order in which they appear in the sdpfrag.
If the recipient of a partial offer had a sent a partial offer
to which it had not yet received a response when the partial
offer was received, then it must take additional steps to ensure
a common view of the media section ordering: the media sections
for the sent partial offer and the received partial answer are
treated as a single list, sorted lexically according to their
respective MID values, and appended to the session in that
order. When that agent receives the corresponding partial
answer, the media section ordering remains the same as was
established by the partial offer.
</t>
<t>
<list style="hanging">
<t>
Note that this requires the ability to re-index media
sections in the case that the remote party rejects the
outstanding partial offer that we sent them (I don't mean
declining the line by setting the port to zero; I mean that
the higher-level protocol actually rejected the offer, like
getting an unrecoverable 400- or 500-class error in SIP).
OPEN ISSUE: This is kinda ugly. Is there maybe some better
way to handle the issue?
</t>
</list>
</t>
</list>
</t>
</section>
<section title="Receiving a Full Offer with a Partial Offer Pending">
<t>
For completeness, this document notes that an agent that receives a
full offer with a sent partial offer pending is in a glare
condition; this is resolved via the higher-level protocol.
</t>
</section>
</section>
<section title="Examples">
<section title="Adding Streams">
<t>
TBD
</t>
</section>
<section title="Removing Streams">
<t>
TBD
</t>
</section>
<section title="Changing a Stream">
<t>
TBD
</t>
</section>
<section title="Both Sides Simultaneously Add Streams">
<t>
TBD
</t>
</section>
<section title="Removing a Stream with Pseudo-Glare">
<t>
TBD (by "pseudo-glare", I mean that one side tries to change a stream
at the same time as the other one tries to remove it)
</t>
</section>
<section title="Changing a Stream with Glare">
<t>
TBD
</t>
</section>
</section>
<section anchor="sec.security" title="Security Considerations">
<t>
TBD
</t>
</section>
<section anchor="sec.iana" title="IANA Considerations">
<t>
TBD -- I don't think we actually need any for this mechanism.
</t>
</section>
</middle>
<back>
<references title="Normative References">
&RFC2119;
&RFC3261;
&RFC3264;
&RFC4566;
&RFC5888;
</references>
<references title="Informative References">
&RFC6086;
&RFC3311;
&sdpfrag;
</references>
</back>
</rfc>
| PAFTECH AB 2003-2026 | 2026-04-24 02:59:33 |