<abstract>This document defines a framework for initiating and managing peer-to-peer multimedia sessions (e.g., voice and video exchanges) between Jabber/XMPP clients in a way that is interoperable with existing Internet standards.</abstract>
<remark><p>Added creator attribute to content element for prevention of race condition; modified spec to use provisional namespace before advancement to Draft (per XEP-0053).</p></remark>
<remark><p>Made several corrections to the state machines and examples.</p></remark>
</revision>
<revision>
<version>0.9</version>
<date>2006-09-08</date>
<initials>ram/psa</initials>
<remark><p>Further cleaned up state machines and state-related actions.</p></remark>
</revision>
<revision>
<version>0.8</version>
<date>2006-08-23</date>
<initials>ram/psa</initials>
<remark><p>Changed channels to components in line with ICE; changed various action names for consistency; added session-extend and session-reduce actions to add and remove description/transport pairs; added description-modify action; added sender attribute to specify directionality.</p></remark>
</revision>
<revision>
<version>0.7</version>
<date>2006-07-17</date>
<initials>psa</initials>
<remark><p>Added implementation note about handling multiple content types.</p></remark>
</revision>
<revision>
<version>0.6</version>
<date>2006-07-12</date>
<initials>se/psa</initials>
<remark><p>Changed media type to content type.</p></remark>
</revision>
<revision>
<version>0.5</version>
<date>2006-03-20</date>
<initials>psa/jb</initials>
<remark><p>Further clarified state machine diagrams; specified that session accept must include agreed-upon media format and transport description; moved deployment notes to appropriate transport spec.</p></remark>
<remark><p>Provided more detail about modify scenarios; defined media-specific and transport-specific actions and adjusted state machine accordingly.</p></remark>
</revision>
<revision>
<version>0.2</version>
<date>2006-02-13</date>
<initials>psa/jb</initials>
<remark><p>Per agreement among the co-authors, moved transport definition to separate specification, simplified state machine, and made other associated changes to the text, examples, and schemas; also harmonized redirect, decline, and terminate processes.</p></remark>
<remark><p>Restructured document flow; provided example of burst mode.</p></remark>
</revision>
<revision>
<version>0.0.8</version>
<date>2005-12-05</date>
<initials>psa/sl/jb</initials>
<remark><p>Distinguished between dribble mode and burst mode, including mode attribute, service discovery features, and implementation notes; provided detailed resource discovery examples; corrected state chart; specified session termination; specified error conditions; specified semantics of informational messages; began to define security considerations; added Joe Beda as co-author.</p></remark>
</revision>
<revision>
<version>0.0.7</version>
<date>2005-11-08</date>
<initials>psa</initials>
<remark><p>Added more detail to basic session flow; harmonized candidate negotiation process with ICE.</p></remark>
<p>There exists no widely-adopted standard for initiating and managing peer-to-peer (p2p) interactions (such as voice, video, or file sharing exchanges) from within Jabber/XMPP clients. Although several large service providers and Jabber/XMPP clients have written and implemented their own proprietary XMPP extensions for p2p signalling (usually only for voice), those technologies are not open and do not always take into account requirements to interoperate with the Public Switched Telephone Network (PSTN) or Voice over Internet Protocol (VoIP) networks based on the IETF's Session Initiation Protocol (SIP) as specified in &rfc3261; and its various extensions.</p>
<p>By contrast, the only existing open protocol has been &xep0111;, which made it possible to initiate and manage p2p sessions, but which did not provide enough of the key signalling semantics to be easily implemented in Jabber/XMPP clients. <note>It is true that TINS made it relatively easy to implement an XMPP-to-SIP gateway; however, in line with the long-time Jabber philosophy of "simple clients, complex servers", it would be better to force complexity onto the server-side gateway and to keep the client as simple as possible.</note></p>
<p>The result has been an unfortunate fragmentation within the XMPP community regarding signalling protocols. There are, essentially, two approaches to solving the problem:</p>
<ol>
<li>Recommend that all client developers implement a dual-stack (XMPP + SIP) solution.</li>
<li>Define a full-featured protocol for XMPP signalling.</li>
<p>Implementation experience indicates that a dual-stack approach may not be feasible on all the computing platforms for which Jabber clients have been written, or even desirable on platforms where it is feasible. Therefore, it seems reasonable to define an XMPP signalling protocol that can provide the necessary signalling semantics while also making it relatively straightforward to interoperate with existing Internet standards.</p>
<p>As a result of feedback received on <cite>XEP-0111</cite>, the original authors of this document (Joe Hildebrand and Peter Saint-Andre) began to define such a signalling protocol, code-named Jingle. Upon communication with members of the Google Talk team, <note>Google Talk is a messaging and voice chat service and client provided by Google; see <<linkurl='http://www.google.com/talk/'>http://www.google.com/talk/</link>>.</note> it was discovered that the emerging Jingle approach was conceptually (and even syntactically) quite similar to the signalling protocol used in the Google Talk application. Therefore, in the interest of interoperability and adoption, we decided to harmonize the two approaches. The signalling protocol specified herein is, therefore, substantially equivalent to the existing Google Talk protocol, with several adjustments based on feedback received from implementors as well as for publication within the Jabber Software Foundation's standards process.</p>
<p>The purpose of Jingle is not to supplant or replace SIP. Because dual-stack XMPP+SIP clients are difficult to build, given that they essentially have two centers of program control, <note>For example, one large ISP recently decided to switch to a pure XMPP approach after having implemented and deployed a dual-stack client for several years.</note> we have designed Jingle as a pure XMPP signalling protocol. However, Jingle is intended to interwork with SIP so that the millions of deployed XMPP clients can be added onto the existing open VoIP networks, rather than limiting XMPP users to a separate and distinct VoIP network.</p>
<li>Make it possible to manage a wide variety of peer-to-peer sessions (not limited to voice and video) within XMPP. <note>Possible other content description formats include file sharing, application casting, application sharing, whiteboarding, torrent broadcasting, shared real-time editing, and distributed musical performance, to name but a few.</note>.</li>
<li>Clearly separate the signalling channel (XMPP) from the data channel (e.g., Real-time Transport Protocol as specified in &rfc3550;).</li>
<li>Clearly separate the content description formats (e.g., for voice chat) from the content transport methods (e.g., User Datagram Protocol as specified in &rfc0768;).</li>
<li>Make it possible to add, modify, and remove content types from an existing session.</li>
<li>Make it relatively easy to implement support for the protocol in standard Jabber/XMPP clients.</li>
<li>Where communication with non-XMPP entities is needed, push as much complexity as possible onto server-side gateways between the XMPP network and the non-XMPP network.</li>
</ol>
<p>This document defines the signalling protocol only. Additional documents specify the following:</p>
<li><p>Various content description formats (audio, video, etc.) and, where possible, mapping those types to the Session Description Protocol (SDP; see &rfc4566;); examples include &xep0167; and &xep0180;.</p></li>
<li><p>Various content transport methods; examples include &xep0176; and &xep0177;.</p></li>
<li><p>Procedures for mapping the Jingle signalling protocol to existing signalling standards such as the IETF's Session Initiation Protocol (SIP; see &rfc3261;) and the ITU's H.323 protocol (see &h323;); these documents are not yet written</p></li>
<td>A number of pairs of negotiated content transport methods and content description formats connecting two entities. It is delimited in time by an initiate request and session ending events. During the lifetime of a session, pairs of content descriptions and content transport methods can be added or removed. A session consists of at least one active negotiated content type at a time.</td>
<td>The format of the content type being established, which formally declares one purpose of the session (e.g., "voice" or "video"). This is the 'what' of the session (i.e., the bits to be transferred), such as the acceptable codecs when establishing a voice conversation. In Jingle XML syntax the content type is the namespace of the &DESCRIPTION; element.</td>
<td>The method for establishing data stream(s) between entities. Possible transports might include ICE, Raw UDP, inband data, etc. This is the 'how' of the session. In Jingle XML syntax this is the namespace of the &TRANSPORT; element. The content transport method defines how to transfer bits from one host to another.</td>
<td>A component is a numbered stream of data which needs to be transmitted between the endpoints for a given content type in the context of a given session. It is up to the transport to negotiate the details of each component. Depending on the content type and the content description, one content description may require multiple components to be communicated (e.g., the audio content type might use two components: one to transmit an RTP stream and another to transmit RTCP timing information).</td>
<p>This document defines the semantics and syntax for overall session management. It also provides pluggable "slots" for content description formats and content transport methods, which are specified in separate documents; however, for the sake of completeness, this document also includes examples for all of the actions related to description formats and transport methods.</p>
<p>At the most basic level, the process for negotiating a Jingle session is as follows:</p>
<ol>
<li>One user (the "initator") sends to another user (the "receiver") a session request with one content type, which includes one content description and (say) two potential transport methods.</li>
<li>If the receiver wants to proceed, it provisionally accepts the request by sending an IQ result.</li>
<li>Both the initiator and receive start exchanging possible transport candidates as quickly as possible (these are sent in quick succession before further negotiation in order to minimize the time required before media data can flow).</li>
<li>These candidates are checked for connectivity.</li>
<li>As soon as the receiver finds a candidate over which media can flow, the receiver sends to the initiator a "session-accept" action.</li>
<li>The parties start sending media over the negotiated candidate.</li>
</ol>
<p>If the parties later discover a better candidate, they perform a "transport-modify" or "content-modify" negotiation and then switch to the better candidate. Naturally they can also modify various other parameters related to the session (e.g., adding video to a voice chat).</p>
<p>When negotiating the session and its content type, the entities involved in the session need to exchange content description formats. The approach taken herein is to specify pure session description information in separate specifications, one for each content description format (audio, video, etc.). <note>While it is possible to send raw Session Description Protocol (SDP) data for the session descriptions (the approach that was taken in <cite>XEP-0111</cite>), this is not necessarily helpful, since SDP is not structured data, not all SDP data is needed or used in the most common use cases, and SDP has been heavily extended in several useful directions that are not captured in the core SDP specification.</note> The session negotiation must contain some content description format(s), which are defined in separate specifications. Those specifications also define the state chart for the content description format in question.</p>
<p>As with the content description formats, the content transport methods are defined in separate specifications. Possible content transport methods include Real-time Transport Protocol (RTP) with Interactive Connectivity Establishment (ICE) and raw UDP. The relevant specifications also define the state chart for the content transport method in question.</p>
<p>In order to initiate a Jingle session, the initiating entity must determine which of the receiver's XMPP resources is best for the desired content description format. If a contact has only one XMPP resource, this task MUST be completed using &xep0030; or the presence-based profile of service discovery specified in &xep0115;.</p>
<p>Naturally, instead of sending service discovery requests to every contact in a user's roster, it is more efficient to use <cite>Entity Capabilities</cite>, whereby support for Jingle and various Jingle content description formats and content transport methods is determined for a client version in general (rather than on a per-JID basis) and then cached. Refer to <cite>XEP-0115</cite> for details.</p>
<p>If a contact has more than one XMPP resource, it may be that only one of the resources supports Jingle and the desired content description format, in which case the user MUST initiate the Jingle signalling with that resource.</p>
<p>If a contact has more than one XMPP resource that supports Jingle and the desired content description format, it is RECOMMENDED for a client to use &xep0168; in order to determine which is the best resource with which to initiate the desired Jingle session.</p>
<p>Once the initiating entity has discovered which of the receiver's XMPP resources is ideal for the desired content description format, it sends a session initiation request to the receiver. This request is an IQ-set containing a &JINGLE; element qualified by the 'http://www.xmpp.org/extensions/xep-0166.html#ns' namespace. The &JINGLE; element MUST possess the 'action', 'initiator', and 'sid' attributes (the latter two uniquely identify the session). For initiation, the 'action' attribute MUST have a value of "session-initiate" and the &JINGLE; element MUST contain one or more &CONTENT; elements, each of which defines a content type to be transferred during the session; each &CONTENT; element in turn contains one &DESCRIPTION; child element that specifies a desired content description format and one or more &TRANSPORT; child elements that specify potential content transport methods.</p>
<p>Note: The syntax and semantics of the &DESCRIPTION; and &TRANSPORT; elements are out of scope for this specification, but are defined in related specifications.</p>
<p>The attributes of the &JINGLE; element are as follows:</p>
<ul>
<li>The 'action' attribute is REQUIRED; it specifies a Jingle action as listed in this document (e.g., "session-initiate").</li>
<li>The 'initiator' attribute is the full JID of the entity that has initiated the session flow (which may be different from the 'from' address on the IQ-set).</li>
<li>The 'responder' attribute (see examples below) is the full JID of the entity that has replied to the initiation (which may be different from the 'to' address on the IQ-set).</li>
<li>The 'sid' attribute is a random session identifier generated by the initiator; this SHOULD match the XML Nmtoken production <note>See <<linkurl='http://www.w3.org/TR/2000/WD-xml-2e-20000814#NT-Nmtoken'>http://www.w3.org/TR/2000/WD-xml-2e-20000814#NT-Nmtoken</link>></note> so that XML character escaping is not needed for characters such as &. (Note: the 'sid' attribute effectively maps to the SIP "Call-ID" parameter.)</li>
</ul>
<p>The attributes of the &CONTENT; element are as follows:</p>
<li>The 'creator' attribute is REQUIRED; it specifies which party originally generated the content description (used to prevent race conditions regarding modifications).</li>
<li>The 'name' attribute is REQUIRED; it specifies a unique name or identifier for the content type (this identifier is opaque and does not have semantic meaning).</li>
<li>The 'senders' attribute is RECOMMENDED; it specifies which entities in the session will be generating content; the allowable values are "initiator", "recipient", or "both" (where "both" is the default).</li>
<p>The initiating entity is unknown to the receiver (e.g., via presence subscription) and the receiver has a policy of not communicating via Jingle with unknown entities, it MUST return a &unavailable; error.</p>
<examplecaption="Initiating Entity Unknown to Receiver"><