A RELOAD Usage for Distributed Conference Control (DisCo)
draft-knauf-p2psip-disco-01

Abstract

This document defines a RELOAD Usage for Distributed Conference Control (DisCo) with SIP. DisCo preserves conference addressing through a single SIP URI by splitting its semantic of identifier and locator using a new Kind data structure. Conference members are enabled to select conference controllers based on proximity awareness and to recover from failures of individual resource instances. DisCo proposes call delegation to balance the load at focus peers. The document also introduces methods to establish security and trust for a shared resource access, as well as for compatibility with conference unaware clients.

Status of this Memo

This Internet-Draft is submitted in full conformance with the provisions of BCP 78 and BCP 79.

Internet-Drafts are working documents of the Internet Engineering Task Force (IETF). Note that other groups may also distribute working documents as Internet-Drafts. The list of current Internet-Drafts is at http://datatracker.ietf.org/drafts/current/.

Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as “work in progress.”

This Internet-Draft will expire on July 3, 2011.

Copyright Notice

Copyright (c) 2010 IETF Trust and the persons identified as the document authors. All rights reserved.

This document is subject to BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents (http://trustee.ietf.org/license-info) in effect on the date of publication of this document. Please review these documents carefully, as they describe your rights and restrictions with respect to this document. Code Components extracted from this document must include Simplified BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Simplified BSD License.

Table of Contents

1.  Introduction
2.  Terminology
3.  Overview of DisCo
    3.1.  Reference Scenario
    3.2.  Initiating a Distributed Conference
    3.3.  Joining a Conference
    3.4.  Conference State Synchronization
    3.5.  Call delegation
    3.6.  Resilience
    3.7.  Topology Awareness
4.  RELOAD Usage for Distributed Conference Control
    4.1.  Kind Data Structure
    4.2.  Determining Coordinates
    4.3.  Conference Creation
    4.4.  Obtaining a Conference Certificate
        4.4.1.  Self-generated
        4.4.2.  Using Enrollment Server
    4.5.  Proximity-aware Conference Participation
    4.6.  Advertising Focus Ability
    4.7.  Configuration Document Extension
5.  Conference State Synchronization
    5.1.  Event Package Overview
    5.2.  <distributed-conference>
    5.3.  <version-vector>/<version>
    5.4.  <conference-description>
    5.5.  <focus>
        5.5.1.  <focus-state>
        5.5.2.  <users>/<user>
        5.5.3.  <relations>/<relation>
    5.6.  Distribution of Change Events
    5.7.  Translation to Conference-Info Event Package
        5.7.1.  <conference-info>
        5.7.2.  <conference-description>
        5.7.3.  <host-info>
        5.7.4.  <conference-state>
        5.7.5.  <users>/<user>
        5.7.6.  <sidebars-by-ref>/<sidebars-by-value>
6.  Distributed Conference Control with SIP
    6.1.  Call delegation
    6.2.  Conference Access
    6.3.  Media Negotiation and Distribution
        6.3.1.  Offer/Answer
    6.4.  Restructuring a Conference
        6.4.1.  On Graceful Leave
        6.4.2.  On Unexpected Leave
7.  DisCo Kind Definition
8.  Access Control Policy USER-CHAIN-MATCH
9.  XML Schema
10.  Relax NG Grammar
11.  Security Considerations
    11.1.  Trust Aspects
12.  IANA Considerations
13.  Acknowledgments
14.  References
    14.1.  Normative References
    14.2.  Informative References
Appendix A.  Change Log
§  Authors' Addresses

1.  Introduction

This document describes a RELOAD Usage for distributed conference control (DisCo) in a tightly coupled model with SIP [RFC3261] (Rosenberg, J., Schulzrinne, H., Camarillo, G., Johnston, A., Peterson, J., Sparks, R., Handley, M., and E. Schooler, “SIP: Session Initiation Protocol,” June 2002.). The Usage provides self-organizing and scalable signaling that allows RELOAD peers, clients and plain SIP user agents to participate in a managed P2P conference. DisCo defines the following functions:

• A protocol scheme for distributed conference control
• RELOAD Usage and definition of conferencing Kind
• RELOAD Usage and definition of a Kind for shared Resources in RELOAD
• Mechanisms for conference synchronization and call delegation
• Mechanism for proximity-aware routing within a conference
• An XML event package for distributed conferences
• Methods of establishing security and trust for shared resource access

In this document, the term distributed conferencing refers to a multiparty conversation in a tightly coupled model in which the point of control (i.e., the focus) is identified by a unique URI, but the focus service is located at many independent entities. Multiple SIP [RFC3261] (Rosenberg, J., Schulzrinne, H., Camarillo, G., Johnston, A., Peterson, J., Sparks, R., Handley, M., and E. Schooler, “SIP: Session Initiation Protocol,” June 2002.) user agents uniformly control and manage a multiparty session. This document defines a new Usage for RELOAD, including an additional Kind code point with a corresponding data structure that complies the demands for distributed conferences. The 'DisCo' data structure stores the mapping of a single conference to multiple conference controllers and thereby separates the conference URI from focus instantiations.

Authorized controllers of a conference are permitted to register their mapping into the DisCo data structure independently. Thus, the DisCo data structure represents a co-managed Resource in RELOAD. To provide trusted and secure access to a co-managed Resource, this document defines a Usage to share a specific RELOAD Resource. A RELOAD user explicitly authorizes RELOAD peers within a RELOAD Kind data structure called 'access list'.

Delay and jitter are critical issues in multimedia communications. The proposed conferencing scheme supports mechanisms to build an optimized interconnecting graph between conference participants and their responsible conference controllers. Conference members will be enabled to select the closest focus with respect to delay or jitter.

DisCo extends conference control mechanisms to provide a consistent and reliable conferencing environment. Controlling peers maintain a consistent view of the entire conference state. The multiparty system can be re-structured based on call delegation operations.

2.  Terminology

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 [RFC2119] (Bradner, S., “Key words for use in RFCs to Indicate Requirement Levels,” March 1997.).

The terminology and definitions from der RELOAD base [I‑D.ietf‑p2psip‑base] (Jennings, C., Lowekamp, B., Rescorla, E., Baset, S., and H. Schulzrinne, “REsource LOcation And Discovery (RELOAD) Base Protocol,” November 2010.), the peer-to-peer SIP concepts draft [I‑D.ietf‑p2psip‑concepts] (Bryan, D., Matthews, P., Shim, E., Willis, D., and S. Dawkins, “Concepts and Terminology for Peer to Peer SIP,” October 2010.) and the terminology formed by the framework for conferencing with SIP [RFC4353] (Rosenberg, J., “A Framework for Conferencing with the Session Initiation Protocol (SIP),” February 2006.). Additionally the following terms are used:

Coordinate Value:

An opaque string that describes a host's relative position in the network topology.

Focus peer:

A RELOAD peer that provides SIP conferencing functions and implements the Usage for distributed conferencing. It is called 'active' if already involved in signaling relation to conference participants. Otherwise, if only registered in a distributed conference data structure, it is referred to as a 'potential' focus peer.

3.  Overview of DisCo

3.1.  Reference Scenario

The reference scenario for the Distributed Conference Control (DisCo) is shown in Figure 1 (Reference Scenario: Focus peers A,B maintain a distributed conference). Peers are connected via a RELOAD [I‑D.ietf‑p2psip‑base] (Jennings, C., Lowekamp, B., Rescorla, E., Baset, S., and H. Schulzrinne, “REsource LOcation And Discovery (RELOAD) Base Protocol,” November 2010.) instance, in which peers A and B are managing a single multiparty conference. The conference is identified by a unique conference URI, but located at peers A and B fulfilling the role of the focus. The mapping of the conference URI to one or more responsible focus peers is stored in a new RELOAD Resource for distributed conferencing within a data structure denoted as DisCo-Registration. The storing peer SP of the distributed conference resource holds this data.

The focus peers A and B maintain SIP signaling relations to conference participants, which may have different conference protocol capabilities. In this example, peer A is the multiparty manager for the RELOAD peer C and the plain SIP user agent E whereas focus peer B serves for RELOAD peer D and the RELOAD client F.

RELOAD peers and clients obtain the contact information for the conference from the storing peer. In contrast, the user agent E receives the conference URI not by RELOAD mechanisms, but resolves the ID and joins the conference by plain SIP negotiation.

Focus peers establish a SIP signaling relation among each other used for notification messages that synchronize the conference focus peers' knowledge about the entire conference state. Additionally, focus peers can transfer calls to each other by a call delegation mechanism.

                                    +----------+
                                    |DisCo Data|
                                    +----------+
                                   /
                             +-------+
                             |Storing|
         # # # # # # # # # # | Peer  | # # # # # # # # # #
        #                    |  O    |                    #
       #                     +-------+                     #
      #                                                     #
     #                                                       #
    #                                                         #
 +----+                                                     +----+
 |Peer| \               RELOAD Instance                     |Peer|
 | C  |  \                                                  | D  |
 +----+   \                                                 +----+
    #     SIP                                                 #
     #      \                                                #
      #      \                                              #
       #     +-------+                       +-------+     #(
        #    | Focus |                       | Focus |    #  )
         # # | Peer  | # # # # # # # # # # # | Peer  | # #  (
             |   A   | <===Conf.Events/====> |   B   |       )
             +-------+    Call delegation    +-------+    Overlay
            /                                        \     Comm.
           /                                          \     (
         SIP                                          SIP    )
         /                                              \   (
        /                                                \   )
   +----------+                                      +--------+
   |User Agent|                                      | Client |
   |    E     |                                      |   F    |
   +----------+                                      +--------+

3.2.  Initiating a Distributed Conference

To create a conference the initiating user agent announces itself as a focus for the conference. It stores its own contact information (Address-of-Record or Destination List) in the RELOAD overlay as a DisCo-Registration Kind (cf., Figure 2 (DisCo Usage generic Call Flow)). The hashed conference URI is used as the Resource-ID. This data structure will later contain the contact IDs of all potential focus peers including optionally topological descriptors.

3.3.  Joining a Conference

A RELOAD-aware node (cf., Bob in Figure 2 (DisCo Usage generic Call Flow)) intending to join an existing conference retrieves the list of potential focus peers stored in the DisCo-Registration under the conference's Resource-ID. To join the conference it selects any of the focus peers (e.g., Alice) and establishes a connection using AppAttach. This transport is then used to send an INVITE to the conference applying the chosen focus as the contact. The selection of the focus peer to contact can optionally be based on proximity information if available.

A conference member proposes as a focus for subsequent participants by storing a mapping of the conference URI to his Address-of-Record or Destination List in the RELOAD overlay in the DisCo-Registration data structure. This decision should incorporate bandwidth, power, and other constraints, but details are beyond the scope of this document.

Alice                          RELOAD                            Bob
(initiating peer)                                     (joining peer)
--------------------------------------------------------------------
  |                               |                               |
  |      Alice stores her mapping to register a conference        |
  | Store mapping(ConfURI, Alice) |                               |
  |------------------------------>|                               |
  |                               |        Lookup ConfURI         |
  |                               |<------------------------------|
  |                               |        Result ConfURI         |
  |                               |------------------------------>|
  |                               |                               |
  |       Bob establishes transport connection to Alice           |
  |                           AppAttach                           |
  |<--------------------------------------------------------------|
  |                           AppAttach                           |
  |-------------------------------------------------------------->|
  |                            INVITE                             |
  |<~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~|
  |                              OK                               |
  |~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~>|
  |                              ACK                              |
  |<~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~|
  |                             Media                             |
  |<=============================================================>|
  |                               |                               |
  |       Bob stores his mapping to become a focus peer too       |
  |                               |  Store mapping(ConfURI, Bob)  |
  |                               |<------------------------------|
  |                               |                               |

3.4.  Conference State Synchronization

Each focus of a conference maintains signaling connections to its related participants independently from other conference controllers. This distributed conference design effects that the entire SIP conference state is jointly held by all conference focus peers. In DisCo, state synchronization is based on a SIP specific event notifications mechanism [RFC3265] (Roach, A., “Session Initiation Protocol (SIP)-Specific Event Notification,” June 2002.).

Each focus peer can complete its view of the entire conference state among the focus peers by subscribing the other focus peers for an XML event package for distributed conferences. It is defined in this document (see section Section 5 (Conference State Synchronization)) and is based on the event package for conference state [RFC4575] (Rosenberg, J., Schulzrinne, H., and O. Levin, “A Session Initiation Protocol (SIP) Event Package for Conference State,” August 2006.). Receivers of event notifications update their local conference state document to regain a valid view of current conference state.

The event notification package for distributed conferences enables focus peers to synchronize the entire conference state. The event package defines additional XML elements and complex types (see Section 9 (XML Schema) for more details), which allow views of the responsibilities of any focus peer in the conference. By providing these views each focus peer is enabled to perform additional load balancing operations and enhances the robustness against departures of focus peers.

3.5.  Call delegation

The call delegation (see Section 4.2 (Determining Coordinates).) is a feature used to transfer an incoming participation request to another focus peer. It can be applied to prevent an overloading of focus peers reaching its limit of serving new clients. Call delegation is realized through SIP REFER requests, which carry signaling and session description information of the callee to be transferred. A focus peer can decide to refer an incoming call to a less loaded remote focus. This feature is achieved transparently to the transferred user agent by using a source routing mechanism at SIP dialog establishment. Descriptions of overload detection are beyond the scope of this document.

3.6.  Resilience

A focus peer can decide to leave the conference or may ungracefully fail. In a traditional conferencing scenario, a loss of the conference controller or the media distributor would cause a complete fail of the multiparty conversation. Distributed conferencing uses the redundancy by multiple focus peers to reconfigure a running multiparty. Participants that lost their entry point to the conference re-invite itself via the remaining focus peers or will be re-invited by the controllers. This option is based on the conference state and call delegation functions.

3.7.  Topology Awareness

DisCo supports landmarking approaches based on an extension for the RELOAD XML configuration document (see Section 4.7 (Configuration Document Extension)) to construct topology-aware connections between focus and peers. Each peer intending to create or participate in a distributed conference MAY determine a topological descriptor that describes its relative position in the network. Focus peers store these coordinate values as additional data field in the DisCo-Registration data structure. This enables peers joining the conference to select the closest focus with respect to its coordinate values.

4.  RELOAD Usage for Distributed Conference Control

4.1.  Kind Data Structure

Each DisCo-Registration data structure stores mappings for one conference instance to many focus peers and for each focus peer the related coordinates value. The data structure uses the RELOAD dictionary type whereas the DictionaryKey value is the Node-ID of the focus peer behind the dictionary entry. This allows a focus peer to update it mapping. The DisCo data structure of type DisCoRegistration is shown as follows:

  enum {
    sip_focus_uri (1),
    sip_focus_node_id (2), (255)
  } DisCoRegistrationtType;

  struct {
    opaque coordinate<0..2^16-1>

    select (DisCoRegistrationtType.type) {
      case sip_focus_uri:
        opaque uri<0..2^16-1>

      case sip_focus_node_id:
        Destination destination_list<0..2^16-1>

      /* This type can be extended */
    }

  } DisCoRegistrationData;

  struct {
    DisCoRegistrationtType type;
    uint16 length;
    DisCoRegistrationData data;
  } DisCoRegistration;

The content of the DisCoRegistrationData structure are as follows:

type
    type of the registration
length
    the length of the registration PDU
data
    the conference registration data

• If the DisCoRegistration is set to "sip_focus_uri", then it contains an Address-of-Record (AoR) as an opaque string and opaque "coordinates" string, that describes the relative network position. See more in section 4.4.
• If registration type is set to "sip_focus_node" then it contains the Destination list for the peer and an opaque string "coordinates" describing the focus' relative network position.

The structure is designed for enabling a peer to contact a focus of the conference that is the nearest to itself. A joining peer MAY select the focus peer, whose coordinate value matches at most to its own. In this manner it reduces the problem of triangle inequality as without this feature a joining peer could choose an inadequate remote conference controller causing large signaling and may streaming delays.

4.2.  Determining Coordinates

Each RELOAD peer within the context of a distributed conference MAY be aware of it's relative position in the network topology. Those position information can support a topology-aware conference construction avoiding long signaling and media delays. Providing this the Usage for distributed conference foresees the coordinates value within the DisCo-Registration data structure that allows focus peers to store a topological descriptor. It is a generic field that describes a peer's relative position in the network. Focus peers store this coordinate value together with their announcement as conference focus. Joining peers likewise MAY determine their coordinates value and then select a focus peer whose relative position matches at most (see section Section 4.5 (Proximity-aware Conference Participation)).

Many algorithms determine topology information by measuring Round-Trip Times (RTT) towards a set on hosts serving as so called landmarks. To support such algorithms this document describes an extension to the RELOAD XML configuration document that allows to configure the set of Landmark hosts that peer must use for position estimation (see section Section 4.7 (Configuration Document Extension)). Once a focus peer has registered its mapping in the DisCo data structure, it also stores the according coordinates in the same mapping. These <Node-ID,coordinates> vectors are used by peers at conference join to select the focus peer that is relatively closest to itself.

Because topology-awareness can be obtained by many different approaches a concrete algorithms is out of scope of this document.

4.3.  Conference Creation

Before a peer registers to a new distributed conference, it is RECOMMENDED to ensure the initiating peer has a most up to date copy of the configuration document. In this way, the conference creator assures that all joining peers will equally determine their coordinates value if such a algorithm is used. The first peer that creates a distributed conference registers it in the RELOAD overlay following the steps as described in Figure 3 (Creation of a Distributed Conference):

   CA      Alice      Peer1     Overlay     PeerN   StoringPeer
------------------------------------------------------------
   |          |   StatReq Res:Conf-URI          |          |
   |          |---------->|--------->|--------->|--------->|
   |          |           StatAns    |          |          |
   |          |<----------|<---------|<---------|<---------|
   |<==Cert===|           |          |          |          |
   |          |           |          |          |          |
   |===Cert==>|   StoreReq Res:Conf-URI Kinds:DisCo[,SIP]  |
   |          |---------->|--------->|--------->|--------->|
   |          |           StoreAns   |          |          |
   |          |<----------|<---------|<---------|<---------|
   |          |           |          |          |          |

1. The peer MAY determine its own coordinate value (if used).
2. The peer SHOULD probe whether the desired conference URI is available. It therefore generates the Resource-ID of the conference URI with the overlay hash function and sends a RELOAD StatReq towards this address. By the corresponding StatAns response the peer knows whether the desired URI is occupied by another a DisCo Kind or even a SIP-Registration Kind [I‑D.ietf‑p2psip‑sip] (Jennings, C., Lowekamp, B., Rescorla, E., Baset, S., and H. Schulzrinne, “A SIP Usage for RELOAD,” July 2010.). If it is, the user MUST choose another URI and repeat the availability checks. If no other DisCo or SIP-Registration Kind are stored at this Resource-ID it proceeds the registration.
3. Storing a conference registration is similar to registering a new virtual user that has the conference URI as its Address-of-Record. Two options for obtaining a conference certificate are available as detailed in section Section 4.4 (Obtaining a Conference Certificate). The first does not require contacting the enrollment server, but only allows a set of conference URIs closely related to the registering peer's user name. The second option allows any form of conference URI but needs to contact the enrollment server. In both cases the resulting certificate contains the conference URI as a user name.
4. The peer finally registers the DisCo data structure signed with the private key corresponding to the above certificate by a Store request towards the storing peer (the owner of the address space for the Resource-ID of the conference URI).

The additional certificate is needed for 2 major purposes:

• It separates the conference creator from the multiparty instance.
• It ensures the conference initiator's privacy. Because the DisCo data structure will be accessed by many peers using the same conference private key. If they were using the conference creators’ key, they were permitted to write non-shared Resources of the creator.

When the conference creator has obtained the conference certificate from the enrollment server, it MAY also register the conference URI as a SIP-Registration Kind as well. In this case, it also MUST sign the Store request with the private key that matches the certificate obtained for the conference URI, since the SIP-Registration Kind uses the USER-NODE-MATCH policy. In case of the conference creator's departure, the remaining focus peers are permitted to redirect the SIP mapping to another focus peer still serving the conference. The SIP-Registration MAY be sent in the same StoreReq as the DisCo registration.

4.4.  Obtaining a Conference Certificate

In RELOAD each node uses a certificate to identify itself when storing data at a specific location. Since the DisCo-Registration needs to be written by multiple nodes, the private key of a group certificate is distributed among all authorized focus peers. This demands the assignment of a certificate for each conference ID which is used for conferencing matters only. This document describes two options for obtaining a conference certificate. The method to be used depends on the desired pattern of the conference ID being registered.

In both cases the certificate should have a sufficiently short lifetime to prevent a compromised certificate from being used continuously. The chance of a conference key leaking and thus being compromised is much larger than for normal RELOAD certificates, since the key is shared among the focus peers of a conference.

4.4.1.  Self-generated

A conference initiator can only register a conference ID that is derived from the user name defined in it's own certificate, without contacting an enrollment server. The conference ID must be closely related to the user name of the creator to prevent malicious peers not associated with the conference from generating a certificate for the same name and registering themselves as a focus.

The building scheme for constructing legitimate conference URIs MUST be specified in the overlay configuration document using a simple pattern matching scheme.

Example:
URI pattern: *-conf-$USER@$DOMAIN
User Name: alice@example.com

XYZ-conf-alice@example.com is allowed
my-pretty-conf-alice@example.com is allowed
alice-conference@example.com is NOT allowed

The conference initiator generates a new public/private key pair and signs the public key with his own private key, thus generating a certificate for the conference. The conference certificate contains the conference URI in the user name field, which MUST comply with the URI pattern specified for DisCo-conferences in the configuration document. Any peer validating the certificate MUST traverse the certificate chain up to the enrollment server and only accept it if the certificates of the creator and of the enrollment server are valid. Additionally the user name in the certificate of the creator MUST match the user name in the conference certificate using the specified pattern.

This conference certificate is subsequently used to sign all entries of the DisCo-Registration kind stored in the overlay. It MUST be accepted by the responsible peer. This results in an extension of the common USER-MATCH access control policy to USER-CHAIN-MATCH, specified in Section 8 (Access Control Policy USER-CHAIN-MATCH).

4.4.2.  Using Enrollment Server

An enrollment server MAY issue certificates for conferences with any URI that is not related to the user name of the conference initiator.

Therefore a user agent intending to register a new conference contacts the enrollment server and requests a certificate using the standard mechanism defined in RELOAD (Jennings, C., Lowekamp, B., Rescorla, E., Baset, S., and H. Schulzrinne, “REsource LOcation And Discovery (RELOAD) Base Protocol,” November 2010.) [I‑D.ietf‑p2psip‑base] Section 10.3.

Conferences with certificates obtained from an enrollment server use the USER-MATCH access control policy for the DisCo-Registration kind.

The enrollment server MUST NOT reuse the Node-ID of the conference initiator for the conference certificate, since this certificate is intended for distribution among the focus peers of a conference. This would allow the focus peers to compromise any private resources stored by the initiator using the NODE-MATCH policy.

4.5.  Proximity-aware Conference Participation

A RELOAD peer intending to join a distributed conference follows the steps shown in Figure 4 (Participation of a Distributed Conference):

  Bob      Peer1     Overlay     PeerN     OwnerOfID     Alice
--------------------------------------------------------------
   |   FetchReq Res:Conf-URI Kind:DisCo        |          |
   |--------->|--------->|--------->|--------->|          |
   |          |FetchAns  |          |          |          |
   |<---------|<---------|<---------|<---------|          |
   |          |          |          |          |          |
   |  Bob calculates Alice as closest Focus    |          |
   |          |          |          |          |          |
   |          |AppAttach--application:5060     |          |
   |--------->|--------->|--------->|--------->|--------->|
   |          |AppAttach--application:5060     |          |
   |<---------|<---------|<---------|<---------|<---------|
   |          |          |          |          |          |
   |<-------------------ICE Checks----------------------->|
   |          |          |          |          |          |
   |          |         INVITE sip:Alice       |          |
   |~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~>|
   |          |         200 OK      |          |          |
   |<~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~|
   |          |         ACK         |          |          |
   |~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~>|
   |          |          |          |          |          |
   |  Optinally, Alice passes writing permission          |
   |          |          |          |          |          |
   |         INFO content:Conf-Key{DisCo-Resource}        |
   |<~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~|
   |          |          |          |          |          |

1. The joining peer MAY determine its own coordinate value (if used).
2. The joining peer sends a FetchReq message for the DisCo Kind to the Resource-ID that corresponds to the hash over the conference URI using the overlays hash-function. The FetchReq SHOULD NOT include any specific dictionary keys thus it will receive all potential -and active focus peers of the conference.
3. Once the joining peer received the Fetch results, it calculates which of the focus peers is the relatively closest to itself.
4. The focus with the smallest distance MAY be chosen for establishing a SIP signaling relation.

Depending on which DisCo-Registration type the selected focus has stored its mapping, the joining Peer has the following 2 possibilities:

1. If the DisCoRegistrationType is sip_focus_node_id, the joining peer uses RELOADs AppAttach request to establish a direct transport connection to the selected focus peer. The application field of the request MUST be set to 5060 indicating for SIP. This transport connection SHOULD be used to a form an ordinary SIP dialog. Further media session establishment is achieved by usual SIP mechanisms.
2. If the DisCoRegistrationType is sip_focus_uri, the joining peer MUST use the SIP-Registration [I‑D.ietf‑p2psip‑sip] (Jennings, C., Lowekamp, B., Rescorla, E., Baset, S., and H. Schulzrinne, “A SIP Usage for RELOAD,” July 2010.) Usage to resolve the URI and form connectivity to the selected focus.

Note that in the second case a focus peer can have multiple locations for its SIP-registration. Therefore a focus MUST assure that its coordinate value corresponds to its current mapping AoR to location. If a focus peer has not determined its relative network position, the coordinates field MUST be left empty.

Regardless of how the focus peer has registered its mapping in the overlay a joining peer MUST add it's coordinate value base64 encoded as URI-parameter in the contact-header field of the SIP INVITE request. An example contact URI is “sip:alice@example.com;coord=YWxpY2VAZXhhbXBsZS5jb20=”. The additional parameter is used by the requested focus peer as it is not capable of serving additional conference participants. It then it MUST delegate the call (see section Section 4.2 (Determining Coordinates)) to the focus peer whose coordinate value matches next best to the coordinates of the joining peer. The focus peer therefore uses the same calculation as described in the joining process.

After the final SIP ACK request completes the signaling relation, a conference focus MAY passes the writing permission to the new participant. It therefore sends a SIP INFO request carrying the conference private key for the DisCo Resource. The decision whether to pass writing permission depends on the selected security model for the distributed conference as described in section Section 6.2 (Conference Access).

4.6.  Advertising Focus Ability

All participants of a distributed conference can become a focus peer for their conference. The decision can depend on the capacities of the joining peer like sufficient processing power (CPU, Memory) for the desired media type and quality of the network connectivity. Additionally, a peer intending to become focus of a conference SHOULD NOT be located behind NAT or its IP SHOULD NOT belong to the private address range. The information whether a participant is behind NAT can be obtained by ICE connectivity checks during the conference joining process. Nevertheless, under special circumstances it might be reasonable to locate a focus peer behind a NAT. For instance within a companies network infrastructure.

If a participant is a candidate to become a focus of the conference it stores its mapping (Destination List or AoR) and coordinate value into the DisCo data structure. Because the DisCo Kind uses either the USER-MATCH or the USER-CHAIN-MATCH access control policy, the shared 'private' key passed by the participant's focus peer is sufficient to permit this peer to write the DisCo Resource. By storing the mapping into the data structure a participant becomes a potential focus.

4.7.  Configuration Document Extension

This section defines an additional parameter for the <configuration> element that extends the RELOAD XML configuration document. The proposed <landmarks> element allows RELOAD provider to publish a set of accessible and reliable hosts that SHOULD be used if RELOAD peers use landmarking algorithms to determine relative position in the network topology.

The <landmarks> element serves as container of the <landmark-host> sub-elements each representing a single host that serves a landmark. The <landmark-host> uses the following attributes:

address:

The IP address (IPv4 or IPv6) of the landmark host.

port:

The port on which the landmark host responses for distance estimation.

More than one landmark hosts SHOULD be present in the configuration document.

5.  Conference State Synchronization

The global knowledge about signaling and media relations among the conference participants and focus peers defines the global state of a distributed conference. It is composed of the union of every local state at the focus peers. To enable focus peers to provide conference control operations that modify and/or require the global state of a conference, this document defines a mechanism for inter-focus state synchronization. It is based on mutual subscriptions for an Event Package for Distributed Conferences and allows to preserve a coherent knowledge of the global conference state. This XML based event package named 'distributed-conference' MUST be supported by each RELOAD peer that is registered within a DisCo-Registration. Participants of a distributed conference MAY support it. To provide backward compatibility to conference members that do not support the distributed-conference event package, this document defines a translation to the Event Package for Conference State [RFC4575] (Rosenberg, J., Schulzrinne, H., and O. Levin, “A Session Initiation Protocol (SIP) Event Package for Conference State,” August 2006.).

5.1.  Event Package Overview

The 'distributed-conference' event package is designed to convey information about roles and relations of the conference participants. Conference controllers obtain the global state of the conference and use this information for load balancing or conference restructuring mechanisms in case of a focus failure. Figure Figure 5 (Overview of the event package for distributed conferences) gives a general overview of the document hierarchy.

distributed-conference
  |
  |-- version-vector
  |    |-- version
  |    |-- version
  |
  |-- conference-description
  |
  |-- focus
  |    |-- focus-state
  |    |    |-- user-count
  |    |    |-- maximum-user-count
  |    |    |-- active
  |    |    |-- locked
  |    |    |-- conf-uris
  |    |    |-- available-media
  |    |
  |    |-- users
  |    |    |-- user
  |    |    |    |-- endpoint
  |    |    |    |    |-- media
  |    |    |    |    |-- call-info
  |    |
  |    |-- relations
  |    |    |-- relation
  |-- focus
  |    |-- ...

The local state for each focus peer is described within a corresponding 'focus' element. Each provides general information about a focus peer, its signaling and media relations to participants and focus peers. Conference participants are aggregated within 'users' elements, respectively, 'user' sub elements.

The document structure of 'distributed-conference' is designed to allow concurrent occurring change events at several focus peers. For that reason, each focus peer has exclusive writing permission to the 'focus' sub element that describes itself. The Event Package for Distributed Conferences therefore imports the mechanisms for partial notification and uses the 'Element Keys' definitions described in [RFC4575] (Rosenberg, J., Schulzrinne, H., and O. Levin, “A Session Initiation Protocol (SIP) Event Package for Conference State,” August 2006.) (see [RFC4575] (Rosenberg, J., Schulzrinne, H., and O. Levin, “A Session Initiation Protocol (SIP) Event Package for Conference State,” August 2006.) sections 4.4-5.). A focus peer is only allowed to update or change that <focus> sub element, whose 'entity' Element Key contains the same AoR as the AoR of RELOAD user that is acting focus peer. This restriction also applies to the child elements of the 'version-vector' element. Each 'version' number is a property of a specific focus peer which is exclusively responsible to increment the version number.

General information about the conference itself, is provided within a 'conference-description' element. In contrast to the 'focus' and 'version-vector' elements, the conference description is not meant for concurrent updating. Instead, it provides static conference descriptions that rarely change during a multiparty session.

More detailed descriptions about the event package and its elements are provided in the following sections. The complete XML schema definition is given in section Section 9 (XML Schema).

5.2.  <distributed-conference>

The <distributed-conference> element is the root of a distributed conference XML document. It uses the following attributes:

entity:

This attribute contains the conference URI that identifies the distributed conference described by the XML document. A SUBSCRIBE request addressed to this URI results in an ordinary subscribe/notify relation between participants and their related focus peer.

state:

In accordance to [RFC4575] (Rosenberg, J., Schulzrinne, H., and O. Levin, “A Session Initiation Protocol (SIP) Event Package for Conference State,” August 2006.), this attribute indicates whether the content of a distributed conference document is a 'full', 'partial' or 'deleted' information.

The <distributed-conference> child elements are <vector-version>, <conference-description> and the <focus> elements. An event notification of state 'full' MUST include all these elements. An event notification of state 'partial' MUST contain at least <version-vector> and its sub elements.

5.3.  <version-vector>/<version>

The Event Package for Distributed Conferences uses the <version-vector>, respectively, <version> elements to enable a coherent versioning scheme based on vector clocks as defined in [timestamps‑acsc88] (Fidge, C., “Timestamps in Message-Passing Systems that Preserve the Partial Ordering,” February 1988.). Each <version> element contains a unsigned integer that describes the local version of a specific focus peer and contains the following attribute:

entity

This attribute contains the AoR of the focus peer. This AoR corresponds to the 'user name' field, that is declared in the certificate that the RELOAD user acting as focus peer obtained while enrollment.

If a focus peer originates a notification for a change event, the focus peer MUST increment its associated <version> element by one. Additionally, a change event notification MUST contain a complete <version-vector> that carries each <version> element known by the focus peer.

The recipient of a change event needs to update the <version> element of the originator of the event notification in its local XML document. All other <version> elements remain constant, although the received <version-vector> might be different. Instead, a comparison of both version vectors can indicate a different knowledge about the global conference state.

As long as each <version> number in the recipients XML document is at least lower than one to the received <version-vector> numbers, there is no need for a subscription refresh. In this case, one or more change event notifications might not reached all subscribers yet.

Otherwise, if any <version> number of the recipient is retarded more than one, the recipient SHOULD refresh the subscription in order to trigger a 'full' state notification. The recipient uses the full state notification to recursively update every <focus> element, whose corresponding <version> element is retarded more than one.

5.4.  <conference-description>

The <conference-description> element provides general information and links to auxiliary services for the conference. The following sub elements provide human-readable text descriptions about the conference:

<display-text>:

Contains a short text description about the conference

<subject>:

Contains the subject of the conference

<free-text>:

Contains a longer textual description about the conference

<keywords>:

Contains a list of keywords that match the conference topic. The XML schema definition and semantic is imported from the 'conference-info' event package [RFC4575] (Rosenberg, J., Schulzrinne, H., and O. Levin, “A Session Initiation Protocol (SIP) Event Package for Conference State,” August 2006.).

In accordance with [RFC4575] (Rosenberg, J., Schulzrinne, H., and O. Levin, “A Session Initiation Protocol (SIP) Event Package for Conference State,” August 2006.) a <service-uris> sub element enables focus peers to advertise auxiliary services for the conference. The XML schema definition and semantic is imported from the 'conference-info' event package [RFC4575] (Rosenberg, J., Schulzrinne, H., and O. Levin, “A Session Initiation Protocol (SIP) Event Package for Conference State,” August 2006.).

The <conference-description> element uses the 'state' Element Key to enable the partial notification mechanism.

5.5.  <focus>

The <focus> element describes an active focus peer as whole. It provides general information about a focus peer (e.g., display-text, languages, etc.). Contains conference specific information about the state of a focus peer (user-count, available media types, etc.). Announces the signaling and media information about participants maintained by this focus peer and describes Signaling or media connections to other focus peers.

The <focus> element uses the following attributes:

entity:

This attribute contains the AoR of the RELOAD user acting as focus peer to the conference. This AoR corresponds to the 'user name' field, that is declared in the certificate the user obtained while RELOAD enrollment [I‑D.ietf‑p2psip‑base] (Jennings, C., Lowekamp, B., Rescorla, E., Baset, S., and H. Schulzrinne, “REsource LOcation And Discovery (RELOAD) Base Protocol,” November 2010.). The AoR in the 'entity' attribute uniquely identifies the focus peer, that is allowed to update or change the sub elements of <focus>. All other focus peers SHOULD NOT update or change sub elements of this <focus> element. A SUBSCRIBE request addressed to this AoR MUST be interpreted as the request for conference state synchronization with another focus peer.

state:

In accordance to RFC4575, this attribute indicates whether the content of the <focus> element is a 'full', 'partial' or 'deleted' information. Since the event package for distributed conferences uniquely restricts modifications on <focus> sub elements to a specific focus peer, a 'partial' notification contains a single <focus> element at maximum.

The following sub elements of <focus> provide general information about a focus peer. The XML schema definition and semantic for <associated-aors>, <roles> and <languages> are imported from the 'conference-info' event package [RFC4575] (Rosenberg, J., Schulzrinne, H., and O. Levin, “A Session Initiation Protocol (SIP) Event Package for Conference State,” August 2006.).

<display-text>:

Contains a short text description about the user acting as focus peer.

<associated-aors>:

In accordance with [RFC4575] (Rosenberg, J., Schulzrinne, H., and O. Levin, “A Session Initiation Protocol (SIP) Event Package for Conference State,” August 2006.), this element contains additional URIS that are associated with this user.

roles:

In accordance with [RFC4575] (Rosenberg, J., Schulzrinne, H., and O. Levin, “A Session Initiation Protocol (SIP) Event Package for Conference State,” August 2006.), this element MAY contain human-readable text descriptions about the roles of the user in the conference.

<languages>:

In accordance with [RFC4575] (Rosenberg, J., Schulzrinne, H., and O. Levin, “A Session Initiation Protocol (SIP) Event Package for Conference State,” August 2006.), this element contains a list of tokens, each describing a language understood by the user.

The following sections describe the remaining sub elements of <focus> more detailed.

5.5.1.  <focus-state>

The <focus-state> element aggregates a set of conference specific information about the RELOAD user acting as focus peer. The following attribute is defined for the <focus-state> element:

status:

This attribute indicates whether the content of the <focus-state> element is a 'full', 'partial' or 'deleted' information.

The <focus-state> element has the following sub elements:

<user-count>:

This element contains the number of participants that are connected to the conference via this focus peer at a certain moment.

<maximum-user-count>:

This number indicates a threshold of participants a focus peer is able to serve. This value MAY changes during a conference, depending on the focus peers current load.

<conf-uris>:

In accordance with the 'conference-info' event package [RFC4575] (Rosenberg, J., Schulzrinne, H., and O. Levin, “A Session Initiation Protocol (SIP) Event Package for Conference State,” August 2006.), this element MAY contains other conference URIs in order to access the conference via different signaling means. The XML schema definition and semantic is imported from [RFC4575] (Rosenberg, J., Schulzrinne, H., and O. Levin, “A Session Initiation Protocol (SIP) Event Package for Conference State,” August 2006.).

<available-media>:

This element is imports the <conference-media-type> type XML scheme definitions from [RFC4575] (Rosenberg, J., Schulzrinne, H., and O. Levin, “A Session Initiation Protocol (SIP) Event Package for Conference State,” August 2006.). It allows a focus peer to list its available media streams.

<active>:

This boolean element indicates whether a focus peer is currently active. Conference participation requests or a call delegation request SHOULD succeed.

<locked>:

In contrast to <active>, this element indicates that a focus peer is not willing to accept any participation or call delegation request.

5.5.2.  <users>/<user>

The <users>, respectively, each <user> element describes a single participant that is connected to the conference via the focus peer which is described by the parent <focus> element. The <users> element XML schema definition and its semantic is imported from the 'conference-info' event package [RFC4575] (Rosenberg, J., Schulzrinne, H., and O. Levin, “A Session Initiation Protocol (SIP) Event Package for Conference State,” August 2006.).

5.5.3.  <relations>/<relation>

The <relations> element serves as container for <relation> elements, each describing a specific connection between to another focus peer. The parent element <relations> uses the 'state' attribute to enable the partial notification mechanism. For the <relation> element the following attribute is defined:

entity:

This attribute contains the AoR of the remote focus.

The content of each <relation> is a comma separated string that describes the tuple <CONNECTION-TYPE,IDENTIFIER>. The CONNECTION-TYPE describes the type of connection (e.g. media, signaling, etc.) and the IDENTIFIER contains a token that uniquely identifies the connection in the conference. It is meant as a generic method to announce any kind of connection to a remote focus. This specification defines following tuples:

media,label:

This tuple identifies a single media stream originated in the focus peer described in the parent <focus> element. The 'media' part indicates that this connections belongs to any kind of media connection. The 'label' part uniquely identifies to the stream within the conference and corresponds to the SDP "label" media attribute defined in [RFC4574] (Levin, O. and G. Camarillo, “The Session Description Protocol (SDP) Label Attribute,” August 2006.).

sync,call-id:

This tuple indicates a subscription for the event package for distributed conferences. The remote focus peer described in the 'entity' attribute is a subscriber for distributed-conference event package for the purpose of conference state synchronization. The 'call-id' part thereby corresponds to the call-id of the SIP subscription dialog.

5.6.  Distribution of Change Events

Maintaining a coherent conference state at each controller of a distributed conference, requires a common protocol scheme to communicate each conference change event to each focus peer. Using SIP specific event notifications [RFC3265] (Roach, A., “Session Initiation Protocol (SIP)-Specific Event Notification,” June 2002.), this requirement would result in an N to M relation (with N=M), between N notifiers and M subscribers to synchronize the conference state. To avoid a 'full meshed' subscription topology, where each focus peer has N subscriptions to all other focus peers, this section describes a 'mutual' subscription scheme that constructs a spanning tree topology among the focus peers.

A member in a distributed conference that accepted an authorized participation request and becomes a new focus peer, SHOULD join the state synchronization process of a conference. Therefore, it sends a SIP SUBSCRIBE to request the Event Package for Distributed Conferences to its own focus peer. The subscription request SHOULD be addressed to AoR of the active focus peer, thus it interprets this subscription as a request for conference state synchronization. The corresponding NOTIFY message contains a 'full' distributed-conference state XML document (see section Section Section 5.1 (Event Package Overview)).

Following, the subscribed focus peer has to subscribe the new upcoming focus peer for the distributed conference event package. The corresponding notification by the new focus carries a 'partial' conference state XML document. It contains the received <version-vector> including a new <version> element for itself and contains a new <focus> element that describes its local state (see section Section 5.5 (<focus>)).

Resulting by this subscription scheme, each focus peer has at least one subscription to obtain updates for the conference state and is a notifier for change events originated itself. In a incrementally increasing conference, the 1st and 2nd focus peer have a mutual subscription for conference change events. A 3rd focus could have a mutual subscription with the 1st focus, a 4th focus to the 2nd focus and so forth. The result is a spanning tree topology among the focus peers in which each focus peer is a possible root for distribution tree for conference change events.

However, the fact that event notifications need to traverse one or more intermediate focus peers until conference-wide delivery, demands a forwarding mechanism for change events. On receiving a change event, a notified focus firstly validates based on the <version-vector> whether the incoming state event is not a duplicate to previews notifications. If its not a duplicate, the received change event, secondly, triggers a new event notification process at the receiver of the change event. It notifies all its subscribers with excepting the sender of the event notification (which is not necessarily the event originator). Thus, the change event will be 'flooded' among the focus peer of a conference.

5.7.  Translation to Conference-Info Event Package

The Event Package for Distributed Conferences imports several XML element definitions of the Event Package for Conference State [RFC4575] (Rosenberg, J., Schulzrinne, H., and O. Levin, “A Session Initiation Protocol (SIP) Event Package for Conference State,” August 2006.). This is caused by two reasons. Firstly, the semantic of these elements are fitting the demands to describe the global state of a distributed conference and, secondly, it facilitates a re-translation to [RFC4575] (Rosenberg, J., Schulzrinne, H., and O. Levin, “A Session Initiation Protocol (SIP) Event Package for Conference State,” August 2006.) to enable a backward compatibility to DisCo-unaware clients. Therefore, each focus peer MAY provide a separate [RFC4575] (Rosenberg, J., Schulzrinne, H., and O. Levin, “A Session Initiation Protocol (SIP) Event Package for Conference State,” August 2006.) conform event notification service to its connected participants.

The following sections describe the translation to the Event Package for Conference State [RFC4575] (Rosenberg, J., Schulzrinne, H., and O. Levin, “A Session Initiation Protocol (SIP) Event Package for Conference State,” August 2006.) by defining translation rules for the root element and its direct sub elements. For a better understanding, the following sections use a notation ci.<ELEMENT> to refer to a sub element of the conference-info element, and disco.<ELEMENT> to refer to an element of the distributed-conference event package.

5.7.1.  <conference-info>

The root element of Event Package for Conference State uses the attributes 'entity', 'state' and 'version' and is the counterpart of the <distributed-conference> root element in the DisCo Event Package. The former two attributes 'entity' and 'state' are equal in both root elements and can be seamlessly translated.

According to [RFC4575] (Rosenberg, J., Schulzrinne, H., and O. Levin, “A Session Initiation Protocol (SIP) Event Package for Conference State,” August 2006.), the version attribute SHOULD be incremented by one at any time a new notification being sent to a subscriber. New notifications SHOULD be triggered by change events that are originated within a focus peer and SHOULD be triggered by reception of a 'distributed-conference' event state of another focus peer.

5.7.2.  <conference-description>

The <conference-description> element exists in both event packages, conference-info and distributed-conference. Thus, the following elements are seamlessly translatable: <keywords>, <display-text>, <subject>, <free-text> and <service-uris>.

The sub elements <conf-uris>, <maximum-user-count> and <available-media> in conference-info have there counterparts below the \focus\focus-state element of the distributed-conference event package. Each describes a local state of a focus peer in the conference. Hence, the intersection of every disco.<conf-uris>, disco.<available-media> and the sum over each disco.<maximum-user-count> element of each disco.<focus> element in distributed-conference, specifies the content of the corresponding conference-info elements.

5.7.3.  <host-info>

According to [RFC4575] (Rosenberg, J., Schulzrinne, H., and O. Levin, “A Session Initiation Protocol (SIP) Event Package for Conference State,” August 2006.) the ci.<host-info> element contains information about the entity hosting the conference. For participants in a distributed conference, the hosting entity is the focus peer through which they are connected to the conference. Thus, the ci.<host-info> element contains information about a focus peer that is serving its participants.

5.7.4.  <conference-state>

The ci.<conference-state> element allows subscribers obtain information about overall state of a conference. Its sub elements ci.<user-count>, ci.<active> and ci.<locked> are reused as sub elements of \focus\focus-state to describe the local state of a focus peer in a distributed conference. The translation rules from the distributed-conference to the conference-info event package are the following:

<user-count>:

The sum over each value of the disco.<user-count> element defines the corresponding ci.<user-count>.

<active>:

The boolean ci.<active> element is defined by the logical concatenation over all disco.<active> elements by an OR-operator.

<locked>

The boolean ci.<locked> element is defined by the logical concatenation over all disco.<locked> elements by an AND-operator.

5.7.5.  <users>/<user>

The distributed-conference event package imports the definitions of the ci.<users> and ci.<user> elements under a parent disco.<focus> element for each focus peer in a conference. Thus, the aggregation over all disco.<users> elements specifies the content of the corresponding ci.<users> element.

5.7.6.  <sidebars-by-ref>/<sidebars-by-value>

In accordance to [RFC4575] (Rosenberg, J., Schulzrinne, H., and O. Levin, “A Session Initiation Protocol (SIP) Event Package for Conference State,” August 2006.), if a participant is connected to a sidebar, its responsible focus peer creates a new <user> by referencing to the corresponding sidebar conference.

6.  Distributed Conference Control with SIP

6.1.  Call delegation

Distributed conference control provides the possibility to delegate the control for a conference participant from one focus to another remote focus for some reason. In contrast to common call transfers that are using the SIP REFER method as described in [RFC3515] (Sparks, R., “The Session Initiation Protocol (SIP) Refer Method,” April 2003.), call delegations are achieved transparently to the transferred party.

However, a focus peer starts a call delegation by sending SIP REFER request containing the URI of the participant in the Refer-To header field. Additionally, the focus peer appends the following parameter to the URI of the participant:

call-id:

Contains the call-ID of the original SIP dialog establishment between the referred participant and the referring focus peer.

sess-id:

Contains the 'session identifier' value of the original SDP 'o=' field of the original offer/answer process between referred participant and referring focus peer.

If the recipient accepts the REFER request, it generates a re-INVITE towards the referred party and sets the SIP call-id header and the SDP 'session-identifier' field in the SDP offer, according to the URI parameter values of the initial REFER request. The From header field and contact header are set to the conference URI with setting the 'isfocus' tag to contact header. This identifies the peer as a focus to the conference and identifies this re-INVITE as a request of the SIP dialog between the party and the conference. To ensure that further signaling messages will be routed correctly, the new focus adds a Record-Route header field that contains the contact information (URI, IP-address,..) of this peer.

An example call flow for call delegation is shown in Figure 6 (Delegating a participant with SIP REFER).

Participant    Referring Focus                         Remote Focus
--------------------------------------------------------------------
     |    Dialog    |                                     |
     |<============>|                                     |
     |              |                                     |
     |  Delegating a participant to remote focus          |
     |              |                                     |
     |              |(1) REFER refer-to:<uri>?call-id=123&sess-id=456
     |              |------------------------------------>|
     |              |(2) 200 OK                           |
     |              |<------------------------------------|
     |              |(3) Notify: pending                  |
     |              |<------------------------------------|
     |              |(4) 200 OK                           |
     |              |------------------------------------>|
     |              |                                     |
     |  Remote focus adds RR-header that carries its URI  |
     |              |                                     |
     |  (5) INVITE sip:<uri> record-route:<rem.focus>     |
     |<---------------------------------------------------|
     |  (6) 200 OK  |                                     |
     |--------------------------------------------------->|
     |  (7) ACK     |                                     |
     |<---------------------------------------------------|
     |              |(8) Notify: active                   |
     |              |<------------------------------------|
     |              |(9) 200 OK                           |
     |              |------------------------------------>|

Note, subscriptions for the event packages 'distributed-conference' and 'conference-info' are in scope of a specific focus peer and its connected participants. Hence, after a successful call delegation, the referring focus peer SHOULD terminate any subscription to the referred participant. The notifier SHOULD include a reason parameter "deactivated" to indicate a migration of the subscription as defined in [RFC3265] (Roach, A., “Session Initiation Protocol (SIP)-Specific Event Notification,” June 2002.). The new SUBSCRIBE request by the party MUST be sent via the SIP dialog to the conference.

6.2.  Conference Access

It is an important issue to define who is allowed to participate a multimedia conference. In many cases, a group conversation can be an open discussion free to participate, while in other occasions a closed privacy of a multiparty session is demanded. Additionally, it is an issue which of the conference participant is allowed to become a controller of the multiparty session.

To provide those constraints for distributed conferences in RELOAD, this document defines a basic set of conference access that decide whether:

• A peer is allowed to participate a conference
• A peer is allowed to become a focus of the conference

Both cases can be regulated with plain SIP authorization mechanisms achieved by the focus peers asking for a participants credentials. Those credentials and the allowed users can be setup at creation of the conference. It is up by the conference to define different access role deciding if a user is allowed to become a focus peer or not.

6.3.  Media Negotiation and Distribution

This section describes a basic scheme for media negotiation and distribution, which is done in an ad-hoc fashion. Each focus peer forwards all media streams it receives from the conference to all connected peers it is responsible for and similarly all streams from its peers to its responsible focus. This results in the media stream naturally following the SIP signalling paths. Implementations MAY choose to use a more sophisticated scheme, e.g. employing cross connections between different sub-trees of the conference, but MUST take measures to prevent loops in media routing.

When a new peer has been attached to a focus, new media steams may be available to the focus, which need to be forwarded to the conference. To accomplish this, the new media streams need to be signalled to the other participants. This is usually done by sending a SIP re-INVITE and modifying the media sessions, adding the new streams to the SDP. This renegotiation can be costly since it needs to be propagated through the whole conference. Also, distributing all media streams separately to all participants can be quite bandwidth intensive. Both problems can partially be mitigated by focus peers performing mixing of media streams, thus trading bandwidth and signalling overheads for computational load on focus peers.

6.3.1.  Offer/Answer

A peer joining a conference negotiates media types and media parameters with its designated focus using the standard SDP offer/ answer protocol [RFC3264] (Rosenberg, J. and H. Schulzrinne, “An Offer/Answer Model with Session Description Protocol (SDP),” June 2002.). The focus SHOULD offer all media streams used in the conference.

A new participant does not necessarily know about all media streams present in a conference beforehand, and thus some of the media streams might not be included in the initial SDP offer sent by the joining peer. An SDP answer sent by the corresponding focus MUST NOT contain any media streams not matching an offer (cf. [RFC3264] (Rosenberg, J. and H. Schulzrinne, “An Offer/Answer Model with Session Description Protocol (SDP),” June 2002.) Section 6). A joining peer which is aware of the distributed nature of the conference, SHOULD NOT send an SDP offer in the initial INVITE message, but instead send an empty INVITE to which the focus replies with an OK, containing the offer. This prevents the need for a second offer/answer-dialog to modify the session. But for compatibility the normal behavior with the INVITE containing the offer MUST be supported.

The focus SHOULD only offer media types and codecs which are already used in the conference and which will probably be accepted when forwarded to neighboring peers, unless the focus is prepared to do transcoding and/or mixing of the received streams.

A focus has two options when distributing media streams from a new participant to the conference. The focus can either mix the new streams into his own, thus averting the need to modify the already established media sessions with neighboring peers or in case the focus is not willing or able to do mixing of the media streams, he SHOULD modify the media sessions with all attached peers by sending a re-INVITE, adding the new media streams coming from the newly joined participant to the SDP. This SHOULD subsequently be done by all other focus peers upon receiving the new stream, resulting in the stream being distributed across the conference.

6.4.  Restructuring a Conference

Distributed conference control provides the possibility to delegate calls to remote focus peers. This feature is used to restructure a conference in case of departure of a focus peer. Following, this section presents restructuring schemes for graceful and unexpected leaves of conference focus peers.

6.4.1.  On Graceful Leave

In a graceful case, the leaving focus peer (LP) accomplishes the following procedure:

• LP deletes its mapping in the DisCo-Registration by storing the "non-existing" value as described in the RELOAD base document [I‑D.ietf‑p2psip‑base] (Jennings, C., Lowekamp, B., Rescorla, E., Baset, S., and H. Schulzrinne, “REsource LOcation And Discovery (RELOAD) Base Protocol,” November 2010.). Afterwards, it fetches the lasted version of the DisCo-Registration to obtain all potential focus peers.
• LP calculates for all its participants the closest focus among all active and potential focus peer using the algorithm described in Section 4.2 (Determining Coordinates). LP then delegates all participants to those focus peers.
• LP then announces its leave by sending a NOTIFY to all its subscribers for the extended conference event package, setting its <focus> state to 'deleted'. Thereafter, it ends its own SIP conference dialog by sending by to its related focus peer.

Since the state synchronization topology in a distributed conference is commonly arranged in a spanning tree, a leave of a focus peer effects a gab in the tree structure. Those focus peers which had the leaving focus peer as their parent focus, are supposed to reconnect to the synchronization graph, by subscribing the leaving peer's parent node.

6.4.2.  On Unexpected Leave

If an unexpected leave is detected by a participant (e.g. missing signaling and/or media packets) it MUST repeat the joining procedure as described in Section 4.5 (Proximity-aware Conference Participation).

7.  DisCo Kind Definition

This section formally defines the DisCo kind.

Name
    DISCO-REGISTRATION

Kind IDs
    The Resource name DISCO-REGISTRATION Kind-ID is the AoR of the
    conference. The data stored is the DisCoRegistrationData, that
    contains a coordinates value describing a peers relative network
    position acting as focus for the conference. Additionally it
    contains either the peers URI or a Destination list.

Data Model
    The data model for the DISCO-REGISTRATION Kind-ID is dictionary.
    The dictionary key is
    the Node-ID of the peer action as focus.

Access Control
    USER-MATCH
     or
    USER-CHAIN-MATCH

The data stored for the Kind-ID DISCO-REGISTRATION is of type
DisCoRegistration. It contains a "coordinates" value, that
describes the peers relative network position and
XOR one of the two following data:

    sip_focus_uri
        the URI of the peer action as focus
    sip_focus_node_id
        the Destination list of the peer acting as focus

8.  Access Control Policy USER-CHAIN-MATCH

The base RELOAD document [I‑D.ietf‑p2psip‑base] (Jennings, C., Lowekamp, B., Rescorla, E., Baset, S., and H. Schulzrinne, “REsource LOcation And Discovery (RELOAD) Base Protocol,” November 2010.) mandates that every kind definition specifies an Access Control Model to use. The base document defines four access control policies, of which none is fitting for the purpose of shared write access to a resource. This document defines a new access control model called USER-CHAIN-MATCH.

In the USER-CHAIN-MATCH policy, a given value MUST be written (or overwritten) if and only if the request is singed with a key associated with a certificate whose user name hashes (using the hash function for the overlay) to the Resource-ID for the resource. Also the user name of the certificate MUST match the user name of an intermediary certificate in the chain to the root certificate, using the matching pattern specified in the configuration document for the corresponding kind.

9.  XML Schema

The XML schema for the event package for distributed conferences is:

<?xml version="1.0" encoding="UTF-8"?>
<xs:schema xmlns:xs="http://www.w3.org/2001/XMLSchema"
           xmlns:ci="urn:ietf:params:xml:ns:conference-info"
     xmlns="urn:ietf:params:xml:ns:distributed-conference"
     targetNamespace="urn:ietf:params:xml:ns:distributed-conference"
     elementFormDefault="qualified"
     attributeFormDefault="unqualified">
<!--
    This imports the definitions in conference-info
-->
<xs:import namespace="urn:ietf:params:xml:ns:conference-info"
        schemaLocation="http://www.iana.org/assignments/xml-registry/
                     schema/conference-info.xsd"/>
<xs:import namespace="http://www.w3.org/XML/1998/namespace"
     schemaLocation="http://www.w3.org/2001/03/xml.xsd"/>
  <!--
   A DISTRIBUTED CONFERENCE ELEMENT
-->
<xs:element name="distributed-conference"
            type="distributed-conference-type"/>
  <!--
    DISTRIBUTED CONFERENCE TYPE
-->
  <xs:complexType name="distributed-conference-type">
    <xs:sequence>
      <xs:element name="version-vector"
            type="version-vector-type" minOccurs="1"/>
      <xs:element name="conference-description"
      type="conference-description-type"
      minOccurs="0" maxOccurs="1"/>
      <xs:element name="focus"
      type="focus-type"
      minOccurs="0"
      maxOccurs="unbounded"/>
       <xs:any namespace="##other" processContents="lax"/>
     </xs:sequence>
     <xs:attribute name="state" type="ci:state-type"/>
     <xs:attribute name="entity" type="xs:anyURI"/>
     <xs:anyAttribute namespace="##other" processContents="lax"/>
  </xs:complexType>
<!--
    VERSION VECTOR TYPE
-->
  <xs:complexType name="version-vector-type">
    <xs:sequence>
      <xs:element name="version"
                  type="version-type"
                  minOccurs="1"
                  maxOccurs="unbounded"/>
      <xs:any namespace="##other" processContents="lax"/>
    </xs:sequence>
    <xs:anyAttribute namespace="##other" processContents="lax"/>
  </xs:complexType>
<!--
    CONFERENCE DESCRIPTION TYPE
-->
  <xs:complexType name="conference-description-type">
    <xs:sequence>
      <xs:element name="display-text"
                  type="xs:string" minOccurs="0"/>
      <xs:element name="subject" type="xs:string" minOccurs="0"/>
      <xs:element name="free" type="xs:string" minOccurs="0"/>
      <xs:element name="keywords"
                  type="ci:keywords-type" minOccurs="0"/>
      <xs:element name="service-uris"
                  type="ci:uris-type" minOccurs="0"/>
      <xs:any namespace="##other" processContents="lax"/>
    </xs:sequence>
    <xs:attribute name="state" type="ci:state-type"/>
    <xs:anyAttribute namespace="##other" processContents="lax"/>
  </xs:complexType>
<!--
    FOCUS TYPE
-->
  <xs:complexType name="focus-type">
    <xs:sequence>
      <xs:element name="display-text"
                  type="xs:string" minOccurs="0"/>
      <xs:element name="associated-aors"
                  type="ci:uris-type" minOccurs="0"/>
      <xs:element name="roles"
                  type="ci:user-roles-type" minOccurs="0"/>
      <xs:element name="languages"
                  type="ci:user-languages-type" minOccurs="0"/>
      <xs:element name="focus-state"
                  type="focus-state-type" minOccurs="0"/>
      <xs:element name="users"
                  type="ci:users-type" minOccurs="0"/>
      <xs:element name="relations"
                  type="relations-type" minOccurs="0"/>
      <xs:any namespace="#other" processContents="lax"/>
    </xs:sequence>
    <xs:attribute name="entity" type="xs:anyURI"/>
    <xs:attribute name="state" type="ci:state-type"/>
    <xs:anyAttribute namespace="##other" processContents="lax"/>
  </xs:complexType>
<!--
    VERSION TYPE
-->
  <xs:complexType name="version-type">
    <xs:simpleContent>
      <xs:extension base="xs:unsignedInt">
        <xs:attribute name="entity" type="xs:anyURI"/>
        <xs:anyAttribute namespace="##other" processContents="lax"/>
      </xs:extension>
    </xs:simpleContent>
  </xs:complexType>
  <!--
    FOCUS STATE TYPE
-->
  <xs:complexType name="focus-state-type">
    <xs:sequence>
      <xs:element name="user-count"
                  type="xs:unsignedInt" minOccurs="0"/>
      <xs:element name="maximal-user-count"
                  type="xs:unsignedInt" minOccurs="0"/>
      <xs:element name="conf-uris"
                  type="ci:uris-type" minOccurs="0"/>
      <xs:element name="available-media"
                  type="ci:conference-media-type" minOccurs="0"/>
      <xs:element name="active" type="xs:boolean" minOccurs="0"/>
      <xs:element name="locked" type="xs:boolean" minOccurs="0"/>
      <xs:any namespace="##other" processContents="lax"/>
    </xs:sequence>
    <xs:attribute name="state" type="ci:state-type"/>
    <xs:anyAttribute namespace="##other" processContents="lax"/>
  </xs:complexType>
  <!--
    RELATIONS TYPE
-->
  <xs:complexType name="relations-type">
    <xs:sequence>
      <xs:element name="relation"
                  type="relation-type"
                  minOccurs="0" maxOccurs="unbounded"/>
      <xs:any namespace="##other" processContents="lax"/>
    </xs:sequence>
    <xs:attribute name="state" type="ci:state-type"/>
    <xs:anyAttribute namespace="##other" processContents="lax"/>
  </xs:complexType>
  <!--
    RELATION TYPE
-->
  <xs:complexType name="relation-type">
    <xs:simpleContent>
      <xs:extension base="xs:string">
        <xs:attribute name="entity" type="xs:anyURI"/>
        <xs:anyAttribute namespace="##other" processContents="lax"/>
      </xs:extension>
    </xs:simpleContent>
  </xs:complexType>
</xs:schema>

10.  Relax NG Grammar

The grammar for the Landmark configuration document extension is:

<!--
    LANDMARKS ELEMENT
-->
parameter &amp;= element landmarks {
    attribute version { xsd:int }
    <!--
        LANDMARK-HOST ELEMENT
    -->
    element landmark-host {
        attribute address { xsd:string },
        attribute port { xsd:int }
    }*
}?

11.  Security Considerations

11.1.  Trust Aspects

TODO: Describing the privacy level for a conference instance; define whether a joining user is allowed to become a member or even focus of a conference.

12.  IANA Considerations

TODO: register Kind-ID code point at the IANA

13.  Acknowledgments

This work was stimulated by fruitful discussions in the P2PSIP working group and SAM research group. We would like to thank all active members for constructive thoughts and feedback. In particular, the authors would like to thank (in alphabetical order) David Bryan, Toerless Eckert, Lothar Grimm, Cullen Jennings, Peter Musgrave, Joerg Ott, Peter Pogrzeba, Brian Rosen, and Jan Seedorf.

14.  References

14.1. Normative References

14.2. Informative References

Appendix A.  Change Log

The following changes have been made from version draft-knauf-p2psip-disco-00.

1. Updated references.
2. Corrected typos.
3. New Section: Conference State Synchronization
4. XML Event Package for Distributed Conferences
5. New mechanism for generating chained conference certificates
6. Allow shared writing of resources using Access Control Policy USER-CHAIN-MATCH
7. Media Negotiation mechanism in a distributed conference
8. New Section: Distributed Conference Control with SIP

Authors' Addresses