SIPPING Working Group                           W. Marshall
Internet Draft                                  AT&T
Document: <draft-dcsgroup-sipping-arch-01.txt>
Category: Informational                         M. Osman

                                                F. Andreasen

                                                D. Evans

                                                January 15, 2003

    Architectural Considerations for Providing Carrier Class Telephony
        Services Utilizing Session Initiation Protocol (SIP)-based
                    Distributed Call Control Mechanisms

Status of this Memo

   This document is an Internet-Draft and is in full conformance with
   all provisions of Section 10 of RFC2026.

   Internet-Drafts are working documents of the Internet Engineering
   Task Force (IETF), its areas, and its working groups. Note that
   other groups may also distribute working documents as Internet-
   Drafts. 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."

   The list of current Internet-Drafts can be accessed at

   The list of Internet-Draft Shadow Directories can be accessed at


   This document provides an overview of a SIP-based Distributed Call
   Signaling (DCS) architecture to support carrier class packet-based
   voice, video, and other real time multimedia services.  Companion
   documents address a specific set of SIP 2.0 protocol extensions and
   usage rules that are necessary to implement the DCS architecture in
   an interoperable fashion.

   The DCS architecture takes advantage of endpoint intelligence in
   supporting telephony services without sacrificing the network's
   ability to provide value through mechanisms such as resource
   management, lookup of directory information and translation
   databases, routing services, security, and privacy enforcement.  At
   the same time, the architecture provides flexibility to allow

DCS Group   Category: Informational - Expiration 04/30/03           1

                           DCS Architecture               October 2002

   evolution in the services that may be provided by endpoints and the

   DCS also takes into account the need to manage access to network
   resources and account for resource usage.  The SIP usage rules
   defined in the accompanying IDs specifically address the
   coordination between Distributed Call Signaling and dynamic quality
   of service control mechanisms for managing resources over the access
   network.  In addition, the DCS architecture defines the interaction
   needed between network provided call controllers, known as a "DCS-
   proxy" for supporting these services.

Table of Contents

   Status of this Memo................................................1
   Table of Contents..................................................2
   1. Introduction....................................................2
   2. Background and Motivation.......................................3
   2.1 Requirements And Design Principles.............................4
   2.2 Distributed Call Signaling Architecture........................6
   2.3 Trust Boundary.................................................9
   2.4 Basic Call Flow................................................9
   3. Resource Management............................................12
   4. Distributed Call State.........................................13
   5. DCS Proxy - DCS Proxy Communications...........................15
   6. Privacy........................................................16
   7. Security Considerations........................................18
   8. Notice Regarding Intellectual Property Rights..................18
   9. Informative References.........................................18
   10. Acknowledgements..............................................19
   11. Author's Addresses............................................19
   Full Copyright Statement..........................................21

1. Introduction

   This document provides an overview of a SIP[6]-based Distributed
   Call Signaling (DCS) architecture to support carrier class packet-
   based voice, video, and other real time multimedia services.  The
   DCS architecture and the corresponding SIP protocol enhancements
   (described in companion documents) are being developed as part of
   the cable industry's PacketCable initiative, managed out of
   CableLabs (see PacketCable is defining a series
   of interface specifications that will enable vendors to develop
   interoperable products for providing internet telephony and other
   multimedia services over DOCSIS-enabled cable data networks.
   The DCS architecture described herein has its roots in the DOSA work
   performed by AT&T Laboratories ["Distributed Open Signaling
   Architecture"; Kalmanek, Marshall, Mishra, Nortz, Ramakrishnan, et
   al.; October, 1998]. A relatively large group of vendors have
   cooperated in an intensive effort to develop the DCS architecture
   and SIP protocol extensions described here and in the accompanying

DCS Group   Category: Informational - Expiration 04/30/03           2

                           DCS Architecture               October 2002

   protocol drafts.  Although DCS was originally designed with cable
   access networks in mind, the SIP signaling enhancements have general
   applicability to carrier class Voice over IP (VoIP) services running
   over Quality of Service (QoS) enabled IP networks.

   The authors have submitted this document to the IETF in order to
   provide general information regarding the DCS architecture and to
   convey the motivation behind the SIP enhancements recommended in the
   accompanying protocol drafts.  We believe that providing SIP
   extensions for the concepts and mechanisms described in this set of
   drafts will significantly enhance SIP's ability to function as a
   carrier-class signaling protocol.  Such an enhancement to SIP would
   undoubtedly aid in its widespread acceptance and deployment. We have
   incorporated several useful comments received from the IETF SIP and
   SIPPING Working groups on earlier versions of this and the other DCS
   related drafts.

   The PacketCable Draft Specification for DCS is available from the
   CableLabs website at:

2. Background and Motivation

   The design of the Distributed Call Signaling (DCS) architecture
   recognizes the trend towards use of packet networks as the
   underlying framework for communications.  These networks will
   provide a broad range of services, including traditional best-effort
   data service as well as enhanced, value-added services, such as
   telephony. At the same time, improvements in silicon will reinforce
   the trend towards increased functionality in endpoints.  These
   intelligent endpoints will take advantage of the widespread
   availability of packet networks to enable a rich set of applications
   and services for users.

   However, when the network is used for real-time telephony
   applications, it is essential to have service differentiation at the
   IP layer.  The ability to control and monitor usage is needed for
   the provider to be able to provide service differentiation and to
   derive revenue from the enhanced services.  At the same time, the
   availability of best effort communications and the migration of
   functionality to the endpoints pose a challenge to the provider to
   find incentives for users to use or pay for enhanced services.

   We see three key functions that a provider can offer, as incentives
   to use enhanced services.  First, the network service provider has
   the unique ability to manage and provide network layer quality of
   service.  When users depend on the quality of the service, as with
   telephony, there is a strong incentive to use the enhanced service,
   rather than a best effort service.  Second, the network service
   provider can play an important role as a trusted intermediary.  This
   includes ensuring the integrity of call routing, as well as ensuring
   both the accuracy and the privacy of information that is exchanged.

DCS Group   Category: Informational - Expiration 04/30/03           3

                           DCS Architecture               October 2002

   The service provider can also add value by ensuring that services
   are provided consistently and reliably, even when an endpoint is
   unavailable.  Finally, there are a number of services that may be
   offered more efficiently by the network service provider rather than
   in endpoints.  For example, conference bridging may be more cost
   effective to implement in a multi-point bridge rather than in every
   endpoint attached to the network.

   A key contribution of the DCS architecture is a recognition of the
   need for coordination between call signaling, which controls access
   to telephony specific services, and resource management, which
   controls access to network-layer resources. This coordination is
   designed to meet the user expectations and human factors associated
   with telephony.   For example, the called party should not be
   alerted until the resources necessary to complete the call are
   available.  If resources were not available when the called party
   picked up, the user would experience a call defect.   In addition,
   users expect to be charged for service only after the called party
   answers the phone.  As a result, usage accounting starts only after
   the called party picks up.  Coordination between call signaling and
   resource management is also needed to prevent fraud and theft of
   service.  The coordination between DCS and Dynamic QoS protocols
   ensures that users are authenticated and authorized before receiving
   access to the enhanced QoS associated with the telephony service.

   It is important to be able to deploy a residential telephone service
   at very large scale, cost-effectively.   To achieve this, DCS
   minimizes the messaging overhead on network call servers, and does
   not require these servers to maintain call state for active calls.
   Once a call is established, call state is maintained only where it
   is needed, in keeping (informally) with the principle of "fate-
   sharing" at the endpoints that are involved in the call, and at the
   Edge Routers in the bearer path that are providing differentiated
   service to the media flow.  This allows the network call servers to
   scale to support more users, and imposes less stringent reliability
   requirements on those servers.

   DCS is also designed so that calling users receive consistent
   service even when a called endpoint is unavailable.  For example,
   when an endpoint is unavailable, service logic in a network call
   server can forward telephone calls to a voice mailbox.

2.1 Requirements And Design Principles

   In this section, we briefly describe the application requirements
   that led to a set of DCS signaling design principles.  In its most
   basic implementation, DCS supports a residential telephone service
   comparable to the local telephone services offered today.  In
   addition to the commonly used service features that need to be
   supported, there are important requirements in the areas of
   reliability, performance, and scalability that influence the
   signaling architecture. Supporting an IP telephony service

DCS Group   Category: Informational - Expiration 04/30/03           4

                           DCS Architecture               October 2002

   comparable to the telephony service offered today requires enhanced
   bearer channel and signaling performance, including:

   . Low delay - end-to-end packet delay must be small enough that it
     does not interfere with normal voice conversations. The ITU
     recommends no greater than 300 ms roundtrip delay for telephony

   . Low packet loss - packet loss must be small enough to not
     perceptibly impede voice quality or performance of fax and voice
     band modems.

   . Short post-dial delay - the delay between the user dialing the
     last digit and receiving positive confirmation from the network
     must be short enough that users do not perceive a difference with
     post-dial delay in the circuit switched network or believe that
     the network has failed.

   . Short post pickup delay - the delay between a user picking up a
     ringing phone and the voice path being cut through must be short
     enough so that the "hello" from either the initiator or the
     receiver of the call is not clipped.

   We identify a number of key design principles that arise from the
   requirements and philosophy outlined above:

   1. It is essential to provide differentiated network-layer quality of
     service, while allowing the provider to derive revenues from the
     use of such differentiated services.

   2. The architecture should allow, and even encourage, implementation
     of services and features in the intelligent endpoints, where
     economically feasible, while still retaining value in the network
     and network-based services.

   3. The architecture must ensure that the network is protected from
     fraud and theft of service. The service provider must be able to
     authenticate users requesting service and ensure that only those
     authorized to receive a particular service be able to obtain it.

   4. The architecture must enable the service provider to add value by
     supporting the functions of a trusted intermediary. This includes
     protecting the privacy of calling and called party information,
     and ensuring the accuracy of the information that is provided in
     messages from the network.

   5. The architecture must enable the service provider to give a
     consistent view of basic services and features even when customer
     premise equipment is unavailable, and allow users to take
     advantage of functionality that is provided in the network, when
     it is cost-effective and easy to use.

DCS Group   Category: Informational - Expiration 04/30/03           5

                           DCS Architecture               October 2002

   6. The architecture must be implementable, cost-effectively, at very
     large scale.

2.2 Distributed Call Signaling Architecture

   The Distributed Call Signaling Architecture follows the principles
   outlined above to support a robust telephony service. Figure 1
   introduces the key components in the architecture.

   The architecture assumes a broad range of DCS-compliant endpoints
   that provide telephony service to the user including Media Terminal
   Adapters (MTAs) that may be integrated with a Cable Modem or is a
   standalone device, as well as other endpoints such as personal
   computers. The access network interfaces to an IP backbone through a
   system we refer to as the Edge Router (ER). The ER is the first
   trusted element within the provider's network and is considered to
   be the edge of the network for providing access to differentiated
   quality of service. We believe that the access network is likely to
   manage resources on a per-flow basis, with associated signaling
   mechanisms (such as RSVP). The ER performs resource management, acts
   as a policy enforcement point and as a source of billing

   DCS-proxies (DPs) process call signaling messages and support number
   translation, call routing, feature support and admission control.
   In the context of SIP, a DCS-proxy is a SIP proxy that is involved
   in processing and forwarding of SIP requests. DPs act as trusted
   decision points for controlling when resources are committed to
   particular users. Media servers represent network-based components
   that operate on media flows to support the service. Media servers
   perform audio bridging, play terminating announcements, provide
   interactive voice response services, etc. Finally, PSTN gateways
   interface to the Public Switched Telephone Network.

DCS Group   Category: Informational - Expiration 04/30/03           6

                           DCS Architecture               October 2002

                 | MTA |                MTA: media terminal adapter
                    |                   ER: Edge Router
    Access Network  |
                 | ER |
                    |    +-----------+
                    |----| DCS Proxy |
                    |    +-----------+
                    |    +------------+
  Backbone Network  |----|Media Server|
                    |    +------------+
                    |    +------------+
                    |----|PSTN Gateway|
                    |    +------------+
                 | ER |
    Access Network  |
                 | MTA |
         Figure 1: A Simple view of Components of an IP Telephony
               Architecture used in a HFC Cable Environment.

   Telephony endpoints are considered to be "clients" of the telephony
   service. Consistent with the design principles, the architecture
   allows a range of services to be implemented by intelligent
   endpoints. They collect dialed digits, participate in signaling and
   contain the service logic required for basic call setup and feature
   support. Endpoints also participate in end-to-end capability
   negotiation. However, endpoints are not trusted to provide accurate
   information to the network or to keep information that is received
   private, except when it is in the endpoint's best interests to do

   Access to network resources on a differentiated basis is likely to
   be controlled by the service provider. The ER receives resource
   management requests from endpoints, and is responsible for ensuring
   that packets are provided the QoS they are authorized to receive
   (either through packet marking, or through routing and queuing the
   packets as a specific QoS assured flow).  The ER requires
   authorization from a network entity (on a call-by-call basis for the
   telephony service) before providing access to enhanced QoS for an
   end-to-end IP flow. The obvious point where this policy and control
   function resides is the DCS-proxy (also called a gate-controller,

DCS Group   Category: Informational - Expiration 04/30/03           7

                           DCS Architecture               October 2002

   because of this responsibility for managing access to enhanced QoS).
   Thus, the ER is able to ensure that enhanced QoS is only provided
   for end-to-end flows that have been authorized and for which usage
   accounting is being done. Since the ER knows about the resource
   usage associated with individual IP flows, it generates the usage
   events that allow a user to be charged for service.

   We introduce the concept of a "gate" in the ER, which manages access
   to enhanced quality of service. The gate is a packet classifier and
   policer that ensures that only those IP flows that have been
   authorized by the DCS-proxy are granted access to enhanced QoS in
   the access and backbone networks. Gates are "opened" selectively for
   a flow. For the telephony service, they are opened for individual
   calls. Opening a gate involves an admission control check that is
   performed when a resource management request is received from the
   endpoint for an individual call, and it may involve resource
   reservation in the network for the call if necessary. The packet
   filter in the gate allows a flow of packets to receive enhanced QoS
   for a call from a specific IP source address and port number to a
   specific IP destination address and port number.

   The DCS-proxy, in addition to implementing many of the call control
   functions, is responsible for the policy decision regarding whether
   the gate should be opened. DCS sets up a gate in advance of a
   resource management message. This allows the policy function, which
   is at the DCS-proxy, to be "stateless" in that it does not need to
   know the state of calls that are already in progress.

   DCS-proxies are typically organized in domains. A DCS-proxy is
   responsible for a set of endpoints and the associated ERs. While
   endpoints are not trusted, there is a trust relationship between the
   ER and its associated DCS-proxy, since the DCS-proxy plays a role as
   a policy server controlling when the ER can provide enhanced QoS
   service. There is also a trust relationship among DCS-proxies.
   Details of the security model are outside the scope of this draft.

   The DCS-proxy is designed as a simple transaction server, so that
   the failure of a DCS-proxy does not affect calls in progress. A
   domain will likely have a primary and one or more secondary DCS-
   proxies. If the primary DCS-proxy fails, only calls in a transient
   state are affected. The endpoints involved in those calls will time
   out and retry. All active calls are unaffected. This is possible
   because the DCS-proxy retains no call state for stable calls. We
   believe this design makes the DCS-proxy efficient and highly
   scalable, and keeps the reliability requirements manageable.

   DCS supports inter-working with the circuit switched telephone
   network through PSTN gateways. A PSTN gateway may be realized as a
   combination of a media gateway controller, media gateway, and a
   signaling gateway. A media gateway acts as the IP peer of an
   endpoint for media packets, converting between the data format used
   over the IP network and the PCM format required for transmission
   over the PSTN. The media gateway controller or the signaling gateway

DCS Group   Category: Informational - Expiration 04/30/03           8

                           DCS Architecture               October 2002

   acts as the IP peer of an endpoint for signaling packets, providing
   signaling inter-working between DCS and conventional telephony
   signaling protocols such as ISUP/SS7. When the media gateway
   controller is the peer, it communicates with the signaling gateway.
   A media gateway control protocol is used to control the operation of
   the media gateway from the media gateway controller.

   There are additional system elements that may be involved in
   providing the telephony service. For example, the DCS-proxy may
   interface with other servers that implement the authorization or
   translation functions. Similarly, three way calling may be supported
   using media servers in the network.

2.3 Trust Boundary

   The DCS architecture defines a trust boundary around the various
   systems and servers that are owned, operated by, and/or controlled
   by the service provider. These trusted systems include the proxies
   and various servers such as bridge servers, voicemail servers,
   announcement servers, etc. Outside of the trust boundary lie the
   customer premises equipment, and various media servers operated by
   third-party service providers.

   Certain subscriber-specific information, such as billing and
   accounting information, stays within the trust boundary. Other
   subscriber-specific information, such as endpoint identity, may be
   presented to untrusted endpoints or may be withheld based on
   subscriber profiles.

   The SIP User Agent (UA) may be either within the trust boundary
   (e.g. PSTN gateway) or outside the trust boundary (e.g. MTA),
   depending on exactly what function is being performed and exactly
   how it is being performed.  Accordingly, the procedures followed by
   a User Agent, as contained in the accompanying drafts, are different
   depending on whether the UA is within the trust boundary or outside
   the trust boundary. A trusted user agent is, in almost all cases,
   equivalent to the combination of an untrusted user agent and a

2.4 Basic Call Flow

   Figure 2 presents a high-level overview of a basic MTA-to-MTA call
   flow in DCS. Each MTA is associated with a DCS-proxy, which acts as
   a SIP proxy. When a user goes off-hook and dials a telephone number,
   the originating MTA (MTA-o) collects the dialed digits and sends the
   initial INVITE message in SIP, to the "originating" DCS-proxy (DP-
   o). This INVITE contains SDP proposing a set of codecs that are
   acceptable to MTA-o (and their implied bandwidth requirements), and
   an indication of the (mandatory) QoS preconditions [7] needed for
   the session. DP-o verifies that MTA-o is a valid subscriber of the
   telephony service (using authentication information in the INVITE
   message) and determines whether this subscriber is authorized to
   place this call. DP-o then translates the dialed number into the

DCS Group   Category: Informational - Expiration 04/30/03           9

                           DCS Architecture               October 2002

   address of a "terminating" DCS-proxy (DP-t) and forwards the INVITE
   message to it.

   We assume that the originating and terminating DCS-proxies trust
   each other. DP-o augments the INVITE message that it forwards with
   additional information, such as billing information containing the
   account number of the caller. DP-t then translates the dialed number
   into the address of the terminating MTA (MTA-t) and forwards the
   INVITE message to MTA-t to notify it about the incoming call.

   The initial INVITE message may invoke call feature handling at the
   terminating MTA, such as call-forwarding. Assuming that the call is
   not forwarded, MTA-t negotiates the coding style and bandwidth
   requirements for the media streams. A reliable provisional 1xx
   response to the initial INVITE is sent back through the DCS-proxies.

DCS Group   Category: Informational - Expiration 04/30/03          10

                           DCS Architecture               October 2002

   MTA-o      ER-o       DP-o       DP-t       ER-t      MTA-t

    | INVITE   |          |          |          |          |
    |----------|--------->| INVITE   |          |          |
    |          |          |--------->|          |          |
    |          |          |          |  INVITE  |          |
    |          |          |          |----------|--------->|
    |          |          |          |          |  183 SDP |
    |          |          |          |<---------|----------|
    |          |          |          |          |          |
    |          |  Gate    |  183 SDP |Gate Setup|          |
    |          |  Setup   |<---------|--------->|          |
    |          |<---------|          |          |          |
    |  183 SDP |          |          |          |          |
    |<---------|----------|          |          |          |
    |          |          | PRACK    |          |          |
    |          |   200 OK (acknowledging PRACK) |          |
    |          |          |          |          |          |
    |<---------|--------Reserve Resources-------|--------->|
    |          |          |          |          |          |
    |          |             UPDATE             |          |
    |----------|--------- |----------|----------|--------->|
    |               200 OK (acknowledging UPDATE)          |
    |          |          |          |          | 180 Ring |
    |          |          | 180 Ring |<---------|----------|
    |          | 180 Ring |<---------|          |          |
    |<---------|----------|          |          |          |
    |          |          | PRACK    |          |          |
    |          |   200 OK (acknowledging PRACK) |          |
    |          |          |          |          |          | User
    |          |          |          |          |  200 OK  | Answers
    |          |          |  200 OK  |<---------|----------|
    |          |  200 OK  |<---------|          |          |
    |<---------|----------|          |          |          |
    |   ACK    |          |          |          |          |
    |          |          |          |          |          |
    |<---------|----------Active  Call----------|--------->|
    |          |          |          |          |          | User
    |          |          |          |          |   BYE    | Hangs Up
    |          |          |          |          |  200 OK  |
   Figure 2: A Basic Call Flow, including Resource Management functions

   In the figure, MTA-t sends a 183 SDP message to DP-t. The 183 SDP
   contains a subset of the capabilities in the INVITE message that are
   acceptable to MTA-t. The SDP also carries the QoS preconditions from

DCS Group   Category: Informational - Expiration 04/30/03          11

                           DCS Architecture               October 2002

   the INVITE. DP-t sends a GATE-SETUP message to the terminating ER
   (ER-t), conveying policy instructions allowing ER-t to open a gate
   for the IP flow associated with this phone call. The GATE_SETUP
   message contains billing information containing the account number
   of the subscriber that will pay for the call.

   DP-t forwards the 183 SDP to DP-o. DP-o sends a GATE-SETUP message
   to the originating ER (ER-o) to indicate that it can open a gate for
   the IP flow associated with the phone call. Finally, DP-o forwards
   200 OK to MTA-o. The initial INVITE request and 183 SDP response
   contain a SIP Contact header to indicate the IP address of the
   remote MTA to be used for subsequent end-to-end SIP signaling
   exchanges. MTA-o acknowledges the 183 SDP by sending a PRACK [5]
   directly to MTA-t.

   Once the initial INVITE/183/PRACK exchange has completed, both MTAs
   reserve the resources that will be needed for the media streams.
   Once MTA-o has successfully made its reservation, it sends an UPDATE
   message [7] to MTA-t, which is immediately acknowledged by MTA-t
   with a 200-OK. MTA-o uses the UPDATE message to communicate the fact
   that the desired pre-conditions necessary for the session as
   perceived by MTA-o are satisfied (e.g., successful reservation of
   resources, as perceived by MTA-o). MTA-t acknowledges the UPDATE
   message with a 200 OK final response directly to MTA-o. However,
   resource reservation from MTA-t's perspective may not be completed
   yet. Thus, the 200 OK acknowledging the UPDATE message does not
   indicate successful resource reservation. Once MTA-t successfully
   reserves the resources needed for the call and starts alerting the
   called user, it sends a 180 Ringing through the proxies to indicate
   that the phone is ringing, and that the calling party should be
   given a ringback call progress tone. We have not described, in
   detail, the messaging involved in resource reservation here, as we
   believe that it is appropriate to allow for a variety of resource
   management mechanisms. Thus, the MTA may use the resource management
   mechanism that is most suitable to the network segment that it is
   attached to. When the called party answers by going off-hook, MTA-t
   sends a 200 OK final response through the proxies, which MTA-o
   acknowledges with an end-to-end ACK. At this point the resources
   that were previously reserved are committed to this conversation,
   and the call is "cut through."

   Either party can terminate the call. An MTA that detects an on-hook
   sends a SIP BYE message to the remote MTA, which is acknowledged.

3. Resource Management

   DCS's resource management protocols distinguish between two phases:
   a "Reserve" phase and a "Commit" phase. During the Reserve phase,
   resources are reserved but are not yet made available to the
   endpoint. This ensures that resources are available before ringing
   the far-end telephone. The Commit phase, which commits the resources
   associated with the flow, is initiated after ringing the far-end
   telephone and after the called party picks up. At this point, the

DCS Group   Category: Informational - Expiration 04/30/03          12

                           DCS Architecture               October 2002

   resources are made available to the endpoint, and usage recording is
   started so that the user can be billed for usage. The use of a two-
   phase approach is essential because of the unique requirements
   associated with human communication, such as telephony. Recognition
   of the need for a two phase resource management approach is a
   significant motivation for the call flow adopted in the previous

   Although we believe that issues of billing ought not to be the
   primary consideration in the design of the protocol, the protocol
   design should not preclude the possibility of usage sensitive
   billing. Therefore, in addition to ensuring that resources are
   available before ringing the phone, the two-phase resource
   management protocol also allows us to preserve the semantics of
   billing that users are accustomed to, whereby usage recording is not
   started until the called party picks up the phone. Backbone
   resources are reserved and allocated in the first phase of the two-
   phase resource reservation protocol.  This is important in order to
   limit the impact of backbone resource management on post-pickup
   delay (this minimizes the likelihood of clipping the first few
   syllables of the conversation).

4. Distributed Call State

   In order to provide enhanced services to millions of endpoints, we
   need an architecture that can be implemented cost-effectively at
   very large scale. Just as we enable flexibility by exploiting
   intelligence at the endpoints, services can be provided in a
   scaleable manner by storing the state associated with applications
   at the endpoints, rather than in network servers. Especially with
   telephony, endpoints are directly involved in handling calls and
   therefore need to maintain and use call state. In contrast, while
   network servers may need to be involved when setting up a call to
   gain access to enhanced QoS, there is no fundamental need for those
   servers to be involved throughout the lifetime of the call.
   Maintaining state for every call at network servers, while
   achievable, increases the reliability requirements and load on the
   servers. The less state kept in the network, the better.

   As a result, the DCS-proxies in DCS are designed to be Call
   stateless transaction servers. The proxy maintains SIP transaction
   state. So, when a DCS-proxy processes a service request from an
   endpoint, it maintains state until the transaction is complete, but
   does not maintain any per-call state about active calls in the
   network. There are two major advantages to this design. First the
   reliability of the service does not depend on the reliability of an
   individual DCS-proxy. A DCS-proxy can fail without affecting calls
   that are currently in progress. Second, it removes many complex
   synchronization problems where two (or more) entities need to have
   simultaneously accurate information. Since interactions with the
   DCS-proxies are simple stateless transactions, it is not necessary
   for consecutive calls to be processed by the same DCS-proxy. DCS-
   proxy crashes affect only the transient calls (the calls that are in

DCS Group   Category: Informational - Expiration 04/30/03          13

                           DCS Architecture               October 2002

   the process of being set up), and not stable conversations.
   Further, it is likely that most calls in a transient state can be
   recovered and successfully established through a backup or spare
   DCS-proxy using endpoint retransmission, with no explicit
   synchronization protocol required between the DCS-proxies. We
   believe this design principle will enable us to operate in very
   large scale, cost effectively. Furthermore it places the function of
   managing the state of a call where it belongs - at the endpoint. An
   existing call can only be affected by failures along the path or by
   failure of the endpoints: there are no unnecessary elements involved
   in a call.

   We note that there are many services that involve the use of servers
   or proxy endpoints that communicate directly with clients. Since
   these endpoints are directly involved in providing service, it is
   necessary and appropriate for them to maintain state. Examples of
   proxy endpoints include application layer firewalls, caching
   servers, transcoders, network-based conference bridges, interactive
   voice response systems, and PSTN gateways. The DCS architecture
   models these as end-points, that maintain appropriate call state.

   We now turn to the mechanisms that allow us to avoid state in the
   DCS-proxies. A number of examples of the need for distributed state
   arise in the implementation of telephony features. These give rise
   to two types of information that a DCS-proxy may present to an
   endpoint that may subsequently be given back to the proxy by the
   endpoint. The first type of information is Remote endpoint
   identification, contained in the "P-Asserted-Identity" header. The
   second type of information is associated with an active session that
   an endpoint is participating in. This latter information, stored in
   the "State" header[3], is information that a service provider or
   proxy may need for methods that are invoked by an endpoint related
   to that session. Thus, a DCS-proxy stores the state information
   about the calls at an endpoint in two new headers, "State" and "P-
   Asserted-Identity". The State header is both encrypted and signed by
   the proxy to ensure the privacy and the integrity of the information
   contained in the header. The information that may be contained in
   State includes resource information (such as Gate information) and
   billing information (such as a billing id). The P-Asserted-Identity
   is only encrypted when privacy is requested by the endpoint (covered
   in detail in the Section 6 below.)

   When needed, the endpoint provides the State to the DCS-proxy that
   generated it, which can use the information to provide additional
   functionality. Because the State header is encrypted and signed by
   the DCS-proxy, the information it contains is trusted by the network
   even though the endpoint itself is not trusted. In addition, DCS-
   proxies store service-specific opaque data associated with a call at
   the edge router. Since charging for telephony services may be tied
   to the use of resources, this information is best stored at the edge
   router, where knowledge of resource usage exists.

DCS Group   Category: Informational - Expiration 04/30/03          14

                           DCS Architecture               October 2002

   The endpoint returns the state (possibly both State and P-Asserted-
   Identity) information to the DCS-proxy when it is needed to
   implement specific features. The endpoint cannot interpret the
   information in the encrypted and signed State header (and P-
   Asserted-Identity if it is also encrypted), and any attempt to
   tamper with it can be detected by the DCS-proxy.

   An example of use of the State information is one where a change in
   coding method in the middle of a call (e.g., upon detection of a fax
   tone) may require the proxies to authorize additional resources.
   Services such as call-transfer and three-way-calling require the
   proxy to be involved in authorizing resources for packet flows to
   the new destination(s).

5. DCS Proxy - DCS Proxy Communications

   DCS-proxies implement a set of service-specific control functions
   required to support the telephony service:

   . Authentication and authorization: Since services are only provided
     to authorized subscribers, DCS-proxies authenticate signaling
     messages and authorize requests for service on a call-by-call

   . Name/number translation and call routing: DCS-proxies translate
     dialed E.164 numbers, or names, to a terminating IP address based
     on call routing logic to support a wide range of call features.

   . Service-specific admission control: DCS-proxies can implement a
     broad range of admission control policies for the telephony
     service.  For example, DCS-proxies may provide precedence for
     particular calls (e.g., 911 calls). Admission control may also be
     used to implement overload control mechanisms, e.g. to restrict
     the number of calls to a particular location or to restrict the
     frequency of call setup to avoid signaling overload.

   . Signaling and service feature support: While many service features
     are implemented by endpoints, the DCS-proxy also plays a role in
     feature support. DCS signaling provides a set of service
     primitives to end-points that are mediated by the DCS-proxy. The
     DCS-proxy is involved in implementing service features that depend
     on the privacy of calling information, e.g., caller-ID blocking.
     It also plays a role in supporting service features that require
     users to receive a consistent view of feature operation even when
     an endpoint is down. For example, while an endpoint may normally
     participate in call forwarding, the DCS-proxy can control call
     forwarding on behalf of an endpoint when the endpoint is

   Endpoints MTA-o and MTA-t communicate through the DCS-Proxies DP-o
   and DP-t, as shown in Figure 2. The interface of concern in this
   section is the one between the DCS-Proxies DP-o and DP-t. In
   contrast to a true stateless SIP proxy, the DCS-Proxy maintains

DCS Group   Category: Informational - Expiration 04/30/03          15

                           DCS Architecture               October 2002

   transaction state. During the interval that a call is being setup, a
   DCS-Proxy keeps state related to a request until a response is

   For each call made to a phone number, DP-o may need to perform the
   functions needed for Local Number Portability (LNP). If a LNP
   database lookup is performed and the resulting dialed string is
   modified, DP-o must modify the Request-URI to include the result of
   the LNP lookup. The originating proxy DP-o generates and stores the
   State header. This information is intended to be sent to endpoint
   MTA-o and included with the first response that is returned to MTA-
   o. The originating DCS-Proxy, DP-o, may then use the call state
   information provided to it in the State header to manipulate call-
   legs when requested by MTA-o.

   As with conventional SIP proxies, DP-o adds its address to the top
   of the Via: header list when forwarding the request. In addition, to
   support billing functions for a carrier, DP-o appends opaque billing
   information in the form of P-DCS-Billing-Info. In addition, to
   support the resource management functions (such as manipulating
   Gates for resource management in concert with call-leg
   manipulation), a P-DCS-Gate: header[2] is included. This allows for
   the subsequent generation of requests for access network QoS by the

   We also depend on originating DCS-Proxy, DP-o to be responsible for
   manipulating call legs. For instance, when a call is being
   forwarded, information about the new destination that the call is
   being forwarded to is provided by DP-t to DP-o. The new INVITE is
   then issued from DP-o. The information exchanged between the DCS-
   proxies enables such a function to be performed.

6. Privacy

   Many conventional telephony systems have the ability to provide
   information about the identity of the calling party to the called
   party before the latter accepts the call (such a capability is
   typically termed "Caller-ID"). Systems that support Caller-ID
   usually provide a mechanism that allows the calling party to
   instruct the network to refrain from delivering this information to
   the destination.

   In order for an IP-based network to provide a caller with a similar
   capability, a new SIP header is needed to signal the desire for
   anonymity to the network elements that would otherwise provide the
   caller's identity to the destination party. If a caller desires to
   remain anonymous, several additional changes to standard SIP are

   The triplet {From:, To:, Call-ID:} is used to (at least in RFC 2543-
   based implementations) identify a call leg in both endpoints and in
   proxies. Because call state information is pushed to the edge of the

DCS Group   Category: Informational - Expiration 04/30/03          16

                           DCS Architecture               October 2002

   network, this information must be delivered unchanged to the
   destination endpoint.

   The SIP From: header normally contains information that identifies
   the caller. In order to hide the identity of the caller, the From:
   header information is provided as anonymous.

   Normally, the SIP Call-ID: header also contains information about
   the caller. In the DCS architecture, to support privacy the value of
   the Call-ID: header is a cryptographic hash string that contains no
   information about the user.

   Since all the normally available mechanisms for passing information
   about the caller are no longer available, a new SIP header, P-
   Asserted-Identity[1], is used to pass the caller's identity to the
   destination.  The P-Asserted-Identity header is primarily used for
   endpoint identification. This header contains the information that
   would normally be present in the From: header; the network passes it
   to the destination endpoint only if the caller has not requested
   anonymity.  If the caller had requested anonymity, then the P-
   Asserted-Identity header contains an encrypted string that can be
   used by the proxy in handling further requests.

   If the user at an endpoint wants to return the last call (e.g., by
   dialing *69 on a traditional telephone) the "call return" function
   is invoked. If the user had subscribed to the caller ID service
   feature, the terminating endpoint could store the information (phone
   number or IP address) associated with the last call.  However, it
   may be the case that the user does not subscribe to the feature, or
   the originator of the previous call may have requested that this
   information be blocked in order to retain privacy. In this case,
   call return can be implemented, while keeping the caller's identity
   private, by using the encrypted P-Asserted-Identity header.

   In addition to the usual privacy elements provided by telephone
   systems, IP-based systems must implement methods of hiding the
   source IP address from the destination if the caller requires
   privacy. The entire address must be obscured, since even a few
   address bits may provide partial location information. Likewise, IP
   addresses of the destination should not be revealed to the caller,
   in order to maintain privacy of transfer destinations.

   IP addresses typically appear in the Contact: header; they also
   appear in SDP descriptions contained in SIP messages. These must all
   be protected. We choose to use an application-level anonymizer that
   inspects the SIP call signaling messages and replaces any
   identifying information contained therein in a consistent manner.
   The identifying information is modified such that when the messages
   are delivered to the destination endpoint, any identifying
   information has been replaced with fields that obscure the identity
   of the party seeking privacy.

DCS Group   Category: Informational - Expiration 04/30/03          17

                           DCS Architecture               October 2002

   This mechanism does not require any modification to the call
   signaling initiated by the endpoints: the application-level
   anonymizer performs these functions silently within the network.

7. Security Considerations

   Building an IP-based system that matches services and matches the
   perceived security present in the PSTN is a daunting challenge.

   Certain mechanisms that are defined in SIP for end-to-end security,
   such as S/MIME, are incompatible with services being offered by the
   proxies inside the network.  It is therefore necessary to implement
   security on a hop-by-hop basis and depend on all the proxies and
   trusted User-Agents on the signaling path to properly secure their
   links.  Billing, accounting, and lawful surveillance information is
   particularly sensitive and needs adequate safeguards. It is
   therefore REQUIRED that all proxies and trusted User Agents
   implement security on a hop-by-hop basis with IPsec or with TLS.

   Careful network administration is required to maintain the trust
   boundary, and interconnection agreements between service providers
   must carefully specify the administration requirements.

   Similarly, the links between each MTA and its DCS-Proxy must be
   protected with IPsec or with TLS.

   See section 6 for a separate discussion of the security
   considerations of a caller-id service, and its associated caller-id-
   blocking service.

Each of the extensions described in this document are more properly and
formally defined in separate Internet-Drafts and RFCs.  Each contains
further security considerations related to their specific extension.

8. Notice Regarding Intellectual Property Rights

   The IETF has been notified of intellectual property rights claimed
   in regard to some or all of the specification contained in this
   document.  For more information consult the online list of claimed

9. Informative References

   1. Jennings, C., Peterson, J., and Watson, M., Private Extensions to
     the Session Initiation Protocol (SIP) for Asserted Identity within
     Trusted Networks, RFC3325, November 2002.

   2. "SIP proxy-to-proxy extensions for supporting Distributed Call
     State", Internet Draft: <draft-dcsgroup-sipping-proxy-proxy-
     01.txt>, October 2002.

   3. "SIP extensions for supporting Distributed Call State", Internet
     Draft: <draft-ietf-sip-state-00.txt>, November 2000.

DCS Group   Category: Informational - Expiration 04/30/03          18

                          DCS Architecture               October 2002

   4. "Private Session Initiation Protocol (SIP) Extensions for Media
     Authorization", RFC3313, February 2003.

   5. "Reliability of Provisional Responses in the Session Initiation
     Protocol", RFC3262, June, 2002.

   6. "SIP: Session Initiation Protocol", RFC3261, June 2002.

   7. "Integration of Resource Management and Session Initiation
     Protocol (SIP)", RFC3312, October 2002.

10. Acknowledgements

   The Distributed Call Signaling work in the PacketCable project is
   the work of a large number of people, representing many different
   companies.  The authors would like to recognize and thank the
   following for their assistance: John Wheeler, Motorola; David
   Boardman, Daniel Paul, Arris Interactive; Bill Blum, Jon Fellows,
   Jay Strater, Jeff Ollis, Clive Holborow, Motorola; Doug Newlin,
   Guido Schuster, Ikhlaq Sidhu, 3Com; Jiri Matousek, Bay Networks;
   Farzi Khazai, Nortel; John Chapman, Bill Guckel, Michael Ramalho,
   Cisco; Chuck Kalmanek, Doug Nortz, John Lawser, James Cheng, Tung-
   Hai Hsiao, Partho Mishra, AT&T; Telcordia Technologies; and Lucent
   Cable Communications.

   Previous versions further acknowledged, as co-authors, several
   people for providing the text of this document.  They are: K. K.
   Ramakrishnan (, TeraOptic Networks; Ed Miller
   (, Terayon; Glenn Russell
   (, CableLabs; Burcak Beser
   (, Juniper Networks; Mike Mannette (Michael_ and Kurt Steinbrenner (Kurt_, 3Com; Dave Oran (, Cisco
   Systems; John Pickens (, Com21; Poornima Lalwaney
   (, Nokia; Jon Fellows
   (, Copper Mountain Networks; and Keith
   Kelly (, NetSpeak.

11. Author's Addresses

   Bill Marshall
   Florham Park, NJ  07932

   Matt Osman
   Louisville, CO  80027

DCS Group   Category: Informational - Expiration 04/30/03          19

                          DCS Architecture               October 2002

   Flemming Andreasen
   Edison, NJ

   Doc Evans
   Boulder, CO  80303

DCS Group   Category: Informational - Expiration 04/30/03          20

                          DCS Architecture               October 2002

Full Copyright Statement

   "Copyright (C) The Internet Society (2002). All Rights Reserved.
   This document and translations of it may be copied and furnished to
   others, and derivative works that comment on or otherwise explain it
   or assist in its implementation may be prepared, copied, published
   and distributed, in whole or in part, without restriction of any
   kind, provided that the above copyright notice and this paragraph
   are included on all such copies and derivative works. However, this
   document itself may not be modified in any way, such as by removing
   the copyright notice or references to the Internet Society or other
   Internet organizations, except as needed for the purpose of
   developing Internet standards in which case the procedures for
   copyrights defined in the Internet Standards process must be
   followed, or as required to translate it into languages other than
   English.  The limited permissions granted above are perpetual and
   will not be revoked by the Internet Society or its successors or
   assigns.  This document and the information contained herein is
   provided on an "AS IS" basis and THE INTERNET SOCIETY AND THE

   This memo is filed as <draft-dcsgroup-sipping-arch-01.txt>, and
   expires June 30, 2003.


   Funding for the RFC Editor function is currently provided by the
   Internet Society.

DCS Group   Category: Informational - Expiration 04/30/03          21