Internet Draft


Internet-Draft 
                                                       Yasuhiro Katsube 
                                                        Ken-ichi Nagami  
                                                         Yoshihiro Ohba
                                                       Shigeo Matsuzawa
                                                          Hiroshi Esaki
                                                  (Toshiba Corporation)

                                                          December 1997

                         Cell Switch Router 
               - Architecture and Protocol Overview - 

                  <draft-katsube-csr-arch-00.txt>


Status of this memo

   This document is an Internet-Draft.  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."

   To learn the current status of any Internet-Draft, please check the
   "1id-abstracts.txt" listing contained in the Internet-Drafts Shadow
   Directories on ftp.is.co.za (Africa), nic.nordu.net (Europe),
   munnari.oz.au (Pacific Rim), ds.internic.net (US East Coast), or
   ftp.isi.edu (US West Coast).


Abstract

   This memorandum describes an internetworking architecture of Cell 
   Switch Router (CSR) and related control protocol overview. Cell 
   Switch Router is an ATM-based label switching router that can 
   provide ATM cut-through paths for packet flows with various levels 
   of granularity while retaining current router-based internetworking
   architecture. The proposed architecture is able to provide the 
   cut-through path in response to the creation of IP forwarding entry
   (topology-driven), the arrival of data packets (traffic-driven), 
   and the reception of control packets such as RSVP (request-driven).
   One important feature that is provided by the proposed architecture
   is interoperability with the emerging ATM network platform, 
   specified by the ATM Forum and/or ITU-T, which provides PVC 
   (Permanent Virtual Channel), VP (Virtual Path), or SVC (Switched 
   Virtual Channel) services. 




Katsube, et al.            Expires June 1998                    [Page  1]

Internet Draft    CSR - Architecture and Protocol -        December 1997



Table of Contents

    1.  Introduction ................................................. 2
    2.  Internetworking Architecture Based on Cell Switch Router ..... 3
    2.1  Architectural Overview ...................................... 3
    2.2  Triggers for Cut-Through Path Establishment ................. 5
    2.3  Interoperation with Standard ATM Network Platform ........... 6
    3.  Cell Switch Router Control Mechanism ......................... 8
    3.1  Neighbor Discovery .......................................... 8
    3.2  Two Modes of Protocol Operation ............................. 8
    3.2.1  Distributed Control Mode (DC-mode) ........................ 8
    3.2.1.1  Operational Overview .................................... 8
    3.2.1.2  Examples of DC-mode Operation ........................... 9
    3.2.2  Ingress Control Mode (IC-mode) ............................10
    3.2.2.1  Operational Overview ....................................10
    3.2.2.2  Examples of IC-mode Operation ...........................12
    3.3  Operations Dependent on the Type of Underlying ATM Networks..12
    3.3.1  PVC-based ATM network .....................................13
    3.3.2  SVC-based ATM network .....................................13
    4.  Security Considerations ......................................14
    5.  Intellectual Property Rights Considerations ..................14
    6.  References ...................................................15
    7.  Authors' Addresses ...........................................15



1. Introduction

    The Internet is growing in both size and traffic volume. In
    addition, emerging applications may require specific bandwidth and
    quality of services (QoSs) in addition to best effort. Such changes
    require conventional routers with more sophisticated processing
    capability, which tends to raise the cost of routers, and accelerate
    investigations of a new internetworking architecture that relies on
    powerful datalink switching capabilities.

    The proposed internetworking architecture is composed of i)CSRs 
    that have ATM label switching capability as well as conventional 
    layer 3 packet forwarding, ii)edge nodes that are located at 
    boundaries between the proposed network and legacy networks (e.g., 
    ATM networks with legacy routers, non-ATM networks), and iii)end hosts
    that are capable of speaking with the CSRs. 

    Direct ATM level connectivities from ingress edge nodes (or hosts)
    to egress edge nodes (or hosts) are provided via intermediate CSRs
    on the path with the ATM label switching capability inside (we refer
    to this ATM level connectivity as an "ATM cut-through path"). The
    cut-through path is controlled by each of the CSRs on the path as



Katsube, et al.            Expires June 1998                    [Page  2]

Internet Draft    CSR - Architecture and Protocol -        December 1997


    well as by ingress/egress edge routers or hosts, which keeps
    conventional hop-by-hop control discipline for managing dynamic
    layer 3 state for unicast and multicast routing, RSVP, and so on.
    The proposed internetworking architecture is designed so as to be
    able to operate over standard ATM networks which are compliant with
    ATM Forum and/or ITU-T that support PVC (Permanent Virtual Channel),
    VP (Virtual Path), or SVC (Switched Virtual Channel) services, as
    well as over point-to- point links.


2.  Internetworking Architecture Based on Cell Switch Router

2.1  Architectural Overview 

    Cell Switch Router(CSR) is a key network element of the proposed
    internetworking architecture[13].  It is interconnected with ATM
    networks through ATM UNI interface[1] and provides cell switching
    functionality in addition to conventional IP packet forwarding. It
    is able to concatenate incoming and outgoing ATM VCs at the ATM
    layer, bypassing packet header processing. By carrying out such ATM
    VC concatenations at multiple CSRs consecutively, ATM level
    cut-through paths composed of multiple VCs, each of which connects
    neighboring CSRs (or CSR and hosts/edge routers), can be provided.

    Two different kinds of VCs for transmitting packets are defined
    between neighboring CSRs or between CSR and hosts/edge routers.

    1) Default-VC : a general-purpose VC used by any communications that
    select conventional hop-by-hop IP forwarding paths. All incoming
    cells received from this VC are assembled into IP packets and
    handled based on their IP headers.

    2) Dedicated-VC : used by a specific packet flow, which is specified
    by, e.g., {dst.IP address, src.IP address} or {dst.IP address,
    src.IP address, protocol, dst.port, src.port}. It can be
    concatenated with other Dedicated-VC(s) that accommodate the same
    packet flow as itself, and can constitute an ATM cut-through path
    for those packet flows.

    The route for a cut-through path follows IP routing information in
    each CSR. As shown in Figure 1, packets from an ingress edge router
    (or source host) X.1 to an egress edge router (or destination host)
    Z.1 are transferred over the route X.1 --> CSR1 --> CSR2 --> Z.1
    regardless of whether the communication is on a hop-by-hop IP
    forwarding basis or cut-through path basis.

    An example of the IP packet transmission mechanism is as follows.

    - The ingress edge X.1 checks an identifier of each IP packet flow,



Katsube, et al.            Expires June 1998                    [Page  3]

Internet Draft    CSR - Architecture and Protocol -        December 1997


      which may be the "destination IP address (or prefix)",
      "source/destination IP address (prefix) pair", "source/destination
      IP address and port" or "source IP address and group address".
      Based on such identifiers, it determines over which VC the packet
      should be transmitted.

    - The CSR1 and CSR2 check the VPI/VCI value of each incoming cell.
      When the mapping from the incoming interface/VPI/VCI to outgoing
      interface/VPI/VCI (which may be plural in the case of multicast) is
      found in an ATM forwarding table, it is directly forwarded to the
      specified interface(s) with the new VPI/VCI value through an ATM
      switch module. When the mapping is not found in the ATM forwarding
      table (or the table shows an IP processing module as an output
      interface), the cell is transmitted to an IP processing module and
      assembled into an IP packet and then forwarded to an appropriate
      outgoing interface/VPI/VCI based on the header of the packet.



          IP subnet X           IP subnet Y          IP subnet Z
    <---------------------> <-----------------> <--------------------->

    +-------+ Default  +-------+ Default   +-------+ Default  +-------+
    |       |     -VC  | CSR 1 |     -VC   | CSR 2 |     -VC  |       |
    | Host +=============+   +===============+   +=============+ Host |
    |  or  +-------------+++++---------------+++++-------------+  or  |
    | Edge +-------------+++++---------------+++++-------------+ Edge |
    |  X.1 +-------------+++++---------------+++++-------------+  Z.1 |
    |       |Dedicated |       | Dedicated |       |Dedicated |       |
    +-------+     -VCs +-------+      -VCs +-------+     -VCs +-------+
           <--------------------------------------------------->
                             Cut-through Path


          Figure 1  Internetworking Architecture based on CSR



    A CSR becomes the termination point (egress edge) of a cut-through
    path when it is an edge of the ATM cloud regarding the end-to-end
    path, it is an edge of the CSR-capable router cloud regarding the
    end-to-end path, or it cannot create a Dedicated-VC toward the
    downstream neighbor for the end-to-end path for some reason. The
    CSR-capable router is required to understand a control protocol that
    exchanges mapping information between each dedicated-VC and a
    specific packet flow that will be transmitted over the VC. 

    Note that the egress edge of the cut-through path can also perform
    the cut-through forwarding which bypasses IP processing. Since the



Katsube, et al.            Expires June 1998                    [Page  4]

Internet Draft    CSR - Architecture and Protocol -        December 1997


    egress edge router can memorize the association between each of the
    cut-through paths it terminates and the specific packet flow which
    is transmitted over the path by using the control protocol described
    in section 3, it can obtain the same information as by referring to
    the IP header, i.e., source IP address and destination IP address
    etc., by referring to the identifier of the incoming dedicated-VC
    (VPI/VCI).  That means that the egress edge router can forward
    packets received from the dedicated-VC to a proper downstream
    neighbor without referring to their IP header even though the
    corresponding outgoing dedicated-VC does not exist.

    In the rest of the document, all the CSR-capable devices including
    CSR, edge router, and end host will be referred to as "CSR" for
    simplicity unless there is a need to distinguish among them.


2.2  Triggers for Cut-Through Path Establishment

    CSR is able to initiate cut-through path establishment in response
    to the creation of IP forwarding table entries (topology-driven),
    the arrival of data packets (traffic-driven), and the reception of
    control traffic such as RSVP resource reservation request (request-
    driven)[2].

    This subsection describes three triggers for cut-through path
    establishment by CSRs, together with possible granularity of packet
    flows (conditions that specify the packet flow conveyed over the
    cut-through path) in each of the cases.


1) Topology-driven path establishment :

    In topology-driven, the cut-through path establishment procedure is
    initiated when a new IP level forwarding table entry is created at a
    CSR.  The cut-through path can naturally accommodate aggregated
    packet flow which is specified by, e.g., {ingress edge router,
    dest.prefix}, {ingress edge router, egress edge router}, {*,
    dest.prefix}, or {*, egress edge router}. The latter two may require
    ATM switches to provide flow merging capability while avoiding AAL5
    frame interleaving.

    Note that the topology-driven aggregated paths should not be
    extended beyond the points where the processing for individual
    end-end flows (e.g., packet filtering or QoS differentiation) should
    be carried out but be terminated at those points.


2) Traffic-driven path establishment : 




Katsube, et al.            Expires June 1998                    [Page  5]

Internet Draft    CSR - Architecture and Protocol -        December 1997


    In traffic-driven, the path setup procedure is initiated when the
    CSR determines that it is worth providing the path for a specific
    packet flow, e.g.,

   * transmission of a packet with specific upper layer protocols
     defined by the port ID of TCP/UDP
   * transmission of TCP SYN packet
   * transmission of a packet with specific source/destination IP
     addresses
   * transmission of more than a certain amount of packets in a 
     predetermined period 

    Granularity of the packet flow for the traffic-driven path could be
    {src.IP_addr, dest.IP_addr} as a default, although aggregated
    cut-through paths specified by, e.g., {ingress edge router,
    dest.prefix}, can also be established with traffic-driven.


3) Request-driven path establishment : 

    In request-driven, the path setup procedure is initiated when the
    CSR receives some control messages (requests) which are not specific
    to CSR network but affect the cut-through operation of the CSR.  An
    example of such a control message is RSVP Reservation (Resv) request
    [3] which requests the CSR to provide specific quality of services
    with regard to the packet forwarding.  The CSR which has received a
    Resv message prepares a dedicated-VC for the requested RSVP flow.

    Granularity of the packet flow for the request(RSVP)-driven path
    could be {src.IP_addr/port, dest.IP_addr/port} as a default although
    {src.IP_addr, dest.IP_addr} may be allowable in some
    cases. Reservation styles shared by multiple senders (Wild Card
    style and Shared Explicit style) should be supported by the
    cut-through paths dedicated to individual senders like a Fixed
    Filter style for simplicity.


2.3  Interoperation with Standard ATM Network Platform 

    Interconnecting multiple CSRs through point-to-point (p-p) links
    such as SONET link is a straightforward configuration that is easy
    to implement. In addition, CSRs are designed to be interconnected
    with each other over the emerging ATM network platform that is
    compliant with the ATM Forum or ITU-T standard UNI. That enables the
    ATM network platform to be utilized for Internet/Intranet services
    as well as other services such as telephony and native ATM services. 
    In addition, Internet/Intranet services based on CSRs can coexist
    with and operate over classical IP over ATM[4] networks, which
    provide intra-subnet communications.



Katsube, et al.            Expires June 1998                    [Page  6]

Internet Draft    CSR - Architecture and Protocol -        December 1997



    CSR operations over three types of ATM networks should be taken into
    account;

    1) VP-based ATM network : provides a VP as a tunnel between 
     neighboring CSRs and picks up an unused VCI in the VP each time 
     a dedicated-VC is needed. 

    2) PVC-based ATM network : provides a number of PVCs between
     neighboring CSRs at an initialization phase and picks up one of
     them each time a dedicated-VC is needed. 

    3) SVC-based ATM network : provides an SVC between neighboring CSRs
     through ATM signaling each time a dedicated-VC is needed. 

    In the case 3), it is possible that a number of SVCs are set up in
    advance through ATM signaling and are registered as a stock, which
    we call "VC pool" approach[5]. Then one of them can be picked up
    each time a dedicated-VC is needed, which is a similar approach to
    the PVC case. Although the necessary VC resources with the VC pool
    approach will be larger than the on-demand SVC setup, the VC pool
    approach can decrease latency for setting up the cut-through path as
    there is no delay for ATM signaling.  In addition, the VC pool
    approach can reduce the processing burden for setting up or
    releasing SVCs when the number of surplus VCs between neighboring
    CSRs is controlled between a predetermined lower bound and an upper
    bound. That is, an additional SVC is set up only when the number of
    surplus SVCs becomes smaller than a predetermined lower bound,
    whereas one of the surplus SVCs is released only when it becomes
    larger than a predetermined upper bound. Whether the "on-demand
    setup approach" or "VC pool approach" should be applied is a matter
    of local decision in each network, taking cost and the system
    responsiveness into account.

    ATM networks can be utilized either as logical point-to-point links
    in which IP addresses are assigned to a CSR for each neighbor CSR,
    or as a multi-access (NBMA) link in which a single IP address is
    assigned to a CSR for each logical IP subnet (LIS) composed of
    several neighbor CSRs. When CSRs are operated over multi-access ATM
    LIS environment, point-to-point (p-p) dedicated-VCs or
    point-to-multipoint (p-mp) dedicated-VCs toward its next-hop(s) are
    utilized by each CSR to provide intra-subnet connectivity.  ATMARP
    server[4] or MARS [6] will be used when the CSR does not know an ATM
    address(es) of its neighbor CSR(s) or edge node(s). Note that this
    does not exclude the use of other methods for unicast/multicast ATM
    address resolution.






Katsube, et al.            Expires June 1998                    [Page  7]

Internet Draft    CSR - Architecture and Protocol -        December 1997


3.  Cell Switch Router Control Mechanism

    This section presents an overview of "FANP (Flow Attribute
    Notification Protocol)", that is a protocol between neighboring
    nodes (CSRs, edge routers, and end hosts) in order to establish,
    maintain, and tear down the cut-through paths.

    There are several changes in an updated version of FANP (FANP
    version 2) from FANP v1[7].  They are i)provision of a neighbor
    discovery protocol[8] that enables CSRs to recognize their
    neighbors, ii)adoption of two modes of cut-through path control
    mechanism; Distributed Control mode (DC-mode)[9] and Ingress Control
    mode (IC-mode)[10], and iii)support of interoperability with variety
    of ATM network platforms.  Each of them will be explained in this
    section.


3.1  Neighbor Discovery 

    The neighbor discovery (ND) is used for recognizing neighbors that
    understand FANPv2, obtaining FANPv2 protocol attributes of the
    neighbors, and checking whether consistent states are maintained by
    the neighboring CSRs. A FANP-capable node recognizes that FANP is
    running on a neighbor node provided it periodically receives ND
    messages from the neighbor.

    The ND procedure takes different actions depending on whether the
    underlying network interface is point-to-point or multi-access. 
    Detailed procedure for each type of interface is described in [8].


3.2  Two Modes of Protocol Operation 

    CSR supports two modes of control operation; Distributed Control
    mode (DC-mode)[9] and Ingress Control mode (IC-mode)[10]. An
    overview and examples of application of each operation mode are
    described below.  Which mode of operation each control message
    belongs to is distinguished by a bit in the common header of the
    FANP message.

3.2.1  Distributed Control Mode (DC-mode) 

3.2.1.1  Operational overview

    In the DC-mode, the cut-through path establishment procedure for a
    packet flow is initiated at individual CSRs on its path in a
    distributed manner.  Each of them transmits control messages to its
    downstream neighbor in order to notify the mapping relationship
    between the packet flow and the outgoing dedicated-VC that will



Katsube, et al.            Expires June 1998                    [Page  8]

Internet Draft    CSR - Architecture and Protocol -        December 1997


    convey the flow.  The downstream CSR that has received the control
    message from its upstream neighbor checks the validity of the
    message, memorizes the received mapping information at its incoming
    interface, and transmits an acknowledgment to the upstream neighbor.

    This message exchange with an upstream neighbor at the CSR does not
    initiate the exchange of any control messages with its downstream
    neighbor in the DC-mode. A control message exchange for a flow
    between a pair of neighboring CSRs is initiated and carried out
    independently from the message exchange for the same flow between
    any other pair of CSRs. As a result of control message exchanges
    performed at individual pair of CSRs, a CSR realizes that both the
    incoming and the outgoing dedicated-VC are associated with the same
    packet flow, and then begins cut-through forwarding.

    The release of the cut-through path is also initiated at individual
    CSRs on the path. A CSR that has detected the trigger to release the
    cut-through path transmits control messages to its downstream
    neighbor to cancel the association between the packet flow and the
    outgoing dedicated-VC that conveys the flow. The reception of the
    control message at a CSR from the upstream neighbor does not
    initiate the transmission of the control message to the downstream
    neighbor.

    As no information regarding the cut-through path with edge-to-edge
    importance can be obtained in the DC-mode, an ingress edge knows
    neither the number of hops for the path nor the constitution of the
    loop in the path. It is easy to change the configuration of
    cut-through paths according to dynamic changes in the router state,
    e.g., unicast routing, multicast group membership/routing, and RSVP
    reservation state.

    The common rule with regard to the trigger selection for the cut-
    through path establishment (the condition to initiate the
    cut-through establishment) and release, and the granularity level of
    the packet flow should be configured to individual CSRs in a domain,
    since no such information is conveyed hop-by-hop by the control 
    messages in the DC-mode.


3.2.1.2  Examples of DC-mode Operation

    The DC-mode operation can be adopted for establishing the traffic-
    driven unicast cut-through paths.  Individual CSRs on the path can
    recognize the flow of the layer 3 level end-to-end granularity by
    referring to the data packets (e.g., {src.IP_addr., dest.IP_addr}). 
    They can also commonly recognize the aggregated flow of 
    {src.IP_addr., dest.prefix} granularity individually as long as
    they have the IP forwarding table entry with the same CIDR prefix



Katsube, et al.            Expires June 1998                    [Page  9]

Internet Draft    CSR - Architecture and Protocol -        December 1997


    given by the routing protocol.

    Control of multicast cut-through paths is suitable for the DC-mode
    operation as the management of group membership, which may change
    frequently, is carried out by individual CSRs on the distribution
    tree. Each CSR is able to add a new leaf to or delete a leaf from
    the active cut-through path in response to detection of join or
    leave of a group member at corresponding interfaces. The granularity
    of {src.IP_addr., multicast group addr.} is straightforward although
    the shared tree defined by, e.g., {Rendezveus Point, multicast group
    addr.} is also possible. Concrete triggers for cut-through
    establishment or release may depend on the routing protocol deployed
    and implementation. Arrival of the data traffic, creation of the
    multicast forwarding cache, or reception of PIM Join messages can be
    the trigger.

    Control of cut-through paths in response to the RSVP reservation
    state (request-driven) at individual CSRs on the path will also be
    appropriate for the DC-mode operation. Reception of the reservation
    (Resv) messages at a CSR from its downstream neighbor initiates the
    control message exchange, which notifies the mapping relationship
    between the flow corresponding to the RSVP session and the 
    dedicated-VC that will convey the flow, with the downstream
    neighbor. Then the CSR transmits the Resv message further upstream.
    Here we assume the use of current standard RSVP message format[3]
    with no additional object defined for this purpose. 


3.2.2  Ingress Control Mode (IC-mode)

3.2.2.1  Operational overview

    In the IC-mode, the cut-through path establishment procedure for a
    packet flow is initiated by a CSR that is hoping to become an
    ingress edge of the cut-through path. The ingress edge transmits
    control messages to its downstream neighbor in order to notify the
    mapping relationship between the packet flow and the Dedicated-VC
    that will convey the flow. The message is composed of i)information
    that identifies the flow, ii)information that identifies the
    dedicated-VC, iii)ingress information that is uniquely created by
    the ingress node for the cut-through path, and iv)a hop count that
    is set to one by the ingress edge. The last two information elements
    are specific to the IC-mode operation, which provides a capability
    to prevent the creation of a potential loop of the cut-through path.

    The CSR that has received the messages for the notification of the
    mapping relationship from the upstream neighbor checks the validity
    of the message, and memorizes the received mapping information at
    its incoming interface. Then the CSR reconstructs and transmits



Katsube, et al.            Expires June 1998                    [Page 10]

Internet Draft    CSR - Architecture and Protocol -        December 1997


    control messages further downstream along the path of the flow to
    notify the mapping relationship between the flow notified by the
    upstream neighbor and a dedicated-VC that will convey the flow.

    The same procedure is performed at every CSR along the path of the
    flow until the notification message reaches the CSR that cannot
    extend the cut-through path any more (e.g., an egress edge of the
    CSR cloud).  The CSR that becomes the egress endpoint of the
    cut-through path returns the acknowledgment message to its upstream
    neighbor, which is forwarded hop-by-hop toward the ingress edge (the
    ingress point of the cut-through path).

    This ingress to egress and egress to ingress message forwarding
    facilitates association of related information about the cut-through
    path. As mentioned above, the control message that notifies the
    mapping relationship contains "a hop count from the ingress edge"
    and "an ingress information that is determined by the ingress edge
    for the cut-through path it originates". The hop count value is
    initially set to one by the ingress edge and is incremented by one
    at each of the CSRs along the cut-through path. When a CSR
    recognizes that the hop-count value in the notification message
    reaches a predetermined threshold, it determines that there is a
    potential loop in the cut-through path. The CSR also determines that
    there is a potential loop when it has received a notification
    message that contains the packet flow identifier and ingress
    information, both of which are the same as those already registered,
    but contains the dedicated-VC identifier that is not the same as has
    been registered. The CSR that detects the potential loop stops
    creation of the cut-through path toward its downstream, and returns
    an error message to the upstream neighbor.

    The ingress edge is able to know the number of hops from itself to
    the egress endpoint of the cut-through path by referring to the
    acknowledgement messages initiated by the egress node. The hop-count
    in the acknowledgement message transmitted by the egress node is set
    to one and incremented by one at each of the CSRs along the reverse
    path of the cut-through. The ingress edge may decrement the TTL
    value of the received packet by the number of hops it learned, or
    may decrement that by one.

    In the IC-mode, the CSR that has failed to extend the cut-through
    path toward its downstream for specific reasons retries path
    establishment after some specified intervals. When the CSR has
    succeeded in extending the cut-through path by the retry procedure,
    the number of hops from the CSR to the egress edge may change, which
    means that the number of hops from the ingress edge to the egress
    edge may also change. The CSR that has succeeded in extending the
    cut-through path transmits a message that notifies an updated
    hop-count from the egress edge toward the ingress edge.



Katsube, et al.            Expires June 1998                    [Page 11]

Internet Draft    CSR - Architecture and Protocol -        December 1997



3.2.2.2  Examples of IC-mode Operation

    The IC-mode operation can be adopted for establishing the unicast
    cut-through paths with aggregated packet flows as well as flows with
    fine granularity level. Examples of the flow granularity are
    {ingress edge's IP_addr., dest.prefix} and {ingress edge's IP_addr.,
    egress edge's IP_addr.}. The trigger to establish the cut-through
    path may be either topology-driven or traffic-driven.  Irrespective
    of the flow granularity and cut-through establishment trigger, the
    notification message is processed at each CSR along the path and
    transmitted hop-by-hop until it arrives at an egress edge.


3.3  Operations Dependent on the Type of Underlying ATM Networks

    As described in 2.3, CSRs should be interconnected over the
    following four types of datalink:

   (a) Point-to-point link that interconnects neighboring CSRs directly
   (b) VP-based ATM network that provides logical point-to-point link
   (c) PVC-based ATM network
   (d) SVC-based ATM network

    The core procedure of the FANPv2 control mechanism performed by the
    neighboring CSRs is a notification of the association between a
    packet flow and a dedicated-VC that will convey the packet flow, and
    an acknowledgment to the notification, which we call "flow
    notification procedure". The use of the VPI/VCI value as an
    identifier of a dedicated-VC in the flow notification procedure
    would suffice in the case (a) and (b).  It, however, does not work
    when CSRs are interconnected over an ATM network that provides PVC
    or SVC services. Since VPI/VCI values at the origination point
    (outgoing interface of the upstream CSR) and the termination point
    (incoming interface of the downstream CSR) of a VC are not the same
    when there are standard ATM switches in between, the VPI/VCI value
    cannot be used as an identifier of a dedicated-VC in the flow
    notification procedure.

    A "VCID (Virtual Connection IDentifier)"[11] is introduced instead,
    which can be uniquely identified by the neighboring CSRs to indicate
    a dedicated-VC in the flow notification procedure.  In the case of
    (c)PVC-based ATM network and (d)SVC-based ATM network, a procedure
    to assign the VCID to each dedicated-VC, which we call "VCID
    notification procedure", should be performed before the flow
    notification procedure.  Note that in the case of (a)the p-p link
    and (b)the VP-based ATM network, no explicit VCID notification
    procedure is needed since the VPI/VCI (or just the VCI) value in the
    cell header can be used as the VCID in those cases.



Katsube, et al.            Expires June 1998                    [Page 12]

Internet Draft    CSR - Architecture and Protocol -        December 1997



    The concrete procedure for the VCID notification differs depending
    on the type of the underlying ATM network: PVC-based ATM or
    SVC-based ATM[12].

3.3.1  PVC-based ATM network

    An upstream CSR transmits a message that includes a VCID value over
    the PVC itself (in-band notification). The VCID value is determined
    by the upstream CSR and should be unique to the pair of CSRs. When
    the downstream CSR transmits the acknowledgment of the above in-band
    message to the upstream CSR, the upstream and downstream CSRs can
    share the same identifier VCID for the PVC. This in-band VCID
    notification procedure can be carried out either immediately after
    the PVC setup or at the time when one of PVCs is picked up for a
    specific cut-through path in response to cut-through establishment
    trigger.

3.3.2  SVC-based ATM network 

    There are two ways for the VCID notification in the case of the SVC-
    based ATM network. One is the in-band notification described above;
    an upstream CSR conveys the VCID value over the SVC that is being
    utilized as a dedicated-VC. This would be adopted when the other way
    described below is not applicable.

    An alternative way for the VCID notification is the use of an
    Information Element (IE) with end-to-end significance in a SETUP
    message of ATM signaling.  When an upstream CSR transmits an ATM SVC
    SETUP message toward its downstream neighbor, it determines a unique
    ID value as the VCID and includes it in the IE that is transparently
    conveyed by the ATM switches in between. After that, the upstream
    CSR is able to perform the flow notification procedure by using the
    above VCID as an identifier of the dedicated-VC.  A "user specified
    layer 3 protocol information field" of the BLLI (Broadband Low Layer
    Information) IE is the most appropriate field for this purpose. An
    issue in this method is that the user specified layer 3 protocol
    information field of the BLLI IE is given just 7 bits, which can
    identify only 128 VCs simultaneously between neighboring CSRs.

    In order to resolve the limitation of the 7-bit BLLI IE, an
    additional message exchange that notifies the mapping between the
    7-bit number conveyed in the BLLI IE (a temporal VCID) and an actual
    VCID, which is determined uniquely by the upstream CSR and is given
    enough bit space, is carried out at the VCID notification phase.
    Namely, after the upstream CSR has set up an SVC with the 7-bit
    temporal ID value, it transmits a control message that replaces the
    temporal ID with the VCID.  At this stage, the temporal ID value is
    released and could be reused by the other SVC. The flow notification



Katsube, et al.            Expires June 1998                    [Page 13]

Internet Draft    CSR - Architecture and Protocol -        December 1997


    procedure will be performed by exchanging the message that conveys
    the association between the VCID and a specific packet flow. The
    procedure is:

      1) SVC SETUP with temporal ID in the BLLI IE 
      2) VCID notification : message exchange that replaces "temporal 
         ID" with "VCID" (temporal ID is released for the other SVCs)
      3) Flow notification : message exchange that notifies association
         between the "VCID" and a specific "packet flow"

    In the case that the "VC pool" approach described in 2.3 is applied,
    the above steps 1) and 2) are carried out when preparing a
    dedicated-VC for future use, and the step 3) is carried out when
    establishing a cut-through path.  In the case that the "on-demand
    setup" approach described in 2.3 is applied, it is possible to
    perform the above two steps 2) and 3) by a single message exchange.
    By making the flow notification message convey the "temporal ID",
    "VCID", and "packet flow", the VCID notification step can be merged
    into the flow notification step. The procedure comes to:

      1) SVC SETUP with temporal ID in the BLLI IE 
      2) Flow notification : message exchange that notifies association
         between the "temporal ID", "VCID" and a specific "packet flow"
         (temporal ID is released for the other SVCs)
     

    The detailed procedure for the cut-through establishment and release
    in individual cases is described in [9].


4.  Security Considerations

    Security issues are not discussed in this document.


5.  Intellectual Property Considerations

    Toshiba Corporation may seek patent or other intellectual property
    protection for some or all of the aspects of the technology
    discussed in this document.  If any standards arising from this
    document are or become protected by one or more patents assigned to
    Toshiba Corporation, Toshiba intends to license them on reasonable
    and non- discriminatory terms.



6.  References 

   [1] The ATM Forum, "ATM User-Network Interface Specification, v3.1",



Katsube, et al.            Expires June 1998                    [Page 14]

Internet Draft    CSR - Architecture and Protocol -        December 1997


       Sept. 1994. 
   [2] R. Callon, et al., "A Framework of Multiprotocol Label Switching",
       IETF Internet-Draft (work in progress), draft-ietf-mpls-framework-
       01.txt, July 1997.
   [3] R. Braden, et al., "Resource ReSerVation Protocol (RSVP), 
       Version 1 Functional Specification", IETF RFC2205, Sept. 1997.
   [4] M. Laubach, "Classical IP and ARP over ATM", IETF RFC1577, 
       Oct. 1993. 
   [5] N. Demizu, et al., "VC Pool", IETF Internet-Draft (work in 
       progress), draft-demizu-mpls-vcpool-00.txt, Oct. 1997.
   [6] G. Armitage, "Support for Multicast over UNI 3.0/3.1 based 
       ATM Networks", IETF RFC2022, Nov. 1996
   [7] K. Nagami, et al., "Toshiba's Flow Attribute Notification 
       Protocol (FANP) Specification", IETF RFC2129, April 1997. 
   [8] K. Nagami, et al., "Flow Attribute Notification Protocol 
       Version 2 (FANPv2) Neighbor Discovery", IETF Internet-Draft
       (work in progress), draft-nagami-fanpv2-nd-00.txt, Dec. 1997
   [9] K. Nagami, et al., "Flow Attribute Notification Protocol 
       Version 2 (FANPv2) Distributed Control Mode", IETF Internet-
       Draft (work in progress), draft-nagami-csr-fanpv2-dcmode-00.txt,
       Dec. 1997.
   [10] Y. Ohba, et al., "Flow Attribute Notification Protocol Version 2
        (FANPv2) Ingress Control Mode", IETF Internet-Draft (work in
        progress), draft-ohba-csr-fanpv2-icmode-00.txt, Dec. 1997.
   [11] N. Demizu, et al., "VC-ID: Virtual Connection Identifier", IETF
        Internet-Draft (work in progress), draft-demizu-mpls-vcid-01.txt,
        Oct. 1997. 
   [12] N. Demizu, et al., "ATM SVC Support for ATM-LSRs", IETF Internet-
        Draft (work in progress), draft-demizu-mpls-atm-svc-00.txt, 
        Oct. 1997.
   [13] Y. Katsube, et al., "Toshiba's Router Architecture Extensions 
        for ATM : Overview", IETF RFC2098, Feb. 1997. 


7.  Authors' Addresses

    Yasuhiro Katsube
    Research and Development Center, Toshiba Corporation
    1, Komukai Toshiba-cho, Saiwai-ku, Kawasaki 210
    Japan
    Phone : +81-44-549-2238
    E-mail : katsube@isl.rdc.toshiba.co.jp

    Ken-ichi Nagami
    Research and Development Center, Toshiba Corporation
    1, Komukai Toshiba-cho, Saiwai-ku, Kawasaki 210
    Japan
    Phone : +81-44-549-2238
    E-mail : nagami@isl.rdc.toshiba.co.jp



Katsube, et al.            Expires June 1998                    [Page 15]

Internet Draft    CSR - Architecture and Protocol -        December 1997



    Yoshihiro Ohba
    Research and Development Center, Toshiba Corporation
    1, Komukai Toshiba-cho, Saiwai-ku, Kawasaki 210
    Japan
    Phone : +81-44-549-2238
    E-mail : ohba@csl.rdc.toshiba.co.jp

    Shigeo Matsuzawa
    Research and Development Center, Toshiba Corporation
    1, Komukai Toshiba-cho, Saiwai-ku, Kawasaki 210
    Japan
    Phone : +81-44-549-2238
    E-mail : shigeom@isl.rdc.toshiba.co.jp

    Hiroshi Esaki
    Computer and Network Division, Toshiba Corporation
    1-1-1 Shibaura, Minato-ku, 105-01, 
    Japan
    Phone : +81-3-3457-2563
    E-mail: hiroshi@isl.rdc.toshiba.co.jp































Katsube, et al.            Expires June 1998                    [Page 16]