Internet DRAFT - draft-woundy-aris-ipswitching

draft-woundy-aris-ipswitching



HTTP/1.1 200 OK
Date: Tue, 09 Apr 2002 12:13:32 GMT
Server: Apache/1.3.20 (Unix)
Last-Modified: Wed, 20 Nov 1996 02:36:31 GMT
ETag: "2e6b54-9e86-32926eaf"
Accept-Ranges: bytes
Content-Length: 40582
Connection: close
Content-Type: text/plain



Internet-Draft                                            Richard Woundy
Expiration Date: May 1997                               Arun Viswanathan
                                                           Nancy Feldman
                                                             Rick Boivie
                                                               IBM Corp.

                                                           November 1996






                ARIS: Aggregate Route-Based IP Switching

                 <draft-woundy-aris-ipswitching-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

   IP based networks use a number of routing protocols, including RIP,
   OSPF, IS-IS, and BGP, to determine how packets ought to be routed.
   Among these protocols, OSPF and BGP are IETF-recommended standards
   that have been extensively deployed and exercised in many networks.
   In this memo, we describe a mechanism which uses these protocols as
   the basis for switching IP datagrams, by the addition of a simple



Woundy, et al.            Expiration: May 1997                  [Page 1]





Internet-Draft                    ARIS                     November 1996


   protocol ("ARIS") that establishes switched paths through a network.
   The ARIS protocol allows us to leverage the advantages of switching
   technologies in an internet network.

   This memo is defined with respect to ATM.  ARIS can be easily
   extended to other switching technologies.


1. Introduction

   In this paper, an Integrated Switch Router (ISR), is a standard IP
   router that has been augmented with ATM virtual circuit (VC)
   switching support.  The ISR at an entry point to the ATM switching
   environment performs standard IP forwarding of datagrams, but the
   "next hop" of the IP forwarding table has been extended to include a
   reference to a VC (switched path).  Each VC may have an endpoint at a
   neighboring router (comparable to today's IP next hops on
   conventional routers), or may traverse a series of ISRs along the
   best IP forwarding path, to an egress ISR endpoint.  This allows
   datagrams to be switched at hardware speeds through an entire ISR
   network.

   The key link between the IP network routing protocols and the ARIS VC
   establishment protocol is the "egress identifier".  The egress
   identifier refers to an egress ISR that forwards traffic either to a
   foreign routing domain, or across an area boundary within the same
   network.  ARIS establishes VCs towards each unique egress identifier.
   Since thousands of IP destinations can map to the same egress
   identifier, ARIS minimizes the number of VCs required in an ISR
   network.  This allows a large network to switch all of its IP
   traffic, resulting in improved aggregate IP throughput.

   Egress ISRs initiate the setup of VCs by sending Establishment
   messages to their upstream neighbors, typically within the same
   domain.  These upstream neighbors forward the messages to their own
   upstream neighbors in Reverse Path Multicast style, after ensuring
   that the VC path is loop-free.  Eventually, all ISRs establish VCs to
   all egress ISRs.

   The VC to an egress point, in general, takes the form of a tree.  A
   tree results because of the "merging" of VCs that occurs at a node
   when multiple upstream VCs for a given egress point are spliced to a
   single downstream VC for that egress point.


2. VC Conservation

   An important goal of the ARIS protocol is to minimize the number of



Woundy, et al.            Expiration: May 1997                  [Page 2]





Internet-Draft                    ARIS                     November 1996


   VCs required by ISRs to switch all IP traffic in the switching cloud.
   Since switches may support only a limited VPI/VCI space, ARIS
   restrains its VC consumption so that VCs are available as needed for
   its own use, as well as for ATM services and other applications, such
   as RSVP.  Further benefits include simplification of network
   management, both for automated tools and for human comprehension and
   analysis, and VC-setup overhead.

   The consumption of VCs is minimized:

        o    by the use of egress routers that may map thousands of IP
             destinations to the same VC, and

        o    by enabling the merging of VCs.

   The merging of VCs enables ARIS to create VC trees, each of which
   connects all of the ingresses to a given egress.  This results in n
   trees, where n is the number of egresses in a network, while still
   providing the benefits of full mesh connectivity (without O(n**2)
   VCs).

   The network routing domain has the greatest performance and VC
   conservation when all routers in the domain are ISRs.  Maximum ARIS
   benefits are also tied closely to an IP network routing topology with
   a high ratio of IP destinations to egresses, as exists in a typical
   IP backbone.  However, ARIS is flexible enough to be highly
   beneficial even in networks with partial ISR deployments or arbitrary
   network routing topologies.

   The ability of the ARIS protocol to conserve the number of VCs
   depends on the hardware capabilities of the ISR.  Some ATM switching
   components can "merge" multiple inbound VCs onto one outbound VC at
   close to standard switching rates.  These merge-capable components
   are able to reassemble cells from the inbound VCs into frames, and
   inject the frames into the outbound VC, without interleaving cells
   from different frames.


3. Loop Prevention

   The ARIS protocol guarantees that VC loops are prevented, even in the
   presence of transient IP routing loops.  With datagram forwarding
   loops, each hop decrements the TTL, so traffic is eventually dropped.
   However, ATM switching does not have a counter similar to the TTL, so
   traffic persists in a VC loop as long as the VC loop exists.  At
   best, the traffic in the VC loop steals bandwidth from other (UBR)
   VCs; at worst, the traffic interferes with IP routing traffic, slows
   down routing convergence, and lengthens the life of the VC loop.



Woundy, et al.            Expiration: May 1997                  [Page 3]





Internet-Draft                    ARIS                     November 1996


   The ARIS protocol avoids creating VC loops by the use of an "ISR ID"
   list, similar in function to the BGP AS_PATH attribute.  Each ISR in
   the VC establishment path appends its own unique ISR ID to each
   establishment message it forwards.  In this way, an ISR is able to
   determine the path a message has traversed, and can ensure that no
   loops are formed.

   Further, if an ISR modifies or deletes an egress due to an IP route
   change, or receives a message that modifies an existing VC to an
   egress, the ISR must unsplice any established upstream VC from the
   downstream VC.  This unsplicing forces inbound traffic to be
   forwarded at the IP network layer, so that transient IP routing
   loops, potentially created by the route change, cannot produce VC
   loops.  The ISR must then re-establish a new VC to the modified
   egress.  Note that ARIS does not attempt to suppress transient IP
   routing protocol loops; it only avoids establishing VC loops with
   this information.


4. ARIS Messages

   The ARIS protocol runs directly over IP, using IP protocol TBD.  The
   following messages are used to manage the ISR switching cloud:

   Init
      This is the first message sent by an ISR to each of its neighbors,
      as notification of its existence.  It is periodically transmitted
      until a positive acknowledgment is received.  The Init message may
      include the neighbor timeout period, acceptable VC label range,
      encapsulation scheme [1], and other adjacency information.

   KeepAlive
      This message is sent by an ISR to inform its neighbors of its
      continued existence.  It is the first message that is transmitted
      after an adjacency has been established.  In order to prevent the
      neighbor timeout period from expiring, ARIS messages must be
      periodically sent to neighbors.  The VC KeepAlive will only be
      sent when no other ARIS messages have been transmitted within the
      periodic interval time.  A recommended transmission interval is
      1/3 of the exchanged neighbor timeout period.

   Establishment
      This message is initiated by the egress ISR, and is periodically
      sent to each upstream neighbor to setup or refresh a VC.  It is
      also sent by any ISR in response to a Trigger message.  Each ISR
      that receives an Establishment message for an egress identifier
      must verify that the path is correct and loop free.  If the
      Establishment message changes a previously known VC path to the



Woundy, et al.            Expiration: May 1997                  [Page 4]





Internet-Draft                    ARIS                     November 1996


      egress identifier, the ISR unsplices the obsolete VC.  The ISR
      creates a downstream VC for the egress identifier, and replies
      with an Acknowledgment message.  It then creates a VC for each of
      its upstream neighbors, forwards the Establishment message to the
      upstream neighbors with its unique ISR ID appended to the ISR ID
      path and the VC label for the upstream neighbor to use for
      forwarding, and waits for an Acknowledgment message.  This pattern
      continues until all ISRs are reached.

   Trigger
      This message is sent by an ISR when it has detected that a local
      IP routing change has modified its path to the egress identifier.
      After unsplicing the obsolete VC, the ISR sends a Trigger message
      to its new downstream neighbor requesting an Establishment
      message.

   Teardown
      This message may be sent when an ISR has lost all connectivity to
      an egress identifier, or when a downstream node to an egress
      identifier has become an upstream node due to routing changes.  In
      the former case, the message traverses the upstream ISR paths of
      the VC, unsplicing each VC along the way.  In the latter case, the
      message is sent single hop to the new upstream (previously
      downstream) node, unsplicing the obsolete VC.

   Acknowledgment
      This message is sent as a response to ARIS messages.  When an ISR
      receives a positive acknowledgment to Init message, it responds
      with a KeepAlive message.  When an ISR receives a positive
      acknowledgment to an Establishment message, it splices the
      upstream VC to the downstream VC.


5. ISR Information Bases

   The ISR needs three logical information bases to compute routes and
   forward datagrams: the routing information base, the forwarding
   information base, and the VC information base.  The first, the
   routing information base (RIB), is used for the computation of best-
   effort routes by various IP routing protocols.  The RIB for the ISR
   is essentially unchanged from the RIB on a standard router.  In the
   ISR context, the RIB is also used to identify egress points and
   egress identifiers for the other two information bases.

   The forwarding information base (FIB) of the ISR has been extended
   beyond the content of the FIB on a standard router to include an
   egress identifier in each next hop entry.  The FIB tends to contain
   many IP destination prefix entries, which point to a small number of



Woundy, et al.            Expiration: May 1997                  [Page 5]





Internet-Draft                    ARIS                     November 1996


   next hop entries that describe the hop-by-hop forwarding
   operation(s).  Next hop entries on the ISR consist of an outgoing
   interface, next hop IP address, egress identifier, and the associated
   established downstream VC.  The association of the next hops with the
   egress identifiers is the responsibility of the routing protocols,
   while the association of the next hop/egress identifiers with the
   established VCs is the responsibility of the ARIS protocol.

   The VC information base (VCIB), which does not exist on a standard
   router, maintains for each egress identifier the upstream to
   downstream VC mappings and related states. This mapping is controlled
   by the ARIS protocol.


6. Egress Identifiers

   The ARIS protocol uses egress identifiers that balance the desire to
   share the same egress identifier among many IP destination prefixes,
   with the desire to maximize switching benefits.  To provide
   flexibility, ARIS supports many types of egress identifiers.  ISRs
   choose the type of egress identifier to use based on routing protocol
   information and local configuration.

   The first type of egress identifier is the IP destination prefix.
   This type results in each IP destination prefix sustaining its own VC
   tree, and thus will not scale in large backbone and enterprise
   networks.  However, this is the only information that some routing
   protocols, such as RIP, can provide.  This type of identifier may
   work well in networks where the number of destination prefixes is
   limited, such as in campus environments, or even in a wide-area
   network of a private enterprise.

   The second type of egress identifier is the egress IP address.  This
   type is used primarily for BGP protocol updates, which carry this
   information in the NEXT_HOP attribute [2].  There are certain types
   of OSPF routes that also use this type.  See "BGP Interaction" and
   "OSPF Interaction" for detailed information.

   The third type of egress identifier is the OSPF Router ID, which
   allows aggregation of traffic on behalf of multiple datagram
   protocols routed by OSPF.  The latest version of OSPF, OSPFv3,
   supports the Router ID for both IP and IPv6 [3].  See "OSPF
   Interaction" for further information.

   The fourth type of egress identifier is the multicast (source, group)
   pair [4], used by multicast protocols, such as DVMRP [5], MOSPF [6]
   and PIM ([7], [8]).  The fifth is the (ingress-of-source, group),
   used for such multicast protocols as MOSPF and PIM.  See "IP



Woundy, et al.            Expiration: May 1997                  [Page 6]





Internet-Draft                    ARIS                     November 1996


   Multicast Interaction" for IP multicast protocol details.

   Other egress identifier types may be defined, including but not
   limited to IS-IS NSAP addresses, NLSP IPX addresses, IPv6 destination
   prefixes, etc.

   A hierarchy amongst the egress identifiers may be introduced to allow
   more flexible control over egress identifier selection.  This allows
   an ISR to autolearn or be configured with non-default egress
   identifiers, and to select which egress identifiers to use in various
   routing situations.

   It should be noted that a network achieves a further performance
   optimization with ARIS when egress identifiers refer to the next hop
   router of the egress ISR.  This allows datagrams to be switched
   entirely from the ingress point in the routing domain to the router
   past the egress ISR.


7. Egress ISRs

   In the ARIS protocol, Establishment messages are originated from the
   egress ISR.  An ISR is considered an egress ISR, with respect to a
   particular egress identifier, under any of the following conditions:

        1.   The egress identifier refers to the ISR itself (including
             one of its directly attached interfaces).

        2.   The egress identifier is reachable via a next hop router
             that is outside the ISR switching infrastructure.

        3.   The egress identifier is reachable by crossing a routing
             domain boundary, such as another area for OSPF summary
             networks, or another autonomous system for OSPF AS
             externals and BGP routes.


8. Establishment Initiation Example


                 +---+
            .... | 2 |
                 +---+ ---> +---+      +--------+
                            | 1 | ---> | Egress | --> ...
                 +---+ ---> +---+      +--------+
            .... | 3 |
                 +---+
              Example: Egress initiates Establishment



Woundy, et al.            Expiration: May 1997                  [Page 7]





Internet-Draft                    ARIS                     November 1996


   a)   The Egress ISR learns of an egress identifier that indicates the
        egress is itself (see "Egress ISRs").  It creates a FIB entry
        for its next hop and egress identifier (itself).

   b)   The Egress creates a VCIB entry with an upstream VC to ISR1, and
        initiates a VC Establishment message with the upstream VC label
        to use, and itself in the ISR ID path.

   c)   ISR1 verifies that the Establish message was received from the
        expected next hop (Egress) by matching its FIB entry, and that
        the ISR ID path is loop free.  It then creates a VCIB entry and
        a downstream VC to the Egress with the given VC label, replaces
        the default VC in the FIB with this new value, and replies to
        the Egress with an Acknowledgment message.

   d)   ISR1 creates an upstream VC to each of its upstream neighbors,
        ISR2 and ISR3, and updates the corresponding VCIB entry.  It
        forwards the Establishment message to each upstream neighbor,
        with its own ISR ID appended to the ISR ID path and with the VC
        label to use.

   e)   When ISR1 receives each acknowledgment from each upstream
        neighbor, it updates the VCIB and splices the corresponding
        upstream VC to its Egress downstream VC.

   All upstream nodes recursively follow the same procedures as ISR1,
   until all Ingress nodes have been added to the VC path to the Egress.

   The Egress ISR is responsible for periodically sending refresh VC
   Establishment messages, to prevent VC timeouts.  If a refresh is not
   received in the allotted time, VCs are unspliced and discarded.


9. Establishment Trigger Example


                                   +---+
                             +-X-> | 2 | ---> ........
                             |     +---+             .
                +---+      +---+                      --> +--------+
           .... | 4 | ---> | 1 |                          | Egress |
                +---+      +---+                      --> +--------+
                             |     +---+             .
                             +---> | 3 | ---> ........
                                   +---+
              Example: ISR1 changes routes from ISR2 to ISR3





Woundy, et al.            Expiration: May 1997                  [Page 8]





Internet-Draft                    ARIS                     November 1996


   a)   ISR1 learns of a new path to the Egress via ISR3 from the
        routing protocols. It removes the FIB and VCIB entries for the
        next hop ISR2/Egress.  ISR1 creates a new FIB entry for the next
        hop ISR3/Egress with a default VC to the next hop.

   b)   ISR1 sends a Trigger message to new downstream node ISR3
        requesting an Establish message for the VC path to the Egress.

   c)   ISR3 creates an upstream VC, updates its corresponding VCIB
        entry, and replies with an Establish message to ISR1, containing
        the full ISR ID path and the VC label.

   d)   ISR1 verifies that the Establish message was received from the
        expected next hop (ISR3), and that the ISR ID path is loop free.
        It then creates a new VCIB entry and downstream VC to ISR3 with
        the given VC label, and replaces the default VC in the FIB with
        this new value.

   e)   ISR1 sends an Acknowledgment message to ISR3.

   f)   ISR3 receives the Acknowledgment, updates the VCIB and splices
        its ISR1 upstream VC to its downstream VC.

   g)   ISR1 appends its ISR ID to the Establish message, and forwards
        the message to ISR4 with the upstream VC label.

   h)   ISR4 verifies the Establish message, updates the VCIB, and
        unsplices the current VC to ISR1/Egress from its upstream
        node(s), and sends an Acknowledgment to ISR1.

   i)   ISR1 receives the Acknowledgment, updates the VCIB and splices
        the ISR4 upstream VC to the ISR3 downstream VC.

   j)   ISR4 appends its ISR ID to the path, and forwards the
        establishment message to its upstream neighbors with a VC label.
        When ISR4 receives an Acknowledgment from an upstream neighbor,
        it updates the VCIB and splices the upstream VC to the ISR1
        downstream VC.

   All upstream nodes recursively follow the same procedure as ISR4,
   until all ingress nodes have been updated.


10. TTL Decrement

   In order to comply with the requirements for IPv4 routers, the IP
   datagram Time-To-Live (TTL) field must be decremented on each hop it
   traverses [9].  Currently, switched packets within ATM cannot



Woundy, et al.            Expiration: May 1997                  [Page 9]





Internet-Draft                    ARIS                     November 1996


   decrement the TTL.  However, ARIS can decrement TTLs appropriately by
   maintaining a hop-count per egress identifier.  This hop-count is
   calculated by including a hop-count field in the Establish message,
   which is incremented at each ISR as it traverses the upstream path.
   Before forwarding a packet on a VC, an ingress ISR decrements the TTL
   by the hop-count plus one.  If the decrement value is greater than or
   equal to the TTL of the packet, the packet is forwarded hop-by-hop.


11. Multipath Considerations

   Many IP routing protocols such as OSPF support the notion of equal-
   cost multipath routes, in which a router maintains multiple next hops
   for one destination prefix when two or more equal-cost paths to the
   prefix exist.  Unfortunately, because of limitations in most ATM
   switching hardware, each path needs its own VC.  Therefore, ingress
   ISRs may maintain a number of VCs to one egress ISR, each VC
   representing a different equal-cost path to the egress.  In this
   case, the ingress ISR will make multipath decisions for traffic on
   behalf of all downstream ISRs.

   Each ISR that receives multiple (legal) Establishment messages from
   downstream ISRs with different paths to the same egress identifier
   can choose one of four different approaches for sending Establishment
   messages upstream.  One approach is to send multiple Establishment
   messages upstream, preserving multiple VCs to the egress ISR.  Each
   Establishment message requires an additional numeric identifier to be
   able to distinguish multiple distinct VCs to the destination, so that
   successive Establishment messages for distinct VCs are not
   misinterpreted as consecutive replacements of the same VC.  When
   multiple Establishment VCs are preserved upstream, they require
   distinct VPI/VCI assignments, which works against conservation of
   VCs.

   Another approach, that conserves VCs at the cost of switching
   performance, is to originate one Establishment message upstream, and
   to forward datagrams at the IP network layer on the multipath point
   ISR.

   A third approach is to propagate only one Establishment message from
   the downstream ISRs to the upstream ISRs, and ignore the content of
   other VC Establishment messages.  This conserves VCs and maintains
   switching performance, but may not balance loads across downstream
   links as well as the first two approaches, even if VCs are
   selectively dropped.

   The final approach is to propagate one Establishment message that
   carries the content of all downstream Establishment messages, so that



Woundy, et al.            Expiration: May 1997                 [Page 10]





Internet-Draft                    ARIS                     November 1996


   only one upstream VC is created to the multipath point.  This
   requires that the ATM switching hardware on the multipath ISR be
   capable of correctly distributing the traffic of upstream VCs onto
   multiple downstream VCs.  Furthermore, the Establishment message to
   send upstream must concatenate the ISR ID lists from downstream
   messages, in order to preserve the VC loop-free property.  The ISR ID
   list concatenation is similar to using AS_SETs for aggregation in the
   BGP protocol.  This final approach has the benefit of both VC
   conservation and performance, although it requires a slightly more
   complex implementation.


12. BGP Interactions with ARIS

   The BGP implementation of the ISR uses the NEXT_HOP attribute as the
   egress identifier.  When the BGP border ISR injects routes into the
   BGP mesh, it may use its own IP address or the address of its
   external BGP peer as the value of the NEXT_HOP attribute.  This
   choice of NEXT_HOP attribute value creates different Establishment
   behaviors with ARIS.

   If the BGP border ISR uses its own IP address as the NEXT_HOP
   attribute in its injected routes, then all of these BGP routes share
   the same egress identifier.  This approach establishes only one tree
   to the BGP border ISR, and the ISR must forward traffic at the IP
   layer towards its external BGP neighbors.

   If the BGP border ISR uses the external BGP peer as the NEXT_HOP
   attribute in its injected routes, then the BGP routes from each
   unique external BGP neighbor share the same egress identifier.  This
   approach establishes one VC tree per external BGP neighbor of the BGP
   border ISR.  The BGP border ISR can switch traffic directly to its
   external BGP neighbors.


13. OSPF Interactions with ARIS

   The OSPF protocol exchanges five types of "link state advertisements"
   to create OSPF routing tables.  All types of advertisements contain
   an "Advertising Router" field, which identifies the OSPF Router ID of
   the router that originates the advertisement.  The ISR uses this OSPF
   Router ID as the egress identifier.

   The use of the OSPF Router ID as an egress identifier allows a new
   level of destination prefix abstraction.  In a typical network, a
   router may be connected to several LANs (Ethernets, Token Rings,
   etc.), and may communicate to remote networks outside of its routing
   domain via adjacent routers.  The remote destination networks may be



Woundy, et al.            Expiration: May 1997                 [Page 11]





Internet-Draft                    ARIS                     November 1996


   injected into the link state routing domain via static configuration,
   or via other routing protocols (such as RIP or BGP).  These local and
   remote networks may be represented in the router forwarding tables as
   many destination prefixes, which cannot be aggregated into shorter
   prefixes (even when using CIDR [10]).  Router labels (OSPF Router ID)
   provide a compact means to represent a number of destination prefixes
   that exit the link state routing domain at the same egress router.
   The association between destination prefixes and router labels is an
   easy by-product of the normal SPF computation.

   The one exception to using the OSPF Router ID is when ISRs receive an
   AS external link advertisement with a non-zero forwarding address.
   The OSPF protocol uses the forwarding address to allow traffic to
   bypass the router that originates the advertisement.  Since the OSPF
   Router ID refers to the bypassed router, it is inadequate as an
   egress identifier in this case.  Instead, the ARIS protocol must use
   the forwarding address as the egress identifier.

   Using the forwarding address as the egress identifier provides
   significant benefits.  Since the AS external forwarding address and
   the BGP NEXT_HOP attribute are both external IP addresses, they are
   compatible types of egress identifiers, which may allow BGP and OSPF
   routes to share the same VC.  Further, the OSPF AS boundary ISR can
   switch traffic directly to its external neighbors, just like BGP.

   The ISR identifies itself as an OSPF egress when the ISR is an area
   border router or an AS boundary router, or when it is directly
   attached to a LAN.


14. IP Multicast Interactions with ARIS

   The ARIS protocol can be used to setup VCs for IP multicast traffic,
   in particular for multicast protocols using Reverse Path Multicasting
   (RPM).  The typical RPM forwarding information base maps a source IP
   network address and multicast group pair, (S,G), to an expected
   incoming interface and a set of outgoing interfaces.  The ISR extends
   the forwarding information base to include one egress identifier per
   (S,G).

   The current choice of egress identifier is the (S,G) pair itself.
   This egress identifier creates one source-based VC tree per source
   address and group pair.  The VC tree carries traffic from the ingress
   ISR to all egress ISRs, using multicast switching within intermediate
   ISRs.  Egress ISRs for multicast are similar to egress ISRs for the
   unicast case, except that multicast egress ISRs are determined by
   group membership location, instead of egress point reachability.  An
   ISR becomes an egress for a particular (S,G) when it forwards traffic



Woundy, et al.            Expiration: May 1997                 [Page 12]





Internet-Draft                    ARIS                     November 1996


   from a source S to a group G over a non-ISR link.

   Having multicast VCs set up on the basis of (S,G) works well with the
   IGMPv3 Group-Source messages, since these IGMP messages can create
   unique trees for each sender within the same group [11].

   An alternative egress identifier choice is to use the "ingress" of
   the source address S in the (S,G) pair.  This choice creates one
   ingress-based VC tree for all of the sources of a group that share a
   given ingress, instead of one for each of the sources.  This permits
   a greater amount of VC aggregation in the ISR cloud.  The ingress of
   a source address is calculated in a similar fashion to calculating an
   egress identifier for a destination prefix. Unfortunately, one cannot
   calculate useful ingress identifiers for DVMRP, for the same reason
   that one cannot calculate useful egress identifiers for RIP.
   Furthermore, since some protocols permit source-specific multicast
   pruning, the multicast distribution tree for a particular group may
   differ according to source address, even if sources share the same
   ingress point.  However, the advantages this approach offers with
   regards to VC conservation on those protocols capable of supporting
   the ingress of source may outweigh the disadvantages of wasting
   bandwidth by sending traffic to leaf networks where a particular
   source may be filtered.

   Based on the topology of the multicast distribution tree, there may
   be multiple egress ISRs for the egress identifier (S,G).  Each ISR
   can send one multicast Establishment message to the one upstream ISR
   on the path back toward the source address.  The ISR ID lists of
   multicast downstream ISRs, with the current ISR ID, are concatenated
   (like BGP AS_SETs) before sending the Establishment message to the
   upstream ISR.

   The observant reader may note that ARIS uses a multicast scheme to
   build unicast VCs, and a unicast scheme to build multicast VCs.


15. Virtual Path Extension

   The ARIS usage of "merged" VC flows requires that ATM switching
   hardware have the capabilty of preventing cell interleaving (see "VC
   Conservation"). Unfortunately, much of the existing ATM switching
   hardware cannot support VC merging.  One solution to this problem is
   to use virtual paths (VPs) to egress points, rather than virtual
   circuits (VCs).  The virtual path extension merges VPs, creating
   trees of VPs to the egress points, instead of merging VCs.  Cell
   interleaving is prevented by the assignment of unique VC identifiers
   (VCIs) within each VP.




Woundy, et al.            Expiration: May 1997                 [Page 13]





Internet-Draft                    ARIS                     November 1996


   The ISRs within a network are assigned unique VCIs to prevent VCI
   collisions when paths from different ISRs are merged.  Each ISR
   requires a block of VCIs as labels to distinguish between cell paths
   to the same egress identifier.  By assigning a unique block of VCIs
   to each ISR, ARIS guarantees that an ISR at a network merge point can
   safely merge upstream VP flows for an egress identifier to a single
   downstream VP without VCI collisions.

   Although the virtual path extension uses VCs much less efficiently
   than a VC merging implementation, it reduces network latency and
   hardware requirements because frame reassembly and re-segmentation is
   not required on intermediate ISRs.  In addition, although this
   variation uses more VC space, the work involved in establishing and
   maintaining switched paths is still O(n).

   An alternative approach to the VC merging problem is to use an end-
   to-end VC label upstream allocation.  This allows the ingress node to
   choose the downstream VC.  In this approach, ISRs acknowledge the
   Establishment message with a label only after they receive an
   Acknowledgment message from their own upstream neighbor. Thus, the
   Establishment message traverses fully to the ingress node before
   being acknowledged.  Ingress ISRs immediately acknowledge the
   Establishment message with the VC label.  These acknowledgements may
   be merged as they travel downstream to the egress node.  This method
   adds latency in the VC set up, and removes the benefits of ARIS VC
   aggregation (O(n**2) versus O(n) VCs).  However, it adds the
   flexibility of performing VC-switching instead of VP-switching, which
   also makes switching possible at the routing boundaries.


16. Multiprotocol Support

   ARIS is capable of multi-protocol support.  Further, the egress
   aggregation feature of ARIS is applicable to those network layer
   topologies (IP, CLNP, IPX) that use a link state routing protocol.
   In particular, integrated IS-IS can calculate routes for CLNP, IP4,
   and IP6 simultaneously (with one Dijkstra calculation), and OSPFv3
   (the new draft of OSPF) can calculate routes for IP4 and IP6
   simultaneously.  Both integrated IS-IS and OSPFv3 use a single router
   label to represent a single router that supports multiple network
   layer protocols.  In this context, ARIS can minimize switching paths
   by using a single switching path for traffic from multiple network
   protocols destined to the same egress multiprotocol router.


17. ARIS And Frame Switching Technology

   The ARIS protocol is easily extendible to other switching



Woundy, et al.            Expiration: May 1997                 [Page 14]





Internet-Draft                    ARIS                     November 1996


   environments.  Though the present document illustrates ARIS in ATM
   cell switching technology, it may later be extended to other
   switching technologies. In fact, ARIS applies well to frame switching
   technology.

   While ARIS solves the problem of cell interleaving in the case of ATM
   by Virtual Path switching, it naturally and easily maps to a frame
   switching environment.  This is due to the fact that multiple
   upstream flows can be merged into a single downstream flow without
   the problems of cell interleaving.


18. Quality of Service

   ARIS can be extended to support Quality of Service (QoS) parameters.
   This will be addressed in a future ARIS revision.


19. ARIS Advantages

   The ARIS approach to IP switching offers the following advantages:

   o    VC aggregation
          - Use of egress identifiers allows many route (CIDR) prefixes
            to map to the same VC
          - Ability to create O(n) ARIS VCs (not O(n**2)) on full mesh,
            where n is the number of nodes in a network.

   o    Low VC setup overhead
          - Few VCs to setup due to routing-based and aggregated VCs
          - VC teardown/setup only on internal route changes

   o    Switches large proportion of traffic
          - All traffic assigned to VCs
          - VCs created independent of traffic patterns

   o    Guaranteed loop-free VC paths

   o    Preserves VC's for RSVP flows or standard ATM connections

   o    Scales to large networks


20. Security Consideration

   Security considerations are not addressed in this memo.





Woundy, et al.            Expiration: May 1997                 [Page 15]





Internet-Draft                    ARIS                     November 1996


21. Intellectual Property Considerations

   International Business Machines Corporation may seek patent or other
   intellectual property protection for some or all of the aspects
   discussed in the forgoing document.


22. Acknowledgements

   The authors wish to acknowledge the following people for their input
   and support: Ed Bowen, Jerry Marin, Wayne Pace, Dean Skidmore, and
   Vijay Srinivasan.


23. References

   [1]  Juha Heinanen, "Multiprotocol Encapsulation over ATM Adaptation
        Layer 5", RFC 1483, Telecom Finland, July 1993

   [2]  Y. Rekhter, T. Li, "A Border Gateway Protocol 4 (BGP-4)", RFC
        1771, IBM Corp, Cisco Systems, March 1995

   [3]  J. Moy, "OSPF Version 2", RFC 1583, Proteon Inc, March 1994

   [4]  S. Deering, "Host extensions for IP multicasting", RFC 1112,
        Stanford University, August 1989

   [5]  D. Waitzman, C. Partridge, S. Deering, "Distance Vector
        Multicast Routing Protocol", RFC 1075, BBN, Stanford University,
        November 1988

   [6]  J. Moy, "Multicast Extensions to OSPF", RFC 1584, Proteon Inc,
        March 1994

   [7]  S. Deering, D. Estrin, D. Farinacci, V. Jacobson, C. Liu, L.
        Wei, P. Sharma, A. Helmy, "Protocol Independent Multicast-Sparse
        Mode (PIM-SM):  Protocol Specification", Internet Draft <draft-
        ietf-idmr-pim-spec-02.txt>, Xerox, Cisco Systems, USC, LBL,
        September 1995

   [8]  D. Estrin, D. Farinacci, V. Jacobson, C. Liu, L. Wei, P. Sharma,
        A. Helmy, "Protocol Independent Multicast-Dense Mode (PIM-DM):
        Protocol Specification", Internet Draft <draft-ietf-idmr-pim-
        dm-spec-01.txt>, USC, Cisco Systems, LBL, January 1996

   [9]  F. Baker (Editor), "Requirements for IP Version 4 Routers", RFC
        1812, Cisco Systems, June 1995




Woundy, et al.            Expiration: May 1997                 [Page 16]





Internet-Draft                    ARIS                     November 1996


   [10] V. Fuller, T. Li, J. Yu, K. Varadhan, "Classless Inter-Domain
        Routing (CIDR):  an Address Assignment and Aggregation
        Strategy", RFC 1519, BARRNET, Cisco Systems, MERIT, OARnet,
        September, 1993

   [11] B. Cain, S. Deering, A. Thyagarajan, "Internet Group Management
        Protocol Version 3", Internet Draft <draft-cain-igmp-00.txt>,
        University of Delaware, Xerox PARC, August 1995


24. Authors' Addresses

   Rick Boivie
   IBM Corp.
   17 Skyline Drive
   Hawthorne, NY 10532
   Phone: +1 914-784-3251
   Email: rboivie@vnet.ibm.com


   Nancy Feldman
   IBM Corp.
   17 Skyline Drive
   Hawthorne, NY 10532
   Phone: +1 914-784-3254
   Email: nkf@vnet.ibm.com


   Arun Viswanathan
   IBM Corp.
   17 Skyline Drive
   Hawthorne, NY 10532
   Phone: +1 914-784-3273
   Email: arunv@vnet.ibm.com


   Richard Woundy
   Continental Cablevision
   The Pilot House - Lewis Wharf
   Boston, MA 02110
   Phone: +1 617-854-3351
   Email: rwoundy@continental.com









Woundy, et al.            Expiration: May 1997                 [Page 17]