path: root/doc/rfc/rfc3488.txt

Network Working Group                                              I. Wu
Request for Comments: 3488                                     T. Eckert
Category: Informational                                    Cisco Systems
                                                           February 2003


                             Cisco Systems
              Router-port Group Management Protocol (RGMP)

Status of this Memo

   This memo provides information for the Internet community.  It does
   not specify an Internet standard of any kind.  Distribution of this
   memo is unlimited.

Copyright Notice

   Copyright (C) The Internet Society (2003).  All Rights Reserved.

Abstract

   This document describes the Router-port Group Management Protocol
   (RGMP).  This protocol was developed by Cisco Systems and is used
   between multicast routers and switches to restrict multicast packet
   forwarding in switches to those routers where the packets may be
   needed.

   RGMP is designed for backbone switched networks where multiple, high
   speed routers are interconnected.

1. Conventions used in this document

   The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT",
   "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this
   document are to be interpreted as described in BCP 14, RFC 2119 [2].

2. Introduction

   IGMP Snooping is a popular, but not well documented mechanism to
   restrict multicast traffic, in switched networks, to those ports that
   want to receive the multicast traffic.  It dynamically establishes
   and terminates multicast group specific forwarding in switches that
   support this feature.








Wu & Eckert                  Informational                      [Page 1]
^L
RFC 3488                  Cisco Systems RGMP               February 2003


   The main limitation of IGMP Snooping is that it can only restrict
   multicast traffic onto switch ports where receiving hosts are
   connected directly or indirectly via other switches.  IGMP Snooping
   can not restrict multicast traffic to ports where at least one
   multicast router is connected.  It must instead flood multicast
   traffic to these ports.  Snooping on IGMP messages alone is an
   intrinsic limitation.  Through it, a switch can only learn which
   multicast flows are being requested by hosts.  A switch can not learn
   through IGMP which traffic flows need to be received by router ports
   to be routed because routers do not report these flows via IGMP.

   In situations where multiple multicast routers are connected to a
   switched backbone, IGMP Snooping will not reduce multicast traffic
   load.  Nor will it assist in increasing internal bandwidth through
   the use of switches in the network.

   In switched backbone networks or exchange points, where predominantly
   routers are connected with each other, a large amount of multicast
   traffic may lead to unexpected congestion.  It also leads to more
   resource consumption in the routers because they must discard the
   unwanted multicast traffic.

   The RGMP protocol described in this document restricts multicast
   traffic to router ports.  To effectively restrict traffic, it must be
   supported by both the switches and the routers in the network.

   The RGMP message format resembles the IGMPv2 message format so that
   existing switches, capable of IGMP Snooping, can easily be enhanced
   with this feature.  Its messages are encapsulated in IPv4 datagrams,
   with a protocol number of 2, the same as that of IGMP.  All RGMP
   messages are sent with TTL 1, to destination address 224.0.0.25. This
   address has been assigned by IANA to carry messages from routers to
   switches [3].

   RGMP is designed to work in conjunction with multicast routing
   protocols where explicit join/prune to the distribution tree is
   performed.  PIM-SM [4] is an example of such a protocol.

   The RGMP protocol specifies operations only for IP version 4
   multicast routing.  IP version 6 is not considered.

   To keep RGMP simple, efficient and easy to implement, it is designed
   for switches to expect RGMP messages from only one source per port.
   For this reason, RGMP only supports a single RGMP enabled router to
   be connected directly to a port of an RGMP enabled switch.  Such a
   topology should be customary when connecting routers to backbone
   switches and thus not pose a limitation on the deployment of RGMP.




Wu & Eckert                  Informational                      [Page 2]
^L
RFC 3488                  Cisco Systems RGMP               February 2003


   All RGMP messages have the following format:

    0                   1                   2                   3
    0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
   +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
   |      Type     |   Reserved    |           Checksum            |
   +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
   |                         Group Address                         |
   +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+

   The reserved field in the message MUST be transmitted as zeros and
   ignored on receipt.

2.1 Type

   There are four types of RGMP messages of concern to the
   router-switch interaction.  The type codes are defined to be the
   highest values in an octet to avoid the re-use of already assigned
   IGMP type codes.

      0xFF = Hello
      0xFE = Bye
      0xFD = Join a group
      0xFC = Leave a group

   Unrecognized message types should be silently ignored.

   Note:

   RGMP and the IANA assignment of address 224.0.0.25 for it predates
   RFC 3228 [9].  RGMP defines Type values which in RFC 3228 are
   assigned to protocol testing and experimentation.  This is not an
   operational issue for RGMP itself because only RGMP packets use the
   IPv4 destination address 224.0.0.25.  The Type values defined above
   are thus ONLY valid in conjunction with the RGMP destination address.

2.2. Checksum

   Checksum covers the RGMP message (the entire IPv4 payload).  The
   algorithm and handling of checksum are the same as those for IGMP
   messages as described in RFC 3376 [5].










Wu & Eckert                  Informational                      [Page 3]
^L
RFC 3488                  Cisco Systems RGMP               February 2003


2.3. Group Address

   In an RGMP Hello or Bye message, the group address field is set to
   zero.

   In an RGMP Join or Leave message, the group address field holds the
   IPv4 multicast group address of the group being joined or left.

2.4 IPv4 header

   RGMP messages are sent by routers to switches. The source IPv4
   address of an RGMP packet is the sending interface IPv4 address of
   the originating router.  The destination IPv4 address of an RGMP
   packet is 224.0.0.25.  Switches supporting RGMP need to listen to
   packets to this group.

3. RGMP Protocol Description

3.1 RGMP Router side Protocol Description

   Backbone switches use RGMP to learn which groups are desired at each
   of their ports.  Multicast routers use RGMP to pass such information
   to the switches.  Only routers send RGMP messages.  They ignore
   received RGMP messages.

   A Router enabled for RGMP on an interface periodically [Hello
   Interval] sends an RGMP Hello message to the attached network to
   indicate that it is RGMP enabled.  When RGMP is disabled on a routers
   interface, it will send out an RGMP Bye message on that interface,
   indicating that it again wishes to receive IPv4 multicast traffic
   promiscuously from that interface.

   When an interface is RGMP enabled, a router sends an RGMP Join
   message out through this interface to each group that it wants to
   receive traffic for from the interface.  The router needs to
   periodically [Join Interval] re-send an RGMP Join for a group to
   indicate its continued desire to receive multicast traffic.

   Routers supporting RGMP MUST NOT send RGMP Join or Leave messages for
   groups 224.0.0.x (x=0...255), 224.0.1.39 and 224.0.1.40.  The latter
   two are known as cisco-rp-announce and cisco-rp-discovery [3].

   When a router no longer needs to receive traffic for a particular
   group, it sends an RGMP Leave message for the group.  For robustness,
   the router MAY send more than one such message.






Wu & Eckert                  Informational                      [Page 4]
^L
RFC 3488                  Cisco Systems RGMP               February 2003


   If IPv4 multicast packets for an undesired group are received at a
   router from a switch, the router MAY send a RGMP Leave message for
   that group to the switch.  These messages are called data-triggered
   RGMP Leave messages and the router SHOULD rate-limit them.  The
   router MAY suppress sending a data triggered RGMP Leave message if it
   has a desired group that has the same MAC destination address as the
   undesired group.  (See RFC 1112 [6] for MAC ambiguity.)  Such
   suppression of data triggered RGMP Leave messages SHOULD be
   configurable if supported.

3.2 RGMP Switch side Protocol Description

   A switch enabled for RGMP on a network consumes RGMP messages
   received from ports of the network and processes them as described
   below.  If enabled for RGMP, the switch must NOT forward/flood
   received RGMP messages out to other ports of the network.

   RGMP on a switch operates on a per port basis, establishing per-group
   forwarding state on RGMP enabled ports.  A port reverts into an RGMP
   enabled port upon receipt of an RGMP Hello message on the port, and a
   timer [5 * Hello Interval] is started.  This timer is restarted by
   each RGMP Hello message arriving on the port.  If this timer expires
   or if it is removed by the arrival of an RGMP Bye message, then the
   port reverts to its prior state of multicast traffic forwarding.

   Correct deployment of RGMP is one RGMP enabled router directly
   connected to a port on a switch that supports RGMP.  The port on the
   switch MAY want to keep track of the IPv4 originator address of the
   RGMP Hello and Bye messages it receives on that port.  In the event
   it receives multiple IPv4 originating addresses in RGMP messages on
   one port, the switch MAY generate an alert to notify the
   administrator.  The switch MAY also have a configuration option that
   will allow for the operator to disable RGMP and have the switch fall
   back to flooding IPv4 multicast on that port, although this is a
   potentially dangerous option.

   By default, connecting two or more RGMP enabled routers to a switch
   port will cause intermittent black holing of IPv4 multicast traffic
   towards these routers.  Black holing occurs when a RGMP Leave is
   received from one router while the other router is still joined.

   This malfunction is not only easily recognized by the actual users
   connected through the routers, but it also adheres to the principle
   that a failure situation causes less traffic than more.  Reverting to
   flooding easily maintains the illusion that everything is working
   perfectly.  The exception is that the traffic constraining benefits





Wu & Eckert                  Informational                      [Page 5]
^L
RFC 3488                  Cisco Systems RGMP               February 2003


   of RGMP are not realized.  This suggests that congestion happens at a
   much later time than the misconfiguration and can then not easily be
   correlated with the cause anymore.

   Because routers supporting RGMP are not required to send RGMP Join or
   Leave messages for groups 224.0.0.x (x=0...255), 224.0.1.39 and
   224.0.1.40, RGMP enabled ports always need to receive traffic for
   these groups.  Traffic for other groups is initially not forwarded to
   an RGMP enabled port.

   RGMP Join and Leave messages are accepted if they arrive on an RGMP
   enabled port, otherwise they will be discarded.  Upon acceptance of
   an RGMP Join message, the switch MUST start forwarding traffic for
   the group to the port.  Upon acceptance of an RGMP Leave message, the
   switch SHOULD stop forwarding traffic for the group to that port.
   The switch's ability to stop forwarding traffic for a group may be
   limited, for example, because of destination MAC based forwarding in
   the switch.  Therefore, it is necessary for the switch to always
   forward traffic for all MAC-ambiguous IPv4 multicast groups (see [6]
   for MAC-ambiguity).

   To stop forwarding of traffic to a group in the event of lost RGMP
   Leave message(s), a switch MAY time out RGMP forwarding state on a
   port for a group [5 * Join Interval] after the last RGMP Join for
   that group has been received on the port.

   Without any layer 2 IPv4 multicast filtering methods running, a
   switch needs to flood multicast traffic to all ports.  If a switch
   does actually run one or more mechanisms beside RGMP to filter IPv4
   multicast traffic, such as IGMP snooping [10], then by default it
   will not flood IPv4 multicast traffic to all ports anymore.  Instead,
   the switch will try to determine which ports still needs to receive
   all IPv4 multicast traffic by default, and which ports do not.

   Compliance with this specification requires that a switch MUST be
   able to elect a port for flooding through the presence of PIM Hello
   messages [4] arriving from the port and also through a manual
   configuration option.  In addition, the switch SHOULD recognize a
   port connected to a router by other appropriate protocol packets or
   dedicated IPv4 multicast router discovery mechanisms such as MRDISC
   [11].  The manual configuration is required to support routers not
   supporting PIM or other methods recognized by the switch.

   Further mechanisms for IPv4 multicast traffic restriction may also be
   used on RGMP enabled ports.  In this case, forwarding for a group on
   the port must be established if either mechanism requires it, and it
   must only be removed if no mechanism requires it anymore.




Wu & Eckert                  Informational                      [Page 6]
^L
RFC 3488                  Cisco Systems RGMP               February 2003


4.   Operational Notes

4.1. Support for networks with multiple switches

   To be simple to implement on switches and resilient in face of
   potential layer 2 network topology changes, RGMP does not specify how
   to restrict multicast traffic on links connecting switches amongst
   each other.  With just RGMP being used, multicast traffic will thus
   be flooded on inter-switch links within a network if at least one
   router is connected to each of the switches.

   This happens implicitly because the switch will not flood/forward
   received RGMP messages out to the inter-switch link and thus the
   switch on the other end will only recognize the port as a router port
   via the PIM Hello messages flooded by the switches.  Manual
   configuration for inter-switch links may be required if non-PIM
   routers are being used, depending on the other capabilities of the
   switch.

   If appropriate, a switch can send out RGMP messages on ports to make
   it look like an RGMP enabled router to a potential switch at the
   other end of the link.  This would constrain IPv4 multicast traffic
   between switches, but this type of "RGMP Spoofing" by the switch is
   outside the scope of this specification.

4.2. Interoperability with RGMP-incapable routers

   Since RGMP messages received at a switch only affect the state of
   their ingress ports, the traffic restriction is applied there only.
   RGMP-incapable routers will receive multicast traffic for all
   multicast groups.

4.3. RGMP and multicast routing protocols

   One result of the simplicity of RGMP are its restrictions in
   supporting specific routing protocols.  The following paragraphs list
   a few known restrictions.

   A router running RGMP on a switched network will not receive traffic
   for a multicast group unless it explicitly requests it via RGMP Join
   messages (besides those group ranges specified to be flooded in 3.1).
   For this reason, it is not possible to run a protocol like PIM
   Dense-Mode or DVMRP across an RGMP enabled network with RGMP enabled
   routers.







Wu & Eckert                  Informational                      [Page 7]
^L
RFC 3488                  Cisco Systems RGMP               February 2003


   In Bidir-PIM, a router elected to be the DF must not be enabled for
   RGMP on the network, because it unconditionally needs to forward
   traffic received from the network towards the RP.  If a router is not
   the DF for any group on the network, it can be enabled for RGMP on
   that network.

   In PIM-SM, directly connected sources on the network can not be
   supported if the elected DR is running RGMP, because this DR needs to
   unconditionally receive traffic from directly connected sources to
   trigger the PIM-SM registering process on the DR.  In PIM-SSM,
   directly connected sources can be supported with RGMP enabled
   routers.

   Both in PIM-SM and PIM-SSM, upstream routers forwarding traffic into
   the switched network need to send RGMP Joins for the group in support
   of the PIM assert process.

5. List of timers and default values

5.1. Hello Interval

   The Hello Interval is the interval between RGMP Hello messages sent
   by an RGMP-enabled router to an RGMP-enabled switch.  Default: 60
   seconds.

5.2. Join Interval

   The Join Interval is the interval between periodic RGMP Join messages
   sent by an RGMP-enabled router to an RGMP-enabled switch for a given
   group address.  Default: 60 seconds.

6. Security Considerations

   The RGMP protocol assumes that physical port security can be
   guaranteed for switch ports from which RGMP messages are accepted.
   Physical port security for RGMP means that physical measures will
   ensure that such ports are dedicatedly connected to one system which
   acts as an RGMP capable router.  This is also the recommended
   configuration to best leverage the benefits of the RGMP protocol
   (e.g., avoiding unwanted third-party IPv4 multicast traffic arriving
   on said ports).

   RGMP specific DoS attacks arise from forged RGMP messages.  If more
   than one system is connected to a port of the RGMP switch, then one
   system may forge RGMP messages and affect the operations of the other
   system(s) on the same port.  This is a potential security risk.





Wu & Eckert                  Informational                      [Page 8]
^L
RFC 3488                  Cisco Systems RGMP               February 2003


   When physical security ensures that only one system is connected to a
   RGMP capable port on a switch, then forged messages from this system
   itself can take effect.  Such forged messages can always be avoided
   by system local measures.

   We consider the ramifications of a forged message of each type:

   Hello Message:

      A forged RGMP Hello message can restrict multicast data towards a
      non-RGMP enabled router on the same port.  This effectively
      introduces a blackholing DoS attack.

   Leave Message:

      A forged RGMP Leave message can restrict IPv4 multicast traffic
      for individual groups toward the port.  The effect is a possible
      blackholing DoS attack similar to an RGMP Hello Message except
      that it does not affect all IPv4 multicast traffic but only that
      of the groups indicated in the forged messages.  It will also only
      affect a port if there officially is only one RGMP enabled router
      connected to it (i.e., if the port is RGMP enabled).

   Bye Message:

      A forged RGMP Bye message can turn the port into being
      RGMP-disabled.  This could, indirectly, cause a DoS attack based
      on the port getting overloaded with IPv4 multicast traffic if the
      network bandwidth of the port was provisioned with the expectation
      that RGMP will suppress unwanted IPv4 multicast messages.

      This type of DoS attack simply re-establishes a port behavior as
      if RGMP was not configured and invalidates the benefit of RGMP.
      This, however, does not introduce an issue that would not have
      been there without RGMP in the first place.

   Join Message:

      A forged RGMP Join message could attract undesired multicast
      packets to the port where it is received from.  The effect is
      similar to an RGMP Bye Message except that it does not affect all
      IPv4 multicast traffic only the groups indicated in the forged
      messages. The message will affect a port only if there officially
      is only one RGMP enabled router connected to it (i.e., if the port
      is RGMP enabled).






Wu & Eckert                  Informational                      [Page 9]
^L
RFC 3488                  Cisco Systems RGMP               February 2003


7.  Normative References

   [1]  Bradner, S., "The Internet Standards Process -- Revision 3", BCP
        9, RFC 2026, October 1996.

   [2]  Bradner, S., "Key words for use in RFCs to Indicate Requirement
        Levels", BCP 14, RFC 2119, March 1997.

   [4]  Estrin, D., Farinacci, D., Helmy, A., Thaler, D., Deering, S.,
        Handley, M., Jacobson, V., Liu, C., Sharma, P. and L. Wei,
        "Protocol Independent Multicast-Sparse Mode (PIM-SM): Protocol
        Specification", RFC 2362, June 1998.

   [5]  Cain, B., Deering, S., Kouvelas, I., Fenner, W. and A.
        Thyagarajan, "Internet Group Management Protocol, Version 3",
        RFC 3376, October 2002.

   [6]  Deering, S., "Host Extensions for IP Multicasting", STD 5, RFC
        1112, August 1989.

   [7]  ANSI/IEEE Std 802.1D 1998 Edition, "Media Access Control (MAC)
        Bridges", 1998.

8. Informative References

   [3]  Internet Multicast Addresses,
        http://www.iana.org/assignments/multicast-addresses

   [8]  Farinacci D., Tweedly D., Speakman T., "Cisco Group Management
        Protocol (CGMP)", 1996/1997
        ftp://ftpeng.cisco.com/ipmulticast/specs/cgmp.txt

   [9]  Fenner, B., "IANA Considerations for IPv4 Internet Group
        Management Protocol (IGMP)", RFC 3228, February 2002.

   [10] Christensen, M. and F. Solensky, "IGMP and MLD snooping
        switches", Work In Progress.

   [11] Biswas, S., Cain, B. and B. Haberman, "IGMP Multicast Router
        Discovery", Work In Progress.

9. Acknowledgments

   The authors would like to thank Gorry Fairhurst, Bill Fenner,
   Giovanni Meo, Mike Norton, Pavlin Radoslavov and Alex Zinin for their
   review of the document and their suggestions.





Wu & Eckert                  Informational                     [Page 10]
^L
RFC 3488                  Cisco Systems RGMP               February 2003


Appendix A. 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
   rights.

Appendix B. Comparison with GARP/GMRP

   This appendix is not part of the RGMP specification but is provided
   for information only.

   GARP/GMRP (defined in IEEE 802.1D [7]) is the ANSI/ISO/IEC/IEEE
   protocol suite to constrain ethernet multicast traffic in bridged
   ethernet networks.  As such it is also a possible alternative to RGMP
   for the purpose of constraining multicast traffic towards router
   ports.  This appendix will explain the motivation not to rely on
   GARP/GMRP and how GARP/GMRP and RGMP differ.

   The key factor in rolling out GARP/GMRP would have been to completely
   replace IGMP Snooping.  This was the design goal of GARP/GMRP.  For
   efficient operations, IGMP Snooping requires hardware filtering
   support in the switch (to differentiate between hosts membership
   reports and actual IPv4 multicast traffic).  Especially in many older
   switches this support does not exist.  Vendors tried to find a way
   around this issue to provide the benefit of constraining IPv4
   multicast traffic in a switched LAN without having to build more
   expensive switch hardware.  GARP/GMRP is one protocol resulting from
   this.  CGMP from Cisco is another one.  While CGMP solves the problem
   without requiring changes to the host stack software, GARP/GMRP
   requires support for it by the host stack.

   Up to date GARP/GMRP has so far not made significant inroads into
   deployed solutions.  IGMP Snooping (and CGMP) are the norm for this
   environment.  In result, GARP/GMRP can not necessarily be expected to
   be supported by layer 2 switches.  In addition, GARP/GMRP does not
   address clearly the issues RGMP tries to solve.  On one hand,
   GARP/GMRP provides much more functionality and as such complexity as
   immediately required.  On the other hand, GARP/GMRP is limited by
   being a standard predominantly for the Ethernet scope.











Wu & Eckert                  Informational                     [Page 11]
^L
RFC 3488                  Cisco Systems RGMP               February 2003


   Beyond the process and applicability reasons, the main differences
   between GARP/GMRP and RGMP are as follows:

   o  GARP/GMRP switches/systems need to send and listen/react to
      GARP/GMRP messages.  In RGMP, routers only need to send RGMP
      messages and switches only need to listen to them.  This protocol
      approach is meant to simplify implementation, operations and
      troubleshooting.

   o  The same switch running RGMP in a backbone network will likely see
      more states then running on the edge only doing IGMP Snooping,
      making it preferable to keep the amount of per group processing
      and memory requirements in RGMP more in bounds than possible in
      IGMP Snooping and GARP/GMRP: In GARP/GMRP, a (multiple) timer
      based state-machines needs to be maintained on a per ethernet
      group address, in RGMP timer maintenance is completely optional
      and there are only two states per group (joined or not joined).

   o  GARP/GMRP is an ethernet level protocol from the IEEE.  It
      supports to constrain traffic for ethernet addresses (groups).
      RGMP does constrain traffic for IPv4 multicast groups.  Today this
      is even beyond the capabilities of typical switch platforms used
      as layer2 switches.  Extensions to support further entities are
      likely easier to come by through extensions to RGMP than to
      GARP/GMRP.

   o  RGMP shares the basic packet format with IGMP (version 2) and is
      as such easy to add to router and switch platforms that already
      support IGMP and IGMP Snooping respectively.  This is especially
      true for switches that in hardware can differentiate between IGMP
      protocol type packets and other IPv4 multicast traffic sent to the
      same (or a MAC ambiguous) group.  In addition, due to the state
      simplicity of RGMP it is easy to integrate IGMP Snooping and RGMP
      operations in the IPv4 multicast control and forwarding plane of a
      switch.

   o  GARP/GMRP supports more than one system (host/router) on a switch
      port which is one reason for its complexity.  In RGMP, this
      configuration is explicitly not supported:  More than one router
      per switched port is not only not a common scenario in today's
      switches layer 2 networks, it is also an undesired configuration
      when unwanted IPv4 multicast traffic is to be kept away from
      routers.








Wu & Eckert                  Informational                     [Page 12]
^L
RFC 3488                  Cisco Systems RGMP               February 2003


   o  GARP/GMRP defines how to constrain multicast traffic between
      switches, another reason for its complexity.  RGMP does not
      explicitly support this as part of the protocol because of the
      following reasons:

      o  It is not necessary to include this function as part of the
         RGMP protocol description because switch implementations can
         transparently decide to support this function (see 4.1 about
         this "RGMP Spoofing").

      o  Important deployments through which large amounts of IPv4
         multicast are moved today are typically single switch
         MIX - Multicast Internet eXchange points.

      o  Avoiding congestion on inter-switch links in general is more
         complex than simply constraining IPv4 multicast traffic to
         paths where it is needed.  With or without IPv4 multicast, the
         aggregate bandwidth needed between switches can easily be the
         aggregate required bandwidth to routers on either sides.  For
         this reason, inter-switch bandwidth is most often appropriately
         over provisioned.  In addition, the likelihood for receiving
         routers to be only on the sources side of an inter-switch link
         is in general deployments rather low.  The cases where traffic
         constrainment on inter-switch links is required and helpful is
         thus limited and can in most cases be avoided or worked around.
         Moving the network to a layer 3 routed network is often the
         best solution, supporting RGMP-Spoofing (see section 4.1) is
         another one.

Appendix C. Possible future extensions / comparison to PIM Snooping

   This appendix is not part of the RGMP specification but is provided
   for information only.

   This appendix presents a discussion of possible extensions to RGMP.
   Included are points on why the extensions are not included and in
   addition a motivation for RGMP in comparison to (PIM) snooping.

   o  Support for multiple switches

      As discussed in "RGMP Spoofing", chapter 4.1 and GARP/GMRP
      comparison in Appendix B.

   o  Support for SSM

      While RGMP works with PIM-SSM, it does not have explicit messages
      for the router to selectively join to (S,G) channels individually.
      Instead the router must RGMP join to all (Si,G) channels by



Wu & Eckert                  Informational                     [Page 13]
^L
RFC 3488                  Cisco Systems RGMP               February 2003


      joining to G.  Extending RGMP to include (S,G) Join/Leaves is
      feasible.  However, currently the majority of switches do not
      support actual traffic constraining on a per channel basis.  In
      addition, the likelihood for actual channel collision (two SSM
      channels using the same group) will only become an issue when SSM
      is fully deployed.

   o  Support for IPv6

      RGMP could easily be extended to support IPv6 by mapping the RGMP
      packet format into the MLD/IPv6 packet format.  This was not done
      for this specification because most switches today do not even
      support MLD snooping.

   o  Support for multiple routers per port

      As discussed in Appendix B.  This is probably one extension that
      should be avoided.  Multiple RGMP router per port are
      inappropriate for efficient multicast traffic constrainment.

   o  Support for non-join based protocols / protocol elements

      For protocols like PIM dense-mode, DVMRP or Bidir-PIM DF routers,
      additional RGMP messages may be added to allow routers to indicate
      that certain group (ranges) traffic need to be flooded from
      (dense-mode) or to (Bidir-PIM) them.

   o  Support for multi-policy switching

      In Multicast Exchange Points (MIXes) environments situations exist
      where different downstream routers for policy reasons need to
      receive the same traffic flow from different upstream routers.

      This problem could be solved by actually providing an upstream
      neighbor field in RGMP Join/Leave messages.  The RGMP switch would
      then forward traffic from one upstream router only to those
      downstream routers who want to have the traffic from exactly this
      upstream router.  This extension would best go in hand with
      changes to the layer 3 routing protocol run between the routers.

   As previously mentioned, RGMP was designed to be easy to implement
   and to support simple layer2 switches.  Implementations could also be
   applied to switches beyond layer 2.  If all the above possible future
   extensions were to be supported by an evolution of RGMP, it would be
   questionable whether such a protocol could be any less complex than
   actually snooping into the layer3 IPv4 routing protocol run between
   routers in a switched LAN.




Wu & Eckert                  Informational                     [Page 14]
^L
RFC 3488                  Cisco Systems RGMP               February 2003


   From the perspective of protocol architecture it is certainly more
   appropriate to have a separate protocol like RGMP or GARP/GMRP for
   this purpose.  Then again, the more complex the requirements are, the
   more duplication of effort is involved and snooping seems to become a
   more attractive option.

   Even though there exists one predominant routing Protocol, PIM, in
   IPv4 multicast, routing with PIM in itself is extremely complex for a
   switch to snoop into.  PIM has two main versions, different
   modes - sparse, dense, bidir, ssm, join / prune / graft messages
   (depending on the mode of the group), various PIM Hello options,
   different versions of asserts, two dynamic mode announcement
   protocols (BSR, AutoRP), and finally supports both IPv4 and IPv6.

   A switch snooping into PIM is very likely to implement just a subset
   of this feature set, making it very hard for the user to determine
   what level of actual traffic constrainment is achieved unless a clear
   specification exists for the implementation (or better the method per
   se.).  In addition, there is always the danger that such a snooping
   implementation may break newer features of the routing protocol that
   it was not designed to handle (likely because they could not have
   been predicted).  For example, this can happen with switches using
   IGMP (v2) snooping implementations that are being subjected to IGMP
   version 3 messages - they break IGMPv3.

   In summary, with PIM still evolving, the approach taken by RGMP is
   the safest one for the immediate problems at hand, and extensions
   like those listed should be considered in time for actual demand.
   (PIM) snooping is a valid alternative once the total amount of
   features that need to be supported makes it an equally attractive
   solution (with respect to complexity) to a dedicated protocol and if
   its functions are well defined to allow predicting its effects - but
   always at the price of possible incompatibilities with upcoming PIM
   protocol extensions unless support for layer 2 switches is explicitly
   considered in moving PIM protocols forward.
















Wu & Eckert                  Informational                     [Page 15]
^L
RFC 3488                  Cisco Systems RGMP               February 2003


Authors' Addresses

   Ishan Wu
   cisco Systems
   170 West Tasman Drive
   San Jose, CA 95134

   Phone: (408) 526-5673
   EMail: iwu@cisco.com


   Toerless Eckert
   cisco Systems
   170 West Tasman Drive
   San Jose, CA 95134

   Phone: (408) 853-5856
   Email: eckert@cisco.com

































Wu & Eckert                  Informational                     [Page 16]
^L
RFC 3488                  Cisco Systems RGMP               February 2003


Full Copyright Statement

   Copyright (C) The Internet Society (2003).  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 INTERNET ENGINEERING
   TASK FORCE DISCLAIMS ALL WARRANTIES, EXPRESS OR IMPLIED, INCLUDING
   BUT NOT LIMITED TO ANY WARRANTY THAT THE USE OF THE INFORMATION
   HEREIN WILL NOT INFRINGE ANY RIGHTS OR ANY IMPLIED WARRANTIES OF
   MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.

Acknowledgement

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



















Wu & Eckert                  Informational                     [Page 17]
^L