Internet Engineering Task Force PIM WG INTERNET-DRAFT Mark Handley/ACIRI draft-ietf-pim-bidir-02.txt Isidor Kouvelas/Cisco Tony Speakman/Cisco Lorenzo Vicisano/Cisco 2 March 2001 Expires: September 2001 Bi-directional Protocol Independent Multicast (BIDIR-PIM) Status of this Document This document is an Internet-Draft and is in full conformance with all provisions of Section 10 of RFC2026. Internet-Drafts are working documents of the Internet Engineering Task Force (IETF), its areas, and its working groups. Note that other groups may also distribute working documents as Internet-Drafts. Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as "work in progress." The list of current Internet-Drafts can be accessed at http://www.ietf.org/ietf/1id-abstracts.txt The list of Internet-Draft Shadow Directories can be accessed at http://www.ietf.org/shadow.html. This document is a product of the IETF PIM WG. Comments should be addressed to the authors, or the WG's mailing list at pim@catarina.usc.edu. Abstract This document discusses Bi-directional PIM, a variant of PIM Sparse-Mode [9] that builds bi-directional shared trees connecting multicast sources and receivers. Bi-directional trees are built using a fail-safe Designated Forwarder (DF) election mechanism operating on each link of a multicast topology. With the assistance of the DF, multicast data is Handley/Kouvelas/Speakman/Vicisano [Page 1] INTERNET-DRAFT Expires: September 2001 March 2001 natively forwarded from sources to the Rendezvous-Point and hence along the shared tree to receivers without requiring source-specific state. The DF election takes place at RP discovery time and provides a default route to the RP thus eliminating the requirement for data-driven protocol events. Note on BIDIR-PIM status The differences between this version of the BIDIR-PIM specification and draft-ietf-pim-bidir-new-00.txt are mostly in the format of the information presented. As BIDIR-PIM has many similarities in operation to Sparse-Mode PIM, the earlier version of this spec relied heavily on the now obsolete PIM-SM [11] specification. This revision removes this dependency and instead references the new Sparse-Mode documentation [9] where necessary. In addition the method in which the protocol specification is presented has been updated to follow the format of [9]. Handley/Kouvelas/Speakman/Vicisano [Page 2] INTERNET-DRAFT Expires: September 2001 March 2001 Table of Contents 1. Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . 5 2. Terminology . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 2.1. Definitions. . . . . . . . . . . . . . . . . . . . . . . . . . 6 2.2. Pseudocode Notation. . . . . . . . . . . . . . . . . . . . . . 7 3. Protocol Specification. . . . . . . . . . . . . . . . . . . . . . 8 3.1. BIDIR-PIM Protocol State . . . . . . . . . . . . . . . . . . . 8 3.1.1. General Purpose State . . . . . . . . . . . . . . . . . . . 9 3.1.2. RP State. . . . . . . . . . . . . . . . . . . . . . . . . . 9 3.1.3. Group State . . . . . . . . . . . . . . . . . . . . . . . . 10 3.1.4. State Summarization Macros. . . . . . . . . . . . . . . . . 11 3.2. PIM Neighbor Discovery . . . . . . . . . . . . . . . . . . . . 12 3.3. Data Packet Forwarding Rules . . . . . . . . . . . . . . . . . 12 3.3.1. Source-Only Branches. . . . . . . . . . . . . . . . . . . . 13 3.4. PIM Join/Prune Messages. . . . . . . . . . . . . . . . . . . . 13 3.4.1. Receiving (*,G) Join/Prune Messages . . . . . . . . . . . . 14 3.4.2. Sending Join/Prune Messages . . . . . . . . . . . . . . . . 16 3.5. Designated Forwarder (DF) Election . . . . . . . . . . . . . . 19 3.5.1. DF Requirements . . . . . . . . . . . . . . . . . . . . . . 19 3.5.2. DF Election description . . . . . . . . . . . . . . . . . . 20 3.5.2.1. Bootstrap Election . . . . . . . . . . . . . . . . . . . 20 3.5.2.2. Loser Metric Changes . . . . . . . . . . . . . . . . . . 21 3.5.2.3. Winner Metric Changes. . . . . . . . . . . . . . . . . . 22 3.5.2.4. Winner Loses Path. . . . . . . . . . . . . . . . . . . . 22 3.5.2.5. Late Router Starting Up. . . . . . . . . . . . . . . . . 22 3.5.2.6. Winner Dies. . . . . . . . . . . . . . . . . . . . . . . 22 3.5.3. Election Protocol Specification . . . . . . . . . . . . . . 23 3.5.3.1. Election State . . . . . . . . . . . . . . . . . . . . . 23 3.5.3.2. Election Messages. . . . . . . . . . . . . . . . . . . . 23 3.5.3.3. Election Events. . . . . . . . . . . . . . . . . . . . . 24 3.5.3.4. Election State Transitions . . . . . . . . . . . . . . . 24 3.6. Timers and Constants . . . . . . . . . . . . . . . . . . . . . 28 3.7. PIM DF Election Packet Formats . . . . . . . . . . . . . . . . 31 3.7.1. Backoff Message . . . . . . . . . . . . . . . . . . . . . . 32 3.7.2. Pass Message. . . . . . . . . . . . . . . . . . . . . . . . 33 4. RP Discovery. . . . . . . . . . . . . . . . . . . . . . . . . . . 33 5. Security Considerations . . . . . . . . . . . . . . . . . . . . . 34 5.1. Appendix A: Election Reliability Enhancements. . . . . . . . . . . . . . . . . . . . . . . . . . . . 34 5.1.1. A.1 Missing Pass. . . . . . . . . . . . . . . . . . . . . . 34 5.1.2. A.2 Periodic Winner Announcement. . . . . . . . . . . . . . 35 5.2. Appendix B: Interoperability with legacy code. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35 5.3. Appendix C: Comparison with PIM-SM . . . . . . . . . . . . . . 35 6. Todo list.... . . . . . . . . . . . . . . . . . . . . . . . . . . 36 7. Authors' Addresses. . . . . . . . . . . . . . . . . . . . . . . . 36 Handley/Kouvelas/Speakman/Vicisano [Page 3] INTERNET-DRAFT Expires: September 2001 March 2001 8. Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . 37 9. References. . . . . . . . . . . . . . . . . . . . . . . . . . . . 37 10. Index. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39 Handley/Kouvelas/Speakman/Vicisano [Page 4] INTERNET-DRAFT Expires: September 2001 March 2001 1. Introduction This document specifies Bi-directional PIM, a variant of PIM Sparse-Mode (PIM-SM) [9] that builds bi-directional shared trees connecting multicast sources and receivers. PIM-SM constructs uni-directional shared trees that are used to forward data from senders to receivers of a multicast group. PIM-SM also allows the construction of source specific trees, but this capability is not related to the protocol described in this document. The shared tree for each multicast group is rooted at a multicast router called the Rendezvous Point (RP). Different multicast group ranges can use separate RPs within a PIM domain. In unidirectional PIM-SM, there are two possible methods for distributing data packets on the shared tree. These differ in the way packets are forwarded from a source to the RP: o Initially when a source starts transmitting, its first hop router encapsulates data packets in special control messages (Registers) which are unicast to the RP. After reaching the RP the packets are decapsulated and distributed on the shared tree. o A transition from the above distribution mode can be made at a later stage. This is achieved by building source specific state on all routers along the path between the source and the RP. This state is then used to natively forward packets from that source. Both these mechanisms suffer from problems. Encapsulation results in significant processing, bandwidth and delay overheads. Forwarding using source specific state has additional protocol and memory requirements. Bi-directional PIM dispenses with both encapsulation and source state by allowing packets to be natively forwarded from a source to the RP using shared tree state. For a complete discussion of the pros and cons of Bi- directional PIM consult appendix C. 2. Terminology In this document, the key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" are to be interpreted as described in RFC 2119 and indicate requirement levels for compliant PIM-SM implementations. Handley/Kouvelas/Speakman/Vicisano Section 2. [Page 5] INTERNET-DRAFT Expires: September 2001 March 2001 2.1. Definitions This specification uses a number of terms to refer to the roles of routers participating in BIDIR-PIM. The following terms have special significance for BIDIR-PIM: MRIB Multicast Routing Information Base. This is the multicast topology table, which is typically derived from the unicast routing table, or routing protocols such as MBGP that carry multicast-specific topology information. It is used by PIM for establishing the RPF interface (used in the forwarding rules). In PIM-SM the MRIB is also used to make decisions regarding where to forward Join/Prune messages whereas in BIDIR-PIM it is used as a source for routing metrics for the DF election process. Rendezvous Point (RP): An RP is a router that has been configured to be used as the root of the distribution tree for a range of multicast groups. Join messages from receivers for a group are sent towards the RP. Upstream Towards the root (Rendezvous-Point) of the tree. The direction used by packets traveling from sources to the RP. Downstream Away from the root of the tree. The direction on which packets travel from the RP to receivers. Designated Forwarder (DF): The protocol presented in this document is largely based on the concept of a Designated Forwarder (DF). A single DF exists for each RP on every link within a BIDIR-PIM domain (this includes both multi-access and point-to-point links). The DF is the router on the link with the best unicast route to the RP. A DF for a given RP is in charge of forwarding downstream traffic onto the link, and forwarding upstream traffic from the link towards the RP. It does this for all the bi-directional groups served by the RP. The DF on a link is also responsible for interpreting IGMP information from local receivers and originating Join messages towards the RP. RPF Interface RPF stands for "Reverse Path Forwarding". The RPF Interface of a router with respect to an address is the interface that the MRIB indicates should be used to forward packets to that address. In the case of a BIDIR-PIM multicast group, the RPF interface is the interface that would be used to send packets to the RP for the group. Handley/Kouvelas/Speakman/Vicisano Section 2.1. [Page 6] INTERNET-DRAFT Expires: September 2001 March 2001 RPF Neighbor The RPF Neighbor of a router with respect to an address is the neighbor that the MRIB indicates should be used to forward packets to that address. Note that in BIDIR-PIM, the RPF neighbor for a group is not necessarily the router on the RPF interface that Join messages for that group would be directed to (Join messages are directed to the DF on the RPF interface for the group). TIB Tree Information Base. This is the collection of state at a PIM router that has been created by receiving PIM Join/Prune messages, PIM DF election messages and IGMP information from local hosts. It essentially stores the state of all multicast distribution trees at that router. MFIB Multicast Forwarding Information Base. The TIB holds all the state that is necessary to forward multicast packets at a router. However, although this specification defines forwarding in terms of the TIB, to actually forward packets using the TIB is very inefficient. Instead a real router implementation will normally build an efficient MFIB from the TIB state to perform forwarding. How this is done is implementation-specific, and is not discussed in this document. 2.2. Pseudocode Notation We use set notation in several places in this specification. A (+) B is the union of two sets A and B. A (-) B is the elements of set A that are not in set B. NULL is the empty set or list. In addition we use C-like syntax: = denotes assignment of a variable. == denotes a comparison for equality. != denotes a comparison for inequality. Braces { and } are used for grouping. Handley/Kouvelas/Speakman/Vicisano Section 2.2. [Page 7] INTERNET-DRAFT Expires: September 2001 March 2001 3. Protocol Specification The specification of BIDIR-PIM is broken into several parts: o Section 3.1 details the protocol state stored. o Section 3.3 specifies the data packet forwarding rules. o Section 3.4 specifies the BIDIR-PIM Join/Prune generation and processing rules. o Designated Forwarder (DF) election is specified in Section 3.5. o PIM packet formats are specified in Section 3.7. o A summary of BIDIR-PIM timers and their default values is given in Section 3.6. 3.1. BIDIR-PIM Protocol State This section specifies all the protocol state that a BIDIR-PIM implementation should maintain in order to function correctly. We term this state the Tree Information Base or TIB, as it holds the state of all the multicast distribution trees at this router. In this specification we define PIM mechanisms in terms of the TIB. However, only a very simple implementation would actually implement packet forwarding operations in terms of this state. Most implementations will use this state to build a multicast forwarding table, which would then be updated when the relevant state in the TIB changes. Although we specify precisely the state to be kept, this does not mean that an implementation of PIM-SM needs to hold the state in this form. This is actually an abstract state definition, which is needed in order to specify the router's behavior. A BIDIR-PIM implementation is free to hold whatever internal state it requires, and will still be conformant with this specification so long as it results in the same externally visible protocol behavior as an abstract router that holds the following state. We divide TIB state into two sections: RP state State that maintains the DF election information for each RP. Group state State that maintains a group-specific tree for groups that map to a given RP. Handley/Kouvelas/Speakman/Vicisano Section 3.1. [Page 8] INTERNET-DRAFT Expires: September 2001 March 2001 The state that should be kept is described below. Of course, implementations will only maintain state when it is relevant to forwarding operations - for example, the "NoInfo" state might be assumed from the lack of other state information, rather than being held explicitly. 3.1.1. General Purpose State A router holds the following state that is not specific to a RP or group: Neighbor State: For each neighbor: o Information from neighbor's Hello o Neighbor's Gen ID. o Neighbor liveness timer (NLT) 3.1.2. RP State A router maintains a multicast-group to RP mapping which is built through static configuration or by using an automatic RP discovery mechanism like BSR or AUTO-RP (see section 4 ). For each BIDIR-PIM RP a router holds the following state: o RP address Designated Forwarder (DF) State: For each router interface: Acting DF information: o DF IP Address o DF metric Election information: o DF Election-Timer (DFT) o Offer-Count (OC) Handley/Kouvelas/Speakman/Vicisano Section 3.1.2. [Page 9] INTERNET-DRAFT Expires: September 2001 March 2001 Current best offer: o IP address of best offering router o Best offering router metric Designated Forwarder state is described in section 3.5. 3.1.3. Group State For every group G a router keeps the following state: Group state: For each interface: Local Membership: o State: One of {"NoInfo", "Include"} PIM Join/Prune State: o State: One of {"NoInfo" (NI), "Join" (J), "PrunePending" (PP)} o Prune Pending Timer (PPT) o Join/Prune Expiry Timer (ET) Not interface specific: o Upstream Join/Prune Timer (JT) o Last RP Used Local membership is the result of the local membership mechanism (such as IGMP) running on that interface. This information is used by the pim_include(*,G) macro described in section 3.1.4. PIM Join/Prune state is the result of receiving PIM (*,G) Join/Prune messages on this interface, and is specified in section 3.4.1. The state is used by the macros that calculate the outgoing interface list in section 3.1.4, and in the JoinDesired(G) macro (defined in section 3.4.2) that is used in deciding whether a Join(*,G) should be sent upstream. Handley/Kouvelas/Speakman/Vicisano Section 3.1.3. [Page 10] INTERNET-DRAFT Expires: September 2001 March 2001 The upstream Join/Prune timer is used to send out periodic Join(*,G) messages, and to override Prune(*,G) messages from peers on an upstream LAN interface. The last RP used must be stored because if the RP Set changes [9] then state must be torn down and rebuilt for groups whose RP changes. 3.1.4. State Summarization Macros Using this state, we define the following "macro" definitions which we will use in the descriptions of the state machines and pseudocode in the following sections. olist(G) = RPF_interface(RP(G)) (+) joins(G) (+) pim_include(G) RPF_interface(RP) is the interface the MRIB indicates would be used to route packets to RP. The olist(G) is the list of interfaces on which packets to group G must be forwarded. The macro pim_include(G) indicates the interfaces to which traffic might be forwarded because of hosts that are local members on that interface. pim_include(G) = { all interfaces I such that: I_am_DF(RP(G),I) AND local_receiver_include(G,I) } The clause "I_am_DF(RP,I)" is TRUE if the router is in the Win or Backoff states in the DF election state machine in section 3.5. Otherwise it is FALSE. The clause "local_receiver_include(G,I)" is true if the IGMP module or other local membership mechanism has determined that there are local members on interface I that desire to receive traffic sent to group G. The set "joins(G)" is the set of all interfaces on which the router has received (*,G) Joins: joins(G) = { all interfaces I such that I_am_DF(RP(G),I) AND DownstreamJPState(G,I) is either Joined or PrunePending } Handley/Kouvelas/Speakman/Vicisano Section 3.1.4. [Page 11] INTERNET-DRAFT Expires: September 2001 March 2001 DownstreamJPState(G,I) is the state of the finite state machine in section 3.4.1. RPF_DF(RP) is the neighbor that Join messages must be sent to in order to reach the RP. This is the Designated-Forwarder on the RPF_interface(RP). 3.2. PIM Neighbor Discovery PIM routers exchange PIM Hello messages with their neighboring PIM routers. These messages are used to update the Neighbor State described in section 3.1. The procedures for generating and processing received Hello messages as well as maintaining Neighbor State are specified in the PIM-SM [9] documentation. 3.3. Data Packet Forwarding Rules For groups mapping to a given RP, the following responsibilities are uniquely assigned to the DF for that RP on each link: o The DF is the only router that forwards packets traveling downstream onto the link. o The DF is the only router that picks-up upstream traveling packets off the link to forward towards the RP. Non-DF routers on a link, that use that link as their RPF interface to reach the RP, may perform the following forwarding actions for bidirectional groups: o Forward packets from the link towards downstream receivers. o Forward packets from downstream sources onto the link (provided they are the DF for the downstream link from which the packet was picked- up). The BIDIR-PIM packet forwarding rules are defined below in pseudocode. iif is the incoming interface of the packet. G is the destination address of the packet (group address). RP is the address of the Rendezvous Point for this group. First we check to see whether the packet should be accepted based on TIB state and the interface that the packet arrived on. A packet is accepted if it arrives on the RPF_interface to reach the RP (downstream traveling packet) or if the router is the DF on the interface the packet arrives (upstream traveling packet). Handley/Kouvelas/Speakman/Vicisano Section 3.3. [Page 12] INTERNET-DRAFT Expires: September 2001 March 2001 If the packet should be forwarded we build an outgoing interface list for the packet. Finally we remove the incoming interface from the outgoing interface list we've created, and if the resulting outgoing interface list is not empty, we forward the packet out of those interfaces. On receipt on a data to G on interface iif: if( iif == RPF_interface(RP) || I_am_DF(RP,I) ) { oiflist = olist(G) (-) iif forward packet on all interfaces in oiflist } Note: A major advantage of using a Designated Forwarder in BIDIR-PIM compared to PIM-SM is that special treatment is no longer required for sources that are directly connected to a router. Data from such sources does not need to be differentiated from other multicast traffic and will automatically be picked up by the DF. This removes the need for performing a directly-connected-source check for data to groups that do not have existing state. 3.3.1. Source-Only Branches Source-only branches of the distribution tree for a group G are branches which do not lead to any receivers, but which are used to forward packets traveling upstream from sources towards the RP. Routers along source-only branches only have the RPF_interface to the RP in their olist for G and hence do not need to maintain any group specific state. Upstream forwarding can be performed using RP state. An implementation may decide to maintain group state for source-only branches for accounting or performance reasons. 3.4. PIM Join/Prune Messages A BIDIR-PIM Join/Prune message consists of a list of Joined and Pruned Groups. When processing a received Join/Prune message, each Joined or Pruned Group is effectively considered individually by applying the following state machines. When considering a Join/Prune message whose PIM Destination field addresses this router, (*,G) Joins and Prunes can affect the downstream state machine. When considering a Join/Prune message whose PIM Destination field addresses another router, most Join or Prune entries could affect the upstream state machine. Handley/Kouvelas/Speakman/Vicisano Section 3.4. [Page 13] INTERNET-DRAFT Expires: September 2001 March 2001 3.4.1. Receiving (*,G) Join/Prune Messages When a router receives a Join(*,G) or Prune(*,G) it must first check to see whether the RP in the message matches RP(G) (the router's idea of who the RP is). If the RP in the message does not match RP(G) the Join or Prune MUST be silently dropped. In addition a router MUST NOT process Join(*,G) messages targeted to itself if it is not the DF for RP(G) on the interface on which the message was received. The per-interface state-machine for receiving (*,G) Join/Prune Messages is given below. There are three states: NoInfo (NI) The interface has no (*,G) Join state and no timers running. Join (J) The interface has (*,G) Join state which will cause us to forward packets destined for G from this interface. PrunePending (PP) The router has received a Prune(*,G) on this interface from a downstream neighbor and is waiting to see whether the prune will be overridden by another downstream router. For forwarding purposes, the PrunePending state functions exactly like the Join state. In addition the state-machine uses two timers: ExpiryTimer (ET) This timer is restarted when a valid Join(*,G) is received. Expiry of the ExpiryTimer causes the interface state to revert to NoInfo for this group. PrunePendingTimer (PPT) This timer is set when a valid Prune(*,G) is received. Expiry of the PrunePendingTimer causes the interface state to revert to NoInfo for this group. +-----------------------------------+ | Figures omitted from text version | +-----------------------------------+ Figure 1: Downstream group per-interface state-machine Handley/Kouvelas/Speakman/Vicisano Section 3.4.1. [Page 14] INTERNET-DRAFT Expires: September 2001 March 2001 In tabular form, the group per-interface state-machine is: +----------+------------------------------------------------------------+ | | Event | | +----------+------------+-----------+------------+-----------+ Prev State |Receive |Receive |Prune |Expiry Stop Being | | |Join(*,G) |Prune(*,G) |Pending |Timer DF on I | | | | |Timer |Expires | | | | | |Expires | | | +----------+----------+------------+-----------+------------+-----------+ | |-> J state|-> NI state |- |- + | NoInfo |start | | | | | (NI) |Expiry | | | | | | |Timer | | | | | +----------+----------+------------+-----------+------------+-----------+ | |-> J state|-> PP state |- |-> NI state +> NI state | Join (J) |restart |start Prune | | | | | |Expiry |Pending | | | | | |Timer |Timer | | | | +----------+----------+------------+-----------+------------+-----------+ | |-> J state|-> PP state |-> NI state|-> NI state +> NI state | | |restart | |Send Prune-| | | Prune |Expiry | |Echo(*,G) | | | Pending |Timer; | | | | | (PP) |stop Prune| | | | | | |Pending | | | | | | |Timer | | | | | +----------+----------+------------+-----------+------------+-----------+ The transition events "Receive Join(*,G)" and "Receive Prune(*,G)" imply receiving a Join or Prune targeted to this router's address on the received interface. If the destination address is not correct, these state transitions in this state machine must not occur, although seeing such a packet may cause state transitions in other state machines. On unnumbered interfaces on point-to-point links, the router's address should be the same as the source address it chose for the hello packet it sent over that interface. However on point-to-point links we also recommend that PIM messages with a 0.0.0.0 destination address are also accepted. The transition event "Stop being DF" implies a DF re-election taking place on this router interface and the router changing status from being the active DF to being a non-DF router (the value of the I_am_DF macro changing to FALSE). When ExpiryTimer is started or restarted, it is set to the HoldTime from the triggering Join/Prune message. Handley/Kouvelas/Speakman/Vicisano Section 3.4.1. [Page 15] INTERNET-DRAFT Expires: September 2001 March 2001 When PrunePendingTimer is started, it is set to the J/P_Override_Interval if the router has more than one neighbor on that interface; otherwise it is set to zero causing it to expire immediately. The action "Send PruneEcho(*,G)" is triggered when the router stops forwarding on an interface as a result of a prune. A PruneEcho(*,G) is simply a Prune(*,G) message sent by the upstream router to itself on a LAN. Its purpose is to add additional reliability so that if a Prune that should have been overridden by another router is lost locally on the LAN, then the PruneEcho may be received and cause the override to happen. A PruneEcho(*,G) need not be sent on a point-to-point interface. 3.4.2. Sending Join/Prune Messages The downstream per-interface state-machines described above hold join state from downstream PIM routers. This state then determines whether a router needs to propagate a Join(*,G) upstream towards the RP. Such Join(*,G) messages are sent on the RPF_interface towards the RP and are targeted at the DF on that interface. If a router wishes to propagate a Join(*,G) upstream, it must also watch for messages on its upstream interface from other routers on that subnet, and these may modify its behavior. If it sees a Join(*,G) to the correct upstream neighbor, it should suppress its own Join(*,G). If it sees a Prune(*,G) to the correct upstream neighbor, it should be prepared to override that prune by sending a Join(*,G) almost immediately. Finally, if it sees the Generation ID (see PIM-SM specification [9]) of the correct upstream neighbor change, it knows that the upstream neighbor has lost state, and it should be prepared to refresh the state by sending a Join(*,G) almost immediately. In addition changes in the next hop towards the RP trigger a prune off from the old next hop, and join towards the new next hop. Such a change can be cause by the following two reasons: o The MRIB indicates that the RPF_interface towards the RP has changed. o There is a DF re-election on the RPF_interface and a new router emerges as the DF. The upstream (*,G) state-machine only contains two states: Not Joined The downstream state-machines indicate that the router does not need to join the RP tree for this group. Handley/Kouvelas/Speakman/Vicisano Section 3.4.2. [Page 16] INTERNET-DRAFT Expires: September 2001 March 2001 Joined The downstream state-machines indicate that the router would like to join the RP tree for this group. In addition, one timer JT(G) is kept which is used to trigger the sending of a Join(*,G) to the upstream next hop towards the RP (the DF on the RPF_interface for RP(G)). +-----------------------------------+ | Figures omitted from text version | +-----------------------------------+ Figure 2: Upstream group state-machine Handley/Kouvelas/Speakman/Vicisano Section 3.4.2. [Page 17] INTERNET-DRAFT Expires: September 2001 March 2001 In tabular form, the state machine is: +----------------------+------------------------------------------------+ | | Event | | Prev State +------------------------+-----------------------+ | | JoinDesired(G) | JoinDesired(G) | | | ->True | ->False | +----------------------+------------------------+-----------------------+ | | -> J state | - | | NotJoined (NJ) | Send Join(*,G); | | | | Set Timer to | | | | t_periodic | | +----------------------+------------------------+-----------------------+ | Joined (J) | - | -> NJ state | | | | Send Prune(*,G) | +----------------------+------------------------+-----------------------+ In addition, we have the following transitions which occur within the Joined state: +-----------------------------------------------------------------------+ | In Joined (J) State | +-----------------+-----------------+-----------------+-----------------+ |Timer Expires | See Join(*,G) | See Prune(*,G) | RPF_DF(RP(G)) | | | to | to | changes | | | RPF_DF(RP(G)) | RPF_DF(RP(G)) | | +-----------------+-----------------+-----------------+-----------------+ |Send | Increase Timer | Decrease Timer | Decrease Timer | |Join(*,G); Set | to | to t_override | to t_override | |Timer to | t_suppressed | | | |t_periodic | | | | +-----------------+-----------------+-----------------+-----------------+ +-----------------------------------------------------------------------+ | In Joined (J) State | +-------------------------------------+---------------------------------+ | Change of RPF_DF(RP(G)) | RPF_DF(RP(G)) GenID | | | changes | +-------------------------------------+---------------------------------+ | Send Join(*,G) to new | Decrease Timer to | | DF; Send Prune(*,G) to | t_override | | old DF; set Timer to | | | t_periodic | | +-------------------------------------+---------------------------------+ Handley/Kouvelas/Speakman/Vicisano Section 3.4.2. [Page 18] INTERNET-DRAFT Expires: September 2001 March 2001 This state machine uses the following macro: bool JoinDesired(G) { if (olist(G) (-) RPF_interface(RP(G))) != NULL return TRUE else return FALSE } 3.5. Designated Forwarder (DF) Election This section presents a fail-safe mechanism for electing a per-RP designated router on each link in a BIDIR-PIM domain. We call this router the Designated Forwarder (DF). 3.5.1. DF Requirements The DF election chooses the best router on a link to assume the responsibility of forwarding traffic between the RP and the link for the range of multicast groups served by the RP. Different multicast groups that share a common RP must use the same bi-directional tree for data forwarding. Hence, the election of an upstream forwarder on each link does not have to be a group specific decision but instead can be RP- specific. As the number of RPs is typically small, the number of elections that have to be performed is significantly reduced by this observation. To optimise tree creation, it is desirable that the winner of the election process should be the router on the link with the "best" unicast routing metric to the RP (as reported by the MRIB). When comparing metrics from different unicast routing protocols, we use the same comparison rules used by the PIM-SM assert process [9]. The election process needs to take place when information on a new RP initially becomes available, and can be re-used as new bidir groups for the same RP are encountered. There are however some conditions where an update to the election is required: o There is a change in unicast metric to reach the RP for any of the routers on the link. o The interface on which the RP is reachable changes to an interface for which the router was previously the DF. o A new PIM neighbor starts up on a link. Handley/Kouvelas/Speakman/Vicisano Section 3.5.1. [Page 19] INTERNET-DRAFT Expires: September 2001 March 2001 o The elected DF dies. The election process has to be robust enough to ensure with very high probability that all routers on the link have a consistent view of the DF. This is because with the forwarding rules described in section 3.3 if multiple routers end-up thinking that they should be responsible for forwarding, loops may result. To reduce the possibility of this occurrence to a minimum, the election algorithm has been biased towards discarding DF information and suspending forwarding during periods of ambiguity. 3.5.2. DF Election description This section does not provide the definitive specification for the DF election process. If any discrepancy exists between section 3.5.3 and this section, the specification in section 3.5.3 is to be assumed correct. To perform the election of the DF for a particular RP, routers on a link need to exchange their unicast routing metric information (as reported by the MRIB) for reaching the RP. In the election protocol described below, many message exchanges are repeated Election_Robustness times for reliability. In all those cases the message retransmissions are spaced in time by a small random interval. 3.5.2.1. Bootstrap Election Initially when no DF has been elected, routers finding out about a new RP start participating in the election by sending Offer messages. Offer messages include the router's metric to reach the RP. Offers are periodically retransmitted with a period of Offer_Interval. If a router hears a better offer than its own from a neighbor, it stops participating in the election for a period of Election_Robustness * Offer_Interval. If during this period no winner is elected, then the router restarts the election from the beginning. If a router receives an offer with worse metrics than its own, then it restarts the election from the beginning. The result should be that all routers except the best candidate stop advertising their offers. A router assumes the role of the DF after having advertised its metrics Election_Robustness times without receiving any offer from any other Handley/Kouvelas/Speakman/Vicisano Section 3.5.2.1. [Page 20] INTERNET-DRAFT Expires: September 2001 March 2001 neighbor. At that point it transmits a Winner message which declares to every other router on the link the identity of the winner and the metrics it is using. Routers hearing a winner message stop participating in the election and record the identity and metrics of the winner. If the local metrics are better than those of the winner then the router records the identity of the winner but reinitiates the election. 3.5.2.2. Loser Metric Changes Whenever the unicast metric to a RP changes for a non-DF router to a value that is better than that previously advertised by the acting DF, the router with the new metric should take action to eventually assume forwarding responsibility. After the metric change is detected, the non- DF router with the now better metric restarts the DF election process by sending Offer messages with this new metric. If no response is received after Election_Robustness retransmissions, the router assumes the role of the DF following the usual Winner announcement procedure. Upon receipt of an offer that is worse than its current metric, the DF will respond with a Winner message declaring its status and advertising its metric. Upon receiving this message, the originator of the Offer records the identity of the DF and aborts the election. Upon receipt of an offer that is better than its current metric, the DF records the identity and metrics of the offering router and responds with a Backoff message. This instructs the offering router to hold off for a short period of time while the unicast routing stabilises. The Backoff message includes the offering router's new metric and address. All routers on the link who have pending offers with metrics worse than those in the backoff message (including the original offering router) will hold further offers for a period of time defined in the Backoff message. If during the Backoff_Period, a third router sends a new better offer, the Backoff message is repeated for the new offer and the Backoff_Period restarted. Before the Backoff_Period expires, the acting DF nominates the router having made the best offer as the new DF using a Pass message. This message includes the IDs and metrics of both the old and new DFs. The old DF stops performing its tasks as soon as the transmission is made. The new DF assumes the role of the DF as soon as it receives the Pass message. All other routers on the link take note of the new DF and its metric. Handley/Kouvelas/Speakman/Vicisano Section 3.5.2.2. [Page 21] INTERNET-DRAFT Expires: September 2001 March 2001 3.5.2.3. Winner Metric Changes If the DF's routing metric to reach the RP changes to a worse value, it sends a set of Election_Robustness randomly spaced Offer messages on the link, advertising the new metric. Routers who receive this announcement but have a better metric may respond with an Offer message which results in the same handoff procedure described above. All routers assume the DF has not changed until they see a Pass or Winner message indicating the change. There is no pressure to make this handoff quickly if the acting DF still has a path to the RP. The old path may now be suboptimal but it will still work while the re-election is in progress. If the routing metric at the DF changes to a better value, a single Winner message is sent advertising the new metric. 3.5.2.4. Winner Loses Path If a router's RPF_interface to the RP switches to be on a link for which it is acting as the DF, then it can no longer provide forwarding services for that link. It therefore immediately stops being the DF and restarts the election. As its path to the RP is through the link, an infinite metric is used in the Offer message it sends. Note: At this stage the old DF will have a new RPF neighbor on the link (indicated by unicast routing) which it could use in a Pass message but this adds unnecessary complication to the election process. 3.5.2.5. Late Router Starting Up A late router starting up after the DF election process has completed will have no immediate knowledge of the election outcome. As a result, it will start advertising its metric in Offer messages. As soon as this happens, the currently elected DF will respond with a Winner message if its metric is better than the metric in the Offer message, or with a Backoff message if its metric worse than the metric in the Offer message. 3.5.2.6. Winner Dies Whenever the DF dies, a new DF has to be elected. The speed at which this can be achieved depends on whether there are any downstream routers on the link. Handley/Kouvelas/Speakman/Vicisano Section 3.5.2.6. [Page 22] INTERNET-DRAFT Expires: September 2001 March 2001 If there are downstream routers, typically their RPF_neighbor as reported by the MRIB before the DF dies will be the DF itself. They will therefore notice either a change in the metric for the route to the RP or a change in RPF_neighbor away from the DF and will restart the election by transmitting Offer messages. If according to the MRIB the RP is now reachable through the same link via another upstream router, an infinite metric will be used in the Offer. If no downstream routers are present, the only way for other upstream routers to detect a DF failure is by the timeout of the PIM neighbor information, which will take significantly longer. 3.5.3. Election Protocol Specification This section provides the definitive specification for the DF election process. If any discrepancy exists between section 3.5.2 and this section, the specification in this section is to be assumed correct. 3.5.3.1. Election State The operation of the election protocol makes use of the variables and timers described below. These are maintained per RP for each multicast enabled interface on the router as introduced in section 3.1: Acting DF information Used to store the election winner who is the currently acting DF. Election-Timer (DFT) Used to schedule transmission of Offer, Winner and Pass messages. Offer-Count (OC) Used to maintain the number of times an Offer or Winner message has been transmitted. Best-Offer Used by the DF to record who has made the last offer for sending the Pass message. 3.5.3.2. Election Messages The election process uses the following PIM control messages the packet format of which is described in section 3.7: Handley/Kouvelas/Speakman/Vicisano Section 3.5.3.2. [Page 23] INTERNET-DRAFT Expires: September 2001 March 2001 Offer (OfferingID, Metric) Sent by routers that believe they have a better metric to the RP than the metric that has been on offer so far. Winner (DF-ID, DF-Metric) Sent by a router when assuming the role of the DF or when re- asserting in response to worse offers. Backoff (DF-ID, DF-Metric, OfferingID, OfferMetric, BackoffInterval) Used by the DF to acknowledge better offers. It instructs other routers with equal or worse offers to wait till the DF passes responsibility to the sender of the offer. Pass (Old-DF-ID, Old-DF-Metric, New-DF-ID, New-DF-Metric) Used by the old DF to pass forwarding responsibility to a router that has previously made an offer. The Old-DF-Metric is the current metric of the DF at the time the pass is sent. 3.5.3.3. Election Events During protocol operation, in addition to the expiration of the Election-Timer and the reception of the four control messages, the following events can take place: o Discovery of new RP o Metric reported by the MRIB to reach the RP changes o DF loses path to RP o Detection of DF failure 3.5.3.4. Election State Transitions In the state machine presented below a router is considered to be an acting DF if it is in the Win or Backoff states. When an action of "set DF to Sender or Dest" is encountered during receipt of a Winner, Pass or Backoff message it means the following: o On receipt of a Winner message set the DF to be the originator of the message and record its metrics. o On receipt of a Pass message set the DF to be the target of the message and record its metrics. Handley/Kouvelas/Speakman/Vicisano Section 3.5.3.4. [Page 24] INTERNET-DRAFT Expires: September 2001 March 2001 o On receipt of a Backoff message set the DF to be the originator of the message and record its metrics. +-----------------------------------+ | Figures omitted from text version | +-----------------------------------+ Figure 3: Designated Forwarder election state-machine In tabular form, the state machine is: +-------------++--------------------------------------------------------+ | || Event | | Prev State ++------------------+------------------+------------------+ | || Recv better | Recv better | Recv better | | || Pass / Win | Backoff | Offer | +-------------++------------------+------------------+------------------+ | || -> Lose | - | - | | || DF = Sender | Set Timer to | Set Timer to | | Offer || | Robustness * | Robustness * | | || | offer_int; Set | offer_int; Set | | || | Count to 0 | Count to 0 | +-------------++------------------+------------------+------------------+ | || - | - | -> Offer | | || DF = Sender or | DF = Sender or | Set Timer to | | Lose || Dest | Dest | Robustness * | | || | | offer_int; Set | | || | | Count to 0 | +-------------++------------------+------------------+------------------+ | || -> Lose | -> Lose | -> Backoff | | || DF = Sender or | DF = Sender or | Set Best to | | Win || Dest; Stop | Dest; Stop | Sender; Send | | || Timer | Timer | Backoff; Set | | || | | Timer to | | || | | backoff_int | +-------------++------------------+------------------+------------------+ | || -> Lose | -> Lose | - | | || DF = Sender or | DF = Sender or | Set Best to | | Backoff || Dest | Dest | Sender; Send | | || | | Backoff; Set | | || | | Timer to | | || | | backoff_int | +-------------++------------------+------------------+------------------+ Handley/Kouvelas/Speakman/Vicisano Section 3.5.3.4. [Page 25] INTERNET-DRAFT Expires: September 2001 March 2001 +-----------++----------------------------------------------------------+ | || Event | | ++-------------+--------------+--------------+--------------+ |Prev State ||Recv Backoff | Recv Pass | Recv Worse | Recv worse | | ||for us | for us | Pass / Win / | Offer | | || | | Backoff | | +-----------++-------------+--------------+--------------+--------------+ | ||- | -> Win | - | - | | ||Set Timer to | Stop Timer | Set DF to | Set/Lower | | ||Hi; Set | | Sender or | Timer to | |Offer ||Count to 0 | | Dest; | Low; Set | | || | | Set/Lower | Count to 0 | | || | | Timer to | | | || | | Low; Set | | | || | | Count to 0 | | +-----------++-------------+--------------+--------------+--------------+ | ||-> Offer | -> Offer | -> Offer | -> Offer | | ||DF = Sender | DF = Sender | DF = Sender | Set Timer to | | ||or Dest; Set | or Dest; Set | or Dest; Set | offer_int; | |Lose ||Timer to | Timer to | Timer to | Set Count to | | ||offer_int; | offer_int; | offer_int; | 0 | | ||Set Count to | Set Count to | Set Count to | | | ||0 | 0 | 0 | | +-----------++-------------+--------------+--------------+--------------+ | ||-> Offer | -> Offer | -> Offer | - | | ||DF = Sender | DF = Sender | DF = Sender | Send Winner | | ||or Dest; Set | or Dest; Set | or Dest; Set | | |Win ||Timer to | Timer to | Timer to | | | ||offer_int; | offer_int; | offer_int; | | | ||Set Count to | Set Count to | Set Count to | | | ||0 | 0 | 0 | | +-----------++-------------+--------------+--------------+--------------+ | ||-> Offer | -> Offer | -> Offer | -> Win | | ||DF = Sender | DF = Sender | DF = Sender | Send Winner; | | ||or Dest; Set | or Dest; Set | or Dest; Set | Stop Timer | |Backoff ||Timer to | Timer to | Timer to | | | ||offer_int; | offer_int; | offer_int; | | | ||Set Count to | Set Count to | Set Count to | | | ||0 | 0 | 0 | | +-----------++-------------+--------------+--------------+--------------+ Handley/Kouvelas/Speakman/Vicisano Section 3.5.3.4. [Page 26] INTERNET-DRAFT Expires: September 2001 March 2001 +-----------------------------------------------------------------------+ | In Offer State | +-----------------------+------------------------+----------------------+ | Timer Expires and | Timer Expires and | Timer Expires and | | Count is less than | Count is equal to | Count is equal to | | Robustness | Robustness and we | Robustness and | | | have path to RP | there is no path | | | | to RP | +-----------------------+------------------------+----------------------+ | - | -> Win | -> Lose | | Send Offer; Set | Send Winner | Set DF to None | | Timer to | | | | offer_int; | | | | Increase Count by | | | | 1 | | | +-----------------------+------------------------+----------------------+ +-----------------------------------------------------------------------+ | In Lose State | +-----------------------------------+-----------------------------------+ | Detect DF Failure | Metric changes and now | | | is better than DF | +-----------------------------------+-----------------------------------+ | -> Offer | -> Offer | | DF = None; Set timer to | Set timer to offer_int; | | offer_int; Set Count to | Set Count to 0 | | 0 | | +-----------------------------------+-----------------------------------+ +-----------------------------------------------------------------------+ | In Win State | +------------------------+------------------------+---------------------+ | Metric changes and | Timer Expires and | No path to RP | | is now worse | Count is less than | | | | Robustness | | +------------------------+------------------------+---------------------+ | - | - | -> Offer | | Set Timer to | Send Winner; Set | Set DF to None; | | offer_int; Set | Timer to | Set Timer to | | Count to 0 | offer_int; | offer_int; Set | | | Increment Count by | Count to 0 | | | 1 | | +------------------------+------------------------+---------------------+ Handley/Kouvelas/Speakman/Vicisano Section 3.5.3.4. [Page 27] INTERNET-DRAFT Expires: September 2001 March 2001 +-----------------------------------------------------------------------+ | In Backoff State | +-----------------------------------+-----------------------------------+ | Metric changes and is | Timer Expires | | now better than Best | | +-----------------------------------+-----------------------------------+ | -> Win | -> Lose | | Stop Timer | Send Pass; Set DF to | | | stored Best | +-----------------------------------+-----------------------------------+ 3.6. Timers and Constants BIDIR-PIM maintains the following timers, as discussed in section 3.1. All timers are countdown timers - they are set to a value and count down to zero, at which point they typically trigger an action. Of course they can just as easily be implemented as count-up timers, where the absolute expiry time is stored and compared against a real-time clock, but the language in this specification assumes that they count downwards to zero. Per Rendezvous-Point (RP): Per interface (I): DF Election Timer: DFT(RP,I) Per Group (G): Upstream Join Timer: JT(G) Per interface (I): Join Expiry Timer: ET(G,I) PrunePending Timer: PPT(G,I) When timers are started or restarted, they are set to default values. This section summarizes those default values. Handley/Kouvelas/Speakman/Vicisano Section 3.6. [Page 28] INTERNET-DRAFT Expires: September 2001 March 2001 Timer Name: DF Election Timer (DFT) +--------------------+-------------------------+------------------------+ | Value Name | Value | Explanation | +--------------------+-------------------------+------------------------+ | Offer_Period | 100 ms | Interval to wait | | | | between repeated | | | | Offer and Winner | | | | messages. | +--------------------+-------------------------+------------------------+ | Backoff_Period | 1 sec | Period that acting | | | | DF waits between | | | | receiving a better | | | | Offer and sending | | | | the Pass message | | | | to transfer DF | | | | responsibility. | +--------------------+-------------------------+------------------------+ | OPLow | rand(0.5, 1) * | Range of actual | | | Offer_Period | randomised value | | | | used between | | | | repeated messages. | +--------------------+-------------------------+------------------------+ | OPHigh | Election_Robustness | Interval to wait | | | * Offer_Period | in order to give a | | | | chance to a router | | | | with a better | | | | Offer to become | | | | the DF. | +--------------------+-------------------------+------------------------+ Timer Names: Join Expiry Timer (ET(G,I)) +----------------+----------------+-------------------------------------+ | Value Name | Value | Explanation | +----------------+----------------+-------------------------------------+ | J/P HoldTime | from message | Hold Time from Join/Prune Message | +----------------+----------------+-------------------------------------+ Handley/Kouvelas/Speakman/Vicisano Section 3.6. [Page 29] INTERNET-DRAFT Expires: September 2001 March 2001 Timer Names: Prune Pending Timer (PPT(G,I)) +--------------------------+--------------------+-----------------------+ | Value Name | Value | Explanation | +--------------------------+--------------------+-----------------------+ | J/P Override Interval | Default: 3 secs | Short period after | | | | a join or prune to | | | | allow other | | | | routers on the LAN | | | | to override the | | | | join or prune | +--------------------------+--------------------+-----------------------+ Timer Names: Upstream Join Timer (JT(G)) +-------------+--------------------+------------------------------------+ |Value Name |Value |Explanation | +-------------+--------------------+------------------------------------+ |t_periodic |Default: 60 secs |Period between Join/Prune Messages | +-------------+--------------------+------------------------------------+ |t_suppressed |rand(1.1 * |Suppression period when someone | | |t_periodic, 1.4 * |else sends a J/P message so we | | |t_periodic) |don't need to do so. | +-------------+--------------------+------------------------------------+ |t_override |rand(0, 0.9 * J/P |Randomized delay to prevent | | |Override Interval) |response implosion when sending a | | | |join message to override someone | | | |else's prune message. | +-------------+--------------------+------------------------------------+ For more information about these values refer to the PIM-SM [9] documentation. Constant Name: DF Election Robustness +--------------------------+-------------------+------------------------+ | Constant Name | Value | Explanation | +--------------------------+-------------------+------------------------+ | Election_Robustness | Default: 3 | Minimum number of | | | | election messages | | | | that must be lost | | | | in order for | | | | election to fail. | +--------------------------+-------------------+------------------------+ Handley/Kouvelas/Speakman/Vicisano Section 3.6. [Page 30] INTERNET-DRAFT Expires: September 2001 March 2001 3.7. PIM DF Election Packet Formats This section describes the details of the packet formats for BIDIR-PIM control messages. BIDIR-PIM shares a number of control messages in common with PIM-SM [9] well as the format for the Encoded-Unicast address. For details on the format of these packets please refer to the PIM-SM documentation. Here we will only define the additional packets that are introduced by BIDIR-PIM. These are the packets used in the DF election process. All PIM control messages have IP protocol number 103. BIDIR-PIM messages are multicast with TTL 1 to the `ALL-PIM-ROUTERS' group `224.0.0.13'. All DF election BIDIR-PIM control messages share the common header below: 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 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |PIM Ver| Type |Subtype| Rsvd | Checksum | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Encoded-Unicast-RP-Address | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Sender Metric Preference | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Sender Metric | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ PIM Ver PIM Version number is 2. Type All DF-Election PIM control messages share the PIM message Type of 10. Subtype Subtypes for DF election messages are: 1 = Offer 2 = Winner 3 = Backoff 4 = Pass Rsvd Set to zero on transmission. Ignored upon receipt. Handley/Kouvelas/Speakman/Vicisano Section 3.7. [Page 31] INTERNET-DRAFT Expires: September 2001 March 2001 Checksum The checksum is standard IP checksum, i.e. the 16-bit one's complement of the one's complement sum of the entire PIM message. For computing the checksum, the checksum field is zeroed. RP-Address The address of the bidir RP for which the election is taking place (note that the length of this field is more than 32 bits). Sender Metric Preference Preference value assigned to the unicast routing protocol that the message sender used to obtain the route to the RP-address. Sender Metric The unicast routing table metric used by the message sender to reach the RP. The metric is in units applicable to the unicast routing protocol used. In addition to the fields defined above the Backoff and Pass messages have the extra fields described below. 3.7.1. Backoff Message The Backoff message uses the following fields in addition to the common election message format described above. 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 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Encoded-Unicast-Offering-Address | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Offering Metric Preference | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Offering Metric | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Interval | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ Offering Address The address of the router that made the last (best) Offer (note that the length of this field is more than 32 bits). Offering Metric Preference Preference value assigned to the unicast routing protocol that the offering router used to obtain the route to RP-address. Handley/Kouvelas/Speakman/Vicisano Section 3.7.1. [Page 32] INTERNET-DRAFT Expires: September 2001 March 2001 Offering Metric The unicast routing table metric used by the offering router to reach the RP. The metric is in units applicable to the unicast routing protocol used. Interval The backoff interval in milliseconds to be used by routers with worse metrics than the offering router. 3.7.2. Pass Message The Pass message uses the following fields in addition to the common election fields described above. 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 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Encoded-Unicast-New-Winner-Address | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | New Winner Metric Preference | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | New Winner Metric | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ New Winner Address The address of the router that made the last (best) Offer (note that the length of this field is more than 32 bits). New Winner Metric Preference Preference value assigned to the unicast routing protocol that the offering router used to obtain the route to RP-address. New Winner Metric The unicast routing table metric used by the offering router to reach the RP. The metric is in units applicable to the unicast routing protocol used. 4. RP Discovery Routers discover that a range of multicast group addresses operates in bi-directional mode and the address of the Rendezvous-Point serving the group range either through static configuration or using an automatic RP discovery mechanism like the PIM Bootsrtap mechanism (BSR). [12]. By default the BSR protocol advertises RPs that operate the PIM-SM protocol. In order to identify a RP as operating in BIDIR mode, the Handley/Kouvelas/Speakman/Vicisano Section 4. [Page 33] INTERNET-DRAFT Expires: September 2001 March 2001 Encoded-Group Address field in Bootstrap and Candidate-RP Advertisement messages has been extended by adding the BIDIR bit (B-bit) as specified below: 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 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Addr Family | Encoding Type |B| Reserved | Mask Len | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Group Multicast Address | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ B-bit When the Bidir-bit is set, all BIDIR capable PIM routers will operate the protocol described in this document for the specified group range. 5. Security Considerations All PIM control messages MAY use IPsec [6] to address security concerns. The authentication methods are addressed in a companion document [7]. Keys may be distributed as described in [8]. 5.1. Appendix A: Election Reliability Enhancements For the correct operation of bi-directional PIM it is very important to avoid situations where two routers consider themselves to be Designated Forwarders for the same link. The two precautions below are not required for correct operation but can help diagnose anomalies and correct them. 5.1.1. A.1 Missing Pass After a DF has been elected, a router whose metrics change to become better than the DF will attempt to take over. If during the re-election the acting DF has a condition that causes it to lose all of the election messages (like a CPU overload), the new candidate will transmit three offers and assume the role of the forwarder resulting in two DFs on the link. This situation is pathological and should be corrected by fixing the overloaded router. It is desirable that such an event can be detected by a network administrator. When a router becomes the DF for a link without receiving a Pass message from the known old DF, the PIM neighbor information for the old DF can be marked to this effect. Upon receiving the next PIM Hello message from the old DF, the router can retransmit Winner messages for all the RPs for which it acting as the DF. The anomaly may also be logged by the router to alert the operator. Handley/Kouvelas/Speakman/Vicisano Section 5.1.1. [Page 34] INTERNET-DRAFT Expires: September 2001 March 2001 5.1.2. A.2 Periodic Winner Announcement An additional degree of safety can be achieved by having the DF for each RP periodically announce its status in a Winner message. Transmission of the periodic Winner message can be restricted to occur only for RPs which have active groups, thus avoiding the periodic control traffic in areas of the network without senders or receivers for a particular RP. 5.2. Appendix B: Interoperability with legacy code The rules provided in [10] for interoperating between legacy PIM-SM routers and new bi-directional capable routers change only slightly to support this new proposal. The only difference is in the definition of a boundary between a bi-directional capable area and a legacy area of the network. In [10], a bidir capable router forwarding upstream, register encapsulates the data packet to the RP if its RPF neighbor is not bidir capable. In our proposal, since all the routers on a link need to co-operate to elect the Designated Forwarder, if even one of the routers on the link is a legacy router, the election cannot take place. As a result register encapsulation is necessary if one or more routers on the RPF interface are not bi-directional capable. As in [10], a Hello option must be used to differentiate between bi- directional capable and legacy routers, and (S,G) state must be created on the router doing the register encapsulation to prevent loops. 5.3. Appendix C: Comparison with PIM-SM This section describes the main differences between Bidir PIM and sparse-mode PIM: o Bidir PIM uses a single shared tree for distributing the data for all the sources of a multicast group. The use of a single tree significantly reduces state requirements on a router. The drawback is that it may produce suboptimal paths from sources to receivers possibly resulting in higher network latency and less efficient bandwidth utilisation. o In Bidir PIM, packets traveling from a source to the RP, are natively forwarded on the shared tree. In contrast sparse-mode PIM uses unicast encapsulation or source-specific state. o In Bidir PIM, sender-only branches do not need to keep group state. Data from the source can be natively forwarded towards the Handley/Kouvelas/Speakman/Vicisano Section 5.3. [Page 35] INTERNET-DRAFT Expires: September 2001 March 2001 RP using RP-specific forwarding state. o The Bidir Designated Forwarder (DF) assumes all the responsibilities of the sparse-mode DR. In a multi-access link, the DF responds to IGMP notifications. Downstream routers on the link use the DF as their upstream neighbor and direct all Join/Prune messages towards it. o To enforce a single forwarder on multi-access links, sparse-mode PIM uses the Assert mechanism which requires data-packets to trigger protocol events. In Bidir PIM, data-driven events are completely eliminated as a correct route is always available at packet forwarding time. The DF election problem is easier than the assert problem because there is a small number of RPs and the per RP DF election can be done in advance. With the assert mechanism, in addition to each RP, a forwarder has to be elected for each possible source to a group. This can not be done before data is available. o With sparse-mode PIM, when forwarding packets using shared-tree (*,G) state, a directly-connected-source check has to be made on every packet. This is done to determine if the packet was originated by a source which is directly connected to the router. For a connected source, source-specific state has to be created to register packets to the RP and prune the source off the shared tree. With Bidir PIM directly connected sources do not need any special handling. The DF for the RP of the group the source is sending to, seamlessly picks-up and forwards upstream traveling packets. 6. Todo list... o Update legacy interoperability section to remove dependency on 10. o Incorporate BSR mods into BSR spec. o In the state machine, perhaps add an arc from NI to NI, labelled "(*,G) Join but not DF" just to make it really clear. Handley/Kouvelas/Speakman/Vicisano Section 7. [Page 36] INTERNET-DRAFT Expires: September 2001 March 2001 7. Authors' Addresses Mark Handley ACIRI/ICSI 1947 Center St, Suite 600 Berkeley, CA 94708 mjh@aciri.org Isidor Kouvelas Cisco Systems kouvelas@cisco.com Tony Speakman Cisco Systems speakman@cisco.com Lorenzo Vicisano Cisco Systems lorenzo@cisco.com 8. Acknowledgments The bidir proposal in this draft is heavily based on the ideas and text presented by Estrin and Farinacci in [10]. The main difference between the two proposals is in the method chosen for upstream forwarding. We would also like to thank Deborah Estrin at ISI/USC as well as Nidhi Bhaskar, Yiqun Cai, Rajitha Sumanasakera and Beau Williamson at cisco for their contributions and comments to this draft. 9. References [1] T. Bates , R. Chandra , D. Katz , Y. Rekhter, "Multiprotocol Extensions for BGP-4", RFC 2283 [2] S.E. Deering, "Host extensions for IP multicasting", RFC 1112, Aug 1989. [3] W. Fenner, "Internet Group Management Protocol, Version 2", RFC 2236. [4] IANA, "Address Family Numbers", linked from http://www.iana.org/numbers.html Handley/Kouvelas/Speakman/Vicisano Section 9. [Page 37] INTERNET-DRAFT Expires: September 2001 March 2001 [5] T. Narten , H. Alvestrand, "Guidelines for Writing an IANA Considerations Section in RFCs", RFC 2434. [6] S. Kent, R. Atkinson, "Security Architecture for the Internet Protocol.", RFC 2401. [7] L. Wei, "Authenticating PIM version 2 messages", draft-ietf-pim- v2-auth-01.txt, work in progress. [8] T. Hardjono, B. Cain, "Simple Key Management Protocol for PIM", draft-ietf-pim-simplekmp-01.txt, work in progress. [9] B. Fenner, M. Handley, H. Holbrook, I. Kouvelas "Protocol Independent Multicast - Sparse Mode (PIM-SM): Protocol Specification (Revised)", Work In Progress, , 2000. [10] D. Estrin, D. Farinacci, "Bi-directional Shared Trees in PIM-SM", Work In Progress, , May 1999. [11] D. Estrin et al, "Protocol Independent Multicast-Sparse Mode (PIM- SM): Protocol Specification", RFC 2362, Nov 1999. [12] W. Fenner, M. Handley, H. Holbrook, I. Kouvelas, "Bootstrap Router (BSR) Mechanism for PIM Sparse Mode", draft-ietf-pim-sm-bsr-00.txt, work in progress. Handley/Kouvelas/Speakman/Vicisano Section 9. [Page 38] INTERNET-DRAFT Expires: September 2001 March 2001 10. Index DownstreamJPState(G,I) . . . . . . . . . . . . . . . . . . . . . . . 11 ET(G,I). . . . . . . . . . . . . . . . . . . . . . . . . . . . .10,14,29 ET(RP,I) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 I_am_DF(RP,I). . . . . . . . . . . . . . . . . . . . . . . . . .11,13,15 J/P_HoldTime . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29 J/P_Override_Interval. . . . . . . . . . . . . . . . . . . . . . . 16,30 JoinDesired(G) . . . . . . . . . . . . . . . . . . . . . . . . . . . 18 joins(G) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11 JT(*,G). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17 JT(G). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10,30 local_receiver_include(G,I). . . . . . . . . . . . . . . . . . . . . 11 NLT(N,I) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 Offer_Period . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29 olist(G) . . . . . . . . . . . . . . . . . . . . . . . . . . . .11,13,18 OT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29 pim_include(G) . . . . . . . . . . . . . . . . . . . . . . . . . . . 11 PPT(G,I) . . . . . . . . . . . . . . . . . . . . . . . . . . . .10,14,30 RPF_interface(RP). . . . . . . . . . . . . . . . . . . . . . . . . 11,13 t_override . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18,30 t_periodic . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18,30 t_suppressed . . . . . . . . . . . . . . . . . . . . . . . . . . . 18,30 Handley/Kouvelas/Speakman/Vicisano Section 10. [Page 39]