PIM Working Group Mark Handley Internet Draft ACIRI Expiration Date: September, 2000 Isidor Kouvelas cisco Systems Lorenzo Vicisano cisco Systems March 1, 2000 Bi-directional Protocol Independent Multicast Status of this Memo This document is an Internet-Draft and is in full conformance with all provisions of Section 10 of RFC2026. Internet-Drafts are working documents of the Internet Engineering Task Force (IETF), its areas, and its working groups. Note that other groups may also distribute working documents as Internet-Drafts. Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as "work in progress." The list of current Internet-Drafts can be accessed at 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. Abstract This document discusses Bi-directional PIM, a variant of PIM Sparse- Mode [1] 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, multi- cast data is natively forwarded from sources to the Rendezvous-Point 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. 1 Introduction Handley, Kouvelas, Vicisano [Page 1] Internet Draft Bidir PIM February 2000 This document discusses Bi-directional PIM, a variant of PIM Sparse Mode [1] that builds bi-directional shared trees connecting multicast sources and receivers. PIM Sparse-Mode (PIM-SM) version 1 and version 2 construct uni- directional shared trees that are used to forward data from senders to receivers of a multicast group. PIM-SM also allows the construc- tion of source specific trees, but this capability is not related to the proposal 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 sparse-mode PIM, 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, it's 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. The ideas presented in this document are similar to those described in [2]. The main difference between the two proposals is in the method used to forward packets traveling upstream from a source to the RP. In particular [2] uses an IP option (UMP option) on data packets to assist with upstream forwarding. The UMP option identifies the next hop router responsible for forwarding the packet upstream. In contrast, this proposal does not alter data packets to embed con- trol information. Instead the identification of the next hop upstream Handley, Kouvelas, Vicisano [Page 2] Internet Draft Bidir PIM February 2000 forwarder is performed at RP discovery time using a fail-safe elec- tion mechanism. This significantly simplifies forwarding procedures and eliminates forwarding loops and packet duplication problems that exist in [2]. Appendix D presents a comparison between the proposal in this document and [2]. The rest of this document is structured as follows. Section 2 defines basic terms. Section 3 describes bidirectional tree formation and forwarding. The new forwarding rules rely heavily on an election mechanism described in section 4. 2 Definitions In the discussion below, the terms upstream, downstream and RPF interface are always referring to the shared tree rooted at the Ren- dezvous Point. Downstream indicates the direction on which packets travel from the RP to receivers along the shared tree. Upstream indi- cates the opposite direction used by packets traveling from sources to the RP. The RPF interface for a group is the interface unicast routing uses to reach the RP. We assume that the reader is familiar with the unidirectional PIM-SM protocol [1], as much of the functionality is common to the version of bidir PIM described below. In particular in the rest of this docu- ment we will use the concepts of (*,G), (S,G) and (*,*,RP) state and their component fields (olist, iif, ...). We will also reference Join and Prune messages whose semantics and packet formats are defined in [1]. In the context of this document, entries in Join and Prune mes- sages always have the RP and WC bits set. Also, default timer values are the ones given in [1]. The protocol presented in this document is largely based on the con- cept of a Designated Forwarder (DF). A single DF exists for each RP on every link within a 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 traffic downstream 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. For those familiar with the DR in PIM-SM, the Bidir DF provides the same support for local receivers. The DF election procedures are described in section 4. 3 Tree Building and Forwarding This section describes how bi-directional tree building procedures and forwarding rules vary from normal PIM-SM operation. A router learns which multicast addresses will be used for sparse- Handley, Kouvelas, Vicisano [Page 3] Internet Draft Bidir PIM February 2000 mode PIM and which will be for bidirectional groups along with the candidate RP information through PIM-SM bootstrap messages. Thus unidirectional and bidirectional groups can coexist in the same domain. Throughout the section it is assumed that on each link, all the routers have a consistent view on which router has the best path to the RP. This router is called the DF for that RP on the link. This assumption rests on the DF election procedures described in section 4. In the procedures described in the rest of this section, if DF infor- mation is required but not available (election is incomplete), then no tree building or forwarding action is taken. 3.1 Tree Building 3.1.1 Joining the (Shared) Tree The procedures for joining the (*,G) shared tree, are almost identi- cal to those used in PIM-SM with the difference that the tasks of the DR are handled by the DF. When a router receives a membership indication from IGMP for a bidirectional group G with rendezvous point RP, and it is the DF for the RP on the link on which the report was received, the following steps are taken: o If no (*,G) state exists for the group, then a (*,G) entry is created and populated with the RP DF information. o If the interface on which the report was received is not in the olist of the entry, then the interface is added to the olist. o According to standard PIM-SM procedures [1], if the olist transi- tioned from null to non-null, a Join message for the group is trig- gered upstream. The Join is directed to the DF for the (*,G) incom- ing interface. When a router receives a Join message addressed to it for a bidir group G with rendezvous point RP, it must determine if it is the DF on the link for this RP. If the router is not the DF, it must ignore the Join message. If it is the DF, then the following steps are taken: o If no (*,G) state exists for the group, then a (*,G) entry is Handley, Kouvelas, Vicisano [Page 4] Internet Draft Bidir PIM February 2000 created and populated with the RP DF information. o If the interface on which the Join was received is not in the olist of the entry, then the interface is added to the olist. o According to standard PIM-SM procedures [1], the expiration timer of the olist interface is updated and if the olist transitioned from null to non-null, a Join message for the group is triggered upstream. The Join is directed to the DF for the (*,G) incoming interface. 3.1.2 Leaving the (shared) tree When the DF for a link receives notification that an interface is no longer required in the olist of a group (either through IGMP or by receiving a Prune), it follows standard PIM-SM procedures except any originated prunes are addressed to the DF on the (*,G) iif. 3.1.3 Designated Forwarder Change When the DF for a RP on a link changes to a different router, tree maintenance has to take place to ensure that traffic is still delivered for all affected groups. 3.1.3.1 Old DF Actions On losing its status as acting DF on a link, the old DF has to take the following actions for existing groups that are affected. o If there were downstream receivers (discovered through IGMP or downstream Joins), the router has to delete the interface to the link from its olist. o If the interface deletion results in a null olist for the (*,G) then the usual actions are taken to propagate a Prune upstream. 3.1.3.2 New DF Actions On assuming the role of the DF for a given link, a router has to take the following actions for each existing group that is affected. If the router has IGMP information from local receivers for a group, the interface to the link must be added to the olist for the (*,G). If the (*,G) entry did not exist then it must be created and populated with the RP DF information. If the (*,G) olist was previously null then the usual actions are taken to propagate a Join upstream. 3.1.3.3 Downstream Router Actions Handley, Kouvelas, Vicisano [Page 5] Internet Draft Bidir PIM February 2000 When learning about a switch to a new DF on the RPF interface, down- stream routers must take the following actions for all affected groups. o If the router has a (*,G) entry with a non-null olist, it must send a Join for the group towards the new DF. o The router may also send a Prune for the group towards the old DF. 3.2 Forwarding Data The following responsibilities are uniquely assigned to the DF of a link: o The DF is the only router that forwards packets traveling down- stream 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 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). When a router receives a multicast packet sent to a bidir group G, it first looks for a (*,G) matching entry. If this is not found, then the matching (*,*,RP) state may be used. Alternatively (*,G) state may be created with a null olist and populated with the RP DF infor- mation. The router must forward the packet if either: o It was received on the RPF interface of the entry (always forward downstream traveling packets) o The router is the Designated Forwarder (DF) for the RP on the interface the packet was received (only the DF forwards upstream). If a decision to forward the packet is made, then it is forwarded on all the interfaces in the olist including the entry's RPF interface Handley, Kouvelas, Vicisano [Page 6] Internet Draft Bidir PIM February 2000 but excluding the interface the packet was received on. Otherwise the packet is discarded. Note: A major advantage of using a Designated Forwarder in bi- directional PIM 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 check for data to groups that do not have existing state. 3.2.1 Source-only Branches Source-only branches of the distribution tree for a group 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 do not have an olist for the group and hence do not need to maintain (*,G) state. Upstream forwarding can be performed using (*,*,RP) state. An implementation may decide to maintain (*,G) state for accounting or performance reasons. 4 Designated Forwarder This section presents a fail-safe mechanism for electing a per-RP designated router on each link in a PIM domain. We call this router the Designated Forwarder (DF). 4.1 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. When comparing metrics from dif- ferent unicast routing protocols, we use the same comparison rules used in the PIM assert process [1]. 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 Handley, Kouvelas, Vicisano [Page 7] Internet Draft Bidir PIM February 2000 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. 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 sec- tion 3.2, if multiple routers end-up thinking that they should be responsible for forwarding, loops may result. To reduce the possibil- ity of this occurrence to a minimum, the election algorithm has been biased towards discarding DF information and suspending forwarding during periods of ambiguity. 4.2 DF Election Description To perform the election of the DF for a particular RP, routers on a link need to exchange their unicast routing metric information for reaching the RP. In the election protocol described below, many message exchanges are repeated 3 times for reliability. In all those cases the message retransmissions are spaced in time by a small random interval. For the purposes of the election, interface specific counters and timers need to be maintained for each RP. When (*,G) entries are created, they inherit information on the elected DF from the corresponding RP database entry. Subsequent changes in the winner of the DF election for a RP are propagated to all dependent (*,G) entries. 4.2.1 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 randomly chosen in the interval [0.5 * Offer-Interval, Offer-Interval]. If a router hears a better offer from a neighbor, it stops Handley, Kouvelas, Vicisano [Page 8] Internet Draft Bidir PIM February 2000 participating in the election for a period of [3 * Offer-Interval]. If during this period no winner is elected, then it restarts the election from the beginning. If a router receives an offer with worse metrics, then it restarts the election from the beginning. The result should be that all routers except the best candidate stop advertising. A router assumes the role of the DF after having advertised its metrics 3 times without receiving any offer from any other 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. 4.2.2 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 DF, the router with the new metric should take action to eventually assume forwarding responsibility. After the metric change is detected, the new candidate restarts participating in the election. If no response is received after 3 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 the 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 sta- bilises. 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. Handley, Kouvelas, Vicisano [Page 9] Internet Draft Bidir PIM February 2000 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. 4.2.3 Winner Metric Changes If the DF's routing metric to reach the RP changes to a worse value, it sends a set of 3 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 indi- cating 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. 4.2.4 Winner Loses Path If a router's path to the RP switches to be through a link for which it is acting as the DF, then it can no longer provide forwarding ser- vices 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 mes- sage but this adds unnecessary complication to the election process. 4.2.5 Late Router Starting Up A late router starting up will have no knowledge of a previous elec- tion outcome. As a result it will start advertising its metric in Offer messages. As soon as this happens, the Winner will respond either with a Winner or with a Backoff message. 4.2.6 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 Handley, Kouvelas, Vicisano [Page 10] Internet Draft Bidir PIM February 2000 routers on the link. If there are downstream routers, typically their RPF neighbor reported by unicast routing will be the DF. They will therefore notice a change in RPF neighbor away from the DF and will restart the election by transmitting Offer messages. If the RP is now reachable through the 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. 4.3 Election Protocol Specification 4.3.1 Protocol State The operation of the election protocol makes use of the variables and timers described below. These are maintained per RP for each multi- cast enabled interface on the router. Offer-Count (O-count) Used to maintain the number of times an Offer or Winner mes- sage has been transmitted. Best-Offer Used by the DF to record who has made the last offer for sending the Pass message. Offer-Timer (O-timer) Used to schedule transmission of Offer and Winner messages. Pass-Timer (P-timer) Used on the DF to schedule transmission of a Pass message. 4.3.2 Message Summary The election uses the following control messages: 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. Handley, Kouvelas, Vicisano [Page 11] Internet Draft Bidir PIM February 2000 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. 4.3.3 Protocol Events During protocol operation, in addition to the expiration of the two timers and reception of the four messages, the following events can take place: o Discovery of new RP o Metric change o DF loses path o Detection of DF failure (unicast routing changed for downstream or Hello expired) 4.3.4 Protocol Operation In the two tables below the following rules and notation apply: o Whenever the notation "?=" is used to assign a value to a timer, the value is assigned only if the timer is not running or the time left running is longer than the new value. o When a new DF is discovered through the receipt of a Winner or Pass message, if it is not already a PIM neighbor, a neighbor entry is created with the default expiration interval. o Whenever the DF is set, the associated metrics are also recorded. o Timers in square brackets are randomly chosen between 0.5 and 1 times the supplied value. o When a router has a path to the RP through the link on which the election is taking place, an infinite metric is used in Offer mes- sages. Handley, Kouvelas, Vicisano [Page 12] Internet Draft Bidir PIM February 2000 Event Condition Non-DF action DF action ======================================================================= Offer |Local metric |O-count = 0 |Send Backoff rcvd |worse |O-timer = [Offer-Int] |Best-Offer = sender | | + 3 * Offer-Int |P-timer = Backoff-Int | | |Stop O-timer |-------------------------------------------------------------- |Local metric |O-count = 0 |Send Winner |better |O-timer ?= [Offer-Int] |Stop P-timer ----------------------------------------------------------------------- Winner |Local metric |O-count = 0 rcvd |worse |Stop O-timer | |DF = sender | |Stop P-timer |-------------------------------------------------------------- |Local metric |O-count = 0 |better |O-timer ?= [Offer-Int] | |DF = sender | |Stop P-timer ----------------------------------------------------------------------- Backoff |Local metric |O-count = 0 rcvd |worse or to us |O-timer = Backoff-Int + [Offer-Int] | |DF = sender | |Stop P-timer |-------------------------------------------------------------- |Local metric |O-count = 0 |better |O-timer ?= [Offer-Int] | |DF = sender | |Stop P-timer ----------------------------------------------------------------------- Pass |Local metric |O-count = 0 rcvd |worse or to us |Stop O-timer | |DF = destination | |Stop P-timer |-------------------------------------------------------------- |Local metric |O-count = 0 |better |0-timer ?= [Offer-Int] | |DF = destination | |Stop P-timer ----------------------------------------------------------------------- Handley, Kouvelas, Vicisano [Page 13] Internet Draft Bidir PIM February 2000 Event Condition Non-DF action DF action ======================================================================= New RP | |O-count = 0 |N/A | |O-timer ?= [Offer-Int] | ----------------------------------------------------------------------- Metric |DF metric |nop |O-count = 0 change |better (*) | |O-timer ?= [Offer-Int] | | |Stop P-timer |-------------------------------------------------------------- |DF metric |O-count = 0 |Send Winner |worse (*) |O-timer ?= [Offer-Int] |Stop P-timer ----------------------------------------------------------------------- No path | |nop |Send Offer (**) to RP | | |O-count = 1 | | |O-timer ?= [Offer-Int] | | |DF = unknown | | |Stop P-timer ----------------------------------------------------------------------- DF | |O-count = 0 |N/A failure | |O-timer ?= [Offer-Int] | ----------------------------------------------------------------------- O-timer |O-count <= 3 |Send Offer expires | |O-count++ | |O-timer ?= [Offer-Int] |---------------|---------------------------------------------- |else |O-count = 0 | |Send Winner (***) | |DF = us ----------------------------------------------------------------------- P-timer | |DF = Best-Offer expires | |Send Pass ----------------------------------------------------------------------- (*) These comparisons are made against the previously stored DF metrics. In the case of the DF, the old local metrics are used to compare against. So "DF metric better" means that the metric has actually become worse. (**) As the path to the RP is now through the link an infinite metric is used in the offer. (***) We only become the DF and send a Winner message if we have a path to the RP (which is not through the link with the ongoing elec- tion). 4.4 Election Message Formats All election messages are sent with a TTL of 1 and are multicast to Handley, Kouvelas, Vicisano [Page 14] Internet Draft Bidir PIM February 2000 the ALL-PIM-ROUTERS group. The structure of Encoded-Unicast addresses is described in [1]. 4.4.1 Common Header The header below is common to all election messages. 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 | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ Type TBD Subtype Used to distinguish between different election messages and is set according to the table below: Message Subtype ----------------------- Offer 1 Winner 2 Backoff 3 Pass 4 Rsvd Set to zero by senders and ignored by receivers. Checksum Calculated as specified in [1]. RP-Address The address of the bidir RP for which the election is taking place. Sender Metric Preference Preference value assigned to the unicast routing protocol that the message sender used to obtain the route to the RP- address. Handley, Kouvelas, Vicisano [Page 15] Internet Draft Bidir PIM February 2000 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. The Backoff and Pass messages have the additional fields described below. 4.4.2 Backoff Message The Backoff message uses the following fields in addition to the com- mon ones described above. +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Encoded-Unicast-Offering-Address | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Offering Metric Preference | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Offering Metric | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Interval | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ Offering Address The address of the router that made the last (best) Offer. Offering Metric Preference Preference value assigned to the unicast routing protocol that the offering router used to obtain the route to RP- address. 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. 4.4.3 Pass Message The Pass message uses the following fields in addition to the common ones described above. Handley, Kouvelas, Vicisano [Page 16] Internet Draft Bidir PIM February 2000 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | 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. 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. 4.5 Timer Values The Offer-Interval is 100 ms. The default Backoff-Interval used in Backoff messages is 1 sec. 5 Advertising Bi-directional Groups Routers discover that a group operates in bi-directional mode from the Encoded-Group Address fields in PIM Bootstrap and Candidate-RP Advertisement messages. The Encoded-Group Address field is modified to include 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 | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ When the Bidir-bit is set, all upgraded bi-directional PIM routers will follow the forwarding rules described in this specification. 6 Security Considerations All PIM control messages MAY use IPsec to address security concerns. Handley, Kouvelas, Vicisano [Page 17] Internet Draft Bidir PIM February 2000 7 References [1] Estrin, et al., "Protocol Independent Multicast-Sparse Mode (PIM- SM): Protocol Specification", RFC 2362, June 1998. [2] D. Estrin, D. Farinacci, "Bi-directional Shared Trees in PIM-SM", Work In Progress, , May 1999. [3] Wei, L., Farinacci, D., "PIM Version 2 DR Election Priority Option", INTERNET-DRAFT, March 1998. 8 Acknowledgments The bidir proposal in this draft is heavily based on the ideas and text presented by Estrin and Farinacci in [2]. The main difference between the two proposals is in the method chosen for upstream for- warding. We would also like to thank Deborah Estrin at ISI/USC as well as Nidhi Bhaskar, Yiqun Cai, Tony Speakman and Rajitha Sumanasakera at cisco for their contributions and comments to this draft. 9 Author Information Mark Handley mjh@aciri.org AT&T Center for Internet Research at ICSI Isidor Kouvelas kouvelas@cisco.com cisco Systems Lorenzo Vicisano lorenzo@cisco.com cisco Systems Handley, Kouvelas, Vicisano [Page 18] Internet Draft Bidir PIM February 2000 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. 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 mes- sage 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. 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. Appendix B: Interoperability with legacy code The rules provided in [2] 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 defini- tion of a boundary between a bi-directional capable area and a legacy area of the network. In [2], 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 Handley, Kouvelas, Vicisano [Page 19] Internet Draft Bidir PIM February 2000 register encapsulation is necessary if one or more routers on the RPF interface are not bi-directional capable. As in [2], 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. 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 signle tree sig- nificantly reduces state requirements on a router. The drawback is that it may produce suboptimal paths from sources to receivers pos- sibly 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 RP using RP-specific forwarding state. o The Bidir Designated Forwarder (DF) assumes all the responsibili- ties 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 com- pletely 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 Handley, Kouvelas, Vicisano [Page 20] Internet Draft Bidir PIM February 2000 every packet. This is done to determine if the packet was ori- ginated 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. Appendix D: Comparison with UMP based bidirectional PIM Using an UMP option for upstream forwarding has the following disad- vantages: o Using the DF election, only routers willing to be forwarders can be elected. In contrast in [2], the downstream router designates the upstream neighbor responsible for forwarding (using Joins and UMP packets). o Using the UMP option, regular data packets are overloaded with con- trol information for the routing protocol. o Inserting the extra option in multicast packets transmitted from a source may result in a packet size exceeding the MTU which will result in fragmentation. o The use of an option complicates the router forwarding mechanism. Additional code to process the new special packet type needs to be written. o The contents of the UMP option have to be rewritten and the packet checksum adjusted on each hop towards the RP at data forwarding time. This introduces additional per packet processing overhead. In bidir PIM [2], if the router elected as the DR is different from that chosen by downstream neighbors for joining the tree, loops can occur. The main shortcoming of the DR is that its election does not take into consideration the location of the RP. To resolve this prob- lem the DR priority draft [3] provides a method for manually confi- guring the DR election winner. Although this provides a solution it has two drawbacks: o It requires a case by case manual configuration. o It cannot solve the problem if there are different RPs in a domain serving separate multicast group ranges. In this scenario the Handley, Kouvelas, Vicisano [Page 21] Internet Draft Bidir PIM February 2000 requirements of each RP for the DR positioning on a particular link may differ. Handley, Kouvelas, Vicisano [Page 22]