Protocol Independent Multicast (PIM) is a collection of multicast routing protocols, each optimized for a different environment. There are two main PIM protocols, PIM Sparse Mode and PIM Dense Mode. A third PIM protocol, Bi-directional PIM, is less widely used.
Typically, either PIM Sparse Mode or PIM Dense Mode will be used throughout a multicast domain. However, they may also be used together within a single domain, using Sparse Mode for some groups and Dense Mode for others. This mixed-mode configuration is known as Sparse-Dense Mode. Similarly, Bi-directional PIM may be used on its own, or it may be used in conjunction with one or both of PIM Sparse Mode and PIM Dense Mode.
All PIM protocols share a common control message format. PIM control messages are sent as raw IP datagrams (protocol number 103), either multicast to the link-local ALL PIM ROUTERS multicast group, or unicast to a specific destination.
PIM Sparse Mode
PIM Sparse Mode (PIM-SM) is a multicast routing protocol designed on the assumption that recipients for any particular multicast group will be sparsely distributed throughout the network. In other words, it is assumed that most subnets in the network will not want any given multicast packet. In order to receive multicast data, routers must explicitly tell their upstream neighbors about their interest in particular groups and sources. Routers use PIM Join and Prune messages to join and leave multicast distribution trees.
PIM-SM by default uses shared trees, which are multicast distribution trees rooted at some selected node (in PIM, this router is called the Rendezvous Point, or RP) and used by all sources sending to the multicast group. To send to the RP, sources must encapsulate data in PIM control messages and send it by unicast to the RP. This is done by the source's Designated Router (DR), which is a router on the source's local network. A single DR is elected from all PIM routers on a network, so that unnecessary control messages are not sent.
One of the important requirements of PIM Sparse Mode, and Bi-directional PIM, is the ability to discover the address of a RP for a multicast group using a shared tree. Various RP discovery mechanisms are used, including static configuration, Bootstrap Router, Auto-RP, Anycast RP, and Embedded RP.
PIM-SM also supports the use of source-based trees, in which a separate multicast distribution tree is built for each source sending data to a multicast group. Each tree is rooted at a router adjacent to the source, and sources send data directly to the root of the tree. Source-based trees enable the use of Source-Specific Multicast (SSM), which allows hosts to specify the source from which they wish to receive data, as well as the multicast group they wish to join. With SSM, a host identifies a multicast data stream with a source and group address pair (S,G), rather than by group address alone (*,G).
PIM-SM may use source-based trees in the following circumstances.
- For SSM, a last-hop router will join a source-based tree from the outset.
- To avoid data sent to an RP having to be encapsulated, the RP may join a source-based tree.
- To optimize the data path, a last-hop router may choose to switch from the shared tree to a source-based tree.
PIM-SM is a soft-state protocol. That is, all state is timed-out a while after receiving the control message that instantiated it. To keep the state alive, all PIM Join messages are periodically retransmitted.
Version 1 of PIM-SM was created in 1995, but was never standardized by the IETF. It is now considered obsolete, though it is still supported by Cisco and Juniper routers. Version 2 of PIM-SM was standardized in RFC 2117 (in 1997) and updated by RFC 2362 (in 1998). Version 2 is significantly different from and incompatible with version 1. However, there were a number of problems with RFC 2362, and a new specification of PIM-SM version 2 is currently being produced by the IETF. There have been many implementations of PIM-SM and it is widely used.
PIM Dense Mode
PIM Dense Mode (PIM-DM) is a multicast routing protocol designed with the opposite assumption to PIM-SM, namely that the receivers for any multicast group are distributed densely throughout the network. That is, it is assumed that most (or at least many) subnets in the network will want any given multicast packet. Multicast data is initially sent to all hosts in the network. Routers that do not have any interested hosts then send PIM Prune messages to remove themselves from the tree.
When a source first starts sending data, each router on the source's LAN receives the data and forwards it to all its PIM neighbors and to all links with directly attached receivers for the data. Each router that receives a forwarded packet also forwards it likewise, but only after checking that the packet arrived on its upstream interface. If not, the packet is dropped. This mechanism prevents forwarding loops from occurring. In this way, the data is flooded to all parts of the network.
Some routers will have no need of the data, either for directly connected receivers or for other PIM neighbors. These routers respond to receipt of the data by sending a PIM Prune message upstream, which instantiates Prune state in the upstream router, causing it to stop forwarding the data to its downstream neighbor. In turn, this may cause the upstream router to have no need of the data, triggering it to send a Prune message to its upstream neighbor. This 'broadcast and prune' behavior means that eventually the data is only sent to those parts of the network that require it.
Eventually, the Prune state at each router will time out, and data will begin to flow back into the parts of the network that were previously pruned. This will trigger further Prune messages to be sent, and the Prune state will be instantiated once more.
PIM-DM only uses source-based trees. As a result, it does not use RPs, which makes it simpler than PIM-SM to implement and deploy. It is an efficient protocol when most receivers are interested in the multicast data, but does not scale well across larger domains in which most receivers are not interested in the data.
The development of PIM-DM has paralleled that of PIM-SM. Version 1 was created in 1995, but was never standardized. It is now considered obsolete, though it is still supported by Cisco and Juniper routers. Version 2 of PIM-DM is currently being standardized by the IETF. As with PIM-SM, version 2 of PIM-DM is significantly different from and incompatible with version 1. PIM Dense Mode (PIM DM) is less common than PIM-SM, and is mostly used for individual small domains.
Bi-directional PIM
Bi-directional PIM (BIDIR-PIM) is a third PIM protocol, based on PIM-SM. The main way BIDIR-PIM differs from PIM-SM is in the method used to send data from a source to the RP. Whereas in PIM-SM data is sent using either encapsulation or a source-based tree, in BIDIR-PIM the data flows to the RP along the shared tree, which is bi-directional - data flows in both directions along any given branch.
BIDIR-PIM's major differences from PIM-SM are as follows.
- There are no source-based trees, and in fact no (S,G) state at all. Therefore there is no option for routers to switch from a shared tree to a source-based tree, and Source-Specific Multicast is not supported.
- To avoid forwarding loops, for each RP one router on each link is elected the Designated Forwarder (DF). This is done at RP discovery time using the DF election message.
- There is no concept of a Designated Router.
- No encapsulation is used.
- The forwarding rules are very much simpler than in PIM-SM, and there are no data-driven events in the control plane at all.
The main advantage of BIDIR-PIM is that it scales very well when there are many sources for each group. However, the lack of source-based trees means that traffic is forced to remain on the possibly inefficient shared tree.
There have been two proposed specifications for Bi-directional PIM. The first was described in draft-farinacci-bidir-pim, which dates from 1999. The protocol described here is a replacement, simpler than and with some improvements over the first. It is described in draft-ietf-pim-bidir.
Mixed-mode PIM Configurations
Typically, PIM-SM, PIM-DM or BIDIR-PIM would be used alone throughout a multicast domain. However it is possible to use a combination of the three by distributing multicast groups between the different protocols. Each group must operate in either sparse, dense or bi-directional mode; it is not possible to use a single group in more than one mode at once. Given such a division, the protocols coexist largely independent of one another.
The one way in which the protocols interact is that the same PIM Hello protocol is used by each, and is only run once on each link. The information learned from the Hello message exchange must be shared among the three routing protocols.
The method used to distribute groups between the three protocols is outside the scope of the PIM protocols and is a matter of local configuration. Note that it is important that every router in the domain has the same assignment of groups to protocols. The following techniques are used.
- The Bootstrap Router (BSR) protocol, used for RP discovery, has been extended to add a "Bi-directional" bit for each group range. This method may be used to assign groups between sparse and bi-directional modes if using BSR.
- Routers may be configured to use dense mode if the RP discovery mechanism (whatever that may be) fails to find an available RP for a group, and to use sparse or bi-directional mode otherwise.
- Router may be manually configured with group ranges for sparse, dense and bi-directional modes.