IP Multicast - Technical Description

Technical Description

IP multicast is a technique for one-to-many and many-to-many real-time communication over an IP infrastructure in a network. It scales to a larger receiver population by requiring neither prior knowledge of a receiver's identity nor prior knowledge of the number of receivers. Multicast uses network infrastructure efficiently by requiring the source to send a packet only once, even if it needs to be delivered to a large number of receivers. The nodes in the network (typically network switches and routers) take care of replicating the packet to reach multiple receivers such that messages are sent over each link of the network only once. The most common low-level protocol to use multicast addressing is User Datagram Protocol (UDP). By its nature, UDP is not reliable—messages may be lost or delivered out of order. Reliable multicast protocols such as Pragmatic General Multicast (PGM) have been developed to add loss detection and retransmission on top of IP multicast.

Key concepts in IP multicast include an IP multicast group address, a multicast distribution tree and receiver driven tree creation.

An IP multicast group address is used by sources and the receivers to send and receive multicast messages. Sources use the group address as the IP destination address in their data packets. Receivers use this group address to inform the network that they are interested in receiving packets sent to that group. For example, if some content is associated with group 239.1.1.1, the source will send data packets destined to 239.1.1.1. Receivers for that content will inform the network that they are interested in receiving data packets sent to the group 239.1.1.1. The receiver joins 239.1.1.1. The protocol typically used by receivers to join a group is called the Internet Group Management Protocol (IGMP).

With routing protocols based on shared trees, once the receivers join a particular IP multicast group, a multicast distribution tree is constructed for that group. The protocol most widely used for this is Protocol Independent Multicast (PIM). It sets up multicast distribution trees such that data packets from senders to a multicast group reach all receivers which have joined the group. For example, all data packets sent to the group 239.1.1.1 are received by receivers who joined 239.1.1.1. There are variations of PIM implementations: Sparse Mode (SM), Dense Mode (DM), Source Specific Mode (SSM) and Bidirectional Mode (Bidir, or Sparse-Dense Mode, SDM). Of these, PIM-SM is the most widely deployed as of 2006; SSM and Bidir are simpler and scalable variations developed more recently and are gaining in popularity.

IP multicast operation does not require an active source to know about the receivers of the group. The multicast tree construction is receiver driven and is initiated by network nodes which are close to the receivers. IP multicast scales to a large receiver population. The IP multicast model has been described by Internet architect Dave Clark as, "You put packets in at one end, and the network conspires to deliver them to anyone who asks."

IP multicast creates state information per multicast distribution tree in the network. If a router is part of 1000 multicast trees, it has 1000 multicast routing and forwarding entries. On the other hand, a multicast router does not need to know how to reach all other multicast trees in the Internet. It only needs to know about multicast trees for which it has downstream receivers. This is key to scaling multicast-addressed services. It is very unlikely that core Internet routers would need to keep state for all multicast distribution trees, they only need to keep state for trees with downstream membership. In contrast, a unicast router needs to know how to reach all other unicast addresses in the Internet, even if it does this using just a default route. For this reason, aggregation is key to scaling unicast routing. Also, there are core routers that carry routes in the hundreds of thousands because they contain the Internet routing table.