Multicast is a one-to-many delivery model in which a source sends a single stream of packets to a group address, and the network replicates those packets only where paths toward interested receivers diverge. In Internet addressing, multicast occupies dedicated address blocks in both IPv4 and IPv6, so a destination address identifies a set of listeners rather than a single interface.
The model sits between unicast, where every receiver gets its own copy from the sender, and broadcast, where every host on a segment receives the packet whether it wants it or not. Its value is bandwidth efficiency at scale: a sender transmits once regardless of audience size, and each network link carries at most one copy of the stream.
How multicast delivery works
Multicast delivery works by separating group membership from packet forwarding. Receivers announce interest in a group to their local router, and routers build a distribution tree that connects the source to every segment with at least one member.
On the local link, hosts use IGMP (Internet Group Management Protocol) in IPv4 and MLD (Multicast Listener Discovery) in IPv6 to join and leave groups. Between routers, PIM (Protocol Independent Multicast) uses the existing unicast routing table to run a reverse path forwarding check: a router accepts a multicast packet only if it arrives on the interface that leads back toward the source, which prevents loops and duplicate copies.
The following table lists the protocols that cooperate in a typical multicast deployment and the role each one plays.
| Protocol | Scope | Function |
|---|---|---|
| IGMPv2 / IGMPv3 | IPv4 host to router | Joins and leaves groups; v3 adds source filtering |
| MLDv1 / MLDv2 | IPv6 host to router | IPv6 equivalent of IGMPv2 and IGMPv3 |
| PIM-SM | Router to router | Builds shared and source trees on explicit join |
| PIM-SSM | Router to router | Builds source trees for (S,G) channels only |
| MSDP | Between domains | Shares active IPv4 sources between rendezvous points |
| IGMP or MLD snooping | Layer 2 switch | Limits flooding to ports with members |
In PIM Sparse Mode, routers first forward traffic along a shared tree rooted at a rendezvous point and can later switch to the shortest-path tree from the source. Source-Specific Multicast skips the rendezvous point entirely, because receivers name both the source and the group, a pair written as (S,G).
Multicast address ranges in IPv4 and IPv6
Multicast addresses are drawn from reserved blocks that can never be assigned to an individual interface as a unicast address. In IPv4 the block is 224.0.0.0/4, the former Class D space; in IPv6 every multicast address begins with ff00::/8, and a 4-bit scope field states how far the packet may travel.
The ranges below show how the address itself signals intended reach and administrative purpose.
| Range | Name or purpose | Typical reach |
|---|---|---|
| 224.0.0.0/24 | Local network control block | Single link, never routed |
| 232.0.0.0/8 | IPv4 source-specific multicast | Global, (S,G) only |
| 233.0.0.0/8 | GLOP block derived from AS numbers | Interdomain |
| 239.0.0.0/8 | Administratively scoped | Inside one organization |
| ff02::/16 | IPv6 link-local scope | Single link |
| ff05::/16 | IPv6 site-local scope | One site |
| ff3x::/32 | IPv6 source-specific prefix | Depends on scope nibble |
Well-known examples include 224.0.0.5 for OSPF routers, 224.0.0.251 and ff02::fb for multicast DNS, and ff02::1 for all IPv6 nodes on a link. IPv6 relies on multicast far more than IPv4 does, because Neighbor Discovery replaces ARP broadcasts with solicited-node multicast addresses.
On Ethernet, an IPv4 group maps to a MAC address beginning 01:00:5e that carries only the lower 23 bits of the group, so 32 IPv4 groups share one MAC address. IPv6 groups map to MAC addresses beginning 33:33. This overlap is a known source of unexpected traffic on hosts that filter by MAC address alone.
Multicast compared with unicast, broadcast and anycast
Multicast differs from the other delivery models in who receives a packet and how many copies the sender produces. The comparison clarifies why each model suits a different class of traffic.
| Model | Receivers | Copies sent by source | Typical use |
|---|---|---|---|
| Unicast | One specific interface | One per receiver | Web, email, most applications |
| Multicast | All members of a group | One in total | IPTV, market data, routing protocols |
| Broadcast | All hosts on a segment | One per segment | ARP, DHCP discovery in IPv4 |
| Anycast | Nearest member of a set | One per request | DNS root servers, CDNs |
IPv6 has no broadcast at all; tasks that used broadcast in IPv4 use link-scoped multicast instead. Anycast shares an address among several servers but delivers each packet to only one of them, whereas multicast delivers to every subscribed receiver.
Where multicast is used and how to deploy it
Multicast is deployed mainly inside managed networks where one operator controls every router between source and receivers. Provider IPTV, financial market data feeds, video surveillance, software imaging in large organizations and media production over SMPTE ST 2110 all depend on it.
The decision table below summarizes common situations and the configuration that usually fits them.
| Situation | Recommended approach | Reason |
|---|---|---|
| Known single source per stream | PIM-SSM with IGMPv3 or MLDv2 | No rendezvous point, simpler and safer |
| Many sources, receivers unaware of sources | PIM-SM with redundant RPs | Shared tree handles source discovery |
| Flat Layer 2 network | IGMP snooping with a querier | Stops flooding to every switch port |
| Delivery over the public Internet | Unicast CDN or AMT tunnels | Native multicast is rarely carried between providers |
Applications typically use UDP for multicast transport, because TCP requires a separate connection per receiver. The TTL in IPv4 or hop limit in IPv6 should be set deliberately; a value of 1 keeps a stream on the local segment.
Limitations and common mistakes
Multicast trades sender efficiency for network state and operational complexity. Every router on a tree holds per-group or per-source entries, and troubleshooting requires tracing trees rather than single paths.
- No built-in reliability: lost UDP packets are not retransmitted unless the application adds forward error correction or a protocol such as NORM.
- Limited interdomain support: most Internet service providers do not forward multicast between networks, so global multicast remains rare.
- Missing IGMP querier: snooping switches without a querier eventually stop forwarding groups, which causes streams to fail after a few minutes.
- Wi-Fi inefficiency: access points often send multicast at the lowest basic data rate, so one stream can consume much of the airtime.
- Security gaps: anyone who can join a group can receive it, so confidential streams need encryption or access control at the router.
The simple explanation
Multicast is like a radio station on a computer network. The station broadcasts one signal, and only the people who tune in hear it. The station does not need to call every listener separately, and people who are not interested are not bothered.
On a network, a computer that wants a stream asks to join a group. The routers then send exactly one copy of the stream down each path that leads to a listener. One transmission can reach thousands of viewers without overloading the sender or the network.
How it is used in practice
A regional broadband provider offers 200 live television channels to 150,000 subscribers. Sending each viewer a separate unicast stream at 8 Mbit/s would require more than a terabit per second of headend capacity during peak evenings. Instead, each channel is assigned an address in the 232.0.0.0/8 source-specific range, and every set-top box sends an IGMPv3 report when the viewer changes the channel.
Access routers run PIM-SSM toward the headend encoders, and each link in the network carries at most one copy of each watched channel. When a viewer switches from news to sports, the box leaves one group and joins another, and the change completes in well under a second. As a result, core traffic depends on the number of channels watched, not the number of viewers, which keeps peak bandwidth predictable even during major sporting events.