Summary

  • RFC 1112 made multicast membership observable through local Queries and Reports, but its randomized report suppression let one host’s message stand for a group rather than enumerate its members.
  • IGMPv1 had no Leave message: routers learned that the final listener was gone only after reports stopped and membership state timed out. RFC 2236 later added a faster leave signal.

A group address was not a guest list

An IP multicast address could be permanent even when no host belonged to it. A host could also send to a group without joining it. In RFC 1112, Stephen Deering separated those facts: the destination address named a group, while membership was maintained per host and network interface. The distinction gave applications a way to ask IP to join or leave, and gave routers a separate question: does this attached network currently have anyone interested?

The standard described two levels of host support. A Level 1 host could send multicast datagrams but did not join groups or implement IGMP. Level 2 added group membership and the protocol needed to tell a neighboring multicast router about it. A group address by itself therefore said little about reception. A router needed a fresh signal from the local segment.

Why the first report did not come alone

When a router sent a Host Membership Query to the all-hosts address 224.0.0.1, each Level 2 host considered its memberships on that interface. Sending every answer immediately would create a burst of duplicate control packets. RFC 1112 instead assigned each membership a randomly chosen report-delay timer. When a host heard a valid report for the same group before its own timer expired, it canceled that pending report. One response could then tell the router that at least one member remained.

The mechanism was a deliberate compression of evidence. A router did not receive a count, a list of machines, or proof that every application was ready to consume traffic. It received a group-level indication on one local network. That reduced repeated messages, especially where many hosts listened to the same group, but discarded information the protocol did not need for forwarding.

Joining was treated differently from answering a periodic query. A host that joined a new group was told to send a report promptly, because it might be the first member on that network. Repeating the report once or twice was recommended to guard against loss. Multicast itself remained best effort: the standard promised neither intact delivery to every member nor a common packet order. Membership signaling could inform a router without making delivery certain.

The silence after the last listener

IGMPv1 specified Queries and Reports, but no Leave Group message. When a host left, its local IP module could stop treating it as a member; the router was not told immediately. Periodic Queries refreshed the router’s view. If no host reported the group, the router’s membership timer eventually expired. The cost of a small control protocol was state that could outlive the last listener.

That was not the same as a router believing a departed host still existed forever. It was a bounded delay, dependent on query and membership timers. But the delay mattered for high-bandwidth groups or networks where membership changed often: traffic could keep being forwarded to a segment with no remaining receiver until the router’s evidence aged out.

In RFC 2236, IGMPv2 added a Leave Group message for faster notification. A querier followed a Leave with group-specific Queries and waited for a response before concluding the segment was empty. The later design did not turn silence into proof; it made the uncertainty interval shorter and tested for another listener.

The historical trade-off is narrow but durable. IGMPv1 optimized the network’s control traffic by treating membership as a presence question. It let a report suppress its neighbors, which kept the protocol quiet while preserving enough information for a router to decide whether a local group needed forwarding. What it could not tell the router was how many members stood behind that single report—or that the final one had just gone.

Sources