How to deal with large rooms. (SPEC-36)

[matrixbot]

For every user in the room we currently have a dedicated state event for them. For large rooms with many 1000s or more members this becomes less than ideal due to the overhead for both clients and servers.

The protocol itself only requires knowledge of which servers are in the room.

(Imported from https://matrix.org/jira/browse/SPEC-36)

(Reported by @erikjohnston)

[matrixbot]

Jira watchers: @erikjohnston @NegativeMjark @richvdh

[matrixbot]

For rooms that have a large number of participants split over a small number of servers we can use the existing full mesh protocol by distributing a list of member servers using membership PDUs for a participant on each server to prove that it should be in the list.

For rooms with a large number of participating servers we might need to look at new protocols for distributing the messages without each server needing to know the full list.

-- @NegativeMjark

[matrixbot]

Is this bug about alternatives to full-mesh routing, or about reducing the number of state events flying around?

-- @richvdh

[matrixbot]

full-mesh alternatives seems to be https://github.com/matrix-org/matrix-doc/issues/469, so I assume it's the latter? is this a real problem?

-- @richvdh