Entity members volunteer to be an entity leader, a central node responsible for communication and group maintenance, as well as running entity-specific application code. Once the entity leader no longer detects the event, it hands-off leadership (and current state) to another node by sending a relinquish leadership message. As the event migrates or expands, the leadership handoff mechanism built into the EMM ensures that the entity and its stored application state migrate with it. The application on a node is informed when the node becomes a leader so that it can pick up computation from where the previous leader left it. The application is also notified when the node ceases to be leader.
A key requirement is to ensure that only one entity is spawned for the same environmental event. The EMM protocol achieves this requirement by announcing the existence of an entity to nearby nodes within a distance called the awareness horizon. By design, the awareness horizon is larger than the sensory horizon. Nodes in the awareness horizon that cannot sense the event are called entity followers, as distinguished from entity members. These nodes are prevented from spawning new entities.
Nodes join the awareness horizon upon the reception of a bounded-hop broadcast (heartbeat) from the nearby EMM leader. Upon receiving a heartbeat, a node sets a corresponding entity timeout timer. The node ceases to be a follower when the timer expires. The timer is re-initialized upon the receipt of each new heartbeat. If the node senses the event signature before the entity timeout timer expires it becomes a member of the existing entity. If the timer had already expired the node is no longer a follower and will create a new entity. This mechanism prevents multiple entities from being spawned for the same environmental event. Figure 2 depicts the node state transition diagram between follower, member, and leader states, as well as the free state in which a node is not cognizant of any entities. For readability, the only transitions shown are those between different states (as opposed to loop-backs from a state to itself).
Note that, while inhibiting entity followers from creating new entities is essential for preventing redundant representation of the same target, it has one unfortunate side effect. Namely, the mechanism may prevent nodes from reporting secondary events in the vicinity of events already reported. Since we assume that events are sparse, the hope is that cases like the above are not common.
