An AP maintains a table of active cluster applications, each entry of which exists as soft state. When a packet arrives with an unknown cluster identifier in its CP header, a new entry will be created in the table and CP probe mechanisms will become active for that application. Similarly, if no CP packet has been seen for a particular cluster identifier , then the entry will time out and be removed from the application table. Use of soft state in this manner is both flexible and lightweight in that it avoids the need for explicit configuration and ongoing administration.
For each cluster application, the AP monitors the number of participating flows, and the number and size of packets received during a given interval. Weighted averages are calculated to dampen the effect of packet bursts. The information is passed back to local cluster endpoints using the CP header whenever a packet arrives from the remote AP on route to a local endpoint. If no such packet arrives within a specified time period, then a report packet is created and ``pushed'' to each endpoint informing them of cluster application membership and bandwidth usage, as well as current network conditions.
An AP also maintains probe state, including a current packet sequence number, estimated round trip time and mean deviation, a loss history and estimated loss rate, and a bandwidth availability calculation. Use of these mechanisms is described below.