Intra-Component Adaptation using
Distributed Events

Each CANS driver and service can incorporate its own adaptation behavior that may or may not be coordinated with adaptation in other components. For example, a frame-dropping component can alter its policies upon detecting different levels of back-pressure on its output buffers. Note that adaptation in a single component is completely isolated as long as its effect is restricted to be within a single semantic segment (see Section 3.1). To trigger adaptation, CANS provides distributed event propagation support, permitting components (including delegate objects for legacy services) to raise arbitrary events as well as listen for specific ones. Event support is realized by a per execution environment Event Manager, which is responsible for catching, firing, and transmitting events across the network. Event raising and firing is implemented using simple method calls and callback functions associated with the relevant component. There are two major types of CANS events: events from the local resource monitor, indicating a change in resource status, and events from components on the data path. The first kind of events are sent only to local components that register themselves as interested listeners. The second kind, issued by components along a data path, are first sent to the plan event delegate (see Section 4.3), which is responsible for propagating the event along the data path as well as handling plan-specific events, such as events to trigger replanning.