Situation awareness is an important application category in cyber-physical systems, and distributed video-based surveillance is a good canonical example of this application class. Such applications are interactive, dynamic, stream-based, computationally demanding, and needing real-time or near real-time guarantees. A sense-process-actuate control loop characterizes the behavior of this application class. ASAP is a scalable distributed architecture for a multi-modal sensor network that caters to the needs of this application class. Features of this architecture include (a) generation of prioritization cues that allow the infrastructure to pay selective attention to data streams of interest; (b) virtual bstraction that allows easy integration of multi-modal sensing capabilities; and (c) dynamic redirection of sensor sources to distributed resources to deal with sudden burstiness in the application. In both empirical and emulated experiments, ASAP shows that it scales up to a thousand of ...