Motivating hierarchical run-time models in measurement and control systems

Measurement and control systems are intrinsically distributed and real-time, as they contain sensor and actuator nodes that interact with the physical world directly. Embedded software in the computational nodes is responsible for timely reaction to sensor data, and for producing actuation. This paper reviews run-time computation models for this kind of real-time embedded software, from message semantics, message acquisition, and the dataflow/control flow perspectives. In general, dataflow centric models are natural for describing measurement and control algorithms and easy to use in distributed systems, but they lack mechanisms for controlling the execution order to fulfill timing constrains. Control-flow centric models are good at handling real-time requirements but are hard to distribute and sometimes hard to analyze. Although most practical run-time models to some extent support both dataflow and control flow, they are hardly universal. In complex applications, it is desirable to use different models in different parts of the system or under different modes of operation. Cleanly integrating multiple run-time models is a challenging task. In this paper we motivate a hierarchical architecture for composing run-time models, based on the Ptolemy II component framework and models of computation. Unlike traditional real-time operating systems that provide only one flat layer of abstraction, the hierarchical architecture enhances flexibility, scalability, and reliability of MC systems by mixing and matching multiple run-time models in a disciplined way.