Predictable cyclic computations in autonomous systems : a computational model and implementation

Sammanfattning:

Cyclic computations under real-time constraints naturally occur in systems which use periodic sampling to approximate continuous signals for processing in a computer. In complex control systems, often with a high degree of autonomy, there is a need to combine this type of processing with symbolic computations for supervision and coordination.

In this thesis we present a computational model for cyclic time-driven computations subjected to run-time modifications initiated from an external system, and formulate conditions for predictable real-time behaviour. We introduce the dual state vector for representing periodically changing data. Computations are viewed as functions from the state represented by the vector at time t to the state one period later. Based on this model, we have implemented a software tool, the Process Layer Executive, which maintains dual state vectors and manages cyclic tasks that perform computations on vectors.

We present the results of a theoretical and practical evaluation of the real-time properties of the tool and give its overhead as a function of application dependent parameters that are automatically derivable from the application description in the Process Layer Configuration Language.