| Abstract: |
Real-time systems need to be correct with respect to both functional and temporal behavior. The use of models to validate specifications and to evaluate design alternatives is nowadays a
necessity for the rapid and efficient development of complex and heterogeneous real-time systems. Many model-based evaluation methods exist. Some use simulation, others analyze the state space or
the numerical parameters of a system. Each method has its individual input model. The evaluation methods and their input descriptions are adapted to specific system characteristics and evaluation
purposes. In the area of schedulability analysis, this work proposes a characterization of task arrival and processor service behavior that unifies several previous methods and extends them
towards the analysis of bursty and utilization-bound behavior. In order to evaluate all relevant aspects of a system or its respective model, several different evaluation methods need to be
applied. This, however, requires multiple input models for the analysis of functional behavior, the documentation of system structure, and the validation of scheduling algorithms and timing
behavior. The use of multiple input models is both error-prone, as the validity with respect to the complete system of results obtained from investigating certain views in isolation is not clear,
as well as inefficient and costly, as several different models need to be built and maintained. This work proposes a novel approach for constructing integrated system models based on Petri nets,
an expressive and clearly defined visual and operational specification formalism. A pragmatic generic real-time system model structure is presented which can be used for scheduling and timing
simulation as well as for formal verification of functionality. Solutions are proposed for the issues that have previously prevented Petri nets from being used for the modeling of systems with
preemptive scheduling. For a model following the proposed generic structure it is shown that certain analytical statements obtained from model abstractions remain valid for the complete model. A
methodology has been developed to systematically construct models that confom to this structure. The model building process makes use of design patterns, a concept that is transfered from the
software engineering domain to the area of modeling with dataflow visual languages such as Petri nets. Petri net design patterns capture behavior instead of structure and are thus orthogonal and
complementary to the traditional approach of reducing model complexity by hierarchical structuring. |
