Performability modeling with stochastic activity networks

Distributed real-time systems are increasingly used in applications such as computer communication networks, industrial process control, automated manufacturing, etc., which usually have strict performance and reliability requirements. To effectively evaluate such complex systems, powerful modeling techniques and tools are needed to provide efficiently useful information about the behavior of these systems. In response to this need, this dissertation considers new models, measures, and methods which can be used to evaluate the performability (performance-reliability) of distributed real-time systems. Particularly, a new class of models, called "stochastic activity networks" (SANs), are developed which permits the representation of concurrency, timeliness, fault-tolerance, and degradable performance in a single model. These models are generalizations of stochastic Petri nets and, which, have some features of queueing networks. They can be used to determine the stochastic state behavior of a complex system. Such state behavior then serves as a base model for the performability evaluation of the system. We also consider new performance variables which are suitable for the evaluation of real-time systems. The introduction of these measures has been stimulated by a growing awareness that the traditional measures of performance are not appropriate for systems which behave in real-time environments. Further, a scheduling mechanism, called "promptness control," is represented and an analytic modeling method is developed to analyze this mechanism. Finally, the use of the above models, measures, and modeling methods is illustrated for performability modeling of distributed real-time systems.