A Specification Framework for Real-Time Scheduling

• Authors: K. Subramani
• Venue: SOFSEM '02 Proceedings of the 29th Conference on Current Trends in Theory and Practice of Informatics: Theory and Practice of Informatics
• Year: 2002

Scheduling in Real-time systems differs from scheduling in conventional models in two principal ways: (a) Execution time variability, and (b) Existence of complex constraints between jobs. There is a third issue that indirectly depends upon the non-constant nature of execution times, viz. the degree of clairvoyance permitted by the application. Whereas traditional scheduling models assume fixed values for job execution time, we model execution times of jobs through convex sets; doing so permits the capturing of situations that could not be modeled otherwise. A feature that is unique to real-time systems is the presence of temporal relationships that constrain job execution. Consider for instance the requirement that job J1 should conclude 10 units before job J2. This can be modeled through a simple, linear relationship, between the start and execution times of jobs J1 and J2, however precedence graphs are inadequate for this purpose. In hard, real-time scheduling, it is important to guarantee a priori, the schedulability of the system, since the violation of a constraint leads to catastrophic consequences. Depending upon the nature of the application involved, there are 3 different schedulability specifications viz. Static (Zero-Clairvoyant), Parametric (Partially Clairvoyant) and Co-Static (Totally Clairvoyant). Each specification affords a different level of flexibility and has different dispatching concerns. In this paper, we present a framework that enables the specification of real-time scheduling problems and discuss the relationship between flexibility and complexity in the proposed model. We motivate each aspect of our model through examples from real-world design.