Using simulated annealing for hard real-time design-to-time scheduling

We present a two-parts scheduling algorithm based on simulated annealing and dynamical dispatching for the design-to-time scheduling of hard real-time applications. A decision-DAG (directed acyclic graph) is calculated and optimized offline using simulated annealing and then interpreted during runtime. The underlying system model consists of non-preemptable, periodic tasks with release times and hard deadlines. Tasks may be data-dependent and they are implemented by exchangeable methods which have nondeterministic execution times and a scalar value that represents the quality of service they deliver The methodology presented improves the quality of service delivered by design-to-time applications while it still guarantees all deadlines being met. We show that simulated annealing delivers much better quality for a generically constructed set of design-to-time applications with execution time variations when the same effort is spent being necessary for optimizing the quality of the applications with only worst-case execution times specified.