Empirical evaluation of weighted and prioritized static scheduling heuristics for real-time multiprocessing

Suboptimal solutions to the NP-complete problem of task scheduling in a multiprocessing system are achievable with the aid of heuristic methods. Static scheduling heuristics for real-time multiprocessing systems are typically based on existing algorithms developed for non-real-time systems. Unfortunately this approach results in the real-rime heuristics inheriting the deficiencies of the non-real-time algorithms as well. Existing scheduling heuristics compromise the results of the scheduling effort by insufficiently representing the task characteristics of an application graph. In this paper we present results of experimentation in which the parameters of the DAG are enhanced to more accurately correspond to those of real-world real-time applications. A method of specifying weighted combinations and priorities of simple scheduling heuristics as the scheduling algorithm is presented. Results of the compound heuristics are compared to the results of previous work in the field with some interesting conclusions