Parallel task scheduling using the order graph method

This paper proposes a new approach for task scheduling of fully specified flow graphs (FSFG) on parallel processing systems. This new approach is based on the enumeration and manipulation of linear extensions identified from a FSFG. The objective is to obtain a multiprocessor schedule which meets a set of admissibility conditions (iteration period, number of processors, precedence, and non-preemption). Feedback and a system-wide iteration period for multiple input data streams are investigated. In addition, flexible models for communications and network topology are introduced to improve the reliability of generated multiprocessor schedules. The resulting order graph method for parallel task scheduling provides an effective, generalized mapping methodology for DSP multiprocessor systems.<
>