Determining effective multiprocessor scheduling policies for repetitive real-time tasks

This work presents an improved solution to the problem of determining effective multiprocessor scheduling polldes for real-time repetitive tasks. In order for this approach to be applicable to some task, the probability density function of its execution time, a value function and a criticality i~ctor are required. When these parameters are available, the proposed method constructs effective schedules for repetitive tasks that yield acceptable intermediate results; in addition, these schedules tend to degrade gracefully under "heavy" load conditions. A three-phase process is proposed: first a balanced allocation of tasks is attempted using the rate-monotone first-fit algorithm; then the schedule for each processing duster is constructed by using a linear programming formulation of the task system. This formulation considers many important parameters not considered by previous approaches, and constitutes the proposed improvement over previous work. Finally, the schedule for the entire system is derived from the partial schedules.