One-level and two-level scheduling for real-time systems

Abstract

This paper proposes and analyzes the one-level and the two-level scheduling approach. The one-level scheduling uses one scheduling policy for scheduling all tasks, whereas the two-level scheduling uses one scheduling policy for the scheduling of high priority tasks and other for the scheduling of low priority tasks. Both scheduling approaches use Earliest Deadline First and Rate Monotonic scheduling algorithms, the two-level scheduling only for scheduling of higher priority tasks. The focus of both approaches is the scheduling of the Imprecise Computation Model tasks. The Imprecise Computation Model trades off the quality of computations with the timeliness. It decomposes each task into a mandatory part and an optional part. Only the mandatory part is required to complete by task's deadline. The scheduling of Imprecise Computation Model tasks that minimizes the total-weighted error is considered. The one-level scheduling is based on an extension of mandatory parts that minimizes the total-weighted error. In order to minimize the total-weighted error the two-level scheduling uses modified Algorithm WNTU by the low level scheduler. The analysis of both approaches is provided including complexity, response time analysis and simulations. Based on the theoretical foundations and the results of simulations the use of one-level and two-level scheduling is recommended under certain assumptions. Single preemptive model is assumed.