Translating off-line schedules into task attributes for fixed priority scheduling

• Published in: Proceedings 22nd IEEE Real-Time Systems Symposium (RTSS 2001) (Cat. No.01PR1420) pp. 225 - 234
• Main Authors: Dobrin, R., Fohler, G., Puschner, P.
• Format: Conference Proceeding
• Language: English
• Published: IEEE 2001
• Subjects: Application software; Contracts; Integer linear programming; Jitter; Job shop scheduling; Processor scheduling; Real time systems; Runtime; Scheduling algorithm; Timing
• ISBN: 0769514200; 9780769514208
• DOI: 10.1109/REAL.2001.990615

Off-line scheduling and fixed priority scheduling (FPS) are often considered as complementing and incompatible paradigms. A number of industrial applications demand temporal properties (predictability, jitter constraints, end-to-end deadlines, etc.) that are typically achieved by using off-line scheduling. The rigid off-line scheduling schemes used, however do not provide for flexibility. FPS has been widely studied and used in a number of applications, mostly due to its simple run-time scheduling, and small overhead. It can provide more flexibility, but is limited with respect to predictability, as actual start and completion times of execution depend on run-time events. In this paper we show how off-line scheduling and FPS run-time scheduling can be combined to get the advantages of both the capability to cope with complex timing constraints and flexibility. The paper assumes that a schedule for a set of tasks with complex constraints has been constructed off-line. It presents a method to analyze the off-line schedule and derive an FPS task set with FPS attributes priority, offset, and period, such that the runtime FPS execution matches the off-line schedule. It does so by analyzing the schedule and setting up inequality relations for the priorities of the tasks under FPS. Integer linear programming (ILP) is then used to find a FPS priority assignment that fulfils the relations. In case the priority relations for the tasks of the off-line schedule are not solvable we split tasks into the number of instances, to obtain a new task set with consistent task attributes. Our schedule translation algorithm keeps the number of newly generated artifact tasks minimal.