Login

Scheduling High Criticality Real-Time Systems
Ref: CISTER-TR-171102       Publication Date: 17, Nov, 2017

Scheduling High Criticality Real-Time Systems

Ref: CISTER-TR-171102       Publication Date: 17, Nov, 2017

Abstract:
Cyclic executives are used to schedule safety-critical real-time systems because of their determinism, simplicity, and efficiency. One major challenge of the cyclic executive model is to produce the cyclic scheduling timetable. This problem is related to the bin-packing problem and is NP-Hard in the strong sense. Unnecessary context switches within the scheduling table can introduce significant overhead; in IMA (Integrated Modular Avionics), cache-related overheads can increase task execution times up to 33%.
Developed in the context of the Software Engineering Master’s Degree at ISEP, the Polytechnic Institute of Engineering in Porto Portugal, this thesis contains two contributions to the scheduling literature. The first is a precise and exact approach to computing the slack of a job set that is schedule policy independent. The method introduces several operations to update and maintain the slack at runtime, ensuring the slack of all jobs is valid and coherent. The second contribution is the definition of a state-of-the-art preemptive scheduling algorithm focused on minimizing the number of system preemptions for real-time safety-critical applications within a reasonable amount of time.
Both contributions have been implemented and extensively tested in scala. Experimental results suggest our scheduling algorithm has similar non-preemptive schedulability ratio than Chain Window RM, yet lower ratio in high utilizations than Chain Window EDF and BB-Moore. For task sets that failed to be scheduled non-preemptively, 98-99% of all jobs are scheduled without preemptions. Considering the fact that our scheduler is preemptive, being able to compete with non-preemptive schedulers is an excellent result indeed. In terms of execution time, our proposal is an order of magnitude faster than the aforementioned algorithms. Both contributions of this work are planned to be presented at future conferences such as RTSS@Work and RTAS.

Authors:
Humberto Carvalho


Master Thesis, Instituto Superior de Engenharia do Porto.
Porto.



Record Date: 22, Nov, 2017