Comparison of Enhancing Methods for Primary/Backup Approach Meant for Fault Tolerant Scheduling

This report explores algorithms aiming at reducing the algorithm run-time and rejection rate when online scheduling tasks on real-time embedded systems consisting of several processors prone to fault occurrence. The authors introduce a new processor scheduling policy and propose new enhancing methods for the primary/backup approach and analyse their performances. The studied techniques are as follows: (i) the method of restricted scheduling windows within which the primary and backup copies can be scheduled, (ii) the method of limitation on the number of comparisons, accounting for the algorithm run-time, when scheduling a task on a system, and (iii) the method of several scheduling attempts. Last but not least, we inject faults to evaluate the impact on scheduling algorithms. Thorough experiments show that the best proposed method is based on the combination of the limitation on the number of comparisons and two scheduling attempts. When it is compared to the primary/backup approach without this method, the algorithm run-time is reduced by 23% (mean value) and 67% (maximum value) and the rejection rate is decreased by 4%. This improvement in the algorithm run-time is significant, especially for embedded systems dealing with hard real-time tasks. Finally, we found out that the studied algorithm performs well in a harsh environment