An Effective RM-Based Scheduling Algorithm for Fault-Tolerant Real-Time Systems

Dependability is the representative property that predominantly distinguishes a hard real-time system from other computer systems besides timeliness. Primary/alternate version technique is a cost-effective means which trades the quality of computation results for promptness to tolerate the software faults. The kernel algorithm proposed in this paper employs the off-line backwards-RM scheme to pre-allocate time intervals to the alternate version and the on-line RM scheme to dispatch the primary version. The methodology is a dual-purpose strategy, which aims to (1) tolerate potential software faults by ensuring the accomplishment of the alternate version once its primary fails to execute or re-execute and (2) achieve better quality of service by maximizing the success rate of primary version.