In real-time systems, tasks must meet their deadlines even in the presence of hardware/software faults. Fault-tolerance in real-time systems refers to the ability of the system to meet task's deadlines in the presence of faults, and is typically achieved by employing redundancy techniques. Redundancy often improves system reliability at the cost of reducing system schedulability. Therefore, there exists a tradeoff between schedulability and reliability. In this paper, we propose a value-based scheduler capturing this tradeoff in multiprocessor real-time systems. The key function of the scheduler is to select a suitable redundancy level for each task so as to improve performance index of the system, where IP is an integrated performance metric that captures this tradeoff. We have conducted extensive simulation studies to evaluate the effectiveness of the proposed scheduler and its variants for a wide range of system parameters. Our studies show that proposed schedulers maintain a high value-ratio for non-trivial tasks sets.
Special Issue Papers