Dual-Token-Based Fault-Tolerant Scheduling for Hard Real-Time Multiprocessor Systems

Real-time multiprocessor systems frequently assume that there exists a dedicated processor for task allocation that never fails. This assumption is, however, too strong in the sense that all the physical objects are subject to failure. Moreover, once the dedicated processor fails, the whole multiprocessor system will fail. As a way to solve this problem, we propose a fault-tolerant scheduling algorithm based on moving dual-token. While the primary processor holding a primary token performs task allocation, the backup processor holding a backup token, in case that the primary processor has failed, does primary processor creation. Since no dedicated processor for task allocation exists in this scheme, failure of the whole multiprocessor system due to that of the dedicated processor can be avoided. In addition, the deadline-friendly scheduling policy used for backup task allocation, compared to heuristic scheduling, allows easier implementation and improved scheduling predictability. Simulation results show that the proposed dualtoken based algorithm yields low rejection rates over those with dedicated processor for task allocation.