Preemptive multiprocessor scheduling anomalies

Preemptive scheduling of periodically arriving tasks on a multiprocessor is considered. We show that many common multiprocessor real-time scheduling algorithms suffer from scheduling anomalies, that is, deadlines are originally met, but a decrease in execution times or an increase in periods of tasks can cause deadlines to be missed. We propose a partitioned multiprocessor fixed-priority scheduling algorithm with the prominent features that (i) it does not suffer from such scheduling anomalies and (ii) if less than 41% of the capacity is used then deadlines are met.