Fault Tolerance in Multiprocessor Real-Time Systems Under Partitioned EDF Scheduling

This paper addresses the problem of scheduling a set of hard real-time tasks on identical multiprocessors in the presence of faults. The solution we describe is named Fault-Tolerant partitioned Earliest Deadline First (FTpEDF. Fault tolerance is employed via task replication, for which task backups (active or passive) are assigned to processors. Passive backups are released upon error detection while active backups execute independently of the occurrence of errors. The described approach is assessed via extensive experiments. When compared to a related approach based on fixed-priority scheduling, FTpEDF has similar performance in terms of required number of processors but generates less active bacukps.